INFECTION CONTROL & HOSPITAL EPIDEMIOLOGY MARCH 2018, VOL. 39, NO. 3

ORIGINAL ARTICLE

Patient, Provider, and Practice Characteristics Associated with Inappropriate Antimicrobial Prescribing in Ambulatory Practices

Monica L. Schmidt, MPH, PhD;1 Melanie D. Spencer, PhD, MBA; 1 Lisa E. Davidson, MD'

OBJECTIVE. To reduce inappropriate antimicrobial prescribing across ambulatory care, understanding the patient-, provider-, and practice­ level characteristics associated with antibiotic prescribing is essential. In this study, we aimed to elucidate factors associated with inappropriate antimicrobial prescribing across urgent care, family medicine, and pediatric and internal medicine ambulatory practices.

DESIGN, SETTING, AND PARTICIPANTS. Data for this retrospective cohort study were collected from outpatient visits for common upper respiratory conditions that should not require antibiotics. The cohort included 448,990 visits between January 2014 and May 2016. Carolinas HealthCare System urgent care, family medicine, internal medicine and pediatric practices were included across 898 providers and 246 practices.

METHODS. Prescribing rates were reported per 1,000 visits. Indications were defined using the International Classification of Disease, Ninth and Tenth Revisions, Clinical Modification (ICD-9/10-CM) criteria. In multivariable models, the risk of receiving an antibiotic prescription was reported with adjustment for practice, provider, and patient characteristics.

RES UL TS. The overall prescribing rate in the study cohort was 407 per 1,000 visits (95o/o confidence interval [CI], 405-408). After adjustment, adult patients seen by an advanced practice practitioner were 15% more likely to receive an antimicrobial than those seen by a physician provider (incident risk ratio [IRR], 1.15; 95% CI, 1.03-1.29). In the pediatric sample, older providers were 4 times more likely to prescribe an antimicrobial than providers aged S30 years (IRR, 4.21; 95°/o CI, 2.96-5.97).

CONCLUSIONS. Our results suggest that patient, practice, and provider characteristics are associated with inappropriate antimicrobial prescribing. Future research should target antibiotic stewardship programs to specific patient and provider populations to reduce inappropriate prescribing compared to a "one size fits all" approach. Infect Control Hosp Epidemiol 20 l 8;39:307-315

Nearly 47 million unnecessary antibiotic prescriptions are Few studies have included practice types, provider, and patient written each year in the outpatient setting. 1 Indications such as characteristics to determine their impact on antimicrobial viral upper respiratory infections, acute bronchitis, and prescribing for common indications in the ambulatory bronchiolitis have clear guidelines that do not support the use care space. 2 3 of antibiotics. ' Overuse of antibiotics has been the primary Understanding characteristics that influence prescribing driver for increasing prevalence of multidrug-resistant rates across different environments, providers, and patients bacterial infections that affect vulnerable populations and will inform strategies for effective antimicrobial stewardship contribute to increased mortality.4-7 and improve patient care in the outpatient setting. The goal To combat increasing resistance to available antibiotics, the of this study was to identify patient, provider and practice White House released a National Action Plan in 2015 that set a characteristics that may contribute to inappropriate antibiotic goal of reducing inappropriate outpatient antibiotic use by prescribing. We targeted 4 clinical conditions where anti­ 50o/o by 2020. 8 Since that time, several studies have been microbials are not indicated: acute upper respiratory infection published that describe baseline prescribing rates in outpatient (URI), acute bronchitis, bronchiolitis, and nonsuppurative 9 14 2 15 17 practices. - Many of these reports focused on a single out­ otitis media. ' - Several patient factors were included: patient setting, such as primary care, or used national data indication for the visit, age, race, gender, Charlson comor­ such as the National Ambulatory Health Care Data or the bidity index (CCI), and average number of visits per patient National Ambulatory Medical Care Survey for their analysis. 12 during the analysis period. At the practice level, characteristics

Affiliations: I. Center for Outcomes Research and Evaluation, Carolinas HealthCare System, Charlotte, North Carolina; 2. Division of Infectious Disease, Carolinas Medical Center, Carolinas HealthCare System, Charlotte, North Carolina. Received May 25, 2017; accepted November 2, 2017; electronically published January 30, 2018 © 2018 by The Society for Healthcare Epidemiology of America. All rights reserved. 0899-823X/2018/3903-0009. DOI: 10. l 017/ice.2017 .263

Downloaded from hnps:/iW'N'N cambridgP..orglu:ire. 26 Oct 2020 at 01:43:46, subject to the Cambridge Core terms of use. 308 INFECTION CONTROL & HOSPITAL EPIDEMIOLOGY MARCH 2018, VOL. 39, NO. 3

included practice type, rural versus urban setting, and year of 448,990 Visits visit. Finally, provider level factors included age of provider I and provider type. Internal Urgent care Famlly Medicine Pediatrics Medicine 76,979 137,807 153,186 81,018

METHODS URI URI URI URI Data Description 54,357 84,324 57,155 122,632

Carolinas HealthCare System (CHS) is a large, integrated Bronchitis Bronchitis Bronchitis Bronchitis network of acute-care hospitals, ambulatory care, urgent care, 19,483 47,063 21,645 16,156 free-standing emergency departments and skilled nursing

facilities. Between 2014 and 2016, the average annual number Bronchiolitis Bronchiolitis Bronchiolitis Bronchiolitis of outpatient visits for common upper respiratory infections 494 1,057 222 ~ 7,006 was 377,617 (95% confidence interval [CI], 376,970-378,264). As an integrated system, sites across the care continuum share NS·Otitis media NS·Otitis media NS·Otitis media NS-Otitis media 2,645 5,363 1,966 7,392 the same electronic medical record (EMR) system where pre­ scriptions and primary, secondary, and tertiary diagnoses for

each outpatient visit are documented. These EMR data are FIG URE 1. Number of patient visits by practice type and diagnosis. captured and updated daily in our data warehouse, which is used for research and quality monitoring purposes. Using this validated data source, ambulatory visits between providers (APP) that included nurse providers and physician January!, 2014, and May 31, 2016, were extracted for this study. assistants or (2) physicians holding a medical doctor or doctor of These data included any outpatient visit where the patient had osteopath degree. The age of the provider at the time of the visit any of the following diagnoses (International Classification of was also captured. Practice-level factors included the year the visit Disease, Ninth and Tenth Revisions, Clinical Modification [ICD-9/ occurred and type ofpractice (internal medicine, family medicine, 9 12 10-CM] diagnostic categories ' ): acute bronchitis, bronchioli­ urgent care, or pediatrics). The year ofthe visit was included in the tis, nonsuppurative otitis media, or viral URI (Supplementary analyses to control for practice-level changes that were not directly 23 12 18 20 Table l). • • • - These diagnoses were selected to reflect captured in our model but occurred over time. For example, new guidelines indicating that antibiotic prescribing is not appro­ policies may have been implemented or diagnosis shifting may priate. 2'15-17 There was no overlap between acute bronchitis, have occurred that could impact prescribing trends. By including bronchiolitis, and URI. For our study, URI included pharyngitis, the calendar year in the model, we adjusted for these changes in nasopharyngitis, acute supraglottis, acute epiglottitis, cough, and prescribing rates due to policy changes. acute tracheitis. Bronchitis and bronchiolitis were specified by acute codes for these indications. Statistical Analyses Only oral antibiotic prescriptions were included for this study. The classes of antibiotics were categorized as follows in All analyses were performed using Stata software version 14.0 order of frequency: macrolides, penicillins, cephalosporins, (StatCorp, College Station, TX). Standardized prescribing rates quinolones, and other less frequently prescribed antibiotics (ie, were calculated by obtaining the total number of visits during tetracyclines or lincomycin derivatives). We captured the route which an antimicrobial was prescribed, divided by the total of delivery for the drug prescribed in our data and included number ofvisits for each ofthe 5 indications, then multiplied by only oral prescriptions written during the outpatient visit. 1,000. Rates were reported across indications by patient demo­ A total of 448,990 visits were extracted for study inclusion, graphics, provider characteristics and practice type (Table 1). which involved 281,315 unique patients seen across 246 practices To adjust for factors that could confound prescribing rates and 898 providers. We included urgent care, family medicine, between visits within a single provider, we constructed 2 internal medicine, and pediatrics practices, and we extracted multivariable models using Poisson regression with robust only visits in which there was any diagnosis of URI, acute errors clustered on providers with incident risk ratios repor­ bronchitis, bronchiolitis or nonsuppurative otitis media ted.21-23 The measured outcome was whether an antibiotic was (Figure 1). The primary outcome of interest was visit-level prescribed at the visit (yes/no). Two models were built to antibiotic prescribing. Prescribing an antibiotic was defined as a separate pediatric patients (ages (}-19 years) from adult visit where ;?:l antibiotic prescription was written. Prescribing patients (20-65 years and older) given the systematic differ­ rates were standardized per 1,000 visits by indication. ences in the care of these 2 populations. Incident risk ratios Data extracted to support the analyses included several (IRRs) were reported to facilitate ease of interpreting the risk patient-level characteristics: age, race, gender, Charlson comor­ of receiving an antibiotic. Patient, provider, and practice bidity index at the time of the visit, and primary insurance type. factors chosen for the models were selected from literature 3 2 26 Providers were divided into 2 categories: ( 1) advanced practice review and guidelines. ' 4-- Once the factors were selected,

