Protecting Public Health at Inland Ohio Beaches: Development of Recreational Water Quality Indicators Predictive of Microbial and Microcystin Exposure

Home , Trophic state index

DISSERTATION

Presented in Partial Fulfillment of the Requirements for the Degree Doctor of Philosophy in the Graduate School of The Ohio State University

Jason W. Marion

Graduate Program in Public Health

The Ohio State University

2011

Dissertation Committee:

Timothy J. Buckley, Advisor

Jiyoung Lee, Co-Advisor

Stanley Lemeshow

John R. Wilkins III

Jason W. Marion

2011

Abstract

Inland lakes are prone to contamination from a variety of sources within their watersheds.

The changing environment can influence transport and fate of fecal indicators and may also influence the growth of harmful cyanobacteria, thereby occasionally creating health- related water quality concerns for recreational water users. To date, epidemiological and limnological studies pertaining to fecal indicators and harmful cyanobacteria have been limited with respect to inland U.S. lakes. The primary goals of this dissertation were to

(1) evaluate illness risks associated with the fecal indicator E. coli, and (2) evaluate predictive tools potentially useful for the rapid prediction of E. coli densities and health- related concentrations of cyanotoxins in inland Ohio lakes. Through an epidemiological study and the collection of water quality data, predictive models for human illness and water quality advisories were developed. The relationship between water quality indicators and reported adverse health outcomes among users an inland Ohio beach were examined. Human health data collected via a prospective cohort study over 26 swimming days during the 2009 swimming season at East Fork Lake demonstrated that wading, playing or swimming in the water was found to be a significant risk factor for GI illness (adjusted odds ratio (aOR) of 3.2; CI=1.1, 9.0). Among water users (n = 806), E.

ii coli density was associated with elevated GI illness risk where the highest E. coli quartile was associated with an aOR of 7.0 (CI=1.5, 32). Upon observing a significant illness association with E. coli densities among swimmers, the need for rapidly estimating E. coli densities was determined to have merit. Current approaches for quantifying E. coli densities rely on culture-based methods that require 18 or more hours to obtain a result.

Using rapidly measured water quality parameters (e.g., total phosphorus, secchi depth, chlorophyll A), univariable models for rapidly estimating health-related E. coli densities were developed and considered for inland Ohio lakes using 182 beach water samples collected from seven Ohio lakes. Univariable logistic regression revealed that deviations in lake-specific water quality as measured by total phosphorus (p < 0.001), phycocyanin pigment (p = 0.018), and trophic state index (TSI) (p = 0.006) were predictive of E. coli levels exceeding recreational water quality criteria. Using the same samples, models were constructed for estimating cyanotoxin concentrations. Microcystin levels exceeding the 4 g/L low risk threshold set by the World Health Organization were detected by

ELISA in 48 of 182 (26.4%) samples. A multivariable logistic regression model using practical and real-time measures of in vivo phycocyanin and secchi depth was constructed to predict beach conditions exceeding the low risk threshold for microcystin. The model

(p = 0.030) predicted microcystin levels >4 g/L with acceptable discrimination as indicated by the area under the ROC curve (0.795). This study indicates a significant health risk for inland beach users and demonstrates the potential to predict health-related hazard levels using practical real-time measures are possible, enabling opportunities for interventions that protect public health.

iii

Dedication

This document is dedicated to my parents Ronald and Patricia Marion.

Acknowledgments

I am extremely grateful to the support I received from my dissertation committee. This work would not have been possible without their encouragement and assistance in supporting my thinking. I am thankful for funding support from my advisor, Dr. Tim

Buckley and my co-advisor, Dr. Jiyoung Lee. I am also particularly grateful to retired

Ohio State Parks Chief, Dan West, and to Scott Fletcher with Ohio State Parks for supporting this research and enabling my work to receive financial support. The financial support provided by the Ohio Water Development Authority and the Public Health

Preparedness for Infectious Diseases Program at OSU provided significant resources that will enable the work contained herein, to have significant impacts in its respective field.

I am also forever grateful for the skills and training I received from my undergraduate mentors, Drs. Brian Reeder and David Smith at Morehead State University, who unknowingly had significant impacts on my thinking, and carrying out of the multiple studies contained in this dissertation.

Vita

1999...... Amelia High School, Batavia, OH

2001 ...... A.A.S. Recreation and Wildlife

Management, Hocking College

2004 ...... B.S. Environmental Science,

Morehead State University

2004 - 2006 ...... Graduate Associate, Institute for Regional

Analysis and Public Policy, Morehead State

University

2006 ...... M.S. Biology, Morehead State University

2006 -2007 ...... Graduate Research Associate, College of

Public Health, The Ohio State University

2008-2010 ...... Doctoral Fellow, Public Health

Preparedness for Infectious Diseases,

The Ohio State University

2010 ...... M.S. Public Health, The Ohio State

University vi

Publications

Marion, J.W., J. Lee, S. Lemeshow, and T.J. Buckley. 2010. Association of gastrointestinal illness and recreational water exposure at an inland U.S. beach. Water Research 44 (16),

4796-4804

Lee, C.S., J. Marion, J. Lee. 2011. A novel genetic marker for the rapid detection of

Bacteroides fragilis in recreational water as a human-specific fecal indicator. Journal of

Water and Health 9 (2), 253-264

Fields of Study

Major Field: Public Health

vii

Table of Contents

Abstract ...... ii

Dedication ...... iv

Acknowledgments...... v

Vita ...... vi

Publications ...... vii

List of Tables ...... x

List of Figures ...... xiv

Chapter 1: Introduction ...... 1

Chapter 2: Background ...... 6

Chapter 3: Association of Gastrointestinal Illness and Recreational Water Exposure at an

Inland U.S. Beach ...... 34

Chapter 4: Carlson‟s Trophic State Index as a Predictor of Advisory-Level E. coli

Densities at Inland Ohio Beaches ...... 65

viii

Chapter 5: In Vivo Phycocyanin Flourometry as a Rapid Screening Tool for Predicting

Elevated Microcystin Concentrations at Inland Beaches...... 88

Chapter 6: Synthesis and Discussion ...... 114

References ...... 122

Appendix A: Human Health Study Beach and Telephone Questionnaire ...... 153

Appendix B: Human Health Study Recruitment Sign ...... 176

Appendix C: Water Quality Data Summary ...... 177

List of Tables

Table 3.1. Individual and household characteristics of persons surveyed at East Fork

Lake (Ohio, United States) with complete surveys and telephone follow-up...... 47

Table 3.2. Descriptive statistics of beach water quality and beach usage during sampling

days (N=26) at East Fork Lake (Ohio, United States)...... 49

Table 3.3. Summary of beach exposures and reported illness by gender and age

classification of respondents during the 2009 swimming season at East Fork Lake

(Ohio, United States)...... 51

Table 3.4. Summary of gastrointestinal illnesses for beach users across various E. coli

density exposure levels at East Fork Lake (Ohio, United States) during the 2009

swimming season...... 52

Table 3.5. Association of beach user attributes with illness outcomes across 26 sampling

days at East Fork Lake (Ohio, United States)...... 54

Table 3.6. Model performance values for three logistic regression models predicting

illness (GI and HCGI illness) across 26 swimming season days at East Fork Lake

(Ohio, United States)...... 56

Table 3.7. Summary statistics of linear regression between GI illness outcomes and E.

coli density across 23 sampling days1 at East Fork Lake (Ohio, United States). .. 57

Table 4.1. Beach sampling locations and general reservoir characteristics)...... 70

Table 4.2. A summary of beach water quality across seven Ohio beaches including with

respect to trophic state index values, and the number of days in which E. coli

exceeded health-relevant densities)...... 77

Table 4.3. A summary of univariable logistic regression models designed for predicting

the odds of beach samples exceeding the E. coli recreational water quality

standard of 235 CFU/100 mL)...... 81

Table 4.4. A summary of performance and assumption measures for the univariable

logistic regression models designed for rapidly predicting the odds of beach

samples exceeding the E. coli recreational water quality standard of 235 CFU/100

mL)...... 84

Table 5.1. Beach sampling locations and general reservoir characteristics...... 94

Table 5.2. Beach water quality characteristics across seven inland Ohio reservoirs over

26 sampling days during summer 2009 ...... 97

Table 5.3. Results of 10 univariable logistic regression models for lake-clustered data

assessing the association between individual water quality parameters and

elevated microcystin concentrations (4 g/L) across seven inland Ohio reservoirs

over 26 sampling days during summer 2009 ...... 103

Table 5.4. Final multivariable logistic regression model for predicting the odds of beach

samples exceeding microcystin concentrations of 4 g/L at Midwestern U.S.

reservoirs...... 106

Table 5.5. A performance analysis of phycocyanin thresholds for predicting microcystin

concentrations with 26% probability of exceeding 4 g/L microcystin.in Ohio

beach waters with varying secchi depths ...... 108

Table C.1. Shapiro-Wilk test for normality results for Alum Creek water quality variables

...... 178

Table C.2. Shapiro-Wilk test for normality results for Buck Creek water quality variables

...... 179

Table C.3. Shapiro-Wilk test for normality results for Deer Creek water quality variables

...... 180

Table C.4. Shapiro-Wilk test for normality results for Delaware Lake water quality

variables ...... 181

Table C.5. Shapiro-Wilk test for normality results for East Fork water quality variables

...... 182

Table C.6. Shapiro-Wilk test for normality results for Lake Logan water quality variables

...... 183

Table C.7. Shapiro-Wilk test for normality results for Madison Lake water quality

variables ...... 184

xii

Table C.8. Spearman‟s rank correlation coefficient for selected Alum Creek water quality

variables ...... 185

Table C.9. Spearman‟s rank correlation coefficient for selected Buck Creek water quality

variables ...... 186

Table C.10. Spearman‟s rank correlation coefficient for selected Deer Creek water

quality variables ...... 187

Table C.11. Spearman‟s rank correlation coefficient for selected Delaware Lake water

quality variables ...... 188

Table C.12. Spearman‟s rank correlation coefficient for selected East Fork water quality

variables ...... 189

Table C.13. Spearman‟s rank correlation coefficient for selected Lake Logan water

quality variables ...... 190

Table C.14. Spearman‟s rank correlation coefficient for selected Madison Lake water

quality variables ...... 191

xiii

List of Figures

Figure 2.1. Total U.S. waterborne disease outbreak cases from untreated recreational

waters for 2005-2006 (Yoder et al. 2008) ...... 9

Figure 2.2. The percent of water samples exceeding recreational water health criteria at

coastal beaches by state and including inland Ohio lakes for the 2007 swimming

season (Dorfman and Rosselot 2008, Ohio Department of Health 2009a)...... 12

Figure 3.1. Regression line estimating highly credible gastrointestinal illness rate among

swimmers at East Fork Lake (Ohio, United States) exposed to varying E. coli

densities across 23 days ...... 57

Figure 4.1. Locations of the beaches for the seven lakes used in the study (Ohio, United

States) ...... 71

Figure 4.2. Box-and-whisker diagram comparing the amount of deviation in daily trophic

state index values away from beach-specific average trophic state index values for

beach water samples determined to be below and above the recreational water

quality criteria (235 CFU E. coli/100 mL)...... 79

xiv

Figure 5.1. The association of median in vivo phycocyanin concentrations and the percent

of samples exceeding the low risk threshold for microcsytins (4 g/L) in the Ohio

lakes during the study period...... 101

Figure 5.2. The relationship between model sensitivity and specificity for multivariable

logistic model predicting beach samples with microcystin concentrations

exceeding the low health risk threshold (4 g/L) ...... 107

Figure 6.1. “Health Effects Data and Trend Lines from the USEPA and Marion Studies”

(as taken directly from U.S. EPA 2010) ...... 117

Figure B.1. Image depicting sign used to recruit East Fork State Park beachgoers to

participate in the human health aspect of the study ...... 176

Figure C.1. Box-and-whiskers plot of in vivo chlorophyll A concentrations ( g/L) by

study beach...... 192

Figure C.2. Box-and-whiskers plot of total phosphorus concentrations ( g/L) by study

beach...... 193

Figure C.3 Box-and-whiskers plot of secchi depths (cm) by study beach...... 194

Figure C.4 Box-and-whiskers plot of water temperatures (°C) by study beach...... 195

Figure C.5. Box-and-whiskers plot of in vivo phycocyanin concentrations ( g/L) by study

beach ...... 196

Figure C.6. Box-and-whiskers plot of daily E. coli densities (CFU/100 mL) by study

beach with EPA criterion depicted at 235 CFU/100 mL ...... 197

Figure C.7. Box-and-whiskers plot of microcystin concentrations ( g/L) measured within

range of the ELISA by study beach...... 198

Figure C.8. Box-and-whiskers plot of trimetric average trophic state index values by

study beach...... 199

xvi

Chapter 1: Introduction

1.1. Thesis Statement

Current approaches for assessing and communicating risk associated with recreational water contact are inadequate for protecting public health. New approaches for rapidly and practically assessing water quality are needed. Accordingly, this dissertation has been designed around the development and evaluation of cost-effective and practical approaches for rapidly assessing infectious disease and algal toxin risks at inland Ohio beaches. This approach relies on rapidly measured indicators and determinants of water quality that can be predictive of fecal indicator and algal toxin densities and ultimately - infectious disease and overall illness risk.

1.2. Statement of the Problem

The link between adverse health outcomes and recreational swimming has been well documented since the 1950s (Stevenson 1953). During the last 30 years, associations between fecal indicators and illness have been established for beach swimmers (Cabelli et al. 1979, Cabelli et al. 1982, Pruss 1998, Wade et al. 2003, Wade et al. 2006, Colford et al. 2007) including the 1984 U.S. EPA study (Dufour 1984) that established Escherichia 1 coli and enterococci as criteria indicators for fecal contamination in freshwater (U.S. EPA

1986). Despite known associations between exposure and water quality, the minimum time to obtain an E. coli result is 18-24 hours after sample collection (U.S. EPA 2006), which is inadequate for communicating same-day water quality risks to the public to prevent exposure. Therefore, there is a need for predictive models and rapid methods to identify undesirable conditions in advance of potential exposure.

Furthermore, the current recreational water quality criteria for fecal indicators are scheduled for revision no later than October 15, 2012 per a consent decree stemming from Natural Resources Defense Council v. Johnson and U.S. EPA (2008). At the present time, there is limited epidemiologic data for recreational water exposure, particularly from inland waters. Additionally, there has been limited epidemiologic research examining non-GI illness human health effects (e.g. skin rash, ear infections, etc.) for recreational water exposure in waters evaluated by rapid chemical, physical and molecular methods for characterizing water quality.

Lastly, there is limited epidemiology and little understanding with respect to human exposure to algal toxins in recreational waters. Some effort has been directed at understanding harmful algal blooms in the coastal environment via the Harmful Algal

Booms and Hypoxia Research and Control Act of 1998 and the HABHRCA amendments of 2004; however, inland waters have received little attention with respect to research and policy. This inadequate attention has led to the U.S. EPA concluding that the available

2 data pertaining to harmful algal blooms in inland waters is insufficient for informing EPA and U.S. policies (Fristachi et al. 2008, Hudnell 2010).

1.3. Hypotheses and Specific Aims

To address the need for practical and effective approaches for assessing human health risks at inland Ohio beaches, our study was designed around the following three strategic hypotheses.

Hypothesis 1: Water-related disease risk is determined by a combination of water quality and exposure behaviors.

Hypothesis 2: Rapid measures of water quality (i.e. phosphorus, trophic state index, turbidity) at inland lakes are predictive of microbial fecal indicator densities (E. coli).

Hypothesis 3: Rapid measures of water quality (i.e. phosphorus, trophic state index, turbidity) at inland lakes are predictive of cyanotoxin concentrations (microcystins)

Therefore, the ultimate research goal was to develop predictive models of recreational water quality and disease risk that are based on practical (rapid, simple, and inexpensive) measures of water characteristics as a means to protect public health.

Research objectives were accomplished through three specific aims.

1. Evaluation of beach water quality. For 7 Ohio inland lakes, measurements of lake- specific hydrological, meteorological, and microbial factors were recorded. Information

3 including E. coli densities, total phosphorus, chlorophyll A, turbidity, pH, dissolved oxygen, specific conductivity, temperature, phycocyanin, microcystin, lake stage and other water quality data were collected permitting beach water quality assessment. Water samples were collected June through September of 2009. Furthermore, water quality index values were calculated. Secchi depth, surface phosphorus and surface chlorophyll

A were used to calculate Carlson‟s trophic state index (TSI) (Carlson 1977). Each metric

(secchi depth, surface phosphorus and chlorophyll A) was used to calculate a trimetric mean TSI value for each sampling period at each lake.

2. Characterizing the health status of beach users before and after exposure. For one Ohio inland beach (East Fork Lake), the parameters in Specific Aim 1 were evaluated in conjunction with a survey of water exposure and water-related illness among beachgoers. Face-to-face surveys regarding beach exposures and health signs/symptoms were conducted on weekends and holidays from June through

September of 2009 at East Fork State Park. Follow-up phone interviews occurred 8 to 10 days after the beach visit and were performed to detect illness symptoms/signs.

3. Development of predictive models. Predictive models relating water quality factors from Specific Aim 1 to human exposure and health outcomes from Specific Aim 2 were employed to model illness risk. Models predicting fecal indicator densities and microcystin concentrations were also constructed. Furthermore, determinants of water quality were evaluated with respect to their predictive potential for estimating disease risk

4 in comparison to the traditional previous day, culture-based fecal indicator density method.

1.4 Dissertation Organization

This dissertation is organized into six chapters. Chapter 1 provides a brief synopsis of the public health relevance of this research pertaining to beach water quality and human health.

Chapter 1 further describes the principal hypotheses and specific aims of this work. Chapter

2 provides the background that informed the hypotheses and specific aims. Chapter 2 is based upon a substantial literature review regarding the epidemiology of recreational water- associated illness, as well as the utility of various fecal and cyanotoxin indicators. The majority of the dissertation is contained in chapters 3, 4 and 5, which were prepared in journal manuscript format to test the hypotheses outlined in the first chapter. The final chapter is a synthesis of chapters 3, 4, and 5 and demonstrates how the culmination of the entire work adds new knowledge in the area of environmental health sciences. Ideas for future research are also discussed.

Chapter 2: Background

2.1. Recreational Water Quality and Human Health Standards

Despite substantial improvements in the Nation‟s water quality following Federal legislation in 1972 with the passage of our Nation‟s first Clean Water Act, a substantial portion of our Nation‟s waters are still failing to meet their designated uses. Furthermore, over 4,000 persons experience recreational water-associated illness annually from outbreaks (Yoder et al. 2008) and there are more than 22,000 beach closing and advisory days annually across the U.S., an estimate that continues to rise as monitoring improves

(Stoner and Dorfman 2004).

Currently, no estimates are available with respect to the number of recreational waterborne illnesses attributable to coastal versus inland waters. It has been accepted since 1984 (Dufour 1984) that marine waters contribute to a higher illness risk than inland waters. Using data from two large epidemiologic studies, Dufour (1984) presents incidence rates of highly credible GI illness (HCGI) among marine water swimmers as

15.2 cases per 1,000 individuals; whereby HCGI is defined as experiencing vomiting and/or diarrhea and fever. The reported HCGI rate for inland beach users is significantly 6 smaller (5.7 cases per 1,000 individuals in Dufour‟s studies). Although these rates have been used for establishing national policy (USEPA 2006), they are based upon epidemiologic studies with a large sample size, but only small beach/water quality variability. For example, the freshwater data are based upon two Pennsylvania Lake Erie

Beaches and one beach from Keystone Lake, Oklahoma. It is important to note that the

HCGI rate at the Lake Erie Beach B was 14.7 per 1000, and 11.0 per 1000 in years 2 and

3 of the 3-year study. The 5.7 cases per 1,000 reported as the national HCGI rate, which is much fewer cases per 1,000 than the Lake Erie beaches, is a national average that includes two study years of low symptom rates from the relatively clean Keystone Lake.

Ultimately, marine or inland, the illness risk associated with marine versus inland waters is not understood and there are limited data to assert that the risk of illness from a marine area is higher or smaller than inland water. It should be noted that in the Dufour (1984) study, there were substantial differences in water quality across the study sites. Sites including more inland water sources with varying levels of contamination are necessary before confident conclusions can be made.

Beyond epidemiologic studies, outbreak analysis has been a tool for understanding whether inland lakes are associated with disease burden. A review of the outbreak data

(Yoder et al. 2008) demonstrates that the majority of recreational waterborne disease outbreaks and outbreak cases for untreated U.S. waters are from lakes. Very few of the reported U.S. waterborne disease outbreaks and outbreak cases are from marine waters.

In the most recent CDC surveillance report on recreational waters, inland waters

7 contributed to 19 of the 20 (95%) recreational waterborne disease outbreaks from non-

Vibrio agents in untreated waters from 2005-2006 (Yoder et al. 2008). This trend has been observed previously in CDC surveillance reports. For example, in 2003-2004 all 19 of 19 outbreaks (100%) were from inland waters (Dziuban et al. 2006). In 2001-2002, all

11 of 11 outbreaks from untreated waters were non-marine (Yoder et al. 2004). Based upon the available outbreak data as reported to CDC, it appears that the inland waters contribute to more infectious disease outbreaks than marine waters (Figure 2.1).

As of 2002, only 60% of the Nation‟s waters were categorized as „good‟ for recreation

(USEPA 2007a). Within the broad classification of waters designated for recreation,

79.37% of the water from lakes, ponds and reservoirs designated for use as contact recreation were impaired. Also impaired among the waters from lakes, ponds and reservoirs were those designated for public bathing (88.61% impaired) and swimming

(62.40% impaired). It is well established that recreational water use can result in significant exposures to pathogens resulting in risk to the populace (Ferley 1989, Pruss

1998). This is not suprising since pathogens are the second leading cause of impairment of our Nation‟s streams and rivers (USEPA 2007a) and are also responsible for over 88% of the beach advisories and closings (Stoner and Dorfman 2004).

Beach water quality is a serious concern in Ohio, and in many other states, particularly because of the size of the population at risk and the degree of uncertainty regarding beach and water quality at Ohio‟s lakes, ponds and reservoirs. Every year 4 million people

8 enjoy the nearly 80 Ohio beaches made available by the Ohio Division of Parks and

Recreation, which is only a very small segment of the national beachgoer population

(Ohio State Parks 2006). Nationally, state park systems alone provide public swimming opportunities at 1,141 state beaches (Ohio State Parks 2007a). These figures are only representative of a small segment of the greater beachgoer community which would also include beachgoers who utilize federal, local and/or private swimming areas and beaches.

If Ohio‟s mean usage per beach was extrapolated to the U.S., a crude estimate of the number of national beachgoers would be over 66 million.

Figure 2.1. Total U.S. waterborne disease outbreak cases from untreated recreational waters for 2005-2006 (Yoder et al. 2008).

In Ohio over 50% of the State‟s assessed streams, rivers and creeks were impaired (U.S.

EPA 2007b); no impairment data are available on the State‟s inland lakes, except for those included in the stream and river assessments. With over half of the State‟s rivers and streams being impaired it is logical to assume potential adverse water quality in the state‟s lakes, ponds and reservoirs as well as an elevated health risk for users. Data are available for Lake Erie (a federally-designated navigable waterway) and a fair amount of support is provided by U.S. EPA to research water quality and health issues associated with Lake Erie and its beaches (U.S. EPA 2007c). Unfortunately, Lake Erie is very different than the State‟s manmade reservoirs. Furthermore, approximately 2 million or

50% of Ohio‟s state park beachgoers utilize the inland lakes (Ohio State Parks 2007b).

Of Ohio‟s 65 inland beaches, 21 had at least one swimming advisories during 2006; whereby the E. coli densities in the previous day sample exceed 235 colony forming units per 100 mL (CFU/100 mL). Six of 13 Lake Erie beaches had at least one advisory with more intensive sampling (Ohio State Parks 2006). The longest advisories, both 15 days, were at Deer Creek and Strouds Run, both inland lakes. During the 2006 season five different state park beaches had values of 4500, 1950, 1800, 1367 and 1100 E. coli colony forming units (CFU) per 100 mL of water (Ohio Department of Health 2006). For the 2007 season, 20 of the state‟s 65 inland beaches experienced advisories. The top five exceeding values were 6900, 2420, 1967, 1733 and 1667 CFU/100 mL (Ohio Department of Health 2007).

The Beaches Environmental Assessment and Coastal Health Act of 2000 (BEACH Act) was established to reduce the risk of illness to users of the Nation‟s recreational waters.

This amendment to the Clean Water Act continues to provide much needed resources to coastal beaches, but prohibits BEACH Act grants to be authorized by the U.S. EPA to support inland beach projects per Section 5 of the BEACH Act.

Despite being excluded from the BEACH Act, many inland lakes may pose health risks comparable to coastal waters. In Ohio, 5.5% of 2007 samples from inland beaches exceeded the 235 CFU/100 ml standard (Ohio Department of Health 2009a), whereas, nationally, 7% of marine water exceeded national human health standards in 2007

(Dorfman and Rosselot 2008). Using the NRDC Report by Dorfman and Rosselot (2008), it can be observed that Ohio‟s inland lakes have an exceedance rate greater than a substantial number (43%) of the coastal states when comparing with their 2007 coastal beach monitoring results (Figure 2.2).

2.2. The History of the E. coli Standard in Freshwater Recreational Waters

The link between adverse health outcomes and recreational swimming and/or bathing has been well documented over the last 100 years. Dufour (1984) citing Moore (1954) reports that as early as 1909, Salmonella was implicated as a swimming-associated disease in Walmer, England, affecting 34 people. Other later documented outbreaks were identified in Connecticut, New York and California in 1921, 1932 and 1942, respectively

(Dufour 1984). By 1950, the U.S. Public Health Service engaged in a comprehensive effort to understand associations between water quality and illness (Stevenson 1953).

Figure 2.2. The percent of water samples exceeding recreational water health criteria at coastal beaches by state and including inland Ohio lakes for the 2007 swimming season

(Dorfman and Rosselot 2008, Ohio Department of Health 2009a).

The U.S. Public Health Service conducted a comprehensive survey at three locations in the early 1950s and the results of these studies indicated a number of significant conclusions. In this first large-scale prospective study, Stevenson (1953) reported that the swimming group had “an appreciably higher over-all illness incidence” than the non-

12 swimming group irrespective of the swimming water quality. It was also noticed that children (under 10 years of age) experienced 100 per cent higher illness prevalence than persons over 10 years of age. Furthermore, observed illnesses were not limited to gastrointestinal disturbances. Ear, nose, throat, and eye ailments represented over half of the observed illnesses. There was also evidence supporting the notion that more polluted beaches as determined by mean coliform density were associated with an increase in illness incidence. In this first health study regarding recreational water quality, weather was suspected as having an impact on the results.

