Introduction to Study Designs and Biostatistics

4/12/2013

WH RESEARCH TRAINING WORKSHOP 2013 Date: Thursdays ,12:30-1:30pm Venue: Auditorium Western Centre for Health Research & Education Sunshine Hospital

Workshop Topic Presenters Date Research Ethics & Governance Dr Tam Nguyen 14-Feb-13 Introduction to Clinical Research Dr Harin Karunajeewa 28-Feb-13 Evaluating the literature A/Professor Kerrie Sanders 14-Mar-13 Writing a research proposal Dr Lizzie Skinner 28-Mar-13 Beginners statistics: Study Design Professor Danny Liew 11-Apr-13 Referencing and EndNote Dr Tam Nguyen & Lynn Higgins 24-Apr-13 Mixed Methods: Quantitative & Qualitative Professor Terrence McCann 9-May-13 Using Excel for research Dr Lizzie Skinner 23-May-13 Making sense of your results Professor Danny Liew 6-Jun-13 Getting your work published A/Professor Kerrie Sanders 20-Jun-13 Writing Abstract for Research Week/ Conferences Dr Debra Kerr 4-Jul-13

Introduction to Study Designs and Biostatistics

Danny Liew

1 4/12/2013

Overview

• overview of study designs

• observational studies

• clinical trials

• basic biostatistics

Classification of Study Designs

observational • [case series, case reports] descriptive • ecological • cross-sectional

• case-control • cohort analytical interventional • clinical trials

2 4/12/2013

Classification of Study Designs

observational • [case series, case reports] non- • ecological longitudinal • cross-sectional • case-control

• cohort longitudinal interventional • clinical trials

Ecological Studies

3 4/12/2013

Ecological Studies

• study of data at population/group level -

no data on individuals

• easily and opportunistically undertaken,

often using routinely collected data

• hypothesis-generating studies

Ecological Study - Hypothetical Example

140

120

100

60 plots of Disease incidence individual countries 40

CANCER 20 20 40 60 80 100 120 140 Average smoking (cig/week) SMOKING

4 4/12/2013

Cross-Sectional Studies

• sample of population selected and information

obtained at one point/period in time

• large studies can take place over years, but each

subject contributes data only once

• that is, there is no follow-up of subjects

5 4/12/2013

Cross-Sectional Studies

• data collected via:

questionnaires ± examinations ± investigations

• mostly descriptive outputs, especially prevalence

eg, of CHD among Australians

Example of Cross-Sectional Study

6 4/12/2013

Case Control Studies

• comparison of previous exposure status between:

–subjects with outcome of interest (cases)

–subjects without outcome of interest (controls)

• controls are often matched with cases, 1:1 or n:1

• matching by confounders - eg: age, sex

7 4/12/2013

Case Control Studies

time

exposure outcome

step 1: define and recruit cases; recruit controls by matching to cases (outcome ascertainment 1st) step 2: determine previous exposure among subjects

Case Control Studies

• explicit knowledge about temporal relationship between exposure and outcome

• useful for studying rare outcomes

• key output: odds ratio, approximation of relative risk of outcome conferred by exposure

8 4/12/2013

Hypothetical Example

Controls: no Cases: Kafoop’s Kafoop’s Syndrome Syndrome

No smoking 200 150

Smoking 100 150

OR = (200*150) / (100*150) = 2.0

Interpretation: smoking doubles likelihood of Kafoop’s Syndrome

Kafoop’s Syndrome

9 4/12/2013

10 4/12/2013

Cohort Studies

• longitudinal, with follow-up of subjects

• collect incidence data

• comparison of outcomes between/among subgroups

eg, not exposed vs exposed to risk factor

• derive relative risks

(recall examples from British Doctor’s Study)

11 4/12/2013

Prospective Cohort Study

time

exposure outcome

Key: explicit knowledge about the temporal relationship between exposure and outcome.

Retrospective Cohort Study

time

exposure outcome

Key: explicit knowledge about the temporal relationship between exposure and outcome.

12 4/12/2013

Cohort Studies

• explicit (often-detailed) knowledge about temporal

relationship between exposure and outcome

• can include multiple exposures and outcomes

• research hypotheses can be addressed post hoc in

established cohorts

The Framingham Heart Study

13 4/12/2013

Framingham Risk Equation

Clinical Trials

14 4/12/2013

“Clinically Proven”

“Is it all a male conspiracy?” The Age 11 July 2002

15 4/12/2013

Clinical Trials

• longitudinal studies designed to assess if an

intervention (removal of exposure) changes

the incidence of an outcome

• most interventions are expected to decrease

the incidence of the outcome

• most involve a control group for comparison

Clinical Trials

intervention A assign intervention placebo / intervention B

prospective follow-up to capture outcomes

16 4/12/2013

Clinical Trials

• ‘gold standard’ for evidence of causality

–active change of exposure status

–tightly controlled study environment

• provides most of the evidence for EBP

17 4/12/2013

Key Outcomes

relative measures of intervention effect: • relative risks • hazard ratios

absolute measures of intervention effect: • absolute risk/rate reduction • number needed to treat

survival analysis

Randomisation

• random allocation of subjects into each

arm of a clinical trial

• objective: treatment groups identical in

all aspects other than the intervention

• rationale: reduce confounding

18 4/12/2013

Confounding

exposure outcome

confounder

Confounding in Clinical Trials

intervention outcome

confounder (age/sex etc...)

19 4/12/2013

JAMA 2002; 288: 321-333.

20 4/12/2013

Basic Biostatistics

Studies and Samples

• studies are undertaken on samples of the population of interest (cf census) • studies are used to make inferences about the population of interest • biostatistics is concerned with the extent to which study (sample) results reflect the ‘truth’

21 4/12/2013

p value

• probability of the study result if it is assumed that the null hypothesis applies - truly no difference between the groups being compared • ie, probability that the study result was a chance finding • p value = conventional cut-off = 0.05 p < 0.05: statistically significant p ≥ 0.05: not statistically significant

p = 0.02 p = 0.01

JAMA 2002; 288: 321-333.

22 4/12/2013

95% Confidence Interval

• interval within which there is 95% confidence that the ‘true’ value lies • if the null value is excluded, result is stat significant

• null value: value if the null hypothesis applies • null value: 1.0 for ratios (eg HR, RR, OR) and 0 for differences (eg absolute risk differences)

JAMA 2002; 288: 321-333.

23 4/12/2013

WH RESEARCH TRAINING WORKSHOP 2013 Date: Thursdays ,12:30-1:30pm Venue: Auditorium Western Centre for Health Research & Education Sunshine Hospital