Downloaded from hrtps://www.c:ambridgP..org/(0re. 26 Oct 2020 at 01:43:46, subject to the Cambridge Core terms of use. INAPPROPRIATE ANTIMICROBIAL PRESCRIBING 309

they were not dropped from our models and the estimates (95% CI, 405-408) (Table I). In the unadjusted analysis, the reported. Clustered errors, using providers as the unit of highest rate of inappropriate prescribing was for acute bron­ clustering) adjusted for similarities in prescribing practices chitis at 703 prescriptions per 1,000 visits (95% Cl, 700-706) within individual providers. 24 (Table 1). Family medicine practices had more visits for acute bronchitis than other practice types, while pediatric practices had the most visits for URI (supplementary table 3). For visits RESULTS with a bronchitis diagnosis for which an antibiotic prescription The overall prescribing rate for the 4 indications evaluated was written, the most frequently prescribed antibiotic classes (adults and pediatrics) was 407 prescriptions per 1,000 visits were macrolides (59.9%) followed by penicillins (17.2%),

TABLE L Antimicrobial Prescribing Rates per 1,000 Visits by Indication"­ Mean Antimicrobial Prescriptions per 1,000 Visits (95o/o CI)

Nonsuppurative Overall Viral URI Bronchitis Bronchiolitis Otitis lvledia Indications Practice Type Urgent care 381 (377-385) 739 (733-745) 358 (315-400) 384 (366-403) 471 (468-475) Family medicine 376 (373-379) 678 (674-682) 364 (335-393) 274 (262-286) 475 (472-478) Internal medicine 407 (403-411) 651 (645-657) 540 (474-606) 248 (230-267) 469 (465-472) Pediatrics 201 (199-203) 801 (795-807) 259 (249-270) 463 (452-474) 279 (277-282) Provider Level Physician 556 (554-559) 866 ( 863-869) 706 (689-723) 620 (610-631) 657 (655-659) Advanced care provider 843 (838-847) 936 (932-940) 953 (935-971) 840 (821-859) 876 (873-879) Age (pediatric) 0-2 y 190 (187-193) 548 (535-562) 236 (226-245) 474 (460-489) 227 (225-230) 3-9 y 227 (224-230) 857 (850-864) 447 (407-487) 483 (466-500) 319 (316-323) 10-19 y 260 (255-265) 835 (827-843) 726 (657-795) 362 (359-385) 373 (368-377) Age (adult) 20-39 y 426 ( 422-431) 732 (725-739) 546 (466-626) 289 (271-306) 508 (504-512) 40-64 y 440 (436-444) 692 (688-697) 580 (527-632) 255 (241-269) 523 (520-526) ~65 y 401 (396-405) 635 (629-641) 502 ( 436-567) 230 (209-251) 483 ( 480-487) Race White 347 (344-350) 706 (702-711) 296 (279-314) 361 (349-373) 440 (438-443) African American 267 (261-272) 686 (675-697) 241 (213-270) 376 (345-408) 353 (348-358) Asian 179 (164-194) 764 (720-807} 220 (132-309) 397 (288-506) 257 (241-272) Other 214 (207-221) 702 (698-705) 287 (275-299) 346 (306-385) 394 (393-396) Year of Visit 2014 321 (318-323) 704 (700-709) 247 (231-263) 291 (280-303) 411 (408-413) 2015 315 (312-317} 706 (701-710) 302 (288-316) 389 (378-400) 406 ( 404-408) 2016 299 (295-304) 690 ( 683-697} 303 (282-324) 454 (438-470) 396 (392-400) Patient's Insurance Provider at Time of Visit Managed care 325 (322-327) 733 (730-737) 277 (264-290) 350 (341-359) 418 (416-420) Medicaid 240 (237-243) 711 (703-719) 275 (261-289) 487 ( 470-504) 310 (307-313) Medicare 394 (389-399) 626 (620-632) 519 (454-584) 223 (202-245) 477 (473-480) Commercial 343 (334-352) 758 (745-770) 324 (264-384) 368 (332-403) 456 (449-464) Self-pay 344 (333-356) 678 (662-695) 314 (242-386) 450 (398-503) 448 (438-457} Other 281 (245-318) 488 (439-537) 280 (100-459) 363 (157-569) 292 (269-314) Age of Provider ~30 y 99 (87-111) 324 (281-367) 21 (0-51) 295 (179-410) 141 (128-154) 31-40 y 283 (280-286) 635 (628-642) 198 (180-217) 380 (364-395) 356 (353-359) 41-50 y 317 (314-320) 684 ( 680-689) 297 (283-311) 327 (315-339) 401 (398-403) 51-60 y 370 (366-373) 693 (688-698) 374 (352-396) 405 (391-420) 471 (468-474} >60 289 (285-293) 650 (643-656) 255 (228-282) 387 (369-405) 404 (401-408) Overall visits NIA NIA NIA NIA 407 (405-408) Overall indications 315 (313-317) 703 (700-706) 284 (275-294) 368 (361-375) NIA

NOTE. NA, not available. "-Standardization of prescribing rates = (total visits where antimicrobial was prescribed/total visits) x 1,000.

Downloaded from hHps;1iwww_:,irr1bridge.org/core. 26 Oct2020 at 01 :43:46, subject to the Cambridge Core terms of use. 310 INFECTION CONTROL & HOSPITAL EPIDEMIOLOGY MARCH 2018, VOL. 39, NO. 3

quinolones (14.5%), cephalosporins (8.0%), and other (0.4%) all indications for patients aged 0 to 64 years (227-523 per (supplementary Table 1). Across all antibiotic classes, the 3 1,000 visits) (Table 1 pediatric and adult overall rates). Pre­ most frequently prescribed antibiotics were azithromycin scribing rates began to decline for patients aged > 64 years. (46.6%) followed by amoxicillin (18.1%), and amoxicillin­ In the adjusted analyses for the pediatric sample, the risk of clavulanate ( l l.8o/o ). Penicillins were the most frequently receiving an antimicrobial at a visit increased as patient age prescribed antibiotic for nonsuppurative otitis media ( 62.5%) increased (Table 2). For example, patients 3-9 years ofage had and bronchiolitis (51.0°/o). Azithromycin was most frequently a 250/o greater risk of receiving an antimicrobial than those prescribed for URI in adults and for acute bronchitis in aged 0-2 years (IRR, 1.25; 95% CI, 1.19-1.32), and this rate pediatric patients (Supplementary Table 4). increased further for those aged 10-19 years (Table 2). Across all practice types the rates of prescribing were greater African-American and Asian pediatric patients were less likely for APPs compared to physician providers (Table 1). At the than white patients to receive an antibiotic at a visit (Table 2). practice level, family medicine practices had the highest rate of Pediatric patients with commercial insurance plans were lOo/o prescribing, while pediatric practices had the lowest rate. There more likely to receive an antibiotic prescription than those was a 133.4o/o increase in the antibiotic prescribing rate across with managed care plans (IRR, 1.10; 95% Cl, 1.00-1.22).

TABLE 2. Pediatrics Poisson Regression Model with Provider Clustered Errors 950;0 CI

Variable IRR Robust SE Low High P Value Patient Factors Indication for Antimicrobial (ref. upper respiratory infection) Bronchitis 3.32 0.21 2.94 3.76 <.OOla Bronchiolitis 1.38 0.12 1.16 1.63 <.OOla N onsuppurative otitis media 2.13 0.19 1.79 2.54 <.001" Age category (ref. 0-2 y) 3-9 y 1.25 0.03 1.19 1.32 <.OOla 10-19 y 1.31 0.05 1.22 1.41 <.OOla Male (ref. female) 0.99 0.01 0.97 1.01 .218 Race (ref. white) African American 0.86 0.03 0.80 0.92 <.001 a Asian 0.69 0.04 0.61 0.78 <.001 a Other 0.99 0.01 0.96 1.02 .515 Charlson comorbidity score at time of visit 1.02 0.02 0.99 1.05 .247 Payer (ref. managed care) Medicaid 1.04 0.05 0.95 l.15 .385 Commercial 1.10 0.05 1.00 1.22 .049a Self-pay 1.03 0.03 0.97 1.10 .313 Other 0.64 0.09 0.48 0.84 .002" Average number of visits per patient 1.03 0.01 1.02 1.04 <.OOla Practice Factors Practice type (ref. urgent care, pediatrics) Family medicine, pediatrics 0.92 0.08 0.78 1.08 .283 Pediatrics 0.84 0.07 0.70 1.00 .045a Year of visit (ref. 2014) 2015 0.98 0.02 0.93 1.02 .301 2016 0.97 0.04 0.90 1.05 .495 Practice in metropolitan service area 1.19 0.10 1.00 1.40 .046a Provider Factors Provider level (ref. physician) Advanced practice practitioner 1.18 0.16 0.91 1.55 .216 Age of provider (ref. ::;30 y) 31-40 y 2.97 0.52 2.11 4.19 <.OOla 41-50 y 3.61 0.66 2.52 5.18 <.OOla 51-60 y 4.21 0.75 2.96 5.97 <.OOla >60 y 2.96 0.50 2.12 4.13 <.OOla

NOTE. IRR, incident risk ratio. aSignificance level, P <.05.