During the 1950s, states, regions and other authorities established water classifications or recommendations despite a high degree of uncertainty and debate with respect to the relationship between water quality and illness. In 1954 the Joint Committee on Bathing

Places of the Conference of State Sanitary Engineers and the American Public Health

Association agreed not to set any standards for outdoor bathing places and argued that it is undesirable to have any absolute standard of safety for outdoor waters, using either bacterial or chemical analysis or sanitary surveys (Lehr and Johnson 1954). There was a general consensus that 1,000 coliform organisms per 100 ml would be considered “fairly acceptable,” unless immediate dangers from human sewage were known. Most states and boards accepted this assumption, but the Tennessee Valley Authority had a different approach. The TVA had the following standards for bathing conditions: waters with less than 50 coliforms per 100 ml were satisfactory, waters with 51 to 500 coliforms per 100 ml were satisfactory with reservations, waters with 501 to 1,000 coliforms were doubtful

13 for use and not recommended, and waters with more than 1,000 coliforms were classified as waters not fit for bathing use.

During the 1950s, there was a significant focus on bacteria and chemical analysis. With regards to diseases, bacteriological analysis and sanitary surveys were the most important methods for determining the safety of swimming areas. The coliform group served as an indicator of pollution for bathing waters, but there was consensus that a better indicator was needed (Lehr and Johnson 1954). Seventeen years later, the argument for better indicators was still being made, as coliform criteria, in most cases, was still “not based on evidence of the epidemiologic probability” of enteric disease transmission

(Krishnaswami 1971).

In 1976 the recently established U.S. Environmental Protection Agency adopted a fecal coliform guideline, which was recommended by the National Technical Advisory

Committee to the Federal Water Pollution Control Administration in 1968. The basis for the fecal coliform guideline was the results from Stevenson (1953). As demonstrated in

Krishnaswami (1971), epidemiologic probability of disease was lacking with this standard. This finding prompted the U.S. Environmental Protection Agency‟s Health

Effects Research Laboratory (HERL) to perform an epidemiological-microbiological study to determine if the new geometric mean fecal coliform density recommendation of

200 CFU per 100 ml of water was appropriate (Cabelli et al. 1979). Also considered was

14 the stringency of the standard that no more than 10 percent of total samples during any

30-period shall exceed 400 per 100 ml of water.

The 1979 HERL study (Cabelli et al. 1979) used two years of New York marine beach data; the results demonstrated a dose-response relationship with regards to an increasing number of adverse health effects with increasing fecal indicator bacteria densities. The results of the study were inconclusive with regards to the ability to conduct a reliable risk analysis based upon any of the indicators measured (total coliforms, fecal coliforms, E. coli, Klebsiella, Enterobacter-Citrobacter, Fecal streptococci and Pseudomonas aeruginosa). The advantage of this study, unlike many previous ones, was that it recorded water quality conditions relative to the exposure of the beach users. Ultimately, the findings of the New York marine beach data coupled with beach data from Louisiana and Massachusetts led to a thorough risk analysis and an EPA-based marine recreational water criterion using enterococci (Cabelli 1983).

The Cabelli (1983) report regarding marine waters raised concern for inland waters. The

1979 and 1981 Cabelli reports demonstrated clear relationships between marine pollution and adverse human health effects. Health effects were obvious in the 1953 study by

Stevenson at the inland waters like the Ohio River, but adverse health effects were not observed in Chicago or Long Island beach goers. The EPA (Dufour 1984) suspected the risk of waterborne illness transmitted via freshwater may be higher since this was the case with Stevenson (1953). Dufour (1984) demonstrated that among the beaches

15 sampled, the gastrointestinal (GI) illness rate for marine waters was 2.67 times higher than the freshwater GI illness rate (15.2 per 1,000 versus 5.7 per 1,000). It is noteworthy though, that Dufour‟s study only examined two study locations, and the illness rates are likely to vary across beaches.

A more in-depth review of the Dufour studies does demonstrate the association between illness and fecal indicators, which was confirmed in the Cabelli (1981) reports. In reviewing the regression lines for total GI and highly credible GI illnesses versus indicator bacteria densities, it was apparent that the strongest associations were with mean E. coli densities, followed by mean enterococcus densities and then mean fecal coliform density (Dufour 1984). The existing recommended fecal indicator organism at the time of the study (fecal coliform) in fact had an inverse relationship with adverse health effects (Dufour 1984). The geometric mean using E. coli (126 colony forming organisms per 100 ml) and any single water sample maximum (235 organisms per 100 ml) correspond to approximately 8 to 14 GI illnesses per 1,000 swimmers. This standard was adopted by EPA in 1986 as the recreational water criterion for bacteria.

2.3. An Assessment of the E. coli Standard for Freshwater

In 2001, a World Health Organization (WHO) report stated “water regulatory agencies have yet to come to terms with the inherent problems resulting from reliance on fecal indicator bacteria as currently determined” (Asbolt et al. 2001). Drawing primary attention is the fact that E. coli has been known to grow in warm waters and soils (Solo-

Gabrielle et al. 2000). Other considerations are that viruses and protozoa are more likely to cause illness than thermotolerant bacteria (Yoder et al. 2008). Fleisher et al. (1996) argues the lack of exposure-response associations to non-GI illnesses, such as respiratory, eye, ear and skin ailments is attributable to infectious agents other than thermotolerant bacteria. Prüss (1998) demonstrated a strong exposure-response for fecal indicators and

GI illnesses in a large meta-analysis of the literature, but Fleisher‟s claim regarding non-

GI illnesses and the lack of exposure-response with other ailments was not addressed.

Fleischer et al. (1996) argue against the use of a single indicator for determining acceptable recreational water quality and the WHO (Absolt et al. 2001) advocates for more indicators of “process efficiency” than reliance on fecal indicators.

In response to a growing number of arguments against the use of enterococci and E. coli standards in the scientific literature, Wade et al. (2003) conducted a comprehensive meta- analysis of the 27 most relevant epidemiologic studies, which included studies ranging from 247 to 26,686 participants. For GI illness (any nausea, diarrhea, or vomiting), the correlation coefficient r was 0.86 for the natural log relative risk of GI illness as a function of continuous E. coli density. The Wade et al. (2003) study validated the U.S.

EPA‟s position on using E. coli as a recreational water standard to reduce GI illness risk among freshwater users. The Wade et al. (2003) study failed to address concerns regarding non-GI illness, which in the CDC‟s comprehensive review of untreated recreational water disease outbreaks for 2005-2006 comprised approximately 28% of all illnesses (Yoder et al. 2008).

Current approaches for assessing beach safety are based on a 1984 health study that relied on dated technology (Darfour 1984). During the 1970s and 1980s, it was determined that fecal indicator bacteria could be used to determine the degree of fecal contamination allowable for protecting public health. The rationale for this approach is based on observed associations between fecal indicator bacteria and pathogens such as adenovirus, norovirus, rotavirus and Camplyobacter (Noble et al. 2003, Jiang et al. 2004, Bates and

Phillips 2004). Additional technologies are now being evaluated as potentially better indicators for waterborne pathogens including male-specific coliphage (Colford et al.

2007) and quantitative polymerase chain reactions (qPCR) for Bacteroidales, enterococci and a number of other indicators (USEPA 2007d., Converse et al. 2009).

2.4. Epidemiologic Studies and Alternative Fecal Indicators

As of 2003, there were 27 significant epidemiologic studies evaluating the association between water quality and adverse human health outcomes following water exposure

(Wade 2003). Among the 27 studies, 19 utilized the cohort design, including all five of the U.S. studies. Among the 27 studies, 10 studies were performed in the freshwater environment, while the other 17 were in the marine environment.

Prior to 2003, the largest three freshwater epidemiology studies in the U.S. were conducted by Stevenson (1953), Dufour (1984) and Calderon et al. (1991) as cited in

Wade (2003), with 5,124, 21,777, and 144 participants, respectively. In all three of these

18 studies, no consideration was given to modern alternative indicators (non-E. coli and non- enterococci), such as Bacteroides spp. Furthermore, associations with physical and chemical water quality parameters were either not considered or not reported. Lastly, associations with quantitative polymerase chain reaction (qPCR) results were not feasible or possible at the time of these studies. Since 2003, there has been limited freshwater epidemiologic research reported in the peer-reviewed literature regarding pathogens. The two peer-reviewed studies focus on the Great Lakes and utilize qPCR methods. In both cases, associations between exposure and illness were observed. In the prospective cohort study performed by Wade et al. (2003) a significant association between qPCR

Enterococcus at the p<0.05 level was observed and for qPCR Bacteroides a marginal association was observed at the p<0.10 level. In another Wade et al. study, a similar relationship with swimming exposure and qPCR Enterococcus was observed (Wade et al.

2008). In the 2008 study, it was also demonstrated that the qPCR Enterococcus results were more strongly associated with illness than the Enterococcus results obtained by membrane filtration. The 2003 study reported results for 5,667 individuals and the 2008 study involved over 21,000 participants.

In addition to the freshwater studies post-2003, Colford and others (2007) conducted a significant marine study with 8,797 beachgoers at Mission, California.

In this study, no associations between illness and exposure to fecal coliforms, total coliforms and Enterococcus were observed. The only significant associations were with male-specific coliphage, in which with increasing coliphage density, GI illness risk

19 increased. The authors suggest their results may have bene influenced by their study site, a non-point source beach.

Given pre-existing literature, substantial gaps exist in the epidemiologic literature pertaining to recreational water exposure. For this reason, of the top 5 priorities listed as

“high” in the “Report of the experts scientific workshop on critical research needs for the development of new or revised recreational water quality criteria of the freshwater beaches,” conducting epidemiologic studies is in this top category. The epidemiology studies of greatest interest are those that examine a broad range of fecal indicators, including human and non-human sources (U.S. EPA 2007d). E. coli, enterococci-qPCR and Bacteroides-PCR are just three of many fecal indicators. Presently, there have been no inland epidemiology cohort studies reported in the peer-reviewed literature that have attempted to elucidate associations between illness and any PCR-based measurement or any type of Bacteroides measurement. Furthermore, there are few, if any, post-1984 inland U.S. epidemiologic studies investigating illness and recreational water exposure.

In addition, simpler methods than qPCR, such as immunomagnetic separation coupled with adenosine triphosphate measurement (IMS/ATP) for E. coli (Lee and Deininger

2004) have not been employed in any epidemiologic study. Lastly, as originally pointed out by Fleisher et al. (1996), little attention has been focused on total illness as opposed to GI illness. Studies from inland lakes regarding any illness and water quality are non- existent in the peer reviewed literature.

2.5. Waterborne Illness Associations beyond Fecal Indicator Bacteria

The majority of historical research regarding recreational water quality has been associated with gastrointestinal illnesses, in which much attention was directed at fecal indicators. The emphasis on fecal indicators is based upon the concept that non- pathogenic indicators serve as a proxy for pathogens that can be transmitted via the fecal- oral route. This concept, however, has been demonstrated to also be useful for predicting illnesses not necessarily explained by fecal-oral transmission routes. For example,

Fleisher et al. (1996), demonstrated elevated risks for respiratory and ear ailments post- exposure to waters with elevated fecal streptococci and fecal coliforms. Fecal streptococci concentrations explained acute febrile respiratory illness much better than fecal coliforms, whereas fecal coliforms were more strongly associated with ear infection outcomes than fecal streptococci. Fleisher et al. (1996) argue there is a need for more than one indicator for optimal protection of public health, despite still focusing only on fecal indicator bacteria.

Few epidemiologic studies of recreational waters have examined physical parameters of water quality that may be associated with illness. Examples of physical parameters include weather conditions, water turbidity, water temperature, air temperature, etc.

Despite evidence demonstrating significant associations between turbidity and indicators, only one peer-reviewed epidemiologic study has reported results in this area. In a study of 25,000 Hong Kong beach users, Kueh et al. (1995) observed significant associations between GI illness and physical parameters; in particular, air temperature and water

21 temperature were significantly associated with self-reported GI illness at the p<0.05 level. Aeromonas spp. and Vibrio parahaemolyticus exposures were also significantly associated with GI illness. Marginal associations (p<0.10) between GI illness and turbidity, Clostridium perfringens and Vibrio cholera were observed. When highly credible GI illness (HCGI) was examined, only turbidity and C. perfringens were significantly associated, with a marginal association existing between HCGI and

Aeromonas spp.

In epidemiology investigations pertaining to drinking water, a significant association between illness and turbidity in drinking water has been repeatedly demonstrated. For example, Schwartz and others (1997) demonstrated statistically significant associations between GI illnesses among children with increasing drinking water turbidity at a

Philadelphia hospital. Furthermore, drinking water turbidity was strongly associated with gastroenteritis among both children and adults in Milwaukee, Wisconsin from 1992-

1993, with children facing increased 2.35 times greater odds of gastroenteritis than children exposed to the water with the least turbidity (< 0.5 NTUs) (Morris et al. 1996).

These results are not too striking as alternative non-biological indicators have been identified to track well with not only illness, but also pathogens. Furthermore, these associations are biologically plausible as demonstrated in research illustrating relationships with pathogens.

Turbidity has been known to track better with Cryptosporidium oocyst densities than fecal and total coliforms (LeChevallier et al. 1991). The correlation between

Cryptosporidium oocysts in surface waters and turbidity was strong (r = 0.748, P < 0.01).

Rainfall, salinity and turbidity have all been determined, in at least one study, to better predict enterovirus in marine waters than fecal coliforms (Gerba et al. 1979).

Additionally, in a study of multiple fecal indicators and pathogens, turbidity was more frequently correlated with all types of indicators and pathogens than any of the other indicators and pathogens. In Ferguson et al. (1996), turbidity was determined to be significantly correlated (p < 0.05) with fecal coliforms, fecal streptococci, C. perfringens,

Aeromonas spp., Giardia spp., and Cryptosporidium spp. With the exception of Giardia spp. and Cryptosporidium spp. being significantly correlated, all other indicators excluding turbidity failed to correlate with these agents. Failing indicators included fecal coliforms, fecal streptococci, C. perfringens, Aeromonas spp. and F-RNA bacteriophage.

Other illness-causing organisms (cyanobacteria or blue-green algae) have been documented and their abundance is often associated with poor water quality. As previously described, generally we associate increases in turbidity with fecal indicators and presumably fecal-oral pathogens. With respect to cyanobacteria, increasing turbidity is also associated increasing blue-green algae density and greater cyantoxin concentrations (Scheffer et al. 1997, Sedmak and Kozi 1998). Likewise, cyanobacteria have been found associated with increased illness risk among recreational water users,

23 where increasing exposure resulted in statistically significant trends in self-reported illness (Pilotto et al. 1997). In this article, the authors recognize and report that the phycocyanin (dominant blue-green algae pigment) may be acting as an indicator for other microbial organisms causing the increased illness incidence among exposed groups.

Phycocyanin and blue-green algae have been strongly associated with phosphorus and water transparency (as measured by secchi depth ) (Trembe and Prepas 1987). In addition, phosphorus has been associated with enabling pathogen persistence and has also been observed in greater concentrations when fecal indicator bacteria concentrations are high. Phosphorus is known to be an important nutrient for microbial survival. ATP,

DNA, phospholipids and other biomolecules in bacterial cells require phosphorus. The presence of phosphorus in water improves bacterial survival even in drinking water

(Juhna et al. 2007) and elevated phosphorus levels are typically observed in agricultural settings where E. coli levels are also elevated (Byers et al. 2005).

2.6. The History of Non-Bacterial Indicators of Recreational Water Quality

Reliance on predictive models utilizing non-bacterial indicators have been gaining interest. Several models have been used to predict E. coli densities or the probability of an exceedance in water quality criteria in the Great Lakes (Nevers and Whitman 2005,

Francy et al. 2003). These two studies only predict or attempt to predict E. coli densities using a variety of parameters including turbidity, wave height, and rainfall, among a few factors in their model. Interestingly, these predictive models predict E. coli densities

24 better than the required previous day E. coli method. No data from these studies have been reported using the following potential indicators: total phosphorus, chlorophyll A and blue-green algae pigments, all of which have been associated with fecal pollution for over 50 years (Hutchinson 1957, Edmondson 1970).

The degree to which these indicators can predict adverse health outcomes has been given limited attention in previous epidemiologic studies. For over 50 years (since Stevenson

1953), the focus has been on indicator bacteria. Few studies have linked non-biological indicators, such as chemical, physical or a combination of indicators to illness rates at any beaches. Now, it is acknowledged that simple measures, like turbidity, should be included in predictive models. Turbidity and transparency impact UV penetration into waters and consequently impact the level of solar disinfection of surface waters. Wet weather phenomena are also associated with turbidity, especially when urban and agricultural runoff contains high nutrient loads. Additionally, turbid water in coastal areas has been associated with human enteric pathogens and is significantly associated with fecal indicators such as coliphage (Lipp et al. 2001). Lastly, a recent study of the Lake

Erie Nowcast model reports that turbidity is the best predictor of bacteria concentrations in their model (Frick et al. 2008).

2.7. Considering Other Water Quality Indicators

For over 30 years, Carlson‟s Trophic State Index (TSI) has been used by lake managers for the management and study of inland lakes. Carlson‟s TSI values represent an

25 estimated doubling in algal biomass for every increase of 10 TSI units. Dr. Carlson developed the index from observations of phosphorus cycling in Minnesota lakes while pursuing doctoral degree at the University of Minnesota. Carlson acknowledged that agricultural and urban runoff, including human sewage, were significant contributors of phosphorus. Phosphorus, serving as the limiting nutrient in most inland waters, was associated with algal growth and turbidity in Carlson‟s studies (Carlson 1977).

As early as the 1950s, phosphorus from human and agricultural waste was associated with increased nutrient loading (Hutchinson 1957). By 1970 it was evident in

Edmondson‟s landmark ecological study that sewage diversions in Lake Washington had contributed to water quality problems, particularly by elevating the phosphorus levels

(Edmondson 1970). Edmondson demonstrated a significant decline in phosphorus after sewage was diverted away from Lake Washington. Edmondson did not evaluate the public health impacts of the diversion.

The metrics used to calculate Carlson‟s TSI are surface phosphorus, chlorophyll A and transparency. All three measurements are fairly simple to assess with the appropriate equipment. The cost, validity and reliability associated with these measurements have made Carlson‟s TSI a standard method for evaluating lake water quality. The three TSI parameters are highly correlated and individual TSI values can be approximated using only one of the three parameters. To accommodate for the temporal fluctuations that can occur among the individual TSI parameters, specific calculations of more precise TSI

26 values can be accomplished by averaging the three equations provided by Carlson (1977).

For the past ten years, the State of Minnesota has used Carlson‟s TSI in the determination of whether its lakes are swimmable (Heiskary 1997). The decision by Minnesota to use

Carlson‟s TSI was related to algal blooms and has not been evaluated with any health outcomes study related to waterborne pathogens.

2.8. Emerging Trends in Beach Exposure Studies

Water quality continues to serve as a single index or tool for estimating disease risk among beach users. When taking an all hazards risk assessment approach, multiple routes of pathogen exposure exist. Furthermore, adverse effects can result from exposures to chemical, physical and non-pathogenic biological agents (cyanotoxins). No research scientist in this area of study would believe all beach-related illnesses are associated with water contact. Potential confounders need to be considered and would enable improved understanding of beach-related illnesses with respect to sand quality and sand exposure, foodborne diseases, direct person-to-person contact, beach restroom hygiene, sun exposure, and other potential sources.

The relationship between sand exposure and illness was recently demonstrated by

Heaney et al. (2009), in which increased sand exposures (e.g. digging and being buried in sand) were associated with elevated risks of developing GI illness and diarrhea. It has been known that E. coli and Enterococci are present in the wet and dry sands along the

27 beach waterfront in concentrations orders of magnitude higher than in the open water

(Whitman et al. 2003, Francy et al. 2002). It has also become more widely accepted that the bacteria-laden beach sands at many beaches act as sources for open water pollution as opposed to sinks (Yamahara et al. 2007, Desmarais et al. 2002, Whitman and Nevers

2003). Exposure to beach sands may influence epidemiologic studies and reported illness associations.

Like beach sands, bather densities are also gaining more attention. A significant amount of data demonstrating that beach users influence their own recreational water quality through pathogen shedding and disturbing sediments is appearing in recent literature.

Graczyk et al. (2007a) demonstrated statistically significant increases in human-virulent microsporidian spores (p = 0.04) on weekend samples where bather densities were significantly higher than weekday samples (p = 0.001). Coupled with these increases was a statistically significant increase in turbidity (p = 0.04). Rainfall, tide, temperature, conductivity, and salinity were not different than weekday samples, suggesting no significant inflows and a potential direct human contribution to the beach water via bather fouling. Bather shedding versus suspension from sediment was not ruled out by the data in the study. Similar results and conclusions were again demonstrated by

Graczyk et al. (2007b) with Cryptosporidium and Giardia.

As noted above, sediment contributions cannot be ruled out; however, other bather densities studies examining pools and timed exposures, demonstrate bather shedding can

28 significantly elevate concentrations of fecal indicator bacteria and Staphyloccus aureus in surrounding waters. Elmir et al. (2007) present data illustrating significant bather shedding in marine water, where bathers shed an average of 3 x 106 CFU of S. aureus per

15 minutes and shed enterococci at a rate of 3 x 105 CFU per 15 minutes. The Elmir team is showing similar results for enterococci and Bacteriodales, an emerging human- associated fecal indicator (Elmir et al. 2009). This phenomenon has been associated with increased illness in at least one older epidemiologic study of only 144 individuals at a freshwater beach (Calderon et al. 1991). Despite this acknowledgement 18 years ago, bather shedding research is not fully understood, which has warranted greater attention.

For this reason, Elmir consistently argues that bather density must be considered when making evaluations of acceptable illness risk associated with beach exposures.

2.9. Issues Warranting New Knowledge – Background Summary

The literature review clearly indicates that there are a great number of research challenges and opportunities in this particular area. The greatest challenges and opportunities involve enhancing the epidemiologic studies and the relevant exposure science.

The biggest problem facing many in environmental health today is a lack of valid and reliable health outcome data. This challenge continues to challenge researchers in the recreational water exposure arena. Epidemiologic studies that evaluate illnesses beyond

GI illness (e.g. infected cuts, skin rash, ear infection, etc.) are warranted, particularly if

29 coupled with new or novel indicators of fecal contamination. Furthermore, epidemiologic studies may need to go beyond associations between fecal indicators and predictors of illness. Predictors of illness may be more relevant than predictors of fecal contamination. Bather density, turbidity, and phycocyanin are associated with pathogens and other disease-causing agents that may not be detected by traditional fecal indicators.

These three factors may be associated with illness, but limited research exists to support hypotheses suggesting their association with illness.

An additional problem is that epidemiologic studies have failed to examine the relationship of test real-time predictors of fecal contamination with respect to water- related illness. Predictive models of water quality may better predict illness than fecal indicator bacteria. To date, no predictive model using solely physical and chemical measurements has been associated with illness data. Other real-time methods, such as qPCR and IMS-ATP methods have had limited to no use in epidemiologic studies thus far. The ability to incorporate these new methods in epidemiologic studies is likely to add substantial value to the field.

The association between human health and water quality deserves much attention, particularly because of the size of the population at risk and the error rate associated with reporting water quality advisories related to single-day maximum exceedances. A Lake

Michigan study on one beach from 1998 to 2001 demonstrates that advisories were posted erroneously 12% of the time (14 of 118 days). More problematic was that

30 postings (based on previous day measurements) only prevented 42% of predicted illnesses associated with advisory conditions. On the 14 days, it is estimated that daily regional economic losses ranged from approximately $1,200 to $37,000 due to lost area sales related to beach visits (Rabinovici et al. 2004). The economic consequences presented here are only based upon expenses associated with visiting the beach and fail to weigh the economic consequences associated with enteric illness.

The failure to adequately assess water quality and subsequently communicate beach risk has real public health and financial implications. Given and others (2006) studied 28

California coastal beaches in Los Angeles and Orange Counties. Using the traditionally delayed culture-based methods for assessing water quality data, total number of beach users and existing enterococci density models, Given et al. (2006) predicted there are

627,800 to 1,479,200 excess enteric illnesses associated with swimming at these 28 beaches per year. The economic costs associated with these enteric illnesses is between

$21 million and $51 million per year depending upon models. The Given et al. (2006) study does not include the other possible non-enteric illnesses.

After three swimming seasons the Ohio "nowcast" model continues to be a valuable tool for predicting recreational water quality at two Lake Erie beaches. Developed in a coordinated effort over six years with nine partner agencies, the 2008 Nowcast model consistently better predicted exceedances in indicator levels at Huntington (Ohio) and

Edgewater Beaches than the traditional previous day E. coli colony count method. Using

31 publicly accessible data from the Nowcast system's web interface, it was observed the model accurately predicted recreational water quality for issuing or not issuing advisories at the two beaches better than the previous day method (Huntington = 85.8% versus

75.5%; Edgewater = 61.7% versus 51.9%). The predictive model outperformed the previous day method with regards to specificity, positive predictive value and negative predictive value. The sensitivity of the model was superior at Huntington but not

Edgewater. At Huntington Beach the model sensitivity was substantially better than traditional method‟s sensitivity (Model = 57.14%; Previous Day Method = 7.1%).

Sensitivity values at Edgewater were low for both approaches (Model = 29.3%; Previous

Day Method = 34.1%). The sensitivity of both the predictive model and the previous day method needs improvement and future methods should continue to attempt to better predict exceedances. The results demonstrate obvious shortfalls in the traditional method for predicting exceedances, but suggest potential in predictive modeling to understand same day recreational water conditions. Predictive models including more or better indicators may produce substantial benefits in estimating water conditions and protecting public health at the nation's beaches.

In reviewing the literature, the model at Huntington Beach (Ohio) using the Nowcast system predicted advisory conditions much better than the traditional method. A 50% improvement in posting advisories when advisories need to be posted could have substantial public health and economic benefits. Models that are constructed with immediate, inexpensive, and simple methods have much promise for preventing illness.

There are two big gaps in the literature that need to be addressed. The first involves no attempts to try to model fecal indicator behavior at inland lakes with simple, inexpensive, immediate methods. The second gap is in the epidemiologic data. To date, there is no evaluation of a predictive model with regards to its ability to predict illness risk associated with exposures at inland or coastal waters. The number of illnesses that could be prevented through the use of a method that can provide beach users and beach managers with more timely information has the potential to prevent millions of recreational water illnesses annually.

Chapter 3: Association of Gastrointestinal Illness and Recreational Water

Exposure at an Inland U.S. Beach

3.1. Introduction

The link between adverse health outcomes and recreational swimming and bathing has been well documented since the 1950s (Stevenson, 1953). During the last 30 years, associations between fecal indicators and illness have been established for beach swimmers (Prüss, 1998; Wade et al., 2003; Wade et al., 2006; Colford et al., 2007) including the 1984 U.S. EPA study (Dufour, 1984) that established Escherichia coli as a criteria indicator for fecal contamination in freshwater (U.S. EPA, 1986).

Furthermore, the current U.S. recreational water quality criteria for fecal indicators are scheduled for revision no later than October 15, 2012 per a consent decree stemming from Natural Resources Defense Council v. Johnson and U.S. EPA (2008). At the present time, there are limited epidemiologic data for recreational water exposure from inland U.S. waters with most of the research in the last twenty-six years having focused

34 on the Great Lakes (Wade et al., 2006; Wade et al., 2008) and coastal marine waters

(Wade et al., 2003). Additionally, there has been limited published epidemiologic research evaluating illness outcomes among users of inland recreational waters.

Beyond epidemiologic studies, outbreak analysis provides a basis for evaluating the public health threat and disease burden. The evidence from reported outbreaks suggests the majority of recreational waterborne diseases from untreated water outbreaks are from untreated inland waters. In the most recent U.S. Centers for Disease Control (CDC) surveillance report on recreational waters, inland waters contributed to 19 of the 20

(95%) of the U.S. recreational waterborne disease outbreaks from non-Vibrio agents in untreated waters from 2005-2006 (Yoder et al., 2008). This finding has been observed in previous CDC surveillance reports. For example in 2003-2004, all 19 of 19 outbreaks

(100%) were from inland waters (Dziuban et al., 2006). In 2001-2002, all 11 of 11 outbreaks from untreated waters were non-coastal (Yoder et al., 2004). This evidence is clear in indicating that U.S. non-coastal waters are a public health concern warranting focused epidemiologic investigation.

Prior to 2003, the three most notable freshwater epidemiology studies in the U.S. were

Stevenson (1953), Dufour (1984) and Calderon et al., (1991) as cited in Wade (2003), with 5,124, 21,777 and 144 participants, respectively. These studies report associations between gastrointestinal (GI) illness and water quality as indicated by fecal indicator bacteria; however, these studies are few and were limited to few bodies of water.

In response to a growing number of arguments against the use of enterococci and E. coli standards in the scientific literature, Wade et al. (2003) engaged in a comprehensive meta-analysis of the 27 most relevant epidemiologic studies, which captured studies ranging from 247 to 26,686 participants. For GI illness, the correlation coefficient was r

= 0.86 for the natural log relative risk of GI illness as a function of E. coli density. The

Wade et al. (2003) study provided evidence in support of the U.S. EPA‟s position on using E. coli as a recreational water standard to reduce GI illness risk among freshwater contact recreation users. Despite good correlation between E. coli density and GI illness relative risk, the correlation was based solely upon three studies.

To contribute knowledge to the understudied issue regarding illness risk among inland recreational water users we engaged in a prospective cohort study. The primary goal of this study was to evaluate the effectiveness of E. coli as an indicator of GI illness risk among recreational water users at East Fork Lake, Ohio, U.S.A. The reservoir, East Fork

Lake, was selected to improve our understanding of illness risk among swimmers exposed to waters of a typical Midwestern U.S. flood-control reservoir used often for recreational activities. This study was designed to identify differences in GI illness risks among swimmers and non-swimmers. By sampling over 26 days, the study design allowed the exploration of GI illness risk for swimmers exposed to various E. coli densities.

3.2. Materials and Methods

3.2.1. Overview of Methods

Beach water samples and human health data were collected from the public beach at East

Fork State Park during the 2009 swimming season. The study design was a prospective cohort design similar to that used by Wade et al (2006) and Wade et al., (2008). In brief, subjects were recruited from the beach on the same day that water quality measurements including E. coli were conducted. Subjects completed a survey at the beach ascertaining demographics and water-related behavior s (e.g. wading, playing or swimming in water).

Subjects were contacted eight to nine days later by phone to ascertain potential water- related illness. The study design, questionnaires and related materials were reviewed and approved by the Institutional Review Board at The Ohio State University (IRB Protocol

#2009H0107).

3.2.2. Sampling Site

Beach water sampling and the human health survey were administered at the public beach at East Fork State Park (39°1‟ 11.2” N; 84° 1‟ 2.8” W) over 26 weekend days starting May 30 and ending August 30, 2009. The 365-m public beach is located along the lake, a human-made 8.7-km2 flood control reservoir providing numerous benefits to the Cincinnati, Ohio, U.S.A., metropolitan/suburban area including full-body contact water recreation. Probable pollution sources for the reservoir are primarily non-point, with dominant watershed land uses being row crop agriculture (37%), light urban/residential (33%) and forestland (25%) (East Fork Watershed Collaborative 2006).

Several small municipal wastewater treatment plants are permitted to discharge into the lake tributaries; however, no discharges are permitted directly into the reservoir. Failing home sewage treatment systems have been identified as likely fecal contamination sources for the lake tributaries and reservoir (East Fork Watershed Collaborative 2006).

3.2.3. Recreational water samples

Samples were collected in accordance with the Ohio Department of Health (2009) standard methods. In brief, a single daily sample was collected using autoclaved 500-mL

Nalgene bottles or 500-mL sterile Whirl-Pak© bags in water of approximately three feet in depth at the approximate center of the 365-m beach. The sample was gathered by sweeping the bottle or bag one foot below the surface of the water. Scum and debris were avoided in the sweeping process. From this single daily sample collection at the beach, all of the analyses and archiving were performed. Within-day sample times varied from 10:50 to 20:25 although most samples (92%) were collected within 4 hours of the median sample time of 13:52. Ancillary water measurements and weather observations were made at the time of sampling. Daily precipitation and the mean daily water inflow for the watershed of the reservoir were provided for each sample collection day by the

Louisville District of the U.S. Army Corps of Engineers (2010). All water quality laboratory analyses were performed within six hours of sample collection. E. coli density was quantified using EPA Method 1603 (U.S. EPA, 2002). Autoclaved deionised water was used as a negative control. The colony counts were converted to colony forming units (CFU)/100 mL after dividing by the filtration volume.

3.2.4. Administration of Beach User Questionnaire

The sampling approach was a cluster design that enrolled households with a household member over 18 years old who was using the East Fork beach and willing to speak on behalf of their household. Subjects were recruited over 26 days across 13 weekends during the summer of 2009. Beachgoers were invited into the study through signage and/or by being approached by study personnel. The invitation for participation followed an approved standard script. The process of consent was administered verbally to persons eligible to consent. Upon receiving consent, households choosing to engage in the full study participated in an initial interview at the beach, a second interview when leaving the beach, and follow-up phone interview 8-9 days after their visit.

The initial interview at the beach captured pertinent enrolment information. This included the identification of household members at the beach as well as phone numbers for the follow-up telephone interview. In addition, information was collected concerning potentially confounding existing symptoms as well as age and sex. Successful completion of the interview resulted in a gift and instructions for providing follow-up information before leaving the beach.

The second interview gathered household member-specific exposure data including: time spent at the beach, time in the water, whether the body or head were submerged, consumption of food at the beach, and others. This second interview concluded with

39 information concerning the telephone interview follow-up survey. Completing the 2nd portion of the survey resulted in gifts for household members.

The telephone interview was the third and final interview. Using previously collected telephone numbers, the participants were contacted 8-9 days following their visit to the East

Fork beach. The information gathered permitted the documentation of any new symptoms observed since the initial interview at the beach 8-9 days prior.

The survey instrument employed was a modified version of the U.S. EPA questionnaire developed for the National Epidemiological and Environmental Assessment of

Recreational Water (NEEAR) study. The full questionnaire has been successfully employed by Wade et al. (2006), Wade et al. (2008) and Heaney et al. (2009). Like the original NEEAR Beach Questionnaire, our modified and reduced length questionnaire retained key questions to ascertain three types of information: exposure status, illness status/symptoms, and demographics. Across all three information domains, the identified adult respondent (> 18 years of age) provided information for themselves and others within their household. Information about exposure status included the amount of time spent at the beach, as well as their activities including the amount of time spent in the water, head immersion, and others. Illness/symptom questions included a variety of possible ailments, not limited to GI symptoms, ear infection, skin rash or skin infection.

Demographic questions were limited to age, gender and ethnicity. Ancillary information

40 was collected regarding the miles users travelled to visit the beach, beach visit frequency, and the importance of cost (free) in their decision to visit the beach.

3.2.5. Classification of Beach Exposures

To characterize beach exposures, dichotomous and categorical classifications were used.

The dichotomous classifications pertained to the identification of exposed (coded as “1”) versus unexposed (“0”). The exposed swimmer group represented respondents that reported to “wade, swim or play in the water”; whereas, the unexposed group indicated no such activity. An additional exposure of interest was the consumption of food at the beach. Respondents reporting to have “consume[d] food while at the beach” the day of the survey were also dichotomously classified as “1” or “0”.

A categorical classification scheme was used to further characterize the level of microbial exposure among the swimmer group. The water quality parameter E. coli was used for class coding. For the logistic regression model, the E. coli classes were 0-3.3, >3.3-11.3,

>11.3-59 and >59-1,538 CFU/100 mL. These values are the quartiles for the E. coli distribution of the beach water samples among the 806 swimmers.

3.2.6. Classification of Health Outcomes

The data were explored to identify any reported illness, GI illness and highly credible gastrointestinal illness (HCGI). Symptoms including but not limited to fever, headache, nausea, diarrhoea, and vomiting were treated as dichotomous variables, where individuals

41 reporting a symptom were coded as “1” for the symptom or “0” if no symptom was reported. GI was defined using the definition established by Dufour (1984), which includes any person reporting any of the following symptoms: nausea, stomachache, diarrhoea or vomiting. GI illness was treated as a dichotomous variable, where individuals with no new GI illness symptoms were coded as “0” and persons with one or more new GI illness symptoms since visiting the beach were coded as “1”. HCGI was defined using the “HCGI-1” definition employed by Colford (2007) and nearly identical to the HCGI illness definition established by Dufour (1984). Persons reporting to exhibit any of the following conditions were coded as “1” for HCGI illness: (1) vomiting; (2) diarrhoea and fever; (3) stomachache or nausea accompanied with a fever.

3.2.7. Data Analysis

GI illness was initially explored by performing tabulations with exposure and demographic data. Initial data analysis focused on the occurrence of specific GI-related symptoms among swimmers and non-swimmers. Further tabulations were performed to compare the frequency of GI and HCGI illnesses among swimmers and non-swimmers.

GI and HCGI illness frequencies were further explored across the four levels of water quality as determined by the distribution of the E. coli density quartiles. Additional tabulations for characterizing the frequency of GI-related symptoms among beach users were performed taking into consideration age, sex and other beach exposures such as food consumption.

Following initial data exploration, multivariable analysis was employed. Logistic regression was used to estimate the risk of GI and HCGI illness via observed adjusted odds ratios. Models were constructed to consider potential confounders and/or effect modifiers such as age and sex. Due to the clustering of participants by household, the data were not considered to be a simple random sample, but instead for the purposes of statistical analysis, the data were treated as clustered by household (Levy and Lemeshow,

1999). Model construction involved the identification of biologically plausible covariates coupled with statistical selection. A reverse stepwise selection procedure was used in model 3 to select variables with some likely association (p < 0.15; Bendel (1977)) with

GI and HCGI illness. Model 3 estimates adjusted odds ratios for swimmers across the four E. coli density levels.

Ultimately, three adjusted logistic regression models were constructed. The first model

(model 1) estimated GI illness risk for swimmers, where swimmers are those who reported wading, playing or swimming in the water. This model used non-swimmers as the reference group and adjusted for only age, gender and the reservoir inflow. With concerns that age and flow were not likely to be linear in the logit for GI and HCGI illness outcomes, age was categorized into six groups that could be classified as young child, older child, teenager, young adult, adult and older adult. Reservoir inflow was categorized using the terciles of exposure for the respondents. The consumption of food at the beach and varying levels of E. coli were not included in this model, despite being associated with the illness outcomes. The second model (model 2) estimates GI and

HCGI illness risk for persons who consumed food at the beach. Model 2 was constructed similar to model 1, with adjustments being made for only age, gender and reservoir inflow. Model 2 did not differentiate swimmers from non-swimmers and did not account for E. coli density, albeit associated with GI illness. The third model (model 3) is the most complex. Model 3 only included swimmers and was developed to determine GI and

HCGI illness risk among swimmers in waters with various densities of E. coli. Model 3 only included age, gender and the statistically-associated covariates (p < 0.15). The dichotomous variable regarding food consumption at the beach and the categorical variable for the 48-h reservoir inflow were both included due to their level of association with GI and HCGI illness in this model. For all models, an adjusted odds ratio (AOR) not including 1.0 in the 95% CI of the AOR was considered significant, where p < 0.05.

The performance of each of the models was evaluated based on fit (Hosmer-Lemeshow

Goodness-of-Fit Test (Hosmer and Lemeshow, 2000) and area under the receiver- operating-characteristic (ROC) curve (Metz, 1978).

Linear regression procedures were also employed to identify the strength of association between GI illness and E. coli density. Linear regression and Pearson correlation analysis was performed as in Dufour (1984) to permit comparison with previous epidemiologic data. Using this methodology, E. coli (CFU/100 mL) was plotted on a logscale versus GI and HCGI illness per 1,000 individuals. Three dates containing less than 10 total observations from swimmers were excluded due to their small sample size and large standardized residual. Additionally, three of the remaining 23 sampling dates

44 had mean E. coli densities less than 1 CFU/100 mL and were converted to 1 CFU/100 mL. Pearson correlations were determined to be significant if p < 0.05.

Associations between illness rates (GI and HCGI) and E. coli density were also evaluated for trend. The Cochran-Armitage Test was employed where GI and HCGI illness rates were on an ordinal scale from lowest mean daily E. coli density to highest. Trends were determined using the Cochran-Armitage Test for Trend where a p < 0.05 indicated a statistically significant trend in the observed data.

Initial tabulations, logistic regression modelling, assessing logistic model fit, and the

Cochran-Armitage Test for Trend were performed with Stata 11 (Stata Corporation,

College Station, TX, USA). Linear regression and Pearson correlation analysis were performed using Minitab 15 (Minitab Inc., State College, PA, USA).

3.3. Results

3.3.1. Enrolment

A total of 682 households consented and completed the initial beach survey. Of these households, 554 returned to complete the beach exit interview. Of those completing the beach exit survey, 300 households were successfully contacted and completed the follow- up phone survey for an overall household study retention rate of 44%. Accordingly, results are presented for the 300 households and 965 individuals who were living at the household for the entire follow-up period. Demographic characteristics of this group are

45 provided in Table 3.1. Our sample was predominantly white (93%), and travelled less than 40 miles to visit the beach (81%).

3.3.2. Characterization of Beach Water Quality

A total of 26 water samples were collected over 13 weekends and all were successfully analyzed. The beach was open and occupied over the 26-day sampling period. For two of the 26 sample days, E. coli densities exceeded the advisory threshold of 235 CFU/100 mL E. coli densities of 1,538 CFU/100 mL and 487 CFU/100 mL were observed these two days. These two observations were the highest E. coli density observations, and

Table 3.2 indicates these results were significantly higher than the majority of observations for E. coli density during the sampling period. These results track with wet weather experienced 24-h and 48-h prior when 6.7 cm of rainfall fell in a 24-h period from 6 A.M. to 6 A.M. the day prior to observing the highest E. coli density. Despite no additional reported rainfall, E. coli densities remained above advisory conditions for a second day. Furthermore, 6 A.M. tributary inflows were 214 m3/s and 18.2 m3/s on these two sample days. The largest 24-h inflow volume reported during the entire study period was 214 m3/s. Median inflow for this reservoir during the sampling period was 1.5 m3/s.

Table 3.2 demonstrates that the water quality at this beach generally exhibited low E. coli densities, as illustrated by a median E. coli value of 9.1 CFU/100 mL.

Table 3.1. Individual and household characteristics of persons surveyed at East Fork

Lake (Ohio, United States) with complete surveys and telephone follow-up.

Household/Respondent Characteristics No. (%)

Household Size (No. of Individuals)

1 51 (17)

2 66 (22)

3 55 (18)

4 64 (21)

5 37 (12)

> 6 27 (9.0)

Distance Travelled to Beach (Miles)

0 – 9.9 170 (18)

10 – 19.9 242 (25)

20 – 29.9 192 (20)

30 – 39.9 177 (18)

> 40 184 (19)

Ethnicity of Individual Household Members

White 895 (93)

Black 22 (2.2)

Hispanic 16 (1.7)

Asian 9 (0.9)

American Indian 5 (0.5)

Other 1 (0.1)

Missing 17 (1.8)

3.3.3. Characterizing Exposures of Beach Users

Characteristics of individuals completing the survey with respect to both demographics and exposure activities are provided in Table 3.3. These data indicate that males and females were similar in their behavior s to wade, play or swim in the water (80 vs. 77%), but females were more likely to consume food at the beach (64 vs. 57%). The youngest age group (0 – 5 years) was most likely to consume food at the beach with 64% of group members having this exposure. Additionally, 90% of the youngest age group members had beach water exposure. The older children (6 – 11 years) and the adolescent/teenager group (12 – 18 years) had the most reported exposure to the water; whereby 98% of 6 –

11 year olds and 93% of 12 – 18 year olds reported to wade, play or swim.. The senior age group (56.0 – 73.9 years) reported the least frequency for food consumption (38%), and wading, playing or swimming in the water (41%).

Table 3.2. Descriptive statistics of beach water quality and beach usage during sampling days (N=26) at East Fork Lake (Ohio, United States)

1st 3rd Beach Parameter N Mean ± S.E. Median Range Quartile Quartile

Temperature (°C) 26 26.7 ± 0.4 25.5 25.9 28.6 24.2 – 30.4

Specific 273 259 – 307

Conductivity 26 276 ± 2.3 268 282

(µS/cm)

Turbidity 14.6 7.0 – 116 26 22.6 ± 4.7 10.6 21.3 (NTUs)

E. coli (CFU/100 9.1 0 – 1538 26 95.1 ± 60.7 0.5 37.8 mL)

48-hr. Reservoir 52 0 – 8184 26 504 ± 313 0 351 Inflow (m3/sec)

Beach User 156 4.0 – 483.0

Density 24 185.0 ± 27.3 57.8 302

(Persons)

3.3.4. Characterizing Illness among Beach Users

Of the 965 individuals included in our sample, there were 109 cases (11.3%) of reported total illness. Of these, 48 (44%) were GI-related. The partitioning of these adverse health outcomes across demographic and exposure factors are presented in Tables 3.3, 3.4 and

3.5. Table 3.3 illustrates that GI-illness incidence was similar among males and females; however, the youngest age group had the highest incidence of GI illness. Among the exposure activities, Table 3.4 demonstrates the group not exposed to water reported 3

(1.9%) cases. Only swimmers in the lowest E. coli density exposure group reported less

GI illness (1.1%), although no statistical difference exists between these groups. GI illness proportions were significantly higher for swimmers in the two highest E. coli density groups, impacting over 8% of swimmers. Stomachaches, diarrhoea, and HCGI illness were also reported more frequently in these highest two groups.

Table 3.3. Summary of beach exposures and reported illness by gender and age classification of respondents during the 2009 swimming season at East Fork Lake (Ohio,

United States)

Beach Exposures Reported Illnesses

No. No.

Consumed Exposed GI Illness HCGI

Food at Body to Illness

Beach Users Attributes No. (%) Beach (%) Water (%)

Gender

Female 540 (56) 333 (64) 444 (82) 23 (4) 10 (2)

Male 425 (44) 241 (57) 362 (85) 25 (6) 3 (1)

Age (years)

0-5 127 (13) 81 (64) 114 (90) 11 (9) 2 (2)

6-11 174 (18) 102 (59) 170 (98) 8 (5) 2 (1)

12-18 137 (14) 85 (62) 128 (93) 5 (4) 3 (2)

19-30 174 (18) 103 (59) 147 (84) 8 (5) 1 (1)

31-55 306 (32) 182 (59) 224 (73) 15 (5) 5 (2)

56-74 32 (3) 12 (38) 13 (41) 1 (3) 0 (0)

Missing 15 (2) 9 (60) 10 (67) 0 (0) 0 (0)

Table 3.4. Summary of gastrointestinal illnesses for beach users across various E. coli density exposure levels at East Fork Lake (Ohio, United States) during the 2009 swimming season.

E. coli Density Levels among Exposed Individuals

Not Exposed 0-3.3 >3.3-11.3 >11.3-59 >59-1,538

toWater CFU/100 mL CFU/100 mL CFU/100 mL CFU/100 mL

No. of Individuals 159 186 178 208 234

No. No. No. No. No.

w/Symptom w/Symptom w/Symptom w/Symptom w/Symptom

GI Symptom (%) (%) (%) (%) (%)

Nausea 0 (0) 0 (0) 3 (1.7) 2 (0.96) 6 (2.6)

Vomit 0 (0) 1 (0.54) 1 (0.56) 3 (1.4) 6 (2.6)

Diarrohea 3 (1.9) 2 (1.1) 4 (2.2) 16 (7.7) 12 (5.1)

Stomachache 0 (0) 0 (0) 0 (2.8) 1 (0.48) 8 (3.4)

Fever 0 (0) 0 (0) 5 (2.8) 3 (1.4) 4 (1.7)

GI Illness 3 (1.9) 2 (1.1) 6 (3.4) 18 (8.7) 19 (8.1)

HCGI Illness 0 (0) 1 (0.54) 1 (0.56) 5 (2.4) 6 (2.6)

3.3.5. GI Illness Risk Estimates for Beach Users

Using logistic regression, we developed three models to evaluate predictors for GI and

HCGI illness risk among East Fork beach users. Table 3.5 provides the adjusted odds ratios (AOR) for three models. The results of model 1, adjusted for age, sex and reservoir inflow, demonstrated a significant AOR for persons who reported wading, swimming or playing in the water (AOR=3.2; CI: 1.1, 9.0) suggesting a 3.2-fold increased odds for GI illness among those who went in the water. The AOR for HCGI was not determined as the non-swimmer reference group reported no HCGI cases. Model

2, adjusted for age, sex and reservoir inflow, demonstrated increased odds for GI illness

(AOR=3.6; CI: 1.4, 9.9) and HCGI illness (AOR=7.2; CI: 1.1-48) for persons reporting to have consumed food while at the beach.

Model 3, which only estimates GI and HCGI illness risk among persons who had contact with the water, demonstrated significant GI illness risk estimates for persons exposed to the highest E. coli levels. After adjusting for age, sex, reservoir inflow, and consuming food at the beach, exposure to beach waters in the second highest E. coli quartile (>11.3 –

59 CFU/100 mL) presented a significant elevated odds ratio for GI illness (AOR=7.2; CI:

1.3, 39), but not for HCGI illness (AOR=6.0; CI: 0.54, 71). Likewise, the highest E. coli quartile (>59 – 1,551 CFU/100 mL) presented similar findings for GI (AOR=7.0; CI: 1.5,

32) and HCGI illness (AOR=3.7; CI: 0.63-77). Although statistical significance was not achieved in the odds ratios for HCGI illness in these two highest E. coli density classes, the respective p-values (p=0.152 and p=0.287) demonstrate that the direction of an

Table 3.5. Association of beach user attributes with illness outcomes across 26 sampling days at East Fork Lake (Ohio, United States).

GI HCGI

No. Ill AOR p No. Ill AOR p

(% Ill) (95% CI) (% Ill) (95% CI) Model No. (%)

Model 1: Exposed Body to Water1

Unexposed 159 (16) 3 (1.9) 0 (0)

Exposed2 806 (84) 45 (5.6) 3.2 (1.1 -9.0) 0.028 13 (1.6) ---- 3

Model 2: Consumed

Food at Beach1 45 (5.6)

Unexposed 379 (40) 8 (2.1) 7 (1.9)

Exposed 574 (60) 40 (7.0) 3.6 (1.4 -9.9) 0.010 36 (6.3) 7.2 (1.1 -48) 0.040

Model 3: E. coli Density

(CFU/100 mL)4 36 (6.3)

0.1 – 3.3 186 (23) 2 (1.1) 1 (0.5)

>3.3 – 11.3 178 (22) 6 (3.4) 3.2 (0.53 -19) 0.204 1 (0.6) 1.6 (0.07 -37) 0.768

>11.3 – 59 208 (26) 18 (8.7) 7.2 (1.3-39) 0.022 5 (2.4) 6.0 (0.54-71) 0.152

>59 – 1551 234 (29) 19 (8.1) 7.0 (1.5-32) 0.013 6 (2.6) 3.7 (0.63-77) 0.287

1 Adjusted for age, sex, reservoir inflow, and clustering within households

2 Exposure is defined as persons who reported to wade, swim or play in the water

3 AOR for HCGI not reported in exposed group since no reference cases in the unexposed group.

4 Adjusted for age, sex, reservoir inflow and consuming food at beach, and clustering within households after excluding persons reporting no wading, swimming or playing in water 54 association, if it exists, is towards increased odds of HCGI illness. Similarly, for the lowest E. coli density class (>3.3 – 11.3 CFU/100 mL) above the reference group (0.1 –

3.3 CFU/100 mL), the adjusted odds ratio did not achieve statistical significance

(p=0.204); however, the likely direction of the association, if one exists, is towards increased odds of GI illness.

3.3.6. Assessing Performance of Models and Trends

The three logistic regression models constructed to predict GI and HCGI illness vary in performance with respect to discrimination. Using the area under the ROC curve (AUC), the value for the GI illness model in model 1 is 0.63 (Table 3.6). Since the AUC value is below 0.70, we conclude the model does not provide acceptable discrimination (Hosmer and Lemeshow 2000). The models used to estimate GI and HCGI risks in model 2 provide better discrimination as indicated by AUC values of 0.69 and 0.76, respectively.

The AUC value of 0.76 provides acceptable discrimination. The most discriminatory models are the GI and HCGI illness risk models labelled as model 3 in Table 3.6. The

AUC value of 0.72 in the GI illness risk model provides fair or acceptable discrimination; however, the AUC value of 0.81 in the HCGI illness risk model provides good discrimination. Additionally, when assessing the fit of each model, all of the p-values are greater than 0.05 when using the Hosmer-Lemeshow Goodness-of-Fit Test. These p- values from the goodness-of-fit test suggest that the probabilities estimated by the model are an accurate representation of the true disease experienced in the data.

Table 3.6. Model performance values for three logistic regression models predicting illness (GI and HCGI illness) across 26 swimming season days at East Fork Lake (Ohio,

United States).

GI HCGI

Hosmer- Hosmer-

Lemeshow Lemeshow Model Goodness-of-Fit Goodness-of-Fit

Area Under Test Area Under Test

ROC Curve (p) ROC Curve (p)

Model 1: Exposed

Body to Water1,2 0.63 0.3605 ---3 ---3

Model 2:

Consumed Food at

Beach1 0.69 0.1772 0.76 .8454

Model 3: E. coli

Density (CFU/100 mL)4 0.72 0.0714 0.81 .7788

1 Adjusted for age, sex, reservoir inflow, and clustering within households

2 Exposure is defined as persons who reported to wade, swim or play in the water

3 Unable to report since no cases of HCGI were reported in the unexposed group.

4 Adjusted for age, sex, reservoir inflow and consuming food at beach, and clustering within households after excluding persons reporting no wading, swimming or playing in water

Figure 3.1. Regression line estimating highly credible gastrointestinal illness rate among swimmers at East Fork Lake (Ohio, United States) exposed to varying E. coli densities across 23 days1.

1 Sample days with less than 10 observations were removed from the analysis due to their large standardized residual, resulting in a total of three of the 26 days being removed from the regression analysis.