Downloaded from ;.,n:ps:l/wv1vv.ca'T'l>ridge.orglcore. 26 Oct 2020 at 01 :43:46, subject to the Cambridge Core terms of use. INAPPROPRIATE ANTIMICROBIAL PRESCRIBING 311

However, patients with other methods of payments that inclu­ sample, providers aged 51-60 years at the time of the visit were ded worker's compensation plans, homeless without insurance, 4 times more likely to prescribe an antimicrobial compared to and community grant coverage were 360/o less likely to receive an providers aged S30 years (IRR, 4.21; 95% CI, 2.96-5.97), but antibiotic prescription (IRR, 0.64; 95% CI, 0.48--0.84). the risk began to decrease for providers aged ::?:60 years (IRR, At the practice level, pediatric practices were 16°/o less likely 2.96; 95% CI, 2.12-4.13) (Table 2). to prescribe an antimicrobial than urgent care practices (IRR, In the adult adjusted model, patient age and race were asso­ 0.84; 95% Cl, 0.70-1.00). Practices seeing pediatric patients ciated with prescribing (Table 3). Patients aged 40-64 years were and residing in an urban setting prescribed more antibiotics 4o/o more likely to receive an antibiotic than patients aged 20-39 than those in rural settings (Table 2). years (IRR, 1.04; 95% CI, 1.02-1.05). All other races were less Among providers, the risk of a patient receiving an anti­ likely to receive an antibiotic than white patients (Table 3). biotic increased as the provider's age increased up to age 61 Patients with Medicaid, Medicare or other payment methods across all indications (Table 1). When adjusted in the pediatric were also less likely to receive an antimicrobial compared to

TABLE 3. Adults Poisson Regression Model with Provider Clustered Errors 950/o CI

Variable IRR Robust SE Low High ?Value Patient Factors Indication for antimicrobial {ref. upper respiratory infection) Bronchitis L60 0.04 1.53 L68 <.001" Bronchiolitis l.30 0.10 l.12 1.52 .001" Nonsuppurative otitis media 0.60 0.03 0.55 0.66 <.OOla Age (ref. 20-39 y) 40-64 y L04 O.Ol L02 L05 <.OOla ~65 y 1.03 0.02 LOO 1.06 .037a Male (ref. female) LOO O.Ol 0.99 1.02 .507 Race (ref. white) African American 0.85 0.02 0.82 0.89 <.OOla Asian 0.85 0.03 0.79 0.92 <.OOla Other 0.95 0.01 0.93 0.97 <.OOla Charlson comorbidity score at time of visit 0.99 0.00 0.98 0.99 <.OOla Payer (ref. managed care) Medicaid 0.91 0.02 0.87 0.96 <.OOla lvledicare 0.95 0.01 0.92 0.97 <.OOla Commercial L05 0.02 I.DO L09 .043a Self-Pay 0.96 0.04 0.88 L06 .435 Other 0.65 0.05 0.57 0.75 <.OOla Average number of visits per patient 0.96 0.01 0.95 0.97 <.OOla Practice Factors Practice type (ref. urgent care, adults) Family medicine, adults 0.96 0.05 0.86 l.07 .458 Internal medicine adults 0.90 0.05 0.80 I.DO .060 Year of visit (ref. 2014) 2015 1.02 0.02 0.99 1.05 .273 2016 I.DO 0.02 0.96 1.05 .988 Practice in metropolitan service area 1.36 0.12 l.15 1.61 <.001" Provider Factors Provider level (ref. physician) Advanced practice practitioner 1.15 O.Q7 1.03 1.29 .014" Age of provider (ref. :530 y) 31-40 y 1.83 0.45 l.13 2.96 .014a 41-50 y 1.88 0.46 1.16 3.03 .OlOa 51-60 y 1.92 0.47 l.19 3.1 I .008" >60 y 1.80 0.45 1.11 2.93 .018a

NOTE. IRR, incident risk ratio. aSignificance level, P <.05.

Downloaded from https:1iwww r:amllridge.orgic0re. 26 Oct 2020 at 01 :43:46, subject to the Cambridge Core terms of use. 312 INFECTION CON"TROL & HOSPITAL EPIDEMIOLOGY MARCH 2018, VOL. 39, NO. 3

those with managed care plans (Table 3). For adults seen in a Some of our findings were consistent across practice settings. metropolitan area, the risk of receiving an antibiotic was 36°/o For example, acute bronchitis was the most common indication greater than in rural practices (IRR, 1.36; 95% Cl, 1.15-1.61). for which an antibiotic was prescribed. The most common drug However, the type of practice was not associated with anti­ prescribed for bronchitis was azithromycin in both adult and microbial prescribing for adult visits. pediatric populations. Previous studies and our results suggest After adjusting for patient and practice factors, APPs were that patient and provider education on appropriate prescribing lSo/o more likely to prescribe an antibiotic than physician for bronchitis, including guidance on correct use ofazithromycin, 26 3 37 providers (IRR, 1.15; 95% Cl, 1.03-1.29) in adult patients may be an effective way to reduce prescribing rates. ' 4­ (Table 3), but this did not hold true for pediatric visits Our analysis found that patient age was strongly associated (Table 2). The age of the prescribing provider was associated with the rates ofantimicrobial prescribing even after adjusting with an increased risk of prescribing an antibiotic and was for other factors, such as comorbidities, gender, race, and similar in the pediatric sample (Table 3 ). indication. The study results demonstrated that, as patient age increased, the risk of receiving an antimicrobial for any of the 4 indications also rose to age 64, with IRRs increasing by age category in both adult and pediatric models (Tables 2 and 3). DISCUSSION After age 64, the rates declined (Table 3). The underlying Antimicrobial stewardship in the acute inpatient care setting reason for this association could not be not identified in this has demonstrated effectiveness in decreasing inappropriate study. Based on previous studies and on qualitative work in antibiotic utilization over the last 20 years and has been asso­ progress, we hypothesize that working-age patients may pres­ 2 29 ciated with decreasing antimicrobial resistance. 6- Studies have sure providers to prescribe antibiotics based on their mis­ identified multiple interventions that have been part of success­ understanding of which illnesses will improve with antibiotic ful inpatient stewardship, such as front-end restriction or treatment and on their need to return quickly to work and 29 30 38 39 post-prescription review and feedback. • In contrast to the family responsibilities. ' These patient dynamics and the outpatient setting, inpatient stewardship focuses on a somewhat pressure they place on provider decisions need to be better static patient population; patients are admitted for several days understood. In-depth qualitative research could assess this and can be followed over time to evaluate clinical progress, to hypothesis and help elucidate the interactions between provi­ review culture data, and to conduct interventions.31 ders and patients of varying ages that lead to unnecessary Antimicrobial stewardship in the outpatient setting seeks to prescribing. This information is essential to informing effec­ address a population that differs in acuity, microbiologic tive antibiotic stewardship interventions. etiology, and patient characteristics. While outpatients are Many publications have evaluated patient characteristics typically less acutely ill than inpatients, national data demon­ and att1tu. des surroun d'ing anhm1cro. . b'ali prescr1'b' 1ng. 363839' ' strate that the volume of antibiotics used in the outpatient However, we found that provider characteristics may also setting is much greater, with up to 30% of all outpatient impact prescribing rates. A study conducted in urgent care, antibiotic prescriptions deemed unnecessary and up to 50% emergency, and primary care outpatient clinics of the Veterans 20 32 inappropriate for the indication. ' Traditional interventions Affairs Health System found significant variation by provider used in the inpatient setting cannot be practically applied to in prescribing for acute respiratory illnesses, but it did not outpatient stewardship. 33 Because of these differences, the evaluate provider-specific factors, such as age and level of large (and diverse) population of patients that need to be training. 40 Our results showed significant variation based on reached, and limited resources available for outpatient anti­ provider age, with younger providers prescribing fewer inap­ microbial stewardship programs, clear evidence is needed to propriate antibiotics than older providers. determine which patient populations should be targeted for In the present study, we also found higher levels of inap­ interventions that can be easily implemented and will yield the propriate prescribing by APPs compared to other providers for greatest reduction in inappropriate prescribing. the adult patient population. For pediatric patients, higher A good starting point for identifying how to reduce inap­ prescribing rates were not associated with APPs. Our findings propriate outpatient prescribing is to evaluate diagnoses for are similar to other previously published literature which have which antibiotics are rarely, if ever, indicated. We chose 4 also found increased associations with prescribing by APPs, common conditions that do not routinely require antibiotics: especially for acute respiratory tract infections.9 Future acute bronchitis, bronchiolitis, nonsuppurative otitis media, national stewardship efforts should target education and and URI. While these conditions are commonly seen across all antimicrobial stewardship interventions for APPs as their role outpatient practices, our findings demonstrate that variation in patient care continues to grow. 9 Unlike other recent studies, in prescribing patterns exists and is associated with several we found that prescribing rates were also higher in urban patient, practice, and provider characteristics. The breadth of versus rural practice settings, after adjusting for other vari­ this study, which included evaluation of prescribing rates for ables.41 Our patient population is limited to the southeastern more than 448,990 patient visits, is a key strength that sup­ region of the United States, which has been well documented 20 1 ported the detailed analysis to detect these differences. to have the highest prescribing rates. .4 Further evaluation is