Significant trends were observed in the data. The Cochran-Armitage Test for Trend demonstrated a significant ordinal trend among swimmers for GI illness (p = 0.0016) and

HCGI illness (p = 0.0151) with increasing E. coli densities. Furthermore, the continuous

E. coli data plotted on a logscale demonstrates a positive association with increasing proportions of GI and HCGI cases among swimmers. Figure 3.1 illustrates the positive association between HCGI cases and increasing E. coli density, which is supported by the correlation coefficient (r = 0.467) and the significant correlation p-value (p=0.025) listed on Table 3.7. Table 3.7 further describes the significant correlation between GI illness frequency and E. coli density where p = 0.031.

Table 3.7. Summary statistics of linear regression between GI illness outcomes and E. coli density across 23 sampling days1 at East Fork Lake (Ohio, United States).

Y- Correlation Symptom Slope Std. Error of Slope Correlation P Intercept Coefficient

GI 27.51 12.57 13.08 0.451 0.031

HCGI 10.36 -0.06 4.28 0.467 0.025

3.4. Discussion

The study showed increased GI illness risk among swimmers when compared to nonswimmers, consistent with inland (Stevenson, 1953; Dufour, 1984; Wade et al. 2006) and marine beach studies (Cabelli et al., 1979; Colford et al. 2007). Furthermore, this study demonstrated the continued effectiveness of E. coli as a fecal indicator for determining GI illness risk among swimmers at the study beach. The association of GI illness and the results of the single daily fecal indicator measurement is an important finding that compliments the results of Wade et al. (2006), in which a single rapid

Enterococcus measurement collected in the morning was useful for determining GI illness risk among swimmers.

This prospective cohort study demonstrated an association between contact recreation with beach water and illness. Overall, swimmers were more likely to report experiencing symptoms than non-swimmers, a finding consistent with Dufour (1984) and Fleisher et al. (1996). Like Dufour (1984), GI symptoms were the most frequently reported. The observed GI illness incidence of 56 cases per 1,000 among swimmers is comparable to the inland reservoir beach study results from Dufour (1984) where the range of GI illness incidence over the two-year study period was 37.9 – 61 cases per 1,000 swimmers. Non- swimmers at the inland reservoir studied by Dufour (1984) exhibited a range in GI illness of 19 – 53 cases per 1,000; whereas, East Fork non-swimmers experienced 19 GI cases per 1,000. Similarly, the proportions of swimmers reporting HCGI illness were

59 comparable. Overall, the rates of GI illness in this study are comparable to previously reported results from the largest epidemiologic study of inland U.S. beaches.

The increased risk of GI illness was also observed to be associated with persons reporting to have consumed food at the beach. This variable was identified through reverse selection when constructing the logistic regression model and presumed to be a variable that serves as a proxy for duration of exposure at the beach while taking into account confounding food-related illnesses.

The logistic regression models constructed demonstrate significant associations with GI and HCGI illness (Table 3.5); however, only after taking into account the E. coli densities does the model achieve acceptable model sensitivity and specificity (AUC > 0.70). Table

3.6 shows that the best performing model is model 3 when the outcome variable is HCGI

(AUC = 0.81). This result, in which the HCGI illness model outperformed the GI illness outcome model, is consistent with the results of Dufour (1984). The most plausible reason is the unmistakeable nature of HCGI illness versus more subjective GI symptoms, which enables improved classification of the truly ill. In all models, the models provided an accurate representation of the reported disease in the data as demonstrated by the

Hosmer-Lemeshow Goodness-of-Fit Test results in Table 3.6. Model 3; however, has a p-value near 0.05 in the GI outcome model, which suggests less than an ideal fit and may suggest that an alternative model may produce a better representation of the true disease

60 reported in the data. To the contrary, the more well-defined HCGI outcome for model 3, suggests a very accurate representation of the reported disease in the data.

The observations of increasing GI and HCGI illness incidence in the higher E. coli density exposure groups when compared with swimmers exposed to the least E. coli suggest that with increasing E. coli density, the risk for GI and HCGI illness increases.

These results are confirmed by significant AORs when GI illness is the outcome of interest. The HCGI illness outcome was unable to achieve significance, likely as an artefact of the small sample size and relatively constant and favourable water quality.

Our models are all limited in that the sample size is small and potentially confounding variables, such as additional recreational water exposures, were not able to be considered due to the abbreviated questionnaire administered. Despite these limitations, using swimmers exposed to the lowest E. coli level as a reference group, the assumption is that exposures away from the beach would be similar across all swimmers, regardless of the day they selected to swim at the study lake. Additionally, linear regression analysis of swimmers demonstrated a significant positive correlation between GI and HCGI incidence with increasing E. coli density (Table 3.7). This strength of association was further upheld for both GI and HCGI incidence in trend tests.

Given the limited resources, we were only able to collect a single water quality sample per day. The time of sample collection was not consistent throughout the day, but was still able to produce a significant association with GI and HCGI illness. It has been

61 determined that tremendous variability can exist in water quality at selected marine beaches (Boehm et al., 2002); however, in this study, we were able to estimate GI illness risk using a single daily sample. These results may not be able to be generalised to reservoirs with different watershed dynamics or pollution sources; however, this study lake had stable reservoir inflows and precipitation during the days in which participants were enrolled. Additional E. coli measurements are not available to challenge our assumption that E. coli densities were stable throughout our individual study days.

Lastly, this study reaffirmed the problem of not having a rapid tool available to prevent exposure to recreational waters when infectious disease risk is great enough to warrant an advisory, as recreational water users were unknowingly exposed to advisory conditions on two occasions. The sample day with the peak reservoir inflow was associated with beach user exposures to waters with E. coli densities exceeding 1500 CFU/100 mL.

Despite having limited data regarding water quality and reservoir inflows at this reservoir, it appears that predictive modelling of E. coli at this reservoir offers an opportunity to inform recreational water users in advance of advisory events in an approach similar to Francy et al. (2003). Predictive modelling efforts of E. coli are underway at this reservoir. Future studies using archived filter membranes from the study period are being planned to enable quantification of alternative indicators using rapid molecular techniques.

3.5. Conclusions

Beach users who reported wading, swimming or playing in the water at this inland

beach were at greater risk for GI illness than persons who did not report these

behavior s.

The risk of GI illness increased among swimmers exposed to increasing densities

of the fecal indicator bacteria, E. coli.

A single E. coli sample per day was effective at estimating infectious disease risk

at our study beach and may be an acceptable approach at similar inland lake

beaches.

The membrane filtration method for E. coli quantification continues to be

inadequate in providing beach users with timely information regarding infectious

disease risk, and rapid or predictive tools are warranted.

3.5. Acknowledgements

Funding for this research was provided by a grant from the Ohio Water Development

Authority. Jason Marion received fellowship support through the Public Health

Preparedness for Infectious Diseases targeted investment in excellence at The Ohio State

University. We are appreciative of the staff with Ohio State Parks, especially Scott

Fletcher, for supporting the study. We are grateful to the East Fork Lake beachgoers who graciously agreed to participate in our survey and to fellow graduate (Michael Lofton,

James Rosenblum, Pei-Yu Chiang, Kaedra Wetzel) and undergraduate (Sara Zolinski and

Matthew Farley) students who gave up their multiple summer weekends to assist in recruiting subjects, completing surveys, and entering data.

Chapter 4: Carlson’s Trophic State Index as a Predictor of Advisory-Level E. coli

Densities at Inland Beaches

4.1. Introduction

Current approaches for assessing human illness risk among recreation waters users at

U.S. freshwater beaches rely on multiple U.S. EPA epidemiological studies (Dufour

1984). These studies demonstrated a significant association between increasing fecal indicator density and highly credible gastrointestinal (HCGI) illness incidence among swimmers (p < 0.05). Among the indicators (E. coli, enterococci, and fecal coliforms) evaluated in the Dufour studies, E. coli demonstrated the strongest positive association with highly credible gastrointestinal illness among swimmers (r = 0.804). From these results, the U.S. EPA established recreational water criteria for freshwater, setting the single-day maximum at 235 colony forming units (CFU)/100 mL (U.S. EPA 1986). The rationale for using a fecal indicator approach is that it was presumed that the fecal indicator would be associated with human pathogen densities and consequently human illness. This approach has been confirmed in recent studies demonstrating indicator associations with multiple pathogens such as adenovirus, norovirus, rotavirus and

Camplyobacter (Noble et al. 2003, Jiang et al. 2004, Bates and Phillips 2004). 65

Furthermore, the epidemiological evidence from the Dufour studies (1984) has since been confirmed. The association between GI illness frequency and E. coli density has been further evaluated among freshwater swimmers, and a meta-analysis of epidemiological studies and a recent inland beach human health study confirmed the usefulness of E. coli as a health-relevant fecal indicator (Wade et al. 2003, Marion et al. 2010).

In recent years the current approach for assessing advisory-level fecal indicators has drawn criticism. The majority of the criticism challenges the usefulness of E. coli or enterococci for beach management purposes as opposed to its relevance as an indicator of human illness risk. This criticism focuses on the one significant limitation of the current approach for assessing fecal contamination at U.S. beaches, which is time obtain results.

Since the traditional U.S. EPA method is culture-based (EPA Method 1603, U.S. EPA

2002), relying on the growth of the indicator organism (E. coli or enterococci) does not allow rapid water quality determination. Consequently, it only allows for the reporting of results 18-24 hours post-sampling, i.e. too late to prevent exposure. This is particularly problematic since beach water conditions can rapidly change during the time period (> 24 hours) between sample collection and reporting water quality results to beach users

(Boehm et al. 2002). This problem has been demonstrated to contribute to increased illness risk for beach goers and has resulted in unnecessarily deterring recreational water use when conditions have improved. The unnecessary closing of beaches can adversely impact the local beach economies due to inaccurate beach water quality reporting. At

California coastal beaches, Kim and Grant (2004) demonstrated that advisory postings

66 were done so in error 40% of the time, largely due to the rapidly changing conditions of the waters (Boehm et al. 2002).

In an effort to generate rapid results related to fecal indicators for the protection of beachgoer health, new methodologies for assessing risk have been considered. Molecular methods such as quantitative polymerase chain reaction (qPCR) and immunomagnetic separation/adenosine triphosphate (IMS/ATP) assays have gained attention for their ability to rapidly estimate fecal indicator densities to allow for result reporting within several hours or less after sample collection (Wade et al. 2006, Colford et al. 2007, Lee and Deininger 2004). However, qPCR specific methods require extensive laboratory equipment and a well-trained staff, which is not likely available in close proximity to the vast majority of inland recreational beaches. Although much less extensive than qPCR laboratory needs and costs, IMS/ATP methods have also been demonstrated to be associated with culture-based indicator densities (Lee and Deininger 2004; 2010, Bushon et al. 2009), but also require some non-traditional laboratory requirements (i.e., luminometer and specific antibodies).

An alternative to molecular approaches is the use of predictive water quality models.

Predictive models rely on real-time measurements of physical conditions and usually include real-time measures of water quality to estimate fecal indicator densities.

Successful models have been implemented at several Lake Erie and Lake Michigan beaches (Francy 2009, Nevers and Whitman 2005, Nevers and Whitman 2011 ). These

67 approaches take advantage of known, easily measurable parameters, such as wave height, turbidity, current direction, chlorophyll concentrations, etc., to estimate E. coli and enterococci densities. Nevers and Whitman (2005) achieved an R2 of 63.5% for estimating E. coli densities for five Lake Michigan beaches using several of the previously mentioned parameters. Predictive models have gained attention for their practicality and ease of use; however, the current approach for predictive modeling often requires a significant data collection effort to enable model development and implementation.

To overcome the need for extensive data collection for modeling efforts, deviations in

Carlson‟s Trophic State Index (TSI) may enable the prediction of recreational water conditions associated with increased illness risk. The TSI has been used by lake managers for the management and study of inland lakes for over 30 years, which has led to a substantial amount of baseline TSI data for many of the Nation‟s and Ohio‟s inland lakes (Davic et al. 1996) since the index was first published by Carlson (1977). The primary intent of Carlson‟s TSI was to estimate a doubling in phosphorus, but the index may be useful in other applications. Phosphorus, serving as the limiting nutrient in most inland waters, is associated with algal growth and decreased water transparency (Carlson

1977). Additionally, the association between phosphorus and fecal sources (human and agricultural) has long been understood, as fecal pollution is associated with nutrient loading and increased ecological productivity (Hutchinson 1957). As early as 1970 it was evident in Edmondson‟s landmark ecological study that sewage diversions into Lake

Washington were responsible for the rapid increase in algal biomass, particularly through the elevation of phosphorus levels (Edmondson 1970). Despite the known association between fecal pollution and phosphorus, the use of phosphorus and TSI for predictive modeling for elevated E. coli densities in recreational waters has previously not been vetted in the scientific literature.

The metrics used to calculate Carlson‟s TSI are surface phosphorus, chlorophyll A and transparency. All three measurements are simple, rapid and affordable compared to molecular approaches for estimating fecal indicator densities. The cost, validity and reliability associated with these measurements have made Carlson‟s TSI a standard method for evaluating lake water quality. The three TSI parameters (total phosphorus, chlorophyll A, and secchi depth) are related and by using any single TSI parameter, the

TSI for a given lake can be estimated (Carlson 1977). Specific calculations of TSI values from actual field measurements can be easily accomplished using three simple equations provided by Carlson (1977). The primary goal of this study was to evaluate the effectiveness of rapidly measured TSI as an indicator of elevated E. coli levels at inland

Ohio beaches. The ability to use TSI as an effective rapid tool for predicting E. coli levels could provide a substantial cost-savings for beach managers desiring results in under one-hour when the use of molecular approaches is not practical.

4.2. Materials and Methods

4.2.1. Sampling Sites

A total of 182 samples were collected from seven inland Ohio beaches starting May 28 and ending August 30, 2009. The lakes represent various sizes and trophic states. Most of the lakes are characterized by Ohio EPA as eutrophic (Davic et al. 1996) and most lakes are at least 4 km2 in size (Table 4.1). The study beaches were selected due to their various sizes, trophic states, proximity to our study laboratory in Columbus, Ohio (Figure

3.1), and their status as state-operated swimming beaches, which enabled unrestricted access for sampling purposes.

Table 4.1. Beach sampling locations and general reservoir characteristics.

Size Trophic Status

Reservoir Latitude/Longitude (Km2) (Davic et al. 1996)

Alum Creek 40.1914°N; -82.9702°W 13.7 Mesotrophic

Buck Creek 39.9501°N; -83.7351°W 8.57 Eutrophic

Deer Creek 39.6194°N; -83.2287°W 5.17 Eutrophic

Delaware 40.3712°N; -83.0591°W 5.26 Eutrophic

East Fork 39.0196°N; -84.1342°W 8.74 Eutrophic

Lake Logan 39.5417°N; -82.4710°W 1.62 Eutrophic

Madison 39.8698°N; -83.3744°W 0.43 Hypereutrophic

Figure 4.1. Locations of the beaches for the seven lakes used in the study (Ohio, United

States).

4.2.2. Recreational Water Samples

A total of 26 samples were taken from each of the seven beaches. Samples were collected twice weekly from each beach in accordance with the Ohio Department of

Health standard methods (ODH 2009) during the Ohio swimming season. In brief, samples were collected by sweeping a 500 mL Nalgene© bottle or 500 mL Whirl-Pak© bag one foot below the surface of the water. All samples were in areas used for 71 swimming, in water approximately three feet in depth, in the approximate center of each beach. The single daily water samples were used to perform all the daily laboratory analyses. In tandem with the water sample collection, field observations regarding weather and real-time water quality (e.g. temperature, dissolved oxygen, pH, etc.) conditions were recorded.

4.2.3. Water Quality

E. coli density was quantified on modified mTEC agar following membrane filtration as presercibed by EPA Method 1603 (U.S. EPA, 2002). Beach water samples were filtered through a 0.45-µM pore size, 47-mm diameter nylon filter membrane (Millipore,

Bedford, Massachusetts). Filters were immediately placed on to modified mTEC agar and incubated for 2-hr at 35OC, and then 44.5OC for an additional 18 to 22-hr. Three water volumes (100mL, 50mL and 20mL) were filtered for all samples, and a negative control plate using autoclaved deionised water was used for each sample batch. After incubation, colonies with magenta coloration were counted by visual observation. These colony counts were converted to CFU/100 mL by dividing by the filtration volume.

For calculating Carlson’s TSI, total phosphorus, chlorophyll A and secchi depth were recorded for each beach water quality sample. All laboratory analyses were performed using sample water within six hours of sample collection. Total phosphorus was quantified using the acid persulfate digestion method with a Hach DR2800 spectrophotometer using the EPA approved Hach Method 8190 (U.S. EPA 1983).

Chlorophyll A was quantified in vivo using raw beach water samples without filtration or acetone extraction. Chlorophyll A was quantified using an Aquaflour™ flourometer

(Turner Designs, Sunnyvale, California). In vivo chlorophyll A (excitation at 460 ± 20 nm, emission > 665 nm) was standardized (R2 = 99.9%) with liquid primary chlorophyll

A standards (catalog number 10-850, Turner Designs). Secchi depth was recorded at the beach in conjunction with each sampling period by lowering a secchi disk and rope in water and recording the minimum distance in which the secchi disk was not visible to the observer (Wetzel and Likens 2000).

Additional water quality parameters were measured, including temperature, pH, specific conductivity and dissolved oxygen, which were recorded in real-time using aYSI 600XL multiprobe data sonde (Yellow Spring Instruments, Yellow Springs, Ohio). Using the same sample water, turbidity and phycocyanin (blue-green algae pigment) were also quantified. Turbidity was quantified using a Hach 2100P portable turbidimeter. In vivo phycocyanin (excitation at 595 nm, emission at 670 nm) was measured using the

Aquaflor™ flourometer, and was standardized (R2 = 99.9%) using C-Phycocyanin standard from Spirulina sp. (Sigma-Aldrich® catalog number P6161, St. Louis,

Missouri).

4.2.4. Calculating Daily Trophic State Index Values

TSI values were recorded each day using a trimetric mean based upon TSI estimates for total phosphorus, chlorophyll A and secchi depth. Using the formulas provided by

Carlson (1977), the phosphorus-derived estimate of TSI = (14.42 ln Total phosphorus

( g/L) + 4.15 ). The chlorophyll A-derived TSI estimate = (9.81 ln Chlorophyll a ( g/L)

+ 30.6). The secchi disk-derived TSI estimate = (60 - 14.41 ln Secchi disk (meters)). We obtained the trimetric mean ((TSITP + TSIChlA + TSISD) / (3)), and used that value as the trimetric mean TSI value for each lake. This approach is common in limnological applications estimating TSI (Wetzel and Likins 2000). One important difference between our chlorophyll A estimation and the approach generally used is that our measurement is based upon in vivo chlorophyll A as opposed to measurements that utilize filtration and acetone extraction.

4.2.5. Data Analysis

Data were initially explored using scatterplots and regression analysis for continuous data to evaluate any associations between E. coli density and other water quality parameters at the individual lakes. Also, given that the advisory threshold for E. coli was set at 235 CFU/100 mL by U.S. EPA (1986), a logistic regression approach using 0/1 coding was also employed for identifying predictors associated with these advisory-level

E. coli densities. The data were not collected at random; therefore, for statistical purposes the data were treated as clustered, with clustering occurring by lake.

4.3. Results and Discussion

4.3.1. Characterization of Beach Water Quality

A total of 18 (9.9%) water samples were found to have E. coli densities exceeding U.S. recreational water criteria (235 CFU/100 mL). These results demonstrate substantially better water quality than Lake Erie coastal beaches in Ohio during the same time period.

Coastal Ohio beaches observed exceedances in 19% of samples during the 2009 swimming season (Dorfman and Rosselot 2010). Among the 18 samples exhibiting action-level E. coli densities, two were reported during heavy rainfall conditions at the beach at the time of sample collection. In total, heavy rainfall occurred at the time of sampling four times (2.1%) during the entire study period. For the purposes of this study, all four samples that were collected at the beach during these rain events were removed from the analysis.

With respect to the overall trophic condition of the seven lakes, the trophic classifications assigned by Davic et al. (1996) in Table 4.1 are still applicable to the trophic condition determined in 2009 (Table 4.2), where Alum Creek is mildly eutrophic and Madison

Lake is firmly hypereutrophic. The remaining lakes are at least firmly eutrophic to hypereutrophic based upon their mean TSIDaily values using the trimetric mean approach.

It is apparent that the chlorophyll A estimates for TSI are lower than the estimates generated by total phosphorus and secchi depth. The likely cause is that chlorophyll A for the purposes of computing Carlson‟s TSI is generally reported upon following the acetone extraction method, which quantifies chlorophyll A lysed from phytoplankton captured during filtration. The in vivo chlorophyll A method does not lyse cells and likely underestimated total chlorophyll A and subsequently the TSIchlorophyll A.

An analysis of the range and standard deviations presented in Table 4.2 demonstrates some variability in the TSI estimates using the three parameters. Fluctuation in TSI has been observed temporally by Knowlton and Jones (2006) in Missouri Lakes and has been viewed as being potentially problematic for lake managers attempting to monitor trends and adequately classify lake condition for long-term management purposes. For the purposes of our studies, the temporal changes in TSI may have some utility for predicting health-relevant water quality events for recreational water users. The association between deviations in lake-specific trophic state index values appears to have applicability for predicting E. coli levels exceeding 235 CFU/100 mL. Figure 4.2 illustrates that for samples with elevated E. coli densities, TSI values are elevated with respect to their average lake-specific TSI values. This difference in deviation in TSI values between advisory-level and “safe” E. coli densities was confirmed with a Mann-

Whitney test (p = 0.0017),where the median for deviation among the advisory level conditions (3.069) was significantly higher than median for deviation in TSI among the non-advisory conditions (-0.135). Only one sample was associated with elevated E. coli densities when the deviation from lake-specific seasonal TSI value was negative. This sample occurred at the hypereutrophic Madison Lake beach. For this sample the

TSIChlorophyll A value was 56.4, which is nearly two standard deviations from the

TSIChlorophyll A season average for this beach. The overall TSI average for this date based upon all three TSI parameters was 70.0, which is among the lowest daily average TSI values recorded for this beach for the 2009 season. It is plausible that with dilution from

76 recent heavy rains that occurred around that date, algal densities were diluted, while phosphorus levels remained constant from runoff inputs.

Table 4.2. A summary of beach water quality across seven Ohio beaches including with respect to trophic state index values, and the number of days in which E. coli exceeded health-relevant densities.

No. Elevated Mean TSI TP ± Mean TSI ChlA

a b Beach Samples E. coli Days SD ± SD

Alum Creek 26 7 51.8 ± 10.5 43.8 ± 8.6

Buck Creek 26 3 63.3 ± 3.8 60.6 ± 2.8

Deer Creek 26 1 68.8 ± 4.0 55.5 ± 4.5

Delaware 25 2 69.0 ± 4.4 61.6 ± 4.3

East Fork 24 1 69.7 ± 5.1 57.1 ± 4.7

Lake Logan 25 0 61.7 ± 4.3 54.7 ± 3.7

Madison 26 2 81.0 ± 2.8 69.6 ± 7.2

continued

Table 4.2 continued

Mean Mean

c TSISecchi ± TSITrimetric ± TSITrimetric

Beach SD SD Range

Alum Creek 58.1 ± 3.9 51.2 ± 5.3 38.1 - 59.9

Buck Creek 65.5 ± 1.8 63.1 ± 2.0 60.0 - 67.8

Deer Creek 69.1 ± 3.3 64.5 ± 3.1 57.8 - 69.0

Delaware 69.7 ± 2.9 66.8 ± 3.0 61.3 - 71.8

East Fork 67 ± 4.0 64.6 ± 3.4 60.0 - 72.9

Lake Logan 65.1 ± 2.3 60.5 ± 2.6 54.7 - 64.1

Madison 75.6 ± 2.3 75.4 ± 3.0 69.7 - 81.1

aThe number of days in which the E. coli density was determined to exceed 235

CFU/100 mL

bStandard deviation

cThe daily trimetric mean for trophic state index (TSI), determined by averaging the

TSI values as estimated using total phosphorus, chlorophyll A, and secchi depth.

Deviationin TSI

-20 -40 Below Criteria Above Criteria

Figure 4.2. Box-and-whisker diagram comparing the amount of deviation in daily

trophic state index values away from beach-specific average trophic state index values for

beach water samples determined to be below and above the recreational water quality

criteria (235 CFU E. coli/100 mL)

4.3.2. Logistic Regression.

Univariable logistic regression models were constructed to evaluate the predictive

capability of water quality parameters for estimating E. coli densities exceeding

recreational water quality criteria. Multivariable models are not reported due to the

limited number of events, as only 16 events occurred among the 178 samples assessed in

79 this study. Although the use of multivariable models generally enhances model performance, Peduzzi et al. (1996) recommends only one variable per ten events in a study. This recommendation was followed, since the use of an additional variable with less than 20 events could introduce bias and reduce the validity of the model.

Furthermore, acceptable univariable models could be constructed, decreasing the need for stressing a model with two or more covariates.

Table 4.3 reports the results of nine univariable models considered in this study. Among the nine variables, the two parameters pertaining to total phosphorus generated statistically significant odds ratios (p < 0.05). Additionally, the TSITrimetric model achieved statistical significance. It is noteworthy to reaffirm that TSITotal Phosphorus is included in the calculation of TSITrimetric. Lastly, phycocyanin was determined to also be statistically associated with the outcome of interest. The statistical findings are supported by biologically plausible explanations. Firstly, phosphorus has the potential to serve as a fecal indicator, particularly because elevated phosphorus levels are typically observed in agricultural settings where E. coli levels are also elevated (Byers et al. 2005).

Additionally, phosphorus levels have historically been associated with human sewage inputs (Edmondson 1970, Jarvie et al. 2006, Neal et al. 2005). Thirdly, phosphorus is known to be an important nutrient for microbial survival. ATP, DNA, phospholipids and other biomolecules in bacterial cells require phosphorus, and increasing concentrations of microbially available phosphorus phosphorus have been demonstrated to improve bacteria survival even in drinking water (Juhna et al. 2007).

Table 4.3. A summary of univariable logistic regression models designed for predicting the odds of beach samples exceeding the E. coli recreational water quality standard of

235 CFU/100 mL.

Constant Odds Ratio, (95%

Univariable Model Term C.I.)

Deviation in Total Phosphorus (ug/L) 0.025 -2.40 1.025 (1.019 -1.032)

Deviation in Chlorophyll A (ug/L) 0.005 -2.28 1.005 (0.980-1.031)

Deviation in Secchi Depth (cm) 0.025 -2.37 1.025 (0.971-1.082)

Deviation in Phycocyanin (ug/L) 0.011 -2.33 1.011 (1.003-1.020)

Deviation in Turbidity (NTU) 0.004 -2.30 1.005 (0.990-1.020)

Deviation in TSISecchi Depth 0.137 -2.37 1.147 (0.878-1.499)

Deviation TSITotal Phosphorus 0.105 -2.44 1.110 (1.080-1.141)

Deviation TSIChlorophyll A 0.105 -2.44 1.114 (0.868-1.429)

Deviation in TSITrimetric 0.172 -2.58 1.188 (1.073-1.314)

continued

Table 4.3 continued

Likelihood

Ratio Test a

Univariable Model S.E. p

Deviation in Total Phosphorus (ug/L) 0.003 0.0001

Deviation in Chlorophyll A (ug/L) 0.010 0.6364

Deviation in Secchi Depth (cm) 0.023 0.3101

Deviation in Phycocyanin (ug/L) 0.003 0.0177

Deviation in Turbidity (NTU) 0.006 0.4612

Deviation in TSISecchi Depth 0.125 0.2569

Deviation TSITotal Phosphorus 0.012 0.0001

Deviation TSIChlorophyll A 0.113 0.3308

Deviation in TSITrimetric 0.049 0.0059

a Models having p < 0.05 shown in bold.

The deviation in TSITrimetric also demonstrated a significant association; whereby, for each one unit increase in deviation from lake-specific TSITrimetric seasonal means, we observe a

19% increased odds of observing E. coli exceeding recreational water quality criteria. An advantage of this model is that it incorporates TSITotal Phosphorus into the calculation, but also considers TSISecchi Depth and TSIChlorophyll A. Although deviation in the later two TSI measurements did not achieve statistical significance in their respective univariable models, when incorporated in the calculation of the TSITrimetric value, they appear to enhance model performance. Transparency, measured as secchi depth, is related to UV penetration into waters, and more transparent water can lead to greater solar intensity, which can disinfect waters (Boyle et al. 2008, Whitman et al. 2004). A disadvantage of this approach is that it requires measuring all three parameters. Additionally, during wet weather phenomena, transparency can decline due to suspended particles, which are known to carry fecal indicators and pathogens. (Gaffield et al. 2003, Lipp et al. 2001).

Overall, the model supports the ecological theory that the samples associated with the least transparency, and most phosphorus would be at the greatest risk for experiencing elevated E. coli densities.

4.3.3. Assessing Model Performance.

The four significant univariable models were evaluated with respect to discrimination and calibration in accordance with Hosmer and Lemeshow (2000). Additionally, the assumption that the model terms were linear in the logit was assessed using the fractional polynomial method (Hosmer and Lemeshow 2000). Table 4.4 demonstrates that all the models perform no differently when using linear terms versus non-linear transformations, which is demonstrated by the large p-values reported. The area under the ROC curves

83 show comparable discrimination among all the terms except deviation in phycocyanin.

The phycocyanin model has the lowest area under the ROC curve (0.6398) and provides the least desirable discrimination.

Table 4.4. A summary of performance and assumption measures for the univariable logistic regression models designed for rapidly predicting the odds of beach samples exceeding the E. coli recreational water quality standard of 235 CFU/100 mL.

Test for

Goodness- Linearity

Area Under of-Fit Test of the logit

Univariable Model ROC Curve p p

Deviation in Total Phosphorus (ug/L) 0.7050 0.1269 0.618

Deviation in Phycocyanin (ug/L) 0.6398 0.5050 0.545

Deviation TSITotal Phosphorus 0.6875 0.3588 0.532

Deviation in TSITrimetric 0.7203 0.3241 1.000

When evaluating all the terms, deviation in TSITrimetric is associated with the best discrimination (72%) and achieves acceptable fit (p = 0.3241) using the Hosmer-

Lemeshow goodness-of-fit test. The most comparable model with respect to discrimination is the model using deviation in total phosphorus (Area under ROC =

70.5%); however, the overall fit of this model is marginal (p = 0.1269).

4.4. Management Implications.

A substantial problem for beach managers is obtaining timely fecal indicator results to enable communication of beach water quality results to current and prospective beachgoers (Whitman and Nevers 2004). This issue deserves much attention, particularly because of the size of the population at risk and error rate associated with reporting water quality advisories related to single-day maximum exceedances. The inadequate monitoring and notification system has resulted in significant human health and economic consequences at coastal beaches (Rabinovici et al. 2004, Given et al. 2006), and is likely a significant, undocumented problem at inland U.S. beaches. To date, empirical predictive models for estimating fecal indicator densities have been explored in a number of studies; however, they have all relied on building reliable databases, and developing, refining and validating models (Nevers and Whitman 2010). In this study, we did a practical data collection effort and used data that could be obtained rapidly to develop useful predictive models.

The trophic state index approach used in this study may offer an opportunity for developing predictive models using existing and reliable databases, thereby significantly reducing the costs and time of model development. In Ohio, historical TSI data are available for several hundred Ohio lakes greater than 0.02 km2 in size (Davic et al. 1996).

For all U.S. states, Section 314 of the Clean Water Act (CWA) requires state reporting of trophic state for public lakes, which suggests an abundance of available data among the states for their respective inland lakes. By using historical TSI values for these lakes when immediate E. coli data is not available, the deviation from mean TSI approach could be employed to serve as a useful approach for estimating adverse water quality for recreation. This is the first study assessing the association between elevated E. coli densities and TSI and total phosphorus. It is presumed that during the swimming season, a sudden change in phosphorus concentrations and overall TSI values could point to health-relevant surface runoff events, which may transport water contaminants of fecal origin. Future studies incorporating hydrological/meterological covariates may generate enhanced models. Epidemiological studies evaluating the association between deviation in TSI and human illness are encouraged.

4.5. Acknowledgment

Funding for this research was provided by a grant from the Ohio Water Development

Authority. Additional funding for this study was provided through fellowship support by

Public Health Preparedness for Infectious Diseases (PHPID) program at The Ohio State

University. We are thankful to Mr. Scott Fletcher with the Ohio Department of Natural

Resources for supporting the study and allowing sample collection at the seven state park beaches. We also are grateful to Pei-Yu Chiang for her assistance in collecting a portion of the field and laboratory data.

Chapter 5: In Vivo Phycocyanin Flourometry as a Rapid Screening Tool for

Predicting Elevated Microcystin Concentrations at Inland Beaches

5.1. Introduction

Harmful cyanobacteria are gaining substantial attention among water quality managers and public health officials around the world. This attention is largely explained by our expanding knowledge regarding cyanotoxin toxicity and a growing body of evidence suggesting an increasing global trend in the frequency of harmful algal bloom (HAB) events (Harvell et al. 2000, Peperzak 2003, Edwards et al. 2006). Although not all cyanobacteria produce cyanotoxins, it is suspected that with increasing water temperatures and nutrient levels, the environment will be more favorable for cyanobacteria blooms that contain harmful genera (Ye et al. 2011, Paerl et al. 2011).

With respect to these blooms, there is some evidence suggesting warmer water temperatures will create selective pressures that will enable Microcystis dominance over non-harmful genera in freshwater U.S. lakes (Davis et al. 2009). The concerns regarding harmful cyanobacteria are warranted, as they are known to produce a variety of

88 cyanotoxins that may act as potent hepatotoxins, neurotoxins, cytotoxins, irritants and/or gastrointestinal toxins (Codd et al. 2005, Carmichael 1997).

The association between elevated cyanobacteria levels and adverse health outcomes was observed by the scientific community as early as 1878 in animals (Francis 1878). More recently (1996) microcystin and cylindrospermopsin were implicated in the deaths of 76 humans in Brazil (Carmichael et al. 2001). Beyond these two reports, Chinese epidemiological studies link microcystin exposure to increased liver cancer incidence (Yu

1995). With respect to recreational exposure, the epidemiology is limited to a handful of cohort studies (Pilotto et al. 1997, Stewart et al. 2006a, Backer et al. 2010, Stewart et al.

2006b) and case reports (Turner et al. 1990) that largely rely on cell densities as opposed to toxin concentrations. Despite these limitations, the World Health Organization

(WHO), relying on mouse and pig toxicity (Fawell et al. 1994, Falconer et al. 1994) studies, established provisional guidelines.

The drinking water provisional guideline for microcystin set at 1.0 g/L (Chorus and

Bartram 1999) was used to establish a “moderate” microcystin health risk guideline of 20

g/L (~100,000 cyanobacteria cells/mL) for recreational waters (WHO 2003).

Recreational exposure to waters in the moderate risk range is presumed to be associated with increased potential for users to develop long-term illness as well as acute symptoms.

The “low” health risk range was derived from human epidemiology and mouse model

(Fawell et al. 1994) studies, recommending advisory signage at 20,000 cyanobacteria

89 cells/mL (WHO 2003), which approximates to 4 g/L of microcystin. This low health risk range is presumed to be associated with a low frequency, but health-relevant increased likelihood of short-term adverse health outcomes, such as skin irritations and

GI illness. The posting of advisory signage is recommended at this level and increased surveillance is encouraged by WHO (Chorus and Bartram 1999, WHO 2003).

Chlorophyll A concentrations of 10 g/L and 50 g/L serve as alternative measures for issuing respective low and moderate risk advisories when cyanobacteria dominate (WHO

2003).

Acknowledging the limited information pertaining to human health effects associated with exposures to harmful algal blooms, the U.S. Environmental Protection Agency

(EPA) recently concluded that the information presently available is insufficient for guiding U.S. policies pertaining to cyanobacteria (Hudnell 2010). Among U.S. states,

Nebraska, New York, Florida, and Ohio all have recently issued numerous human contact advisories for freshwater environments (Walker 2008, Burns 2008, Boyer 2008, Ohio

EPA 2011).

Our ability to measure or predict harmful algal blooms for the purpose of evaluating their densities in individual lakes, trends in growth, exposure assessment, epidemiology, or beach management suffers from methodological limitations. The available analytical methods include enzyme-linked immunosorbent assays (ELISA), protein phosphatase type 2 inhibition assays (PP2IA), and high-performance liquid chromatography coupled

90 with ultraviolet light detection (HPLC/UV) or mass spectrometry (HPLC/MS) (Chorus

2005). Other measures of blooms or harmful exposures involve algal identification and counting as prescribed by the WHO (WHO 2003). The analytical methods require laboratory resources not readily available to most beach managers. Therefore, evaluation of HAB health risks requires sending samples out for analysis, which creates practical constraints with respect to cost and time to receive a result. Because none of the needed methods are real-time or rapid, it is practically impossible for health officials to adequately notify recreational water users in a timely manner of potential risks from cyanotoxins (Chorus 2005).

Given the relatively high relative abundance of microcystin compared to other toxins

(Carmichael 1997), a substantial amount of research (on identifying harmful blooms) has focused on microcystin. Methods for the rapid detection of microcystin have been developed and have been deemed effective in quantifying microcystin concentrations.

Methods such as microcystin-specific ELISAs have been developed (Chu et al. 1989), improved upon (McDermott et al. 1995), and are now commonly applied for research and regulation (Ingrid and Bartram 1999, Chorus 2005, Carmichael and An 1999). Newer approaches, such as quantitative polymerase chain reaction (qPCR) amplification of toxin-producing synthetase genes (Ouellette and Wilhelm 2003 Rinta-Kanto et al. 2005) and ELISA-based sensor applications for estimating microcystin concentrations have gained attention (Ding and Mutharasan 2011). Collectively, these methods provide substantial improvement in estimating microcystin concentrations; however, practical

91 limitations pertaining to laboratory resources, sample transport, timeliness of reporting, and cost are a concern.

An alternative non-molecular approach involves methods for identifying harmful blooms that focus on quantification of phycocyanin pigment via flourometry (Lee et al. 1994).

The positive association between cyanobacteria, including specifically Microcystis cell density, and phycocyanin pigment has been well documented (Lee et al. 1994, Ahn et al.

2007, Ahn et al. 2002, Izydorczyk et al. 2005), and has also been shown to outperform the Chlorophyll A quantification methods recommended by WHO to predict harmful blooms (Chorus 1999, WHO 2003). Recently, it was demonstrated in inland northern

U.S. waters that flourometric procedures focusing on phycocyanins not only were associated with cell densities, but also microcystin concentrations (Makarewicz et al.

2009, Murby 2009, Lehman 2007). This association with the toxin was much stronger for phycocyanin measurements than chlorophyll A, the WHO recommended method

(Chorus 1999, WHO 2003).

To address the public health need for a rapid and practical means for predicting microcystin, we evaluated the effectiveness of rapidly measured in vivo phycocyanin and in vivo chlorophyll A as indicators of elevated microcystin concentrations at seven inland

Ohio beaches.

5.2. Materials and Methods

5.2.1. Sampling Sites

We evaluated the water quality at beaches of seven Ohio inland lakes during the 2009 swimming season (May 28 to August 30). A total of 26 samples were collected from each beach. Five beaches were located on reservoirs operated by the U.S. Army Corps of

Engineers primarily for flood control with a secondary recreational use designation.

Among these five Corps lakes, their beaches attract over 575,000 swimmers annually

(U.S. Army Corps of Engineers, 2011). The other two lakes are state-owned recreation reservoirs (swimmer usage data not available). The trophic statuses of the lakes range from mesotrophic to hypereutrophic (Davic et al. 1996), with most lakes being classified as eutrophic (see Table 5.1). The Corps lakes are substantially larger than the state- owned lakes.

5.2.2. Sampling Procedure

The procedure employed is described in Marion et al. (2010). In brief samples were collected bi-weekly from the public beaches at the seven study lakes in accordance with

Ohio Department of Health standard methods (ODH 2009). All analyses that were performed on the beach water were generated from these single bi-weekly samples. Field observations of weather and real-time measurements of water quality were recorded in conjunction with sample collection. From each sample, approximately 12 mL of water was stored at -80°C in 15 mL plastic centrifuge tubes for later analysis of microcystins.

Table 5.1. Beach sampling locations and general reservoir characteristics.

Trophic Status

Size (Davic et al. Corps Annual

Reservoir Latitude/Longitude (Acres) 1996) Lake Swimmers1

Alum Creek 40.1914°N; -82.9702°W 3,387 Mesotrophic Yes 420,528

Buck Creek 39.9501°N; -83.7351°W 2,120 Eutrophic Yes 51,727

Deer Creek 39.6194°N; -83.2287°W 1,277 Eutrophic Yes 19,366

Delaware 40.3712°N; -83.0591°W 1,300 Eutrophic Yes 90,784

East Fork 39.0196°N; -84.1342°W 2,160 Eutrophic Yes 83,077

Lake Logan 39.5417°N; -82.4710°W 400 Eutrophic No NA

Madison 39.8698°N; -83.3744°W 106 Hypereutrophic No NA

1U.S. Army Corps of Engineers (2011)

5.2.3. Water Quality and Flourometry

Temperature, pH, specific conductivity and dissolved oxygen were recorded at the time of sampling using a multisensor probe (YSI 600XL multiprobe data sonde (Yellow

Springs, Ohio). Total phosphorus was quantified with the acid persulfate digestion method using Hach Method 8190 (Hach Company 2007, US EPA 1983) and a Hach 94

DR2800 spectrophotometer (Hach®, Loveland, Colorado). Turbidity was quantified using the Hach 2100P portable turbidimeter (U.S. EPA 1983). Secchi depth

(transparency) was measured using a submerged secchi disk (Wetzel 2000). Due to thunderstorm activity, secchi depth was not measured for one Delaware Lake sample.

Chlorophyll A and phycocyanin were both quantified in vivo, using the intact cells without filtration or extraction. Both chlorophyll A and phycocyanin were quantified using a two-channel handheld Aquaflour™ flourometer (Turner Designs®, Sunnyvale,

California). Chlorophyll A (excitation at 460 ± 20 nm, emission > 665 nm) was standardized (R2 = 99.9%) with liquid primary chlorophyll A standards (catalog number

10-850, Turner Designs®). Phycocyanin (excitation at 595 nm, emission at 670 nm) was standardized (R2 = 99.9%) using C-Phycocyanin standard from Spirulina sp. (catalog number P6161, Sigma-Aldrich®, St. Louis, Missouri). The mean trophic state index

(TSI) was calculated by averaging the TSI estimate from total phosphorus, secchi depth and in vivo chlrophyll A. The mean TSI calculation slightly differs from Carlson (1977) in that it used in vivo chlorophyll A as opposed to extracted chlorophyll A.

5.2.4. Microcystin Detection and Quantification

The concentration of microcystin was measured using the EPA-validated

Microcystins/Nodularins (ADDA) ES, ELISA kit in 96-well format (catalog number

PN520011ES, Abraxis ®,Warminster, Pennsylvania) (James et al. 2010). The samples were thawed and then underwent a freeze/thaw procedure two times at -20°C to effectively rupture cells and release the toxins (Lehman 2007). After final thawing,

95 samples were used in the ELISA per manufacturer instructions. The assay detects and quantifies numerous variants of currently known microcystins with polyclonal antibodies providing a cumulative concentration for total microcystins (James et al. 2010). Per manufacturer instructions, the optical density was recorded at 450 nm using a plate photometer (Dynex Technologies MRX TC Revelation) after stopping the color reaction at the specified time. For each multi-well plate, a total of six standards were used in duplicate for developing the plate-specific standard curves (mean R2 = 98.0%, R2 Range

= 96.2% - 99.1%). Additionally, positive and negative controls were used, and all samples were evaluated in duplicate. There was good agreement between the two measured optical densities for each sample (Pearson correlation = 0.966, p < 0.001).

Samples exceeding 5 g/L microcystin exceeded the assay‟s specified range of detection.

Samples below 0.15 g/L were below the specified range of quantification. Samples below 0.10 g/L were below the specified limit of detection.

5.2.5. Data Analysis

Data were initially explored using scatterplots and regression analysis for continuous data. Logistic regression was used with 0/1 coding for the event variable, where a positive event was defined as any sample exceeding 4 g/L microcystin, corresponding to the WHO low health risk range of 20,000 cyanobacteria cells/mL.

Modeling was performed taking into account clustering due to the repeated measures from each lake. Numerous models were constructed using investigator-controlled

96 backward selection procedures. Using this backward selection procedure, covariates with

Wald test p-values < 0.15 were considered for the full model (Bendel and Affifi 1977).

After including all covariates into a full model, the least significant terms were individually removed, one at a time, until the most parsimonious significant model could be obtained. For the multivariable models, continuous covariates were evaluated for linearity in the logit via the fractional polynomial method (Hosmer and Lemeshow,

2000). For covariates observed to be non-linear in the logit, appropriate transformations were used.

5.3. Results and Discussion

5.3.1. Characterization of Beach Water Quality

Water quality characteristics at the seven in-land lakes are described in Table 5.2. The water quality observed in our study was consistent with previous observations by OEPA

(1996). The trophic state index places all the lakes except Madison Lake in the eutrophic classification, which was observed to be hypereutrophic. Alum Creek, characterized as mesotrophic by many measures in 1996 also was characterized in some reports as being mildly eutrophic, a condition which we observed with a mean TSI of 51.3.

Table 5.2. Beach water quality characteristics across seven inland Ohio reservoirs over

26 sampling days during summer 2009.

Lake Abbreviation

n = 26 n = 26 n = 26

Parameter (mean ± S.E.) AC1 BC2 DC3

Temperature 25.4 ± 0.38 25.4 ± 0.41 26.0 ± 0.33

Specific Conductivity (mS) 432 ± 1.86 435 ± 10.5 459 ± 7.39

Dissolved Oxygen (mg/L) 9.72 ± 0.66 11.6 ± 0.78 11.9 ± 0.67

pH 8.48 ± 0.33 8.66 ± 0.02 8.85 ± 0.03

Secchi Depth (cm) 118 ± 6.37 68.7 ± 1.70 54.5 ± 2.56

Total Phosphorus (mg/L) 35.9 ± 6.62 64.6 ± 4.26 90.8 ± 5.20

Chlorophyll A (mg/L) 4.67 ± 0.40 22.2 ± 1.31 13.5 ± 1.20

Turbidity (NTU) 6.15 ± 0.51 10.0 ± 0.75 20.5 ± 1.56

Mean Trophic State Index 51.3 ± 1.03 63.3 ± 0.38 64.3 ± 0.65

Phycocyanin ( g/L) 6.04 ± 1.18 73.7 ± 3.41 160 ± 13.1

No. Samples > 4 g/L

Microcystin (% of samples) 6 (23.1) 8 (30.8) 18 (69.2)

1Alum Creek Lake, 2Buck Creek Lake, 3Deer Creek Lake

continued 98

Table 5.2 continued

n = 26 n = 26 n = 26

Parameter (mean ± S.E.) DEL4 EF5 LL6

Temperature 25.8 ± 0.38 26.7 ± 0.36 27.0 ± 0.41

Specific Conductivity (mS) 474 ± 3.96 276 ± 2.29 209 ± 5.31

Dissolved Oxygen (mg/L) 12.1 ± 1.06 10.2 ± 0.51 10.4 ± 0.51

pH 8.67 ± 0.04 9.10 ± 0.04 8.82 ± 0.05

Secchi Depth (cm) 53.1 ± 2.22 62.1 ± 3.09 72.0 ± 2.43

Total Phosphorus (mg/L) 95.4 ± 5.64 107 ± 11.2 55.3 ± 3.40

Chlorophyll A (mg/L) 25.7 ± 2.20 16.6 ± 1.78 12.3 ± 0.78

Turbidity (NTU) 18.3 ± 1.70 22.6 ± 4.70 9.15 ± 0.57

Mean Trophic State Index 66.8 ± 0.61 64.9 ± 0.73 60.3 ± 0.54

Phycocyanin ( g/L) 59.9 ± 8.15 57.1 ± 4.08 28.1 ± 1.35

No. Samples > 4 g/L

Microcystin (% of samples) 4 (15.4) 0 (0) 3 (11.5)

4Delaware Lake, 5East Fork Lake, 6Lake Logan

continued

Table 5.2 continued

n = 26

Parameter (mean ± S.E.) MAD7

Temperature 25.9 ± 0.31

Specific Conductivity (mS) 486 ± 10.9

Dissolved Oxygen (mg/L) 13.0 ± 1.21

pH 8.73 ± 0.05

Secchi Depth (cm) 34.3 ± 1.06

Total Phosphorus (mg/L) 210 ± 8.20

Chlorophyll A (mg/L) 67.7 ± 9.10

Turbidity (NTU) 40.7 ± 4.18

Mean Trophic State Index 75.4 ± 0.58

Phycocyanin ( g/L) 130 ± 15.8

No. Samples > 4 g/L

Microcystin (% of samples) 9 (34.6)

7Madison Lake

100

i 70 t % Exceeding = 0.783 + 0.3967 (PC [Median])

s Deer Creek y

c R-Sq. = 66.9%

o 60

i M

4 40

n Buck Creek Madison i

d 30

e e

c Alum Creek x

E 20

s Delaware

e l

p 10

m Lake Logan

a S

0 East Fork %

0 20 40 60 80 100 120 140 160 Median Phycocyanin Concentration ( g/L)

Figure 5.1. The association of median in vivo phycocyanin concentrations and the percent of samples exceeding the low risk threshold for microcsytins (4 g/L) in the Ohio lakes during the study period.

The analysis comparing median lake concentrations of phycocyanin and the percent of samples exceeding 4 g/L microcystin (Figure 5.2) demonstrated a positive association (r

= 0.818, Pearson p = 0.025), whereas no association was observed when performing the same analysis between median chlorophyll A by lake and percent of microcystin samples exceeding the low risk threshold (r = 0.158; Pearson p = 0.733). An analysis of the mean concentrations for phycocyanin and the percent of samples exceeding 4 g/L microcystin

101 demonstrates a similar association (r = 0.778, Pearson p = 0.039). No other associations

(based on Pearson correlation p-values < 0.05) with microcystin levels and other water quality metrics such as temperature, pH, turbidity and specific conductivity were observed.

5.3.2. Univariable Logistic Regression. Univariable models were considered before multivariable models. Among 10 water quality indicators (see Table 5.3), only 3 demonstrated some association (p < 0.15) with elevated microcystin levels (specific conductivity, dissolved oxygen, and phycocyanin as determined using the Wald statistic).

An evaluation of these crude models demonstrated that phycocyanin is superior as an individual indicator.

Using the area under the receiver-operating-characteristic (ROC) curve to assess discrimination of each significant crude model, phycocyanin provides acceptable discrimination (73.3%), whereas dissolved oxygen exhibits poor discrimination

(58.81%). Specific conductivity performs similarly to phycocyanin with respect to discrimination (Area under ROC curve = 69.4%). An assessment of linearity via the fractional polynomial method (Hosmer and Lemeshow 2000) demonstrated that phycocyanin does not appear to be related to elevated microcystin as a linear function in the logit as demonstrated by the deviance associated with this single term that was significantly higher than the deviance by the best second order polynomial (p < 0.001).

102

Given this observation, a second order polynomial transformation (see table 5.4) was used.

Table 5.3. Results of 10 univariable logistic regression models for lake-clustered data assessing the association between individual water quality parameters and elevated microcystin concentrations (4 g/L) across seven inland Ohio reservoirs over 26 sampling days during summer 2009.

Univariable Models OR (95% CI) SE OR Wald (p)

Temperature 0.94 (0.75 - 1.15) 0.080 0.470

Specific Conductivity ( S) 1.01 (1.00 - 1.02) 0.003 0.040

Dissolved Oxygen (mg/L) 1.08 (1.00 - 1.16) 0.032 0.047

pH 0.60 (0.03 - 12.8) 0.746 0.694

Secchi Depth (cm) 0.99 (0.97 - 1.02) 0.010 0.585

Total Phosphorus ( g/L) 1.00 (1.00 - 1.00) 0.001 0.943

in vivo Chlorophyll A ( g/L) 0.99 (0.99 - 1.01) 0.005 0.827

Turbidity (NTU) 1.00 (0.98 - 1.03) 0.009 0.780

Trophic State Index (mean) 1.02 (0.98 - 1.05) 0.014 0.302

in vivo Phycocyanin ( g/L) 1.01 (1.00 - 1.03) 0.005 0.053

103

Other variables, such as chlorophyll A, total phosphorus, turbidity, and temperature did not have any apparent initial association with elevated microcystin concentrations (p >

0.15). The lack of association with chlorophyll A (the WHO standard indicator) is consistent with the Watzin et al. study results (2006). Additionally, our observation of a stronger association between microcystin and phycocyanin than with chlorophyll A is also consistent with the study results reported by Rinta-Kanto et al. (2009)

5.3.3. Multivariable Logistic Regression

A multivariable model initially containing all three of the significant covariates (i.e. DO,

SC, and PC, p<0.15) from the univariable models was constructed. The dissolved oxygen term was associated with the highest p-value (p = 0.131) and therefore it was removed from the model resulting in a two-variable model that included SC (p = 0.108) and PC (p = 0.100). PC was determined to be a non-linear in the logit and it was decided to include its squared term in the model. Using the model terms (PC and PC2), specific conductivity became less significant (p = 0.177) and was therefore removed.

Individually, additional covariates were added to form the most significant model as determined by the likelihood ratio test, while ensuring acceptable discrimination and model fit. The covariates turbidity (p = 0.017), secchi depth (p = 0.012), and mean TSI

(p = 0.004) enhanced model performance when placed in the model individually. The model containing secchi depth was determined to provide the most discrimination and had the best fit according to the Hosmer-Lemeshow Goodness-of-Fit Test.