Downloaded from h1':ps:/fv.NvV'J cambrioq?..org/cor~. 26 Oct 2020 at 01 :43:46, subject to the Cambridge Core terms of use. INAPPROPRIATE ANTIMICROBIAL PRESCRIBING 313

needed to better identify socioeconomic factors contributing demonstrated that antibiotics are underprescribed for black 45 46 to prescribing in outpatient population. adults, but the reasons for this are not fully understood. ' We detected variation in prescribing rates by patient race. This study had several limitations. First, while data were In both pediatric and adult samples, white patients received analyzed at the visit level, administrative billing data were used significantly more antibiotics than other races. These study to identify visits where any of the 4 indications were present as results are consistent with those in many other studies, for a diagnosis for a single visit. This strategy assumed that the 42 44 both pediatric and adult populations. - Several studies have antibiotic was given for the indication identified, which may

Bronchiolitis 358 URI 381 Urgent Care Non·suppurative Otitis Media 384 Bronchitis 739

Non·suppurativa Otitis Media 274 Bronchiolitis 36'1 Family Medicine URI 376 Bronchitis 6?8

Non-suppurative Otitis Media 2·18 URI 407 Internal Medicine-Adults Bronchiolitis S.10 Bronchi1is 651

URI 201 Bronchiolitis 259 Pedia1rics Non-suppura1ive Otitis Media 463 Bronchitis 80 ' ' ' 0 200 400 600 800'

FIGURE 2. Prescribing rates per 1000 patient visits by practice type and diagnosis.

Physician tl!liillillllilllll URI Advanced Care Provider lli111111111111JllllliJlllillJlll a·13

Phy5ician llllll••••llllllllllJ ass Bronchitis Advanced Care Provider [llJlllllllllll!IJ!ll'llJ 936

Physician llllJlll!lllilllll! Bronchiolitis Advanced Care Provider 111111111111111111•111111!1 953

Non-suppuratiote Oti1is Media Advanced Care Provider 840

0 200 400 600 800 1,000

FIG URE 3. Prescribing rates per 1000 patient visits by diagnosis and provider type.

Downloaded from hnps://vl\vw.r.ambricg<0.org/core. 26 Oct 2020 at 01 :43:46, subject to the Cambridge Core terms of use. 314 INFECTION CONTROL & HOSPITAL EPIDEMIOLOGY MARCH 2018, VOL. 39, NO. 3

have affected the accuracy of prescribing rates. Although we 3. Harris AM, Hicks LA, Qaseem A, High Value Care Task Force of also recognize limitations when using billing codes to define the American College of P, for the Centers for Disease C, the study cohort, we followed prior strategies used by multiple Prevention. Appropriate antibiotic use for acute respiratory tract 20 47 publications to identify our visits to include in the analysis. ' infection in adults: advice for high-value care from the American Despite these limitations, our study results are consistent with College of Physicians and the Centers for Disease Control and previously published work reporting that acute bronchitis is a Prevention. Ann Intern lvled 2016;164:425-434. 10 8 4. Napolitano LM. Emerging issues in the diagnosis and manage­ frequent indication for antibiotic misuse. ,4 Additionally, ment of infections caused by multi-drug-resistant, gram­ our findings support prior results that identified higher anti­ 9 positive cocci. Surg Infect (Larchmt) 2005;6(Suppl 2):S5-S22. biotic prescribing rates for APPs than for physician providers. 5. Branski LK, Al-Mousawi A, Rivero H, Jeschke MG, Sanford AP, Our study identified variation in patient-, provider-, and Herndon DN. Emerging infections in burns. Surg Infect practice-level factors associated with inappropriate antibiotic (Larchmt) 2009;10:389--397. prescribing. Antibiotics are not recommended for any of the 6. Davoudi A, Najafi N, Alian S, et al. Resistance pattern of anti­ 2 15 16 49 indications selected for inclusion in this study. ' ' , biotics in patient underwent open heart surgery with nosocomial Understanding the factors that impact prescribing is critical infection in north of Iran. Glob JHealth Sci 2015;8:288-297. to determining how to reduce the misuse of antibiotics. A "one 7. Bassetti M, Carnelutti A, Peghin M. Patient specific risk stratifi­ size fits all" approach to antibiotic stewardship interventions cation for antimicrobial resistance and possible treatment stra­ may not be the best strategy to meet aggressive goals for tegies in gram-negative bacterial infections. Expert Rev Anti Infect reducing inappropriate prescribing. This study suggests that Ther 2017;15:55--<55. 8. National action plan for combating antibiotic-resistant bacteria. opportunities exist to tailor interventions to specific settings of The White House website. https://www.whitehouse.gov/sites/ care, provider types, and patient characteristics that could be default/files/docs/national_action_plan_for_combating_antibotic­ more effective and efficient in improving appropriate pre­ resistant_bacteria.pdf. Published 2015. Accessed October 20, 2016. scribing and, ultimately, in reducing antibiotic resistance. 9. Sanchez GV, Hersh AL, Shapiro DJ, Cawley JF, Hicks LA. Outpatient antibiotic prescribing among United States nurse ACKNOWLEDGMENTS practitioners and physician assistants. Open Forum Infect Dis 2016;3:ofwl68. Financial support: Support for this study was received from a grant awarded 10. Roberts RM, Hicks LA, Bartoces tvL Variation in US outpatient by The Duke Endowment to extend an existing acute-care antimicrobial antibiotic prescribing quality measures according to health plan stewardship program to the outpatient setting. The Duke Endowment pro­ and geography. Am Manage Care 2016;22:519-523. vided Carolinas HealthCare with funding to collect baseline data, to perform J qualitative research, and to provide education to patients, providers, and 11. Shamsuddin S, Akkawi tvIE, Zaidi ST, Ming LC, Ma nan MM. practices regarding antimicrobial stewardship. This paper reflects information Antimicrobial drug use in primary healthcare clinics: a retro­ collected prior to the roll out of an antimicrobial stewardship program to spective evaluation. Int JInfect Dis 2016;52:16-22. inform and tailor interventions to our ambulatory care setting. The Duke 12. Fleming-Dutra KE, Hersh AL, Shapiro DJ, et al. Prevalence of Endowment did not participate in analysis, preparation, review, or any aspect inappropriate antibiotic prescriptions among US ambulatory care of the research performed at Carolinas HealthCare. visits, 2010-2011. JAMA 2016;315:1864-1873. Potential conflicts ofinterest: All authors report no conflicts of interest rele­ 13. Dyar OJ, Beovic B, Vlahovic-Palcevski V, Verheij T, Pulcini C. vant to this article. How can we improve antibiotic prescribing in primary care? Expert Rev Anti-infect Ther 2016;14:403-413. Address correspondence to Melanie D. Spencer, PhD, MBA, Center for 14. Tamma PD, Cosgrove SE. Addressing the appropriateness of Outcomes Research and Evaluation, 1540 Garden Terrace Suite 406, Carolinas outpatient antibiotic prescribing in the United States: an impor­ HealthCare System, Charlotte, North Carolina 28203 (melanie.spencer@ tant first step. JAMA 2016;315:1839-1841. carolinashealthcare.org). 15. American Academy of Family Physicians; American Academy of Otolaryngology-Head and Neck Surgery; American Academy of SUPPLEMENTARY MATERIAL Pediatrics Subcommittee on Otitis tviedia With Effusion. Otitis To view supplementary material for this article, please visit media with effusion. Pediatrics 2004;113:1412-1429. https://doi.org/ I 0.1017/ice.2017.263 16. Ralston SL, Lieberthal AS, ivleissner HC, et al. Clinical practice guideline: the diagnosis, management, and prevention of bronchiolitis. pediatrics. 2014; 134:el 474--el502. Pediatrics 2015; 136:782. REFERENCES 17. Hersh AL, Jackson ivlA, Hicks LA, American Academy of Pedia­ 1. Talkington K, Hyun D, Zetts R, Kothari P. Antibiotic use in trics Committee on Infectious Diseases. Principles of judicious outpatient settings. The Pew Charitable Trusts website. http:// antibiotic prescribing for upper respiratory tract infections in www.pewtrusts.org/ ~ /media/assets/2016/05/antibioticuseinout pediatrics. Pediatrics 2013;132: 1146-1154. patientsettings.pdf; Published May 2016. Accessed November 5, 18. Snow V, Mottur-Pilson C, Gonzales R, et al. Principles of 2017. appropriate antibiotic use for treatment of acute bronchitis 2. Snow V, Mottur-Pilson C, Gonzales R, et al. Principles of appro­ in adults. Ann Intern Med 2001;134:518-520. priate antibiotic use for treatment of nonspecific upper respiratory 19. Subcommittee on Urinary Tract Infection, Steering Committee tract infections in adults. Ann Intern Med 2001;134:487-489. on Quality Improvement and Management, Roberts KB. Urinary