104

Overall, the logistic model containing PC ( g/L), PC2 ( g/L)2, and secchi depth (cm) provided the best model for estimating the probability of elevated microcystin levels (via the Likelihood Ratio Test (p = 0.0298)) (see Table 5.4). Interaction between secchi depth and PC was found to be non-significant (p =0.511). An examination of previous literature suggests that increasing secchi depth (more transparent water) is generally associated with a lower likelihood for harmful microcystin concentrations (Graham et al.

2004); however, our multivariable model failed to confirm this. Although our data supports the interpretation that an increasing secchi depth is negatively correlated with phycocyanin concentrations (r = -0.539, p < 0.001), the more transparent water is associated with a greater probability for advisory level microcystin concentrations when there are increases in phycocyanin concentrations. Given that the -value for phycocyanin increases 67% from 0.03 to 0.05 with the addition of the secchi depth term, we conclude that secchi depth is a confounder. This is biologically plausible as background levels of phycocyanin at more turbid, low secchi depth lakes are likely to be higher, suggesting greater competition among cyanobacteria. However, in more transparent lakes, where background phycocyanin

105

Table 5.4. Final multivariable logistic regression model for predicting the odds of beach samples exceeding microcystin concentrations of 4 g/L at Midwestern U.S. reservoirs.

Covariate SE b Wald (p) OR (95% CI)

in vivo Phycocyanin ( g/L) 0.0507 0.0108 0.003 1.05 (1.02 - 1.08) in vivo Phycocyanin2 ( g/L) -0.0001 0.0000 0.008 1.00 (1.00 - 1.00)

Secchi Depth (cm) 0.0300 0.0051 0.001 1.03 (1.02 - 1.04)

Constant Term -5.9812

LR Test (p = 0.0298), Hosmer- Lemeshow Goodness of Fit Test (p = 0.32923 )

levels are low, a small increase in phycocyanin is potentially associated with rapidly growing bloom-formers in response to favorable conditions and less competition.

106

1.00

0.75

0.50

Sensitivity/Specificity

0.25 0.00 0.00 0.25 0.50 0.75 1.00 Probability cutoff

Sensitivity Specificity

Figure 5.2. The relationship between model sensitivity and specificity for multivariable

logistic model predicting beach samples with microcystin concentrations exceeding the

low health risk threshold (4 g/L).

107

Table 5.5. A performance analysis of phycocyanin thresholds for predicting microcystin concentrations with 26% probability of exceeding 4 g/L microcystin in Ohio beach waters with varying secchi depths.

Median Secchi Phycocyanin Correct Above Correct Below

Depth (cm) by Threshold PC Threshold PC Threshold

SD Decile N ( g/L) (Sensitivity %) (Specificity %)

32 21 98 4 (57.1) 8 (57.1)

41 19 90 9 (100) 6 (60.0)

47 15 84 5 (100) 6 (60.0)

54 19 78 4 (80.0) 10 (71.4)

61 19 72 3 (100) 13 (81.3)

65 17 69 2 (40.0) 9 (75.0)

70 18 65 5 (100) 9 (69.2)

76 17 60 0 (0.0) 11 (78.6)

87 18 51 1 (100) 14 (82.4)

124 18 26 1 (20.0) 12 (92.3)

Total 181 Varies 34 (70.8) 98 (73.7)

continued

108

Table 5.5 continued

Median Secchi Positive Negative Correctly

Depth (cm) by False False Predictive Predictive Classified

SD Decile N Positives Negatives Value % Value % %

32 21 6 3 40.0 72.7 57.1

41 19 4 0 69.2 100 78.9

47 15 4 0 55.5 100 73.3

54 19 4 1 50.0 90.9 73.7

61 19 3 0 50.0 100 84.2

65 17 3 3 40.0 75.0 64.7

70 18 4 0 55.6 100 77.8

76 17 3 3 0.0 78.6 64.7

87 18 3 0 25.0 100 77.8

124 18 1 4 50.0 75.0 72.2

Total 181 35 14 49.3 87.5 72.9

109

5.3.4. Assessing Model Performance

The multivariable model provided very good discrimination where an area under the

ROC curve of 79.49% was observed compare with 73.3% in the univariable model). The

Hosmer-Lemeshow Goodness of Fit Test indicated that the probabilities produced by the model reflected the true outcomes experienced in the data across the deciles of risk (p =

0.3292).

5.4. Management Implications

The results of this study demonstrated the potential value of predictive models for rapidly and practically screening microcystin risk at inland beaches using simple indicators. This study focused on predicting water quality events exceeding the “relatively mild and/or low probability of adverse health effects” risk level of 4 g/L microcystin, which approximates to 20,000 cyanobacteria cells per mL using WHO assumptions (Chorus and

Bartram 1999). Due to methodological limitations, this study does not focus solely on samples exceeding the “moderate probability of adverse health effects” risk level of 20

g/L microcystin, which recommends restricting bathing when exceeded. In our study, any sample exceeding 4 g/L microcystin was determined to be an event warranting management action. Among our defined events, some samples may have exceeded 20

g/L microcystin; however, toxin concentrations were not quantified in this study when above 5 g/L.

110

This particular probability of exceeding the low risk threshold of 4 g/L microcystin was selected to optimize sensitivity and specificity. The sensitivity and specificity curves cross when the probability for exceeding the WHO low risk threshold is at approximately

0.26 (Figure 5.2). Although it is desirable to minimize false negatives to ensure beach signage is posted and persons are knowledgeable when using waters that are associated with increased risk for adverse health effects, it is also important to ensure that confidence is maintained in the beach advisories by not issuing too many advisories that are false positives. Using the proposed approach, the false positive rate is 19.2% (35 FP of 182 samples) and the false negative rate is 7.7% (14 FN of 182 samples). Although it would be desirable to have a lower false negative rate, to accomplish that goal would require using a lower threshold probability for risk. By setting the bar lower for this low risk advisory, we would increase our false positive rate and subsequently the amount of days our beaches would be under advisories. From this study, we recommend taking management action when the probability of exceeding the low risk threshold exceeds

0.26. Based upon this recommendation, 69 of the 182 (37.9%) samples assessed would have resulted in management action (i.e. posting an advisory). A likely consequence of the selection of a more conservative probability for risk characterization is not decreased use of swimming beaches, but desensitization to advisory signage. This desensitization could be compounded by the presence of advisory signage related to fecal indicator densities. The use of more conservative probabilities are not yet recommended; however, upon the development of models predictive of moderate adverse health risks associated with recreational water exposure (> 20 g/L microcystin), more conservative approaches

111 may be advantageous. Ultimately, this model is not intended to supplant beach manager judgment, but merely serve as a tool for determining if further testing or advisory signage is warranted for reducing risks associated with microcystin.

Although improved sensitivity is desired to minimize false negatives, other predictive methods for beach waters have been implemented and perform with similar or slightly better sensitivity for predicting health-relevant fecal indicator densities for beach management decisions (Morrison et al. 2003; Nevers and Whitman 2011). These models, as well as the model presented in this study, are useful in that they can provide real-time estimates of health-relevant water quality at reduced cost and rapidly while waiting for laboratory results.

A novel approach for the real-time estimation of health-relevant microcystin risk levels at inland recreational beaches is presented in this study. The practical advantages (i.e. on- site measurement, cost, timeliness) of estimating microcystin risk levels via in vivo phycocyanin and secchi depth are considerable. Furthermore, our results are based upon exceeding the low health risk estimate for 20,000 cyanobacteria cells/mL as set by WHO

(2003), which is approximately 4 g/L . Additional research is needed to examine this approach for predicting the WHO‟s no contact risk level of 20 g/L microcystin.. Future studies using non-ELISA based endpoints for microcystin concentration estimation may also produce improved results. Lastly, this study again reconfirms the results of

Markarewicz et al. 2009, Murby 2009, and Lehman 2007, who demonstrated a significant

112 association between microcystin and phycocyanin. Beach managers and regulatory entities should consider this approach or similar screening approaches for the rapid estimation of microcystin risk for the protection of human health in recreational waters when immediate sample results are not available.

5.5. Acknowledgment

Funding for this research was provided by a grant from the Ohio Water Development

Authority. We thank Scott Fletcher with Ohio State Parks for supporting the study and permitting sample collection at the seven state park beaches. We are grateful to Dr. Glen

Needham at The Ohio State University for allowing us to use his plate photometer. We also thank Pei-Yu Chiang who assisted with a portion of the field and laboratory data collection effort.

113

Chapter 6: Synthesis and Discussion

6.1. Major Findings

Upon accomplishing the work set out in the specific aims, we can evaluate the principal hypotheses. In doing so, we see that all the hypotheses were able to be answered in the affirmative.

For hypothesis 1, it was believed that water-related disease risk would be determined by a combination of water quality and water exposure behaviors. We were able to determine that both water quality and water exposure behaviors were associated with increased odds of illness. Persons who exposed their body to the water had 3.2 times greater odds for reporting GI illness than persons who reported that they did not expose their body to water. Food consumption at the beach was also associated with illness and was a significant confounder. People consuming food at the beach had 3.6 times greater odds for reporting GI illness than those reporting no food consumption.

Lastly, with respect to E. coli densities in the water, swimmers in the highest two quartiles of E. coli density had 7 times greater odds for reporting GI illness than

114 swimmers who used the cleanest water (first quartile of the E. coli distribution).

Based on these statistically significant odds ratios (as reported in chapter 3), our data supports the conclusion that water-related disease risk is determined by a combination of water quality and water exposure behaviors.

In reviewing hypothesis 2, it was expected that water quality determinants and water quality index values at inland lakes would be predictive of E. coli densities. Based upon the results presented in chapter 4, it is apparent that changes in phosphorus levels and trophic state index values hold some value. In the cases of total phosphorus and trophic state index, positive changes in those respective water quality values were associated with statistically significant increased odds of detecting elevated E. coli densities above 235 CFU/100 mL. Based upon an assessment of model discrimination and calibration, it was determined that deviation in trophic state index performed better with our dataset than the other evaluated water quality parameters.

With respect to hypothesis 3, it was believed that water quality determinants and water quality index values at inland lakes would be predictive of algal toxin concentrations as measured by microcystin. Chapter 5 demonstrates that several water quality parameters were associated with elevated levels of microcystin, including dissolved oxygen and specific conductivity; however, when compared against the cyanobacteria pigment, phycocyanin, the phycocyanin term was most useful in predicting harmful concentrations of microcystin. The phycocyanin term

115 becomes more useful in a multivariable model that incorporates secchi depth.

Overall, it was determined that several terms, particularly phycocyanin pigment, could be measured to predict elevated levels of the harmful toxin, microcystin.

6.2. Public Health Implications

The most significant implication of this work is the affirmation of E. coli as a health- relevant indicator for recreational water quality at freshwater beaches. In recent years, the recreational water quality criteria have drawn criticism for reasons outlined in chapter

2. Since 1984, no epidemiology study evaluating the effectiveness of the E. coli standard has been performed in U.S. waters. Furthermore, previous epidemiology studies were limited to the Ohio River (Stevenson 1953) and marine waters (see Chapter 2). No epidemiology studies evaluating the current indicator for “small” human-made reservoirs have been performed, despite their significant utility as recreational destinations. Our study results confirmed the effectiveness of E. coli as a health-relevant water quality indicator at our inland study beach. Additionally, in a comparative analysis performed by

U.S. EPA, our observations regarding health effects following recreational water exposure to E. coli were determined to be similar to previous observations by Dufour

(1984), suggesting the usefulness of the E. coli indicator across multiple freshwater bodies. The comparison of our study with the Dufour (1984) studies was able to be performed due to the experimental design used, which includes the same prospective cohort approach and definitions illness (EPA 2010a). U.S. EPA (2010b) in a comparison study found no statistical difference in the health effects curves (Figure 6.1) as

116 determined by ANCOVA (slope p = 0.85, intercept p = 0.10). These results confirm the use of E. coli as a health-relevant indicator.

Figure 6.1. “Health Effects Data and Trend Lines from the USEPA and Marion Studies”

(as taken directly from U.S. EPA 2010).

117

Other significant findings are the frequency of occurrence of elevated E. coli and microcystin levels at inland Ohio beaches. For example, 18 of 182 (9.9%) beach water samples exceeded recreational water quality for pathogen risks as determined by the E. coli indicator. With respect to microcystin, 48 of 182 (26.4%) of the beach water samples exceeded 4 g/L. We were not able to determine how many of those samples would have exceeded 20 g/L. These simple descriptive statistics point to the fact that recreational water quality in Ohio periodically poses health risks that warrant communication to beach users. The current approach in Ohio of sampling our beaches bi-weekly is inadequate for monitoring and advising beach users of pathogen and cyanotoxin risks.

The basic monitoring techniques described for screening or predicting pathogen and microcystin risks at these inland lakes do not require significant laboratory resources or a highly trained set of laboratory or field personnel. Our results from chapters 4 and 5 demonstrate that practical and real-time measurements can be performed on our beach waters to generate useful results that can be used for screening or predicting water quality when immediate data collection is not possible. The tools presented in this dissertation can provide beach managers and health departments substantially more knowledge regarding recreational water quality risks than current approaches, which often involve making no observations most days during the recreational season. The use of the approaches presented in this document could be used in tandem with current monitoring and/or could be used as part of an advisory notification system to protect public health.

118

6.3. Future Studies

The opportunities for future studies pertaining to this work are substantial. The areas of future studies can be binned into three categories: epidemiological research, enhanced predictive modeling, and improved risk communication.

6.3.1 Future Epidemiology Studies

One major epidemiology study needs to be performed in the freshwater environment as soon as possible. This study would explore the association between microcystin exposure and human illness following recreational water exposure. Our results from chapter 5 indicate that microcystin is quite prevalent in Ohio‟s inland lakes. An understanding of human illness as it pertains to both microcystin and pathogen risk (using

E. coli as an indicator) is warranted. First, we need to determine if an illness association with microcystin exists in freshwater environments. Secondly, with respect to illness risks among recreational water users, is the illness association with fecal indicators in the recreational water environment confounded by cyanotoxin exposure? This assessment of confounding has never been performed.

To date, a total of six epidemiology studies pertaining to cyanobacteria exposure in recreational waters have been described in the literature, comprising four cross-sectional and two prospective cohort studies (Stewart et al. 2006a). In the first cohort study

(Pilotto et al. 1997), persons swimming for at least 60 minutes with exposure to waters with >5,000 cyanobacteria cells/mL had 3.4 times greater odds of reporting any illness 7- days post-exposure than non-swimmers. In the second cohort study (Stewart et al. 119

2006b), swimmers in waters with the greatest cyanobacteria cell densities had 1.7 times greater odds for reporting any illness and 2.1 times greater odds for reporting respiratory illness than the swimmers using the waters with the lowest cyanobacteria cell density. In the four cross-sectional studies described in Stewart et al. (2006a), no statistically significant results were reported. None of the studies provide a statistical evaluation of measured cyanotoxins and human health outcomes; however, some indication of an association with illness and anatoxin-a exposure is described in Stewart et al. (2006b.), but due to the small exposed population, no statistically significant results were reported.

Additionally, an assessment of confounding by illness associations with fecal indicators is not described or reported.

The same desired epidemiology study could also explore some additional questions. Can chemical and physical indicators of fecal contamination events serve as health-relevant indicators of illness risk? In our study we demonstrate that changes in Carlson‟s TSI are predictive of elevated E. coli levels. We might consider exploring whether positive change in Carlson‟s TSI is associated with increased probabilities for illness following recreational water exposure. Some of this work is possible using our existing dataset; however, given the small size of our dataset and not many cases, only several water quality indicators can be evaluated per model. Future studies using large epidemiological datasets on recreational water users should consider the impact of rainfall events or water quality indicators, such as total phosphorus, on predicting adverse health outcomes in study populations.

120

6.3.2 Enhanced Predictive Modeling

The models evaluated in our study predicting elevated E. coli densities were limited.

Since only 16 events occurred, our models were univariable. With more cases, which could be obtained through increased sampling, it is possible that multivariable models with greater discriminatory power could be developed.

With regards to the models developed for predicting elevated microcystin levels, very beneficial models could likely be developed to predict microcystin levels exceeding 20

g/L. Future studies using the ELISA approach employed in the study, incorporating dilutions, may be able to accurately quantify Ohio beach water samples in these high ranges. As described in chapter 5, the 20 g/L concentration of toxin is when no contact is advised and long-term adverse health effects are possible. It is believed that phycocyanin would still be an important indicator of microcystin risk in those studies.

121

References Cited

Ahn, C.-Y., Chung, A.-S., Oh H.-M. 2002. Rainfall, phycocyanin, and N:P ratios related

to cyanobacterial blooms in a Korean large reservoir Hydrobiologia 474, 117-124.

Ahn, C.-Y., Joung S.-H., Yoon S.-K., Oh H.-M. 2007. Alternative alert system for

cyanobacterial bloom, using phycocyanin as a level determinant J. Microbiol. 45(2),

98-104.

Ashbolt NJ, Grabow WOK and Snozzi M. 2001. Indicators of microbial water quality.

In Water Quality: Guidelines, Standards and Health. Eds. Fewtrell, L and Bartram J.

World Health Organization. IWA Publishing, London, UK.

Backer, L.C., McNeel, S.V., Barber, T., Kirkpatrick, B., Williams, C., Irvin, M., Zhou,

Y., Johnson, T.B., Nierenberg, K., Aubel, M., LePrell, R., Chapman, A., Foss, A.,

Corum, S., Hill, V.R., Kieszak, S.M., Yung-Sung, C. 2010. Recreational exposure to

microcystins during algal blooms in two California lakes Toxicon. 55, 909-921.

122

Bates P and Phillips CA. 2005. Agricultural practices as a source of Camplyobacter

spp. in river water. Journal of Environmental Health Research 4(1):17-23.

Beaches Environmental Assessment and Coastal Health Act. 2000. Public Law 106-

284.

Bendel, R.B., Afifi, A.A. 1977. Comparison of stopping rules in forward stepwise

regression J. Am. Stat. Assoc. 72(357), 46-53

Beversdorf LJ, Bornstein-Forst SM, and McLellan SL. 2007. The potential for beach

sand to serve as a reservoir for Escherichia coli and the physical influences on cell die-

off. Journal of Applied Microbiology 102(5):1372-1381.

Boehm AB, Grant SB, Kim JH, Mowbray SL, McGee CD, Clark CD, et al. 2003.

Decadal and shorter period variability of surf zone water quality at Huntington Beach,

California. Environmental Science and Technology 36(18):3885–3892.

Boyer, G.L. 2008. Chapter 7: Cyanobacterial toxins in New York and the lower Great

Lakes ecosystems. In: H.K. Hudnell, Ed., International Symposium on Cyanobacterial

Harmful Algal Blooms Adv. Exp. Med. Biol. 616, p.949

123

Boyle M, Sichel C, Fernández-Ibáñez P, Aria-Quiroz GB, Iriarte-Puña M, Mercado A,

Ubomba-Jaswa E, and McGuigan KG. 2008. Bactericidal effect of solar water

disinfection under real sunlight conditions. Applied and Environmental Microbiology

74(10):2997-3001.

Brittain, S.M., Wang, J., Babcock-Jackson, L., Carmichael, W.W., Rinehart, K.L., and

Culver, D.A. 2000. Isolation and characterization of microcystins, cyclic heptapeptide

hepatotoxins from a Lake Erie strain of Microcystis aeruginosa J. Great Lakes Res.

26(3), 241-249.

Burns, J. 2008. Chapter 5: Toxic cyanobacteria in Florida Waters. In: H.K. Hudnell, Ed.,

International Symposium on Cyanobacterial Harmful Algal Blooms Adv. Exp. Med.

Biol. 616, p.949

Bushon RN, Brady AM, Likirdopulos CA, and Cireddu JV. 2009. Rapid detection of

Escherichia coli and enterococci in recreational water using an immunomagnetic

separation/adenosine triphosphate technique. Journal of Applied Microbiology

106(2):432-441.

124

Byers HL, Cabrera ML, Matthews MK, Franklin DH, Andrae JG, Radcliffe DE,McCann

MA, Kuykendall HA, Hoveland CS and Calvert VH. 2005. Phosphorus, sediment,

and Escherichia coli loads in unfenced streams of the Georgia Piedmont, USA.

Journal of Environmental Quality 34:2293-2300.

Cabelli, V.J., A.P. Dufour, M.A. Levin, L.J. McCabe and P.W. Haberman. 1979.

Relationship of microbial indicators to health effects at marine bathing beaches.

American Journal of Public Health 69(7):690-696

Cabelli VJ, Dufour AP, McCabe LJ, Levin MA. 1982. Swimming-associated

gastroenteritis and water quality. American Journal of Epidemiology 115:606-

616.

Cabelli VJ. 1983. Health Effects Criteria for Marine Recreational Waters. U.S.

Environmental Protection Agency Report EPA-600/1-80-031. Washington, DC.

Calderon R, Mood E, Dufour A. 1991. Health effects of swimmers and nonpoint

sources of contaminated water. International Journal of Environmental Health

Research 1:21-31.

125

Canteras JC, Juanes JA, Perez L, Koev KN. 1995. Modelling the coliforms inactivation

rates in the Cantabrian Sea (Bay of Biscay) from in situ and laboratory determinations

of T90. Water Science and Technology 32(2):37-44.

Carlson RE. 1977. A trophic state index for lakes. Limnology and Oceanography

22(2):361-369.

Carmichael, W.W. The cyanotoxins Adv. Bot. Res. 1997. 27, 211-256.

Carmichael, W.W., An, J. 1999. Using an enzyme linked immunosorbent assay (ELISA)

and a protein phosphatase inhibition assay (PPIA) for the detection of microcystins

and nodularins Nat. Toxins. 7(6), 377-385.

Carmichael, W.W., Azevedo, S.M.F.O., An, J.S., Molica R.J.R., Jochimsen E.M., Lau,

S., Rinehart, K.L., Shaw, G.R., and Eaglesham, G.K. 2001. Human fatalities from

cyanobacteria: chemical and biological evidence for cyanotoxins Environ. Health.

Perspect. 109(7), 663-668.

Chorus, I., Ed. 2005. Current approaches to cyanotoxin risk assessment, risk

management and regulation in different countries. Federal Environment Agency

(Umweltbundesamt). Berlin. p. 117.

126

Chorus, I., Bartram, J. 1999. Toxic Cyanobacteria in Water: A guide to their public

health consequences, monitoring and management. E & FN Spon on behalf of WHO,

London. p. 416.

Chu, F.S., Huang, X., Wei, D., Carmichael, W.W. 1989. Production and characterization

of antibodies against microcystins Appl. Environ. Micro. 55(8), 1928-1933.

Codd, G.A., Morrison, L.F., Metcalf, J.S. 2005. Cyanobacterial toxins: risk management

for health protection Toxicol. Appl. Pharmacol. 2005. 203, 264-272.

Colford JM, Wade TJ, Schiff KC, Wright CC, Griffith JF, Sandhu SK, Burns S, Sobsey

M, Lovelace G and Weisberg S. 2007. Water quality indicators and the risk of illness

at beaches with nonpoint sources of fecal contamination. Epidemiology 18(1):27-35.

Correll DL. 1998. The role of phosphorus in the eutrophication of receiving waters: a

review. Journal of Environmental Quality 27:261-266.

Converse RR, Blackwood AD, Kirs M, Griffith JF and Noble RT. 2009. Rapid QPCR-

based assay for fecal Bacteroides spp. as a tool for assessing fecal contamination in

recreational waters. In Press, Water Research doi:10.1016/j.wateres.2009.06.036

127

Davic, R.D., Eicher, D., DeShon, J. 1996. Ohio Water Resources Inventory, Volume 3:

Ohio’s Public Lakes, Ponds and Reservoirs. Ohio Environmental Protection Agency,

Division of Surface Water. Columbus, OH. p. 296.

Davis, T.W., Berry, D.L., Boyer, G.L., Gobler, C.J. 2009. The effects of temperature and

nutrients on the growth and dynamics of toxic and non-toxic strains of Microcystis

during cyanobacteria blooms Harmful Algae. 8, 715-725.

Desmarais TR, Solo-Gabriele HM and Palmer CJ. 2002. Influence of soil on fecal

indicator organisms in a tidally influenced subtropical environment. Applied and

Enviornmental Microbiology 68(3):1165-1172.

Ding, Y., Mutharasan, R. 2011. Highly sensitive and rapid detection of microcystin-LR

in source and finished water samples using cantilever sensors Environ. Sci. Technol.

45, 1490-1496.

Dorfman M and Rosselot KS. 2008. Testing the Waters, 18th Ed. Natural Resources

Defense Council.

Dorfman M and Rosselot KS. 2010. Testing the Waters, 20th Ed. Natural Resources

Defense Council.

128

Dufour A. 1984. Health Effects Criteria for Fresh Recreational Waters. U.S.

Environmental Protecion Agency Report EPA-600-1-84-2004. Cincinnati, OH

Dziuban EJ, Liang JL, Craun GF, Hill V, Yu PA, Painter J, Moore MR, Calderon RL,

Roy RL and Beach MJ. 2006. Surveillance for waterborne-disease outbreaks

associated with recreational water – United States, 2003-2004. Mortality and

Morbidity Weekly Report 55(12):1-24.

East Fork Watershed Collaborative, 2006. East Fork Lake Tributaries Watershed

Management Plan. Online Document. URL:

ftp://ftp.co.clermont.oh.us/oeq/EastForkLakeTributariesWAP_draft_9_25_06.pdf

Edmondson WT. 1970. Phosphorus, nitrogen, and algae in Lake Washington after

diversion of sewage. Science 169:690-691.

Edwards, M., Johns, D.G., Leterme, S.C., Svendsen, E., Richardson, A.J. Regional

climate change and harmful algal blooms in the northeast Atlantic Limnol. Oceanogr.

2006. 51(2), 820-829.

129

Elmir SM, Wright HM, Solo-Gabriele HM, Abdelzaher A, Fleming LE, Miller G,

Rybolowik M, Shih M-TP, Pillai SP, Cooper JA and Quaye EA. 2007. Quantitative

evaluation of bacteria released by bathers in a marine water. Water Research

41(2007):3-10.

Elmir SM, Shibata T, Solo-Gabriele HM, Sinigalliano CD, Gidley ML, Miller G, Plano

LRW, Kish J, Withum K, Fleming LE. 2009. Quantitative evaluation of enterocci

and Bacteroidales released by adults and toddlers in marine water. Water Research

43(18):4610-4616

Falconer, I.R., Burch, M.D., Steffensen, D.A., Choice, M. and Coverdale, O.R. 1994.

Toxicity of the blue-green alga (cyanobacterium) Microcystis aeruginosa in

drinking water to growing pigs, as an animal model for human injury and risk

assessment. J.Environ. Toxicol. Water Qual. 9, 131-139.

Fawell, J.K., James, C.P. and James, H.A. 1994. Toxins from Blue-Green Algae:

Toxicological Assessment of Microcystin-LR and a Method for its Determination in

Water, Water Research Centre, Medmenham, UK, 1-46.

130

Ferey, J.P., D. Zmlrou, F. Balducci, B. Baleux, P. Fera, G. Larbaight, E. Jacq, B.