Downloaded from :nq;~;:/iv·!':,w cambridg;o,nrgicnre. 26 Oct 2020 at 01:43:45, subject to the Cambridge Core terms of use. INAPPROPRIATE ANTIMICROBIAL PRESCRIBING 315

tract infection: clinical practice guideline for the diagnosis and better clinical outcomes: a randomized, double-blinded, and management of the initial UTI in febrile infants and children 2 to placebo-controlled clinical trial.] Pediatr 2012;161: 1104-1108. 24 months. Pediatrics 2011;128:595-610. 36. Kuzujanakis M, Kleinman K, Rifas-Shiman S, Finkelstein TA. 20. Fleming-Dutra KE, Hersh AL, Shapiro DJ, et al. Prevalence of Correlates of parental antibiotic knowledge, demand, and inappropriate antibiotic prescriptions among US ambulatory care reported use. Ambul Pediatr 2003;3:203-210. visits, 2010-2011. JAMA 2016;315:1864-1873. 37. Altiner A, Brockmann S, Sielk M, Wilm S, Wegscheider K, 21. Cameron AC, Pravin K. Microeconomics Using Stata. College Abholz HH. Reducing antibiotic prescriptions for acute cough by Station, TX: Stata Press; 2010. motivating GPs to change their attitudes to communication and 22. Zou GY, Donner A. Extension of the modified Poisson regression empowering patients: a cluster-randomized intervention study. model to prospective studies with correlated binary data. Statist JAntimicrob Chemother 2007;60:638-644. Method Med Res 2013;22:661-670. 38. Holmes JH, Metlay J, Holmes WC, Mikanatha N. Developing a 23. Zou G. A modified poisson regression approach to prospective patient intervention to reduce antibiotic overuse. AMIA Annu studies with binary data. Am JEpidemiol 2004;159:702-706. Symp Proc 2003:864. 24. NicKay R, Mah A, Law MR, NicGrail K, Patrick DNI. Systematic 39. Pechere JC. Patients' interviews and misuse of antibiotics. Clin review of factors associated with antibiotic prescribing for Infect Dis 2001;33(5uppl 3):5170-5173. respiratory tract infections. Antimicrob Agents Chemother 40. Tones BE, Sauer B, Jones MM, et al. Variation in outpatient 2016;60:4106-4118. antibiotic prescribing for acute respiratory infections in the 25. Ivanovska V, Hek K, Mantel Teeuwisse AK, Leufkens HG, Nielen veteran population: a cross-sectional study. Ann Intern Med MM, van Dijk L. Antibiotic prescribing for children in primary 2015;163:73-80. care and adherence to treatment guidelines. J Antimicrob 41. Antibiotic prescription fill rates declining in the United States. In: Chemother 2016;71: 1707-1714. The Health of America Report. BlueCross BlueShield website. 26. Ubhi H, Patel M, Ludwig L. How well do outpatient prescriptions h ttps://www.bcbs.com/ the-health-o f-america/reports/ antibiotic­ adhere to good antimicrobial stewardship? Arch Dis Child 2016; prescription-rates-declining-in-the-US. Published August 2017. 10l:e2. Accessed November 5, 2017. 27. Drekonja DM, Filice GA, Greer N, et al. Antimicrobial steward­ 42. Gerber JS, Prasad PA, Localio AR, et al. Racial differences in ship in outpatient settings: a systematic review. Infect Control antibiotic prescribing by primary care pediatricians. Pediatrics Hosp Epidemiol 2015;36:142-152. 2013;131:677-684. 28. Sanchez GV, Fleming-Dutra KE, Roberts RM, Hicks LA. Core 43. Gahbauer AM, Gonzales NIL, Guglielmo BT. Patterns of anti­ elements of outpatient antibiotic stewardship. MMWR bacterial use and impact of age, race/ethnicity, and geographic 2016;65:1-12. region on antibacterial use in an outpatient medicaid cohort. 29. Dobson EL, Klepser ME, Pogue TM, et al. Outpatient antibiotic Pharmacotherapy 2014;34:677-685. stewardship: Interventions and opportunities. JAPhA 2017;57: 44. Mangione-Smith R, Elliott MN, Stivers T, McDonald L, 464-473. Heritage J, McGlynn EA. Racial/ethnic variation in parent 30. Drekonja D, Filice G, Greer N, et al. Antimicrobial Stewardship expectations for antibiotics: implications for public health cam­ Programs in Outpatient Settings: A Systematic Review. Washington, paigns. Pediatrics 2004;113:e385-e394. DC: Department of Veterans Affairs; 2014. 45. Kornblith AE, Fahimi J, Kanzaria HK, Wang RC. Predictors for 31. Pollack LA, Srinivasan A. Core elements of hospital antibiotic under-prescribing antibiotics in children with respiratory infec­ stewardship programs from the Centers for Disease Control and tions requiring antibiotics. Am J Emerg Med 2017 Tul 28. pii: Prevention. Clin Infect Dis 2014;59(Suppl 3):S97-S100. 50735-6757(17)30632-0. doi: 10.1016/j.ajem.2017.07.081. 32. Centers for Disease Control and Prevention. Office-related anti­ 46. Fleming-Dutra KE, Shapiro DJ, Hicks IA, Gerber JS, Hersh AL. Race, biotic prescribing for persons aged ::;;14 years-United States, otitis media, and antibiotic selection. Pediatrics 2014;134:1059-1066. 1993-1994 to 2007-2008. MMWR 2011;60:1153-1156. 47. Watson TR, Wang L, Klima J, et al. Healthcare claims data: an 33. Gangat NIA, Hsu JL. Antibiotic stewardship: a focus on underutilized tool for pediatric outpatient antimicrobial stew­ ambulatory care. S Dak Med 2015;Spec No:44-48. ardship. Clin Infect Dis 2017;64:1479-1485. 34. Fleming-Dutra KE, Demirjian A, Bartoces NI, Roberts RM, 48. McCullough AR, Pollack AJ, Plejdrup Hansen M, et al. Taylor TH Jr, Hicks LA. Variations in antibiotic and azithromycin Antibiotics for acute respiratory infections in general practice: prescribing for children by geography and specialty-United States, comparison of prescribing rates with guideline recommenda­ 2013. Pediatr Infect Dis J 2017. PMID: 28746259; doi: 10.1097/ tions. lvled] Aust 2017;207:65--69. INF.0000000000001708. 49. Chow AW, Benninger MS, Brook I, et al. IDSA clinical practice 35. Pinto LA, Pitrez PM, Luisi F, et al. Azithromycin therapy in guideline for acute bacterial rhinosinusitis in children and adults. hospitalized infants with acute bronchiolitis is not associated with Clin Infect Dis 2012;54:e72-el 12.