Moissonnier, A. Blineau and J. Boudout. 1989. Epidemiologic significance of

Microbial pollution criteria for river recreational waters. International Journal of

Epidemiology 18(1):198-205.

Ferguson CM, Coote BG, Ashbolt NJ and Stevenson IM. 1996. Relationships between

indicators, pathogens and water quality in an estuarine system. Water Research

30(9):2045-2054.

Fleisher JM, Day D, Salmon RL, Jones F, Wyer MD, Godfree AF. 1996. Marine waters

contaminated with domestic sewage: nonenteric illnesses associated with bather

exposure in the United Kingdom. American Journal of Public Health 86:1228-1234.

Francis, G. Poisonous Australian lake Nature. 1878. May, 11-12.

Francy, DS, Gifford AM, Darner RA. 2003. Escherichia coli at Ohio bathing beaches –

distribution, sources, wastewater indicators, and predictice modeling. U.S.

Geological Survey, Water-Resources Investigations Report 02-4285. Columbus, OH.

Francy, DS, and Darner RA. 2007. Nowcasting beach advisories at Ohio Lake Erie

beaches: U.S. Geological Survey Open-File Report 2007–1427.

131

Francy DS. 2009. Use of predictive models and rapid methods to nowcast bacteria levels

at coastal beaches. Aquatic Ecosystem Health and Management 12(2):177-182.

Frick WE, Zhongfu G and Zepp RG. 2008. Nowcasting and forecasting concentrations

of biological contaminants at beaches: a feasibility case study. Environmental

Science and Technology 42:4819-4824.

Gaffield SJ, Goo RL, Richards LA, and Jackson RJ. 2003. Public health effects of

inadequately managed stormwater runoff. American Journal of Public Health

93(9):1527-1533.

Gerba GP, Goyal SM, LaBelle RL, Cech I, Bodgan GF. 1979. Failure of indicator

bacteria to reflect the occurrence of enteroviruses in marine waters. American

Journal of Public Health 69(11):1116-1119.

Given S, Pendleton LH, Boehm AB. 2006. Regional public health cost estimates of

contaminated coastal waters: a case study of gastroenteritis at Southern California

beaches. Environmental Science and Technology 40(16):4851-4858.

132

Graczyk TK, Sunderland D, Tamang L, Shields TM, Lucy FE and Breysse PN. 2007a.

Quantitative evaluation of the impact of bather density on levels of human-virulent

microsporidian spores in recreational water. Applied and Environmental

Microbiology (73)13:4095-4099

Graczyk TK, Sunderland D, Tamang L, Lucy FE and Breysse PN. 2007b. Bather

density and levels of Cryptosporidium, Giardia and pathogenic microsporidian spores

in recreational bather water. Parasitology Research 101:1729-1731.

Hach Company. DR 2800 Spectrophotometer Procedures Manual, Edition 2. Germany.

2007. p. 814.

Harvell, C.D., Kim, K., Burkholder, J.M., Colwell, R.R., Epstein, P.R., Grimes, D.J.,

Hoffman, E.E., Lipp, E.K., Osterhaus A.D.M.E., Overstreet, R.M., Porter, J.W.,

Smith, G.W., Vasta, G.R. Emerging marine diseases – climate links and anthropogenic

factors Science. 2000, 285, 1505-1510.

Heaney, C.D., Sams E., Wing, S., Marshall, S., Brenner, K., Dufour, A.P., Wade, T.J.,

2009. Contact with beach sand among beachgoers and risk of illness. American

Journal of Epidemiology 170(2), 164-72.

133

Heiskary, S. 1997. Lake Prioritization for Protecting Swimmable Use. Minnesota

Pollution Control Agency, Water Quality Division. October 1997. Minneapolis, MN.

Hosmer, D.W., Lemeshow, S. Applied logistic regression, second edition. New York:

Wiley and Sons, Inc. 2000. p. 375.

Hudnell, H.K. The state of freshwater harmful algal blooms assessments, policy and

legislation Toxicon. 2010. 55, 1024-1034.

Hunter, P.D., Tyler, A.N., Gilvear, D.J., Willby, N.J. Using remote sensing to aid the

assessment of human health risks from blooms of potentially toxic cyanobacteria

Environ. Sci. Technol. 2009. 43, 2627-2633.

Hutchinson GE. 1957. A Treatise on Limnology, Volume 1. Wiley.

Ishii S, Hansen DL, Hicks RE, and Sadowsky MJ. 2007. Beach sand and sediments are

temporal sinks and sources of Escherichia coli in Lake Superior. Environmental

Science and Technology 41(7):2203-2209.

Izydorczyk, K., Tarczynska, M., Jurczak, T., Mrowczynski, J., Zalewski, M.

Measurement of phycocyanin fluorescence as an online early warning system for

cyanobacteria in reservoir intake water Environ. Toxicol. 2005. 20(4), 425-430.

134

James, R., Gregg, A., Dindal, A., McKernan, J. Environmental technology verification

report: Abraxis microcystin test kits. Online document. Accessed online: June 22,

2011 URL: http://www.epa.gov/etv/pubs/600r10158/600r10158.pdf

Jarvie HP, Neal C, and Withers PJA. 2006. Sewage-effluent phosphorus: a greater risk to

river eutrophication than agricultural phosphorus? Science and the Total

Environment 360(1-3):246-253.

Jiang SC and Chu W. 2004. PCR detection of viruses in southern California urban

rivers. Journal of Applied Microbiology 97(1):17-28.

Juhna T, Birzniece D and Rubulis J. 2007. Effects of phosphorus on survival of

Escherichia coli in drinking water biofilms. Applied and Environmental

Microbiology 73(11):3755-3758.

Kelsey JL, Whittemore AS, Evans AS, Thompson WD. 1996. Methods in Observational

Epidemiology. Oxford University Press.

Kim JH, Grant SB. Public mis-notification of coastal water quality: a probabilistic

evaluation of posting errors at Huntington Beach, California. Environmental Science

and Technology 38(9):2497–2504.

135

Knowlton MF and Jones JR. 2006. Temporal variation and assessment of trophic state

indicators in Missouri reservoirs: implication for lake monitoring and management.

Lake and Reservoir Management 22(3):261-271.

Krishnaswami, S.K. 1971. Health aspects of water quality. American Journal of Public

Health 61(11):2259-2268

Kueh CSW, Tam TY, Lee T, Wong SL, Lloyd OL, Yu ITS, Wong TW, Tam JS, Bassett

DCJ. 1995. Epidemiologic study of swimming-associated illnesses relating to

bathing-beach water quality. Water Science and Technology 31(5-6):1-4.

LeChevallier MW, Norton WD, Lee RG. 1991. Occurrence of Giardia and

Cryptosporidium spp. in surface water supplies. Applied and Environmental

Microbiology 57(9):2610-2616.

Lee J and Deininger RA. 2004. Detection of E. coli in beach water within 1 hour using

immunomagnetic separation and ATP bioluminescence. Luminescence 19:31-36.

Lee, J and Deininger RA. 2010. Rapid detection of enterococci in marine beach water by

immunomagnetic capture and bioluminescence and its comparison with conventional

methods. Journal of Environmental Health 72:9:8-13

136

Lee, T., Tsuzuki, M., Takeuchi, T., Yokoyamal, K., Karubel, I. 1994. In vivo

fluorometric method for early detection of cyanobacterial waterblooms J. Appl.

Phycol. 6, 489-495.

Lehman, E.M. Seasonal occurrence and toxicity of Microcystis in impoundments of the

Huron River, Michigan, USA Wat. Res. 2007. 41, 795-802.

Lehr E.L. and Johnson CC. 1954. Water quality of swimming places. Public Health

Reports 69(8):742-748.

Lipp EK, Farrah SA and Rose JB. 2001. Assessment and impact of microbial fecal

pollution and human enteric pathogens in a coastal community. Marine Pollution

Bulletin 42(4):286-293

Malek MA, Cums AT, Holman RC, Fischer TK, Bresee JS, Glass RI, Steiner CA and

Parashar UD. Diarrhea- and Rotavirus-associated hospitalizations among children

less than 5 years of age: United States, 1997 and 2000. Pediatrics 117(6):1887-

1892.

Marion JW, Lee J, Lemeshow S, and Buckley TJ. 2010. Association of gastrointestinal

illness and recreational water exposure at an inland U.S. beach. Water Research

44(16):4796-4804

137

Makarewicz, J.C., Boyer, G.L., Lewis, T.W., Guenther, W., Atkinson, J., Arnold, M.

2009. Spatial and temporal distribution of the cyanotoxin microcystin-LR in the Lake

Ontario ecosystem: coastal embayments, rivers, nearshore and offshore, and upland

lakes J. Great Lakes Res. 35, 83-89.

McDermott, C.M., Feola, R., Plude, J. 1995. Detection of cyanobacterial toxins

(microcystins) in waters of northeastern Wisconsin by a new immunoassay technique

Toxicon. 33(11), 1433-1442.

Metz, C.E., 1978. Basic principles of ROC analysis. Seminars in Nuclear Medicine 8,

283-298.

Morris RD, Naumova EN, Levin R, Munasinghe RL. 1996. Temporal variation in

drinking water turbidity and diagnosed gastroenteritis in Milwaukee. American

Journal of Public Health 86(2):237-239.

Morrison, A.M., Coughlin, K., Shine, J.P., Coull, B.A., Rex, A.C. 2003. Receiver

operating characteristic curve analysis of beach water quality variables Appl. Environ.

Micro. 2003. 69(11), 6405-6411.

138

Murby, A.L. Assessing spatial distributions of cyanobacteria and microcystins in N.H.

lakes with implications for lake monitoring. University of New Hampshire. 2009.

p.89.

Natural Resources Defense Council v. Stephen L. Johnson and U.S. EPA., 2008. United

States District Court for the Central District of California. No. CV06-4843 PSG

(JTLx). Filed 04 September 2008.

Neal C, Jarvie HP, Neal M, Love AJ, Hill L, and Wickham H. 2005. Water quality of

treated sewage effluent in a rural area of the upper Thames Basin, southern

England, and the impacts of such effluents on riverine phosphorus concentrations.

Journal of Hydrology 304(1-4):103-117.

Nevers MB and Whitman RL. 2005. Nowcast modeling of Escherichia coli at multiple

urban beaches of southern Lake Michigan. Water Research 39(20):5250-5260.

Nevers MB and Whitman RL. 2010. Policies and practices of beach monitoring in the

Great Lakes, USA: a critical review. Journal of Environmental Monitoring 12:581-

590.

139

Nevers, MB and Whitman RL. 2011. Efficacy of monitoring and empirical predictive

modeling at improving public health protection at Chicago beaches. Water

Research 45: 1659-1668.

Noble, RT, Allen SM, Blackwood AD, Weiping C, Jiang SC, Lovelace GL, Sobsey, MD,

Stewart JR and Wait DA. 2003. Use of viral pathogens and indicators to

differentiate between human and non-human fecal contamination in a microbial

source tracking comparison study. Journal of Water and Health 1(4):195-207

Ohio Department of Health. 2006. 2006 Inland State Park Beach Sample Results.

Online document.

http://www.odh.ohio.gov/ASSETS/5B25DBE72F58498AAF67CF514FB7EDE1/Inla

nd2005.pdf. Ohio Department of Health, Columbus, Ohio.

Ohio Department of Health. 2007. 2007 Inland State Park Beach Sample Results.

Online document.

http://www.odh.ohio.gov/ASSETS/E5ED23AFF0D64F0A8A52F20CF4A27E70/Inla

nd-07.pdf. Ohio Department of Health, Columbus, Ohio.

Ohio Department of Health. 2009a. Beach Monitoring. Online document.

http://www.odh.ohio.gov/odhprograms/eh/bbeach/beachmon.aspx

140

Ohio Department of Health. 2009b. Sampling Procedures – Beach Monitoring. Online

document. http://www.odh.ohio.gov/odhprograms/eh/bbeach/beach_procedures.aspx

Ohio EPA. Ohio harmful algal bloom advisories in 2010. Online document. Accessed:

May 3, 2011. p. 1.

Ohio State Parks. 2006a. Ohio State Parks: 2006 Annual Report. Online document.

http://www.dnr.state.oh.us/Portals/2/explore/annualrpt06.pdf Ohio Department of

Natural Resources, Division of Parks and Recreation, Columbus, Ohio.

Ohio State Parks. 2007a. Nationwide Matrix Summary. (Internal Online Document).

http://www.dnr.state.oh.us/parknet/fsg/service/BeachOperations/matrixsum/tabid/160

52/Default.aspx. Ohio Department of Natural Resources, Division of Parks and

Recreation, Columbus, Ohio.

Ohio State Parks. 2007b. Beach safety talking points for 2007. (Internal Online

Document).

http://www.dnr.state.oh.us/Portals/21/files/pr/Beach%20Talk%20Points%208-07.doc

Ohio Department of Natural Resources, Division of Parks and Recreation, Columbus,

Ohio.

141

Natural Resources Defense Council (NRDC) 2007. Testing the Waters 2007: A guide to

water quality at vacation beaches. Online document.

http://www.nrdc.org/water/oceans/ttw/exesum.pdf.

Natural Resources Defense Council v. Stephen L. Johnson and U.S. EPA. 2008.

United States District Court for the Central District of California. No. CV06-4843

PSG (JTLx). Filed 04 September 2008.

Nevers MB and Whitman RL. 2005. Nowcast modeling of Escherichia coli

concentrations at multiple urban beaches of southern Lake Michigan. Water

Research 39:5250-5260.

Nevers MB and Whitman RL. 2008. Coastal strategies to predict Escherichia coli

concentrations for beaches along a 35 km stretch of southern Lake Michigan.

Environmental Science and Technology 42(12):4454-4460.

Nevers, M.B., Whitman, R.L. 2011. Efficacy of monitoring and empirical predictive

modeling at improving public health protection at Chicago beaches Wat. Res. 45,

1659-1668.

142

Noble RT, Allen SM, Blackwood AD, Weiping C, Jiang SC, Lovelace GL, Sobsey, MD,

Stewart JR and Wait DA. 2003. Use of viral pathogens and indicators to differentiate

between human and non-human fecal contamination in a microbial source tracking

comparison study. Journal of Water and Health 1(4):195-207

Ouellette, A.J.A., Wilhelm, S.W. 2003. Toxic cyanobacteria: the evolving molecular

toolbox Front. Ecol. Environ. 1(7), 359-366.

Paerl, H.W., Hall, N.S., Calandrino, E.S. 2011. Controlling harmful cyanobacterial

blooms in a world experiencing anthropogenic and climatic-induced change Sci. Total

Environ. 409, 1739-1745.

Peduzzi P, Concato J, Kemper E, Holford TR, and Feinstein A. 1996. A simulation study

of the number of events per variable in logistic regression analysis. Journal of

Clinical Epidemiology 49(12):1373-1379.

Peperzak, L. Climate change and harmful algal blooms in the North Sea Acta Oecol.

2003. 24, 139-144.

143

Pilotto LS, Douglas RM, Burch MD, Cameron S, Beers M, Rouch G, Robinson P, Kirk

M, Cowie CT, Hardiman S, Moore C and Attewell RG. 1997. Health effects of

exposure to cyanobacteria (blue-green algae) during recreational water-related

activites. Australian and New Zealand Journal of Public Health 21:562-566.

Prüss A., 1998. Review of epidemiological studies on health effects from exposure to

recreational water. International Journal of Epidemiology 27(1), 1-9.

Rabinovici SJM, Bernknopf RL, Wein AM, Coursey DL, Whitman RL. 2004.

Economic and health risk trade-offs of swim closures at a Lake Michigan Beach.

Environmental Science and Technology 38(10):2737-2745.

Rinta-Kanto, J.M., Ouellette, A.J.A., Boyer, G.L., Twiss, M.R., Bridgeman, T.B.,

Wilhelm, S.W. Quantification of toxic Microcystis spp. during the 2003 and 2004

blooms in western Lake Erie using quantitative real-time PCR Environ. Sci. Technol.

2005. 39, 4198-4205.

Rinta-Kanto, J.M., Konopko, E.A., DeBruyn, J.M., Bourbonniere, R.A., Boyer, G.L.,

Wilhelm, S.W. 2009. Lake Erie Microcystis: relationship between microcystin

production dynamics of genotypes and environmental parameters in a large lake

Harmful Algae. 8, 665-673.

144

Scheffer M, Rinaldi S, Gragnani A, Mur LR, van Nes EH. 1997. Dominance of

filamentous cyanobacteria in shallow, turbid lakes. Ecology 78(1):272-282.

Schwartz J, Levin R, Hodge K. 1997. Drinking water turbidity and pediatric hospital use

for gastrointestinal illness in Philadelphia. Epidemiology 8(6):615-620.

Sedmak B, Kozi G. 1998. The role of microcystins in heavy cyanobacterial bloom

formation. Journal of Plankton Research 20(4):691-708.

Solo-Gabriele, HM, Wolfert MA, Desmarais TR and Palmer CJ. 2000. Sources of

Escherichia coli in a coastal subtropical environment. Applied and Environmental

Microbiology 66:230-237.

Sridhar, B.B.M., Vincent, R.K. 2007. Spectral reflectance measurements of a

Microcystis bloom in Upper Klamath Lake, Oregon J. Great Lakes Res.. 33, 279-284.

Stevenson, A.H. 1953. Studies of bathing water and health. American Journal of Public

Health 43:529-538.

145

Stewart, I., Webb, P.M., Schluter, P.J., Shaw G.R. 2006a. Recreational and occupational

field exposure to freshwater cyanobacteria – a review of anecdotal and case reports,

epidemiological studies and the challenges for epidemiological assessment Environ.

Health. 5:6

Stewart, I., Webb, P.M., Schluter, P.J., Fleming, L.E., Burns, J.W. Jr., Gantar, M.,

Backer, L.C., Shaw, G.R. 2006b. Epidemiology of recreational exposure to freshwater

cyanobacteria BMC Public Health. 6:93.

Stoner, N. and M. Dorfman. 2004. Testing the waters: a guide to water quality at

vacation beaches. Natural Resources Defense Council.

Stoner, N., M. Merkel and M. Dorfman. 2004. Swimming in sewage. Natural

Resources Defense Council and Environmental Integrity Project.

Trimbee AM and Prepas EE. 1987. Evaluation of total phosphorus as a predictor of the

relative biomass of blue-green algae with emphasis on Alberta lakes. Canadian

Journal of Fisheries and Aquatic Sciences 44(7):1337-1342.

146

U.S. Army Corps of Engineers, Louisville District, 2010. Yearly Lake Reports. Online

database. URL: http://www.lrl.usace.army.mil/wcds/dlbrpt.htm

U.S. Army Corps of Engineers. Value to the Nation: fast facts for Ohio. Accessed

online: May 3, 2011. URL: http://www.corpsresults.us/recreation/state.asp?state=OH#

U.S. Environmental Protection Agency. 1983. Methods for Chemical Analysis of Water

and Wastes. U.S. Environmental Protection Agency Report EPA-600/4-79-020,

Cincinnati, OH.

U.S. Environmental Protection Agency. 1986. Ambient Water Quality Criteria for

Bacteria. U.S. Environmental Protection Agency Report EPA-440/5-84-002.

Washington D.C.

U.S. Environmental Protection Agency. 1993. Methods for the determination of

inorganic substances in environmental samples. U.S. Environmental Protection

Agency Report EPA/600/R-93/100, Cincinnati, OH.

U.S. Environmental Protection Agency. 2002. Method 1603: Escherichia coli (E. coli)

in Water by Membrane Filtration Using Modified membrane-Thermotolerant

Escherichia coli Agar (Modified mTEC). U.S. Environmental Protection Agency

Report EPA-821-R-02-023, Washington D.C.

147

U.S. Environmental Protection Agency. 2006. Water quality standards for coastal

recreation waters: considerations for states as they select appropriate risk levels.

U.S. Environmental Protection Report EPA-823-F-06-012, Washington D.C.

U.S. Environmental Protection Agency. 2007a. National Water Quality Inventory:

Report to Congress, 2002 Reporting Cycle. U.S. Environmental Protection Agency

Report EPA-841-R-07-001, Washington, D.C.

U.S. Environmental Protection Agency. 2007b. 2004 National Assessment Database .

Online Application: http://www.epa.gov/waters/305b/index_2004.html

U.S. Environmental Protection Agency, Washington, D.C.

U.S. Environmental Protection Agency. 2007c. Beaches. Online Website:

http://www.epa.gov/beaches/. U.S. Environmental Protection Agency, Washington,

D.C.

U.S. EPA. 2007d. Report of the experts scientific workshop on critical research needs

for the development of new or revised recreational water quality criteria. U.S.

Environmental Protection Agency Report EPA-823-R-07-006. Warrenton, VA.

148

U.S. EPA. 2010a. Applicability of Great Lakes NEEAR Dataset to Inland Recreational

Water Criteria: Summary of Key Studies. U.S. Environmental Protection Agency

Report EPA-823-R-1-0002.

U.S. EPA. 2010b. Evaluation of multiple indicator combinations to develop quantifiable relationships. U.S. Environmental Protection Agency Report EPA-822-R-10-004.

Vincent, R.K., Qin, X., McKay, R.M.L., Miner, J., Czajkowski, K., Savino, J.,

Bridgeman, T. 2004. Phycocyanin detection from LANDSAT TM data for mapping

cyanobacterial blooms in Lake Erie Remote Sens. Environ. 89, 381-392

Wade TJ, Pai N, Eisenberg JN, Colford JMJ. 2003. Do U.S. Environmental Protection

Agency water quality guidelines for recreational waters prevent gastrointestinal

illness? A systematic review and meta-analysis. Environmental Health Perspectives

111:1102–1109.

Wade TJ, Calderon RL, Sams E, Beach M, Brenner KP, Williams AH and Dufour AP.

2006. Rapidly measured indicators of recreational water quality are predictive of

swimming-associated gastrointestinal illness. Environmental Health Perspectives

114(1):24-28

149

Wade TJ, Calderon RL, Brenner KP, Sams E, Beach M, Haugland R, Wymer L, Dufour

AP. 2008. High sensitivity of children to swimming-associated gastrointestinal

illness: results using a rapid assay of recreational water quality. Epidemiology

19(3):375-83.

Walker, S., Lund, J.C., Schumacher, D.G., Brakhage, P.A., McManus, B.C., Miller, J.D.,

Augustine, M.M., Carney, J.J., Holland, R.S., Hoagland, K.D., Holz, J.C., Barrow,

T.M., Rundquist, D.C., Gitelson, A.A. 2008. Chapter 6: Nebraska Experience. In: H.K.

Hudnell, Ed., International Symposium on Cyanobacterial Harmful Algal Blooms Adv.

Exp. Med. Biol.616, p.949

Watzin, M.C., Miller, E.B., Shambaugh, A.D., Kreider, M.A. Application of the WHO

alert level framework to cyanobacterial monitoring of Lake Champlain, Vermont

Environ. Toxicol. 2006. 21(3), 278-288.

Wetzel RG and Likens GE. 2000. Limnological Analyses, 3rd ed. Springer-Verlag.

Whitman RL and Nevers MB. 2004. Escherichia coli sampling reliability at a frequently

closed Chicago beach: monitoring and management implications. Environmental

Science and Technology 38(16): 4241-4246.

150

Whitman RL and Nevers MB. 2003. Foreshore sand as a source of Escherichia coli in

nearshore water of a Lake Michigan beach. Applied and Environmental Microbiology

69(9):5555-5562.

Whitman RL, Neveres MB, Korinek GC, and Byappanahalli MN. 2004. Solar and

temporal effects on Escherichia coli concentration at Lake Michigan swimming beach.

Applied and Environmental Microbiology 70(7):4276-4285.

Whitman RL, Shively DA, Pawlik H, Nevers MB and Byappanahalli MN. 2003.

Occurrence of Escherichia coli and enterococci in Cladophora (Chlorophyta) in

nearshore water and beach sand of Lake Michigan. Applied and Environmental

Microbiology 69(8):4714-4719.

World Health Organization. Guidelines for safe recreational water environments;

Volume 1: coastal and fresh waters. Geneva. 2003. p.219.

Yamahara, KM, Layton BA, Santoro AE and Boehm AB. 2007. Beach sands along the

California coast are diffuse sources of fecal bacteria to coastal waters.

Environmental Science and Technology 41(13):4515-4521.

Ye, C., Shen, Z., Zhang, T., Maohong, F., Lei, Y., Zhang, J. 2011. Long-term joint effect

of nutrients and temperature increase on algal growth in Lake Taihu, China J. Environ.

Sci. 23(2), 222-227.

151

Yoder JS, Blackburn BG, Craun GF, Hill V, Levy DA, Chen N, Lee SH, Calderon RL

and Beach MJ. 2004. Surveillance for waterborne-disease outbreaks associated with

recreational water – United States, 2001-2002. Mortality and Morbidity Weekly

Report 53(8):1-22.

Yoder JS, Hlavsa MC, Craun GF, Hill V, Roberts V, Yu PA, Hicks LA, Alexander NT,

Calderon RL, Roy SL, Beach MJ. 2008. Surveillance for waterborne diseases and

outbreaks associated with recreational water use and other aquatic facility-associated

health events – United States, 2005-2006. Mortality and Morbidity Weekly Report

57(9):1-29.

152

Appendix A: Human Health Study Beach and Telephone Questionnaire

153

154

155

156

157

158

159

160

161

162

163

164

165

166

167

168

169

170

171

172

173

174

175

Appendix B: Human Health Study Recruitment Sign

Figure B.1. Image depicting sign used to recruit East Fork State Park beachgoers to participate in the human health aspect of the study.

176

Appendix C: Water Quality Data Summary

177

Table C.1. Shapiro-Wilk test for normality results for Alum Creek water quality variables.

Variable W V z Prob>z Temperature 0.958 1.201 0.376 0.353 Sp. Conductivty 0.741 7.416 4.106 0.000 Dissolved Oxygen 0.828 4.912 3.262 0.001 pH 0.939 1.740 1.135 0.128 Secchi Depth 0.955 1.274 0.496 0.310 Wave Length 0.949 1.466 0.783 0.217 Wave Height 0.977 0.657 -0.862 0.806 Total Phosphorus 0.642 10.227 4.765 0.000 Chlorophyll A 0.978 0.627 -0.955 0.830 Phycocyanin 0.637 10.374 4.794 0.000 Turbidity 0.898 2.907 2.187 0.014 E. coli CFU/100 mL 0.801 5.691 3.563 0.000 TSI - Secchi Depth 0.984 0.448 -1.644 0.950 TSI - Total Phosphorus 0.912 2.526 1.899 0.029 TSI - Chlorophyll A 0.659 9.763 4.669 0.000 TSI - Trimetric Mean 0.955 1.294 0.527 0.299 Previous Day Inflow Mean (CFS) 0.960 1.141 0.270 0.394 Sample Day Inflow Average (CFS) 0.929 2.019 1.440 0.075 Mean 2-Day Inflow Average (CFS) 0.964 1.035 0.070 0.472 Deviation from Mean TSI - Trimetric 0.900 2.858 2.152 0.016 Deviation from Mean Total P 0.642 10.224 4.764 0.000 Deviation from Mean Phycocyanin 0.970 0.847 -0.340 0.633 Deviation from Mean Chlorophyll A 0.244 21.615 6.298 0.000 Deviation from Mean Turbidity 0.967 0.937 -0.132 0.553 Deviation from Mean 2 Day Inflow 0.874 3.605 2.628 0.004

178

Table C.2. Shapiro-Wilk test for normality results for Buck Creek water quality variables.