Downloaded from hups:/ivvww carnbridge.rng/core. 26 Oct 2020 at 01 :43:46, subject to the Cambridge Core terms of use. Centers for Disease Control and Prevention

R Morbidity and Mortality Weekly Report Weekly I Vol. 68 I No. 2 January 18, 2019

Opioid Prescribing Rates in Nonmetropolitan and Metropolitan Counties Among Primary Care Providers Using an Electronic Health Record System ­ United States, 2014-2017

Nlacarena C. Garcia, DrPHI; Charles M. Heilig, PhD1; Scott H. Lee, PhDl; J\r1ark Faul, PhD2; Gery Guy, PhD2; lvlichael F. lademarco, MD1; Katherine Hempstead, PhD3; Dorrie Raymond, MA4; Josh Gray, MBA4

Drug overdose is the leading cause of unintentional injury­ trends in opioid prescribing and other areas of public health associated death in the United States. Among 70,237 fatal importance, with minimal lag time under ideal conditions. As drug overdoses in 2017, prescription opioids were involved less densely populated areas appear to indicate both substantial in 17,029 (24.2%) (J). Higher rates of opioid-related deaths progress in decreasing opioid prescribing and ongoing need have been recorded in nonmetropolitan (rural) areas (2). In for reduction, community health care practices and interven­ 2017, 14 rural counties were among the 15 counties with the tion programs must continue to be tailored to community highest opioid prescribing rates.* Higher opioid prescribing characteristics. rates put patients at risk for addiction and overdose (3). Using Athenahealth is a commercial vendor and developer ofcloud­ deidentified data from the Athenahealth electronic health based practice management and EHR systems for physician record (EHR) system, opioid prescribing rates among 31,422 practices and hospitals. Approximately 100,000 health pro­ primary care providers t in the United States were analyzed viders, serving about 86 million patients in the United States, to evaluate trends from January 2014 to March 2017. This use Athenahealth's applications. This retrospective study used analysis assessed how prescribing practices varied among six deidentified Athenahealth EHR prescription data from 31,422 urban-rural classification categories of counties, before and primary health care providers serving approximately 17 million after the March 2016 release ofCDC's Guideline far Prescribing Opioids for Chronic Pain (Guideline) (4). Patients in non­ core (the most rural) counties had an 87% higher chance of INSIDE receiving an opioid prescription compared with persons in 31 Gastroschisis Trends and Ecologic Link to Opioid large central metropolitan counties during the study period. Prescription Rates ­ United States, 2006-2015 Across all six county groups, the odds of receiving an opioid 37 Overdose Deaths Involving Fentanyl and Fentanyl prescription decreased significantly after March 2016. This Analogs ­ New York City, 2000-2017 decrease followed a flat trend during the preceding period in 41 Notes from the Field: Fentanyl Drug Submissions ­ micropolitan and large central metropolitan county groups; in United States, 2010-2017 contrast, the decrease continued previous downward trends in 44 Notes from the Field: Typhoid Fever Outbreak ­ the other four county groups. Data from EHRs can effectively Harare, Zimbabwe, October 2017-February 2018 supplement traditional surveillance methods for monitoring 46 Notes from the Field: Tuberculosis Control in the Aftermath of Hurricane Maria ­ Puerto Rico, 2017

*U.S. Opioid Prescribing Rate Maps. https://v,,ww.cdc.gov/drugoverdose/maps/ 48 QuickStats rxrate-maps.html. t Primary care providers in an ambulatory setting; limited to family medicine, family practice, or general practice, or providers who have an internal medicine specialty with no subspecialty. Nurse practitioners and physician assistants are Continuing Education examination available at included among primary care providers. https://www.cdc.gov/m mwrI cme/conted_i nfo.htm l#weekly.

U.S. Department of Health and Human Services Centers for Disease Control and Prevention IVlorbidity and Mortality Weekly Report patients. Patient-level data were aggregated by week over the CDC's National Center for Health Statistics urban-rural clas­ 166 weeks from January 5, 2014, through March 11, 2017. For sification scheme.~ From most to least densely populated, the each week during which a patient had at least oneAthenahealth six categories include large central metropolitan, large fringe record, that patient contributed one patient-week to this analy­ metropolitan, medium metropolitan, small metropolitan, sis. For each patient-week, it vvas noted whether primary care micropolitan, and noncore counties. providers using Athenahealth's EHR system prescribed one or This analysis includes three components. First, the period­ more opioids {Supplementary Table I, https://stacks.cdc.gov/ speciflc percentage of patients with opioid prescriptions was view/cdc/61743) _§ Percentage ofpatient-weeks during which estimated empirically and with seasonal adjustment using an opioid prescription was written was considered equivalent logistic regression. Second, smooth temporal trends were statis­ to the percentage of patients receiving an opioid prescription tically separated from annual seasonal components using locally during that time. weighted regression (5). Third, to quantify the period-specific For comparisons over time, data were divided into three annual rate ofincrease or decrease in prescribing rates, a second periods. Period I comprises 52 weeks from January 5, 2014, logistic regression model estimated the seasonally adjusted through January 3, 2015; period 2 includes the next 63 weeks, annual percent change (APC) in the odds of receiving an opi­ ending March 19, 2016; and period 3 covers the final 51 weeks, oid prescription. Statistical software was used for all analyses; through March 11, 2017. The first cutpoint allows compari­ statistical tests and confidence intervals (Cis) are presented as sons benveen the first and second years' data, and the second simultaneous procedures adjusted for multiple comparisons. cutpoint supports comparisons before and after the publica­ Overall, 128,194,491 patient-weeks of data are included tion of the CDC Guideline. For comparison by population in the analysis; at least one opioid was prescribed dur­ density, data were stratified by providers' counties according to ing 8,810,237 (6.9%) of these patient-weeks, decreas­ ing from 7.4% during period 1 to 6.4% during period 3

§Short and long acting opioid drugs in this study included buprenorphine, (Table I) (Supplementary Table 2, https://stacks.cdc.gov/view/ butorphanol, codeine, dihydrocodeine, fentanyl, hydrocodone, cdc/61744). Buprenorphine prescribed for pain and opioid hydromorphone, levorphanol, meperidine, methadone, morphine, naltrexone, use disorder treatment represented only 0.02o/o of all opioid nalbuphine, naloxone, oxycodone, oxymorphone, pentazocine, propoxyphene, tapentadol, and tramadol. The study does not count cough and cold medications containing opioids. ~ https://www.alc.gov/nchs/data_access/urban_rutal.htm; https://www.cdc.gov/ nchs/data!series/sr_O2/sr02_166.pdf.

The }.fjtfV(IR series ofpublications is published by the Center for Survdllance, Epide1niology, and Laboratory Services, Centers for Disease Control and Prev.ontion (CDC), U.S. Department of Health and Human Services, Atlantl, GA j{).)29-4027. Suggested citation: [Author names; fh~t three, then et a!., if1nore than six.] [Reporr ride]. NHvIWR }..:forb lvfortal \'V'kly Rep 2019;68:[indusive pagt: numbers]. Centers for Disease Control and Prevention Robert R. Redfield, }vID, Director Anne Schuchat. ivlD. Principal Deputy Dire,:tor Lt0slie Dauphin, PhD, A,:ring A,:wci1lt<' DirectorjOr Sdt11ce Barba1a Ellis, Ph.D, 1\:1$, Act;ng Dfrector, Office ofScience Q"alit)' Chesley L. Rich.trds, MD, l\:fPH, Dep1tty Dircrtor far Pnhlic Health Scient~fic Strvfrcs Mkhael E ladem.uco, lvfD, ivfPH, Di;e,;tor, CenttrjO; Sun·ei.l/,mr:<'. Epidemiolagy. anti Labomto~V Seri'ice> MMWR Editorial and Production Staff (Weekly} Charlone K. Kent, PhD, i'v1PH, A.rti11g Editor in Chief, Ewcutive Editor i'vlarth..1 E Boyd, LMd Visual liifOmltitioJJ j""'peda/ist Jacqueline Gindler, 1\10, Editor Maureen A. Leahy, Julia C. i\farrinroe, 1-iary Oott, MD, iv1PH, Onlb11' Editor Stephen R. Spriggs, Tong Yang, Teresa F. Rutledge, J\fmlrlging Editor Visual Information Speci,di.sts Douglas \V'. \'•(/eatherw.i.x. Lead Technical 'W'"ri.ter-Editor Qu.mg i\t Doan, lviBA, Phyllis H. King, Gl<'nn Damon. Soumya Dunworrh, PhD. Teresa rvr. Hood, 1'.-15. Terraye M. Srn_rr, lvloua Yang, Ttchnicril \'llritcr-Editan fnjOnn,uion Tfchnology Sp<'cittLi;ts MMWR Editorial Board Timothy F. Jones, 1\:iD, Chairman lvlatthew L. Boulron, 1v1D. JY1PH Robin Ikeda, i'viD, l\:iPH Stephen C. Redd, 1\-tD, Virginia A. Caine. 110 Phyllis ivleadows. PhD, J\1SN, RN Patrick L Remington, iVID, lv1PH Katherine Lyon Daniel, PhD Jewel ~:iul!en, LvlD. J'vlPH. MPA Carlos Roig, !v!S, J\:L&.. Jonath,u1 E. Fielding, iv!D, 1vlPl-l, NIBA Jeff Niederdeppe, PhD \'(!illiam Schaffner, IvfD David W. Fk1ning, i'viD P,trricia Quinlisk, tvfD, i\1PH lviorgan Bobb Swanson, BS William E. Halperin, lvlD, DrPH, tv1PH