Variable W V z Prob>z Temperature 0.960 1.144 0.275 0.392 Sp. Conductivty 0.666 9.543 4.623 0.000 Dissolved Oxygen 0.906 2.616 1.966 0.025 pH 0.949 1.457 0.771 0.220 Secchi Depth 0.952 1.368 0.642 0.261 Wave Length 0.971 0.827 -0.388 0.651 Wave Height 0.894 3.034 2.274 0.011 Total Phosphorus 0.772 6.522 3.843 0.000 Chlorophyll A 0.933 1.910 1.326 0.092 Phycocyanin 0.939 1.750 1.146 0.126 Turbidity 0.863 3.932 2.806 0.003 E. coli CFU/100 mL 0.829 4.893 3.254 0.001 TSI - Secchi Depth 0.964 1.017 0.035 0.486 TSI - Total Phosphorus 0.925 2.132 1.551 0.060 TSI - Chlorophyll A 0.978 0.637 -0.925 0.823 TSI - Trimetric Mean 0.956 1.244 0.448 0.327 Previous Day Inflow Mean (CFS) 0.966 0.972 -0.057 0.523 Sample Day Inflow Average (CFS) 0.840 4.580 3.119 0.001 Mean 2-Day Inflow Average (CFS) 0.956 1.269 0.488 0.313 Deviation from Mean TSI - Trimetric 0.963 1.067 0.134 0.447 Deviation from Mean Total P 0.978 0.637 -0.925 0.823 Deviation from Mean Phycocyanin 0.938 1.785 1.187 0.118 Deviation from Mean Chlorophyll A 0.963 1.072 0.142 0.444 Deviation from Mean Turbidity 0.981 0.553 -1.215 0.888 Deviation from Mean 2 Day Inflow 0.425 16.452 5.739 0.000

179

Table C.3. Shapiro-Wilk test for normality results for Deer Creek water quality variables.

Variable W V z Prob>z Temperature 0.979 0.605 -1.029 0.848 Sp. Conductivty 0.906 2.694 2.031 0.021 Dissolved Oxygen 0.891 3.117 2.330 0.010 pH 0.963 1.068 0.135 0.446 Secchi Depth 0.908 2.623 1.976 0.024 Wave Length 0.945 1.563 0.916 0.180 Wave Height 0.985 0.430 -1.729 0.958 Total Phosphorus 0.962 1.084 0.165 0.434 Chlorophyll A 0.958 1.191 0.358 0.360 Phycocyanin 0.9071 2.656 2.002 0.023 Turbidity 0.946 1.545 0.891 0.186 E. coli CFU/100 mL 0.743 7.350 4.088 0.000 TSI - Secchi Depth 0.941 1.676 1.059 0.145 TSI - Total Phosphorus 0.969 0.889 -0.242 0.596 TSI - Chlorophyll A 0.957 1.216 0.400 0.345 TSI - Trimetric Mean 0.950 1.444 0.753 0.226 Previous Day Inflow Mean (CFS) 0.591 11.701 5.041 0.000 Sample Day Inflow Average (CFS) 0.775 6.448 3.820 0.000 Mean 2-Day Inflow Average (CFS) 0.690 8.874 4.474 0.000 Deviation from Mean TSI - Trimetric 0.943 1.619 0.987 0.162 Deviation from Mean Total P 0.942 1.668 1.049 0.147 Deviation from Mean Phycocyanin 0.744 7.319 4.079 0.000 Deviation from Mean Chlorophyll A 0.893 3.071 2.299 0.011 Deviation from Mean Turbidity 0.870 3.728 2.697 0.004 Deviation from Mean 2 Day Inflow 0.918 2.345 1.746 0.040

180

Table C.4. Shapiro-Wilk test for normality results for Delaware Lake water quality variables.

Variable W V z Prob>z Temperature 0.928 1.998 1.415 0.079 Sp. Conductivty 0.862 3.823 2.741 0.003 Dissolved Oxygen 0.854 3.951 2.801 0.003 pH 0.921 2.186 1.599 0.055 Secchi Depth 0.924 2.107 1.523 0.064 Wave Length 0.900 2.766 2.079 0.019 Wave Height 0.883 3.241 2.404 0.008 Total Phosphorus 0.972 0.772 -0.528 0.701 Chlorophyll A 0.930 1.951 1.367 0.086 Phycocyanin 0.821 5.106 3.34 0.000 Turbidity 0.886 3.174 2.361 0.009 E. coli CFU/100 mL 0.455 15.154 5.557 0.000 TSI - Secchi Depth 0.958 1.159 0.301 0.382 TSI - Total Phosphorus 0.966 0.945 -0.116 0.546 TSI - Chlorophyll A 0.947 1.462 0.777 0.219 TSI - Trimetric Mean 0.967 0.927 -0.155 0.562 Previous Day Inflow Mean (CFS) 0.410 16.395 5.718 0.000 Sample Day Inflow Average (CFS) 0.438 15.630 5.620 0.000 Mean 2-Day Inflow Average (CFS) 0.479 14.476 5.463 0.000 Deviation from Mean TSI - Trimetric 0.966 0.933 -0.142 0.557 Deviation from Mean Total P 0.925 2.093 1.510 0.066 Deviation from Mean Phycocyanin 0.875 3.464 2.540 0.006 Deviation from Mean Chlorophyll A 0.734 7.394 4.090 0.000 Deviation from Mean Turbidity 0.942 1.602 0.963 0.168 Deviation from Mean 2 Day Inflow 0.939 1.686 1.067 0.143

181

Table C.5. Shapiro-Wilk test for normality results for East Fork water quality variables.

Variable W V z Prob>z Temperature 0.843 4.103 2.871 0.002 Sp. Conductivty 0.934 1.718 1.100 0.136 Dissolved Oxygen 0.950 1.310 0.549 0.292 pH 0.933 1.755 1.144 0.126 Secchi Depth 0.982 0.474 -1.516 0.935 Wave Length 0.902 2.575 1.924 0.027 Wave Height 0.930 1.819 1.216 0.112 Total Phosphorus 0.798 5.286 3.386 0.000 Chlorophyll A 0.790 5.485 3.461 0.000 Phycocyanin 0.963 1.04 0.089 0.464 Turbidity 0.709 7.605 4.126 0.000 E. coli CFU/100 mL 0.416 15.265 5.542 0.000 TSI - Secchi Depth 0.925 1.958 1.367 0.086 TSI - Total Phosphorus 0.938 1.634 0.998 0.159 TSI - Chlorophyll A 0.920 2.082 1.491 0.068 TSI - Trimetric Mean 0.930 1.837 1.236 0.108 Previous Day Inflow Mean (CFS) 0.651 9.118 4.494 0.000 Sample Day Inflow Average (CFS) 0.673 8.557 4.365 0.000 Mean 2-Day Inflow Average (CFS) 0.647 9.229 4.519 0.000 Deviation from Mean TSI - Trimetric 0.943 1.500 0.824 0.205 Deviation from Mean Total P 0.987 0.337 -2.213 0.987 Deviation from Mean Phycocyanin 0.937 1.645 1.012 0.156 Deviation from Mean Chlorophyll A 0.962 1.002 0.003 0.499 Deviation from Mean Turbidity 0.977 0.592 -1.067 0.857 Deviation from Mean 2 Day Inflow 0.691 8.091 4.251 0.000

182

Table C.6. Shapiro-Wilk test for normality results for Lake Logan water quality variables.

Variable W V z Prob>z Temperature 0.962 1.048 0.096 0.462 Sp. Conductivty 0.774 6.288 3.759 0.000 Dissolved Oxygen 0.905 2.635 1.981 0.024 pH 0.966 0.934 -0.139 0.555 Secchi Depth 0.926 2.053 1.470 0.071 Wave Length 0.741 7.206 4.037 0.000 Wave Height 0.839 4.472 3.062 0.001 Total Phosphorus 0.931 1.922 1.335 0.091 Chlorophyll A 0.969 0.860 -0.308 0.621 Phycocyanin 0.970 0.865 -0.297 0.617 Turbidity 0.815 5.145 3.348 0.000 E. coli CFU/100 mL 0.734 7.401 4.092 0.000 TSI - Secchi Depth 0.954 1.273 0.494 0.311 TSI - Total Phosphorus 0.978 0.617 -0.986 0.838 TSI - Chlorophyll A 0.908 2.569 1.929 0.027 TSI - Trimetric Mean 0.921 2.184 1.597 0.055 Previous Day Inflow Mean (CFS) Sample Day Inflow Average (CFS) Mean 2-Day Inflow Average (CFS) Deviation from Mean TSI - Trimetric 0.901 2.743 2.063 0.020 Deviation from Mean Total P 0.912 2.453 1.835 0.033 Deviation from Mean Phycocyanin 0.916 2.346 1.743 0.041 Deviation from Mean Chlorophyll A 0.784 6.009 3.666 0.000 Deviation from Mean Turbidity 0.955 1.260 0.473 0.318 Deviation from Mean 2 Day Inflow

183

Table C.7. Shapiro-Wilk test for normality results for Madison Lake water quality variables.

Variable W V z Prob>z Temperature 0.958 1.193 0.361 0.359 Sp. Conductivty 0.889 3.164 2.360 0.009 Dissolved Oxygen 0.864 3.897 2.788 0.003 pH 0.963 1.052 0.104 0.459 Secchi Depth 0.982 0.506 -1.395 0.918 Wave Length 0.815 5.276 3.408 0.000 Wave Height 0.888 3.191 2.378 0.009 Total Phosphorus 0.956 1.260 0.474 0.318 Chlorophyll A 0.898 2.919 2.195 0.014 Phycocyanin 0.788 6.075 3.697 0.000 Turbidity 0.895 3.015 2.262 0.012 E. coli CFU/100 mL 0.719 8.037 4.271 0.000 TSI - Secchi Depth 0.992 0.225 -3.060 0.999 TSI - Total Phosphorus 0.973 0.770 -0.537 0.704 TSI - Chlorophyll A 0.953 1.357 0.625 0.266 TSI - Trimetric Mean 0.978 0.638 -0.921 0.821 Previous Day Inflow Mean (CFS) Sample Day Inflow Average (CFS) Mean 2-Day Inflow Average Deviation from Mean TSI - 0.974 0.752 -0.583 0.720 (CFS) Deviation from Mean Total P 0.964 1.019 0.040 0.484 Trimetric Deviation from Mean Phycocyanin 0.972 0.793 -0.476 0.683 Deviation from Mean Chlorophyll 0.874 3.592 2.621 0.004 Deviation from Mean Turbidity 0.944 1.591 0.952 0.171 A Deviation from Mean 2 Day

Inflow

184

Table C.8. Spearman‟s rank correlation coefficient for selected Alum Creek water quality variables.

Variable Temp SpCond D.O. pH Secchi_cm WaveLen Temp 1.000 SpCond -0.195 1.000 D.O. -0.027 -0.003 1.000 pH 0.610 -0.026 -0.344 1.000 Secchi_cm -0.303 -0.052 -0.206 -0.121 1.000 WaveLen 0.551 -0.068 -0.118 0.440 -0.332 1.000 WaveHt 0.099 -0.068 -0.058 0.209 -0.241 0.674 Total P -0.003 0.406 -0.543 0.370 0.038 0.206 ChlA 0.295 -0.133 -0.369 0.586 -0.277 0.304 Turbidity 0.118 0.126 0.411 -0.003 -0.510 0.139 PhycoCyn -0.076 0.157 -0.284 0.224 0.035 0.223 E_coli 0.249 0.162 -0.353 0.307 -0.144 0.234 Secchi_TSI 0.303 0.052 0.206 0.121 -1.000 0.332 TP_TSI -0.003 0.406 -0.543 0.370 0.038 0.206 ChlA_TSI 0.295 -0.133 -0.369 0.586 -0.277 0.304 Mean_TSI 0.213 0.225 -0.414 0.459 -0.359 0.332

Variable WaveHt Total P ChlA Turbidity PhycoCyn E_coli WaveHt 1.000 Total P 0.264 1.000 ChlA 0.254 0.431 1.000 Turbidity 0.252 0.038 0.249 1.000 PhycoCyn 0.096 0.258 0.268 0.059 1.000 E_coli -0.007 0.417 0.415 -0.023 0.174 1.000 SD_TSI 0.241 -0.038 0.277 0.510 -0.035 0.144 TP_TSI 0.264 1.000 0.431 0.038 0.258 0.417 ChlA_TSI 0.254 0.431 1.000 0.249 0.268 0.415 Mean_TSI 0.386 0.801 0.700 0.286 0.106 0.446

Variable SD_TSI TP_TSI ChlATSI MeanTSI SD_TSI 1.000 TP_TSI -0.038 1.000 ChlATSI 0.277 0.431 1.000 MeanTSI 0.359 0.801 0.700 1.000 185

Table C.9. Spearman‟s rank correlation coefficient for selected Buck Creek water quality variables.

Variable Temp SpCond D.O. pH Secchi_cm WaveLen Temp 1.000 SpCond -0.124 1.000 D.O. -0.194 -0.150 1.000 pH 0.221 -0.250 -0.316 1.000 Secchi_cm -0.143 0.348 -0.327 -0.144 1.000 WaveLen 0.263 -0.033 -0.398 0.194 -0.190 1.000 WaveHt 0.123 0.087 -0.281 0.000 -0.157 0.881 Total P -0.323 -0.165 -0.105 0.142 0.023 0.352 ChlA -0.265 -0.419 0.574 -0.082 -0.453 -0.115 Turbidity -0.200 -0.326 0.752 -0.133 -0.648 -0.086 PhycoCyn -0.521 -0.048 0.524 -0.424 -0.222 -0.233 E_coli 0.214 -0.043 -0.224 0.264 -0.258 0.276 Secchi_TSI 0.143 -0.348 0.327 0.144 -1.000 0.190 TP_TSI -0.323 -0.165 -0.105 0.142 0.023 0.352 ChlA_TSI -0.265 -0.419 0.574 -0.082 -0.453 -0.115 Mean_TSI -0.323 -0.376 0.378 0.092 -0.492 0.167

Variable WaveHt Total P ChlA Turbidity PhycoCyn E_coli WaveHt 1.000 Total P 0.281 1.000 ChlA -0.149 0.069 1.000 Turbidity -0.071 0.098 0.760 1.000 PhycoCyn -0.033 0.089 0.545 0.560 1.000 E_coli 0.097 -0.104 0.192 0.136 -0.057 1.000 SD_TSI 0.157 -0.023 0.453 0.648 0.222 0.258 TP_TSI 0.281 1.000 0.069 0.098 0.089 -0.104 ChlA_TSI -0.149 0.069 1.000 0.760 0.545 0.192 Mean_TSI 0.116 0.692 0.659 0.631 0.328 0.055

Variable SD_TSI TP_TSI ChlATSI MeanTSI SD_TSI 1.000 TP_TSI -0.023 1.000 ChlATSI 0.453 0.069 1.000 MeanTSI 0.492 0.692 0.659 1.000

186

Table C.10. Spearman‟s rank correlation coefficient for selected Deer Creek water quality variables.

Variable Temp SpCond D.O. pH Secchi_cm WaveLen Temp 1.000 SpCond -0.208 1.000 D.O. -0.205 0.476 1.000 pH 0.381 -0.567 -0.600 1.000 Secchi_cm -0.561 0.475 0.490 -0.663 1.000 WaveLen 0.591 -0.042 -0.261 0.295 -0.419 1.000 WaveHt 0.509 0.117 -0.127 0.220 -0.268 0.779 Total P 0.175 -0.398 -0.600 0.429 -0.312 0.061 ChlA 0.300 -0.509 -0.652 0.744 -0.621 0.246 Turbidity 0.536 -0.471 -0.476 0.731 -0.819 0.502 PhycoCyn 0.397 -0.483 -0.595 0.727 -0.741 0.311 E_coli 0.245 -0.241 -0.041 0.117 -0.371 0.197 Secchi_TSI 0.561 -0.475 -0.490 0.663 -1.000 0.419 TP_TSI 0.175 -0.398 -0.600 0.429 -0.312 0.061 ChlA_TSI 0.300 -0.509 -0.652 0.744 -0.621 0.246 Mean_TSI 0.391 -0.506 -0.722 0.763 -0.721 0.267

Variable WaveHt Total P ChlA Turbidity PhycoCyn E_coli WaveHt 1.000 Total P 0.122 1.000 ChlA 0.239 0.493 1.000 Turbidity 0.399 0.376 0.689 1.000 PhycoCyn 0.178 0.517 0.791 0.746 1.000 E_coli 0.138 0.135 0.064 0.276 0.229 1.000 SD_TSI 0.268 0.312 0.621 0.819 0.741 0.371 TP_TSI 0.122 1.000 0.493 0.376 0.517 0.135 ChlA_TSI 0.239 0.493 1.000 0.689 0.791 0.064 Mean_TSI 0.254 0.757 0.866 0.741 0.884 0.303

Variable SD_TSI TP_TSI ChlATSI MeanTSI SD_TSI 1.000 TP_TSI 0.312 1.000 ChlATSI 0.621 0.493 1.000 MeanTSI 0.721 0.757 0.866 1.000

187

Table C.11. Spearman‟s rank correlation coefficient for selected Delaware Lake water quality variables.

Variable Temp SpCond D.O. pH Secchi_cm WaveLen Temp 1.000 SpCond -0.401 1.000 D.O. 0.077 -0.040 1.000 pH 0.204 -0.252 -0.060 1.000 Secchi_cm -0.223 0.468 0.521 -0.151 1.000 WaveLen 0.183 -0.030 -0.359 0.041 -0.308 1.000 WaveHt 0.054 0.100 -0.400 -0.116 -0.114 0.855 Total P 0.003 -0.501 -0.304 0.187 -0.660 0.238 ChlA 0.120 -0.409 0.083 0.276 -0.251 0.121 Turbidity 0.101 -0.286 -0.528 0.245 -0.904 0.208 PhycoCyn 0.151 -0.494 -0.324 0.369 -0.684 0.266 E_coli 0.357 -0.241 0.336 0.152 -0.156 -0.276 Secchi_TSI 0.223 -0.468 -0.521 0.151 -1.000 0.308 TP_TSI 0.003 -0.501 -0.304 0.187 -0.660 0.238 ChlA_TSI 0.120 -0.409 0.083 0.276 -0.251 0.121 Mean_TSI 0.222 -0.600 -0.295 0.307 -0.765 0.342

Variable WaveHt Total P ChlA Turbidity PhycoCyn E_coli WaveHt 1.000 Total P 0.031 1.000 ChlA 0.020 0.417 1.000 Turbidity 0.021 0.643 0.435 1.000 PhycoCyn 0.104 0.929 0.461 0.688 1.000 E_coli -0.422 0.108 0.053 0.117 0.106 1.000 SD_TSI 0.114 0.660 0.251 0.904 0.684 0.156 TP_TSI 0.031 1.000 0.417 0.643 0.929 0.108 ChlA_TSI 0.020 0.417 1.000 0.435 0.461 0.053 Mean_TSI 0.109 0.876 0.690 0.778 0.876 0.200

Variable SD_TSI TP_TSI ChlATSI MeanTSI SD_TSI 1.000 TP_TSI 0.660 1.000 ChlATSI 0.251 0.417 1.000 MeanTSI 0.765 0.876 0.690 1.000

188

Table C.12. Spearman‟s rank correlation coefficient for selected East Fork water quality variables.

Variable Temp SpCond D.O. pH Secchi_cm WaveLen Temp 1.000 SpCond -0.372 1.000 D.O. -0.135 0.477 1.000 pH 0.382 -0.301 0.213 1.000 Secchi_cm -0.294 0.035 0.237 -0.086 1.000 WaveLen 0.222 -0.063 0.110 0.285 -0.205 1.000 WaveHt 0.299 0.018 0.080 0.397 -0.207 0.879 Total P 0.043 0.201 0.074 0.283 -0.694 0.334 ChlA -0.466 0.320 0.278 -0.152 -0.188 -0.024 Turbidity 0.300 0.011 -0.105 0.106 -0.912 0.315 PhycoCyn 0.171 0.254 0.328 0.024 -0.153 -0.327 E_coli -0.077 -0.160 -0.194 0.051 -0.492 0.452 Secchi_TSI 0.294 -0.035 -0.237 0.086 -1.000 0.205 TP_TSI 0.043 0.201 0.074 0.283 -0.694 0.334 ChlA_TSI -0.466 0.320 0.278 -0.152 -0.188 -0.024 Mean_TSI -0.081 0.238 0.011 0.024 -0.826 0.165

Variable WaveHt Total P ChlA Turbidity PhycoCyn E_coli WaveHt 1.000 Total P 0.323 1.000 ChlA -0.140 0.276 1.000 Turbidity 0.281 0.750 0.146 1.000 PhycoCyn -0.210 -0.042 0.255 0.114 1.000 E_coli 0.207 0.540 0.174 0.468 -0.420 1.000 SD_TSI 0.207 0.694 0.188 0.912 0.153 0.492 TP_TSI 0.323 1.000 0.276 0.750 -0.042 0.540 ChlA_TSI -0.140 0.276 1.000 0.146 0.255 0.174 Mean_TSI 0.116 0.853 0.580 0.780 0.141 0.515

Variable SD_TSI TP_TSI ChlATSI MeanTSI SD_TSI 1.000 TP_TSI 0.694 1.000 ChlATSI 0.188 0.276 1.000 MeanTSI 0.826 0.853 0.580 1.000

189

Table C.13. Spearman‟s rank correlation coefficient for selected Madison Lake water quality variables.

Variable Temp SpCond D.O. pH Secchi_cm WaveLen Temp 1.000 SpCond -0.095 1.000 D.O. 0.278 -0.156 1.000 pH 0.548 0.106 -0.208 1.000 Secchi_cm -0.210 -0.298 0.025 0.085 1.000 WaveLen 0.339 0.058 0.175 0.191 0.205 1.000 WaveHt 0.258 0.116 0.200 0.057 0.180 0.923 Total P -0.032 0.261 -0.370 0.194 -0.068 0.056 ChlA -0.204 0.266 -0.370 -0.132 -0.265 0.158 Turbidity -0.039 0.002 -0.099 -0.254 -0.361 0.103 PhycoCyn 0.019 0.225 -0.410 0.032 -0.469 -0.022 E_coli -0.064 0.010 0.077 -0.311 -0.194 0.007 Secchi_TSI 0.210 0.298 -0.025 -0.085 -1.000 -0.205 TP_TSI -0.032 0.261 -0.370 0.194 -0.068 0.056 ChlA_TSI -0.204 0.266 -0.370 -0.132 -0.265 0.158 Mean_TSI -0.069 0.344 -0.423 0.026 -0.411 0.021

Variable WaveHt Total P ChlA Turbidity PhycoCyn E_coli WaveHt 1.000 Total P 0.106 1.000 ChlA 0.300 0.554 1.000 Turbidity 0.077 0.540 0.497 1.000 PhycoCyn 0.076 0.598 0.620 0.358 1.000 E_coli -0.085 -0.318 -0.120 0.182 -0.213 1.000 SD_TSI -0.180 0.068 0.265 0.361 0.469 0.194 TP_TSI 0.106 1.000 0.554 0.540 0.598 -0.318 ChlA_TSI 0.300 0.554 1.000 0.497 0.620 -0.120 Mean_TSI 0.126 0.837 0.822 0.641 0.705 -0.227

Variable SD_TSI TP_TSI ChlATSI MeanTSI SD_TSI 1.000 TP_TSI 0.068 1.000 ChlATSI 0.265 0.554 1.000 MeanTSI 0.411 0.837 0.822 1.000

190

Table C.14. Spearman‟s rank correlation coefficient for selected Lake Logan water quality variables.

Variable Temp SpCond D.O. pH Secchi_cm WaveLen Temp 1.000 SpCond -0.285 1.000 D.O. -0.081 0.354 1.000 pH 0.058 -0.103 0.078 1.000 Secchi_cm -0.045 0.063 -0.095 -0.093 1.000 WaveLen 0.277 -0.052 -0.130 0.169 0.098 1.000 WaveHt 0.135 -0.044 -0.308 0.191 0.118 0.876 Total P -0.131 -0.222 -0.181 0.216 -0.495 0.122 ChlA -0.190 0.296 -0.195 -0.552 -0.178 0.121 Turbidity 0.133 -0.152 -0.135 -0.030 -0.717 0.094 PhycoCyn -0.216 -0.215 0.376 0.168 -0.261 -0.071 E_coli 0.056 -0.140 0.210 -0.231 0.069 -0.351 Secchi_TSI 0.045 -0.063 0.095 0.093 -1.000 -0.098 TP_TSI -0.131 -0.222 -0.181 0.216 -0.495 0.122 ChlA_TSI -0.190 0.296 -0.195 -0.552 -0.178 0.121 Mean_TSI -0.147 0.115 -0.275 -0.363 -0.529 0.122

Variable WaveHt Total P ChlA Turbidity PhycoCyn E_coli WaveHt 1.000 Total P 0.103 1.000 ChlA 0.216 0.058 1.000 Turbidity 0.106 0.709 0.240 1.000 PhycoCyn -0.188 0.280 -0.273 0.238 1.000 E_coli -0.500 -0.168 -0.281 -0.277 0.263 1.000 SD_TSI -0.118 0.495 0.178 0.717 0.261 -0.069 TP_TSI 0.103 1.000 0.058 0.709 0.280 -0.168 ChlA_TSI 0.216 0.058 1.000 0.240 -0.273 -0.281 Mean_TSI 0.206 0.486 0.841 0.607 -0.100 -0.239

Variable SD_TSI TP_TSI ChlATSI MeanTSI SD_TSI 1.000 TP_TSI 0.495 1.000 ChlATSI 0.178 0.058 1.000 MeanTSI 0.529 0.486 0.841 1.000

191

Figure C.1. Box-and-whiskers plot of in vivo chlorophyll A concentrations ( g/L) by study beach.

192

Figure C.2. Box-and-whiskers plot of total phosphorus concentrations ( g/L) by study beach.

193

Figure C.3. Box-and-whiskers plot of secchi depths (cm) by study beach.

194

Figure C.4. Box-and-whiskers plot of water temperatures (°C) by study beach.

195

Figure C.5. Box-and-whiskers plot of in vivo phycocyanin concentrations ( g/L) by study beach.

196

Figure C.6. Box-and-whiskers plot of daily E. coli densities (CFU/100 mL) by study beach with EPA criterion depicted at 235 CFU/100 mL .

197

Figure C.7. Box-and-whiskers plot of microcystin concentrations ( g/L) measured within range of the ELISA by study beach.

198

Figure C.8. Box-and-whiskers plot of trimetric average trophic state index values by study beach.

199