26 MIVlWR I January 18, 2019 I Vol. 68 I No. 2 US Department of Health and Human Services/Centers for Disease Control and Prevention Morbidity and Mortality Weekly Report

TABLE 1. Number and percentage of patient-weeks with at least one opioid prescription - Athenahealth, United States, January 2014­ March 2017 Percentage receiving opioid prescription No. receiving opioid Urban-rural category* No. of patient-weeks prescription Overall Period 1t Period 2t Period 3t Non core 8,979,403 864,364 9.6 10.3 9.9 9.0 Micropolitan 16,342,824 1,532,747 9.4 9.4 9.6 9.1 Small metro 18,860,569 1,443,246 7.7 8.0 7.7 7.4 Medium metro 32,045,592 2,158,111 6.7 7.3 6.9 6.2 Large fringe metro 31 ,430,958 1,753,802 5.6 6.4 5.8 5.0 Large central metro 20,535,145 1,057,967 5.2 5.4 5.2 5.0 All counties 128,194,491 8,810,237 6.9 7.4 7.0 6.4 *National Center for Health Statistics urban-rural classlfication scheme for counties. https://www.cdc.gov/nchs/data_access/urban_rural.htm. t Period 1: January 5, 2014-January 3, 2015; period 2: January 4, 2015-March 19, 2016; period 3: March 20, 2016-March 11, 2017. Period-specific percentages are based on raw counts rather than statistical models.

prescriptions. By county classification, the overall percentage in period 1, and percentages in period 3 differed significantly ofpatients with opioid prescriptions ranged from 5.2% in large from those in period 2 (p<0.003). central metropolitan counties to 9.6o/o in noncore counties Visual inspection of the prescribing trends by urban­ during the study period. Patients in noncore counties had an rural status and by period revealed patterns in both the 87% higher chance of receiving an opioid prescription than raw (Supplementary Figure I, https://stacks.cdc.gov/view/ did patients in large central metropolitan areas during the cdc/61741) and seasonally adjusted (Supplementary Figure 2, study period. https://stacks.cdc.gov/view/cdc/61742) data. During period 1, The lowest period-specific percentages ofpatient-weeks with before release of the CDC Guideline, the odds of receiving an opioid prescription occurred in large central metropolitan an opioid prescription increased 6.4% per year in noncore counties (5.0%-5.4%) (p<0.001, multiplicity-adjusted Wald counties (95% multiplicity-adjusted Wald CI= 2.1-10.8), tests), except during period 3, when percentages in large and 9.7% per year in micropolitan counties (95% CI= 6.5­ metropolitan counties (5.0°/o) were the same as those in large 13.0) (Table 2) (Figure). During period 3, afrer release of the fringe metropolitan counties (5.0%) (Supplementary Table 2, CDC Guideline, the odds of receiving an opioid prescrip­ https://stacks.cdc.gov/view/ cdc/61744). In contrast, the high­ tion decreased significantly in all county groups. Comparing est period-specific percentages (9.0°/o-10.3o/o) were in noncore trends benveen periods, the APC increased in large central counties (p<0.02), except in period 3, when percentages in metropolitan counties in period 2 compared with period 1 noncore counties (9.0o/o) were similar to those in micropolitan (p<0.001) and decre

TABLE 2. Annual percent change {APC) in odds of receiving at least one opioid prescription - Athenahealth, United States, January 2014-March 2017 Period 1 t Period 2t Period 3t p-value (direction of change}§

Urban-rural category* APC (95°,.U Cl) APC {95o/o Cl) APC (95%CI) Period 1 versus period 2 Period 2 versus period 3

Non core 6.4 (2. 1 to 10.8)~ -10.1 (-12.2to-8.0)~ -7.5 (-10.7 to -4.2)~ <0.001 (decrease) 0.713 (-) Micropolitan 9.7 (6.5 to 13.0)11 -0.8 (-2.6 to 0.9) -13.3 (-15.6 to -10.9)~ <0.001 (decrease) <0.001 (decrease) Small metro 0.2 (-2.8 to 3.2) -4.5 {-6.2 to -2.7)11 -5.8 (-8.4 to -3.2)11 0.013 (decrease) 0.977 (-) Medium metro -2.5 (-4.8 to -0.1)** -8.7 (-10.1 to-7.4)11 -9.2 {-11 .2 to -7.2)11 <0.001 (decrease) 0.999 (-) Large fringe metro -2.0 (-4.7 to 0.8) -14.9 (-16.2 to-135)~ -13.1 (-15.1 to-10.9)~ <0.001 (decrease) 0.616(-) Large central metro -9.9 (-13.2 to -6.4)11 1.8 (-0.3 to 3.9) -11.7 {-14.3 to -8.9)11 <0.001 (increase) <0.001 (decrease) All counties -1.4 {-10.6 to 8.8) -8.2 (-13.3 to -2.7)tt -10.4 {-17.9 to -2.2) tt 0.371 (-) 0.856(-) Abbreviation: Cl= confidence interval. *National Center for Health Statistics urban-rural classification scheme for counties. https://www.cdc.gov/nchs/data_access/urban_rural.htm. t Period 1: January 5, 2014-January 3, 2015; period 2: January 4, 2015-March 19, 2016; period 3: March 20, 2016-March 11, 2017. § p-values from multiplicity-adjusted Wald tests;(-) indicates a nonsignificant difference (p>0.05) between APCs in adjacent periods. ~ p<0.001. ** p<0.05. tt p<0.01.

US Department of Health and Human Services/Centers for Disease Control and Prevention MMWR I January 18, 2019 ' Vol. 68 I No. 2 27 Morbidity and Mortality Weekly Report

FIG.URE. Model-based trends in percentage of patient-weeks with at least one opioid prescription, by urban-rural category - Athenahealth United States, January 2014-March 2017 '

100..,.-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Smooth trend, large central metro Model-based trend, large central metro Smooth trend, large fringe metro Model-based trend, large fringe metro Smooth trend, medium metro Model-based trend, medium metro Smooth trend, small metro Model-based trend, small metro Smooth trend, micro Model-based trend, micro Smooth trend, noncore Model-based trend, noncore

10.5

10.0 ..

9.5

9.0 ..

8.5

8.0 - ...... :• .. "' . . . "' ...... "' ...... , . 7.5 ...... • • • i ; I I 0 0 • • •••• .. • • • • • • • !! !! ! ! ! ! ! : : : : :i: ! ! ! •• ...... 7.0 ... -­ ··········­ ··;;;;;;;;;; ...... ~. 6.5 ...... ' ...... ·------!ill!!!~~=:::::..... _ ...... ,-_' _..._ ...... • ~ I I I I ; :; 6.0 (' -- "ii;i-..o;;.:-...... ~::,-... '.;;;­ ..;;;-....._ 5.5 .....>. 5.0 - 4.5 -'-,------,----__;T------,------.-_;____T""'____~_ _J Jan Jul Jan Jul Jan Jul Jan - 2014 2015 2016 2017

Date of prescription

Discussion in all urban-rural county groups after the March 2016 pub­ Throughout the analysis period, opioid prescribing rates by lication of the CDC Guideline. Those trends represented primary care providers were significantly higher in nonmetro­ significant decreases in the micropolitan and large central politan counties than in metropolitan counties. Whereas the metropolitan categories. In the other four county groups, prescribing rate increased from January 2014 through January however, the significant decreases after March 2016 represented 2015 (period I) in both micropolitan and noncore counties, a continuation of previously decreasing trends. those trends halted, and rates became flat or declined through Higher odds ofopioid prescribing in nonmetropolitan coun­ mid-March 2016 (period 2). Trends in all other urban-rural ties might be attributed in part to prescription drug use and categories were flat or decreasing over the same t\vo periods. misuse at an earlier age as well as higher prevalences ofchronic The odds ofa patient receiving an opioid prescription decreased pain among persons living in rural areas (6,7). Nonmetropolitan

28 MMWR I January 18, 2019 I Vol. 68 I No. 2 US Department of Health and Human Services/Centers for Disease Control and Prevention IV\orbidity and Mortality Weekly Report counties also tend to have larger populations of older adults who have higher prevalences of conditions associated with Summary pain ( 6). Opioid prescribing in rural (nonmetropolitan) areas What is already known about this topic? is strongly influenced by providers' individual relationships Opioid prescribing rates vary by county urbanization level and are declining overall. with their patients (8), and can be inconsistent with opioid prescribing guidelines. As well, access to medication-assisted What is added by this report? treatment facilities and alternative therapies are limited in Analysis of patient opioid prescription data from a national electronic health record vendor during 2014-2017 found that rural areas (8). Variations in the implementation of state-run the percentage of patients prescribed an opioid was higher in prescription drug monitoring programs and state-based laws rural than in urban areas. Significant decreases in opioid (9), such as the regulation of pain-management clinics, might prescribing occurred across all urban-rural categories after the also differ in urban and rural communities. March 2016 release of the CDC Guideline for Prescribing Opioids Despite reductions in opioid prescribing in recent years (1), for Chronic Pain. opioid-involved overdose death rates have increased, largely What are the implications for public health practice? driven by heroin and illicitly manufactured fentanyl (2). As less densely populated areas indicate both progress in Many persons who self-report heroin use have a history of decreasing opioid prescribing and need for ongoing reduction, misusing prescription opioids (10). Addressing prescription tailoring community health care practices and intervention opioid use is an important step in curbing opioid-involved programs to community characteristics will remain important. overdose deaths. Interventions such as using Prescription Drug 1"1onitoring Programs and practices that align with evidence­ over time. & less densely populated areas appear to indicate based adoption ofthe CDC Guideline can improve prescribing both substantial progress in decreasing opioid prescribing and decisions.** The Guideline can help providers and patients ongoing need for reduction, community health care practices weigh the benefits and risks of prescribing opioids according and intervention programs must continue to be tailored to to best available evidence and individual patient needs (4). community characteristics. This srudy demonstrates that data from EHRs can effectively Acknowledgment supplement traditional surveillance methods for monitoring trends in opioid prescribing and other areas of public health Philip Galebach, formerly of Athenahealth, AthenaResearch, importance. The lag between the collection of the data and Watertown, Massachusetts. this analysis could potentially be reduced to a matter ofweeks Corresponding author: Macarena C. Garcia, [email protected], 404--539-4410. \Vith optimized workflows. 1Center for Surveillance, Epidemiology, and Laboratory Services, CDC; The findings in this report are subject to at least three 2National Center for Injury Prevention and Control, CDC; 3Robert Wood limitations. First, the conclusions drawn from the records Johnson Foundation, Princeton, New Jersey; 4Athenahealth, AthenaResearch, Watertown, Massachusetts. provided by Athenahealth might not be generalizable to all patients in primary care. Second, although the data include All authors have completed and submitted the ICMJE form for disclosure of potential conflicts of interest. No potential conflicts of all patients with an opioid prescription, they do not include interest were disclosed. other characteristics ofeach prescription, including indication (e.g., chronic versus acute pain or opioid use disorder treated References with buprenorphine [although this drug accounted for a small 1. Scholl L, Seth P, I

US Department of Health and Human Services/Centers for Disease Control and Prevention MMWR I January 18, 2019 I Vol. 68 / i\Jo. 2 29 fVlorbidity and Mortality Weekly Report

6. Keyes KM, Cerda M, Brady JE, Havens JR, Galea S. Understanding the 9. Rutkow L, Chang HY, Daubresse M, Webster DW, Stuart EA, Alexander rural-urban differences in nonmedical prescription opioid use and abuse CC. Effect of Florida's prescription drug monitoring program and pill in the United States. AmJ Public Health 2014;104:e52-9. https://doi. mill laws on opioid prescribing and use. JAMA Intern Nied mg/ 10.2105/AJPH.20 13.30 \ 709 2015; 175: 1642-9. https://doi.org/ 10.100l/jamainternmed.2015 .3931 7. Monnat SM, Rigg KK. Examining rtual/urban differences in prescription 10. Compton WM, Jones CM, Baldwin GT. Relationship between opioid misuse among US adolescents. J Rural Health 2016;32:204--18. nonmedical prescription-opioid use and heroin use. N Engl J Med hrtps:// doi.org/ 10.11 l 1/jrh.12141 2016;374;154-63. httpdldoi.otgl 10.1056/NE)Mtal508490 8. Click IA, Basden JA, Bohannon JNI, Anderson H, Tudiver F. Opioid prescribing in rural family practices: a qualitative study. Subst Use Misuse 20 18;53;533-40. httpdIdoi.org/ 10.1080/ I 0826084.2017.1342659

30 MMVVR I January 18, 20 19 I Vol. 68 I No. 2 US Department of Health and Human Services/Centers for Disease Control and Prevention Policy Research Perspectives

Medical Liability Claim Frequency Among U.S. Physicians

By Jose R. Guardado, PhD

Introduction

Medical liability claims impose costs to society-monetary and non-monetary-so examining their prevalence is important. Using new data from the American Medical Association's (AMA) 2016 Physician Practice Benchmark Survey, this paper presents estimates of claim frequency for all physicians and explores whether the likelihood of claims varies by age, gender, specialty, practice type and ownership status.

The 2016 Benchmark Survey offers a unique opportunity to shed light on physicians' risk of liability. Other data sources provide measures of claim frequency, but they lack the total number of physicians which is needed to calculate the relative frequency (or risk) of getting sued. For example, while the PIAA-a trade association of medical liability insurance companies-has very good data on the number of closed claims, it does not report the number of physicians that are insured. 1 Schaffer et al. (2017) use data on the number of lawsuits from the National Practitioner Data Bank (NPDB). However, because the NPDB data do not report the number of physicians, they have to rely on the AMA Masterfile data to obtain ii. In contrast, our data allows us to ask questions such as: out of a nationally representative sample of physicians, how many of those physicians have been sued?

As a preview, we find that liability claims against physicians is not a rare event. Thirty-four percent of physicians have had a claim filed against them at some point in their careers. We also find that claim frequency varies by certain factors, particularly age, specialty and gender. Older physicians had a higher incidence of claims than did younger ones. There was also wide variation in claim incidence by specialty. General surgeons and obstetricians/gynecologists (OB/GYN) were the physicians most likely to be sued.

Data and Methods

The 2016 Benchmark Survey is a nationally representative survey of post-residency physicians who provide at least 20 hours of patient care per week, are not employed by the federal government, and

1 PIM. Closed Claims Comparative, 2016

© 2017 American Medical Association. All rights reserved. 2 practice in one of the 50 U.S. states or the District of Columbia (Kane 2017). The data include 3500 physicians with a response rate of 36 percent. 2

This paper relies on three questions in the 2016 Benchmark Survey aimed at collecting information on medical liability claim frequency: 1) whether any claims have been filed against the physician during his/her career, ii) the number of claims filed against the physician during his/her career and iii) the number of claims filed in the twelve months prior to the survey.3 This paper uses the terms "lawsuits" and "claims" interchangeably.

It should be noted that getting sued is not necessarily indicative of medical error. To shed light on this, consider that data from the PIAA show that 68 percent of claims that closed in 2015 were dropped, dismissed or withdrawn, and out of the 7 percent of claims that were decided by a trial verdict, 88 percent were won by the defendant.4 Thus, this paper's focus is not on whether physicians make medical errors, but rather simply on whether they had liability claims filed against them.

The analysis begins with a presentation of population estimates of claim frequency. We report descriptive statistics measuring the extent of claims against all physicians as well as by age, gender, specialty, practice type and physician ownership status. This descriptive analysis also gives us a first glimpse at whether there is any variation in claim incidence by those different categories.

Although population estimates are very useful in their own right, we complement them with analyses that allow us to look at the correlation between a specific characteristic (e.g. gender) and claim incidence, while controlling for other important factors (e.g. specialty). While the descriptive analysis looks at the correlations between claim frequency and individual characteristics separately, this more systematic regression analysis looks at all of those correlations simultaneously.

Results

Descriptive Results - Population Estimates

In this section, we present population estimates of medical liability claim frequency for all physicians as well as by age, gender, specialty and practice arrangements. The results show similar patterns across age, gender and specialty as those found in Kane (2010), who used data from 2007-2008. However, due to significant differences in survey methodologies, we cannot compare changes in claim incidence over time.

Table 1 presents the results for all physicians and by age and gender. It shows the percentage of physicians who had any medical liability claims filed against them. It also reports the fraction of

2 The survey includes weights to account for the probability of selection into the sample, to adjust for non-resolution of eligibility status, non-response, and differences between respondents and the population. All population estimates presented here are weighted. For a more detailed description of the survey methodology, see Kane (2017). 3 About 8 percent of the final 3500 sample resulted in missing observations-"don't know" or "prefer not to answer" responses to the liability questions, resulting in Ns ranging from 3145 to 3211. 4 PIM. Closed Claims Comparative, 2016 and author's calculations of data from that report. 3 physicians who have been sued two or more times, the share who have been sued during the twelve months prior to the survey, and the average number of liability claims filed per 100 physicians.5

The results show that getting sued is not an uncommon event for physicians. Thirty-four percent of all physicians have been sued, and 16.8 percent have been sued two or more times. On average 68 liability claims were filed per every 100 physicians. Because of the narrower time frame, the fraction of physicians who have been sued recently is much lower. Column 3 shows that only 2.3 percent of physicians were sued in the last year.

The probability of getting sued increases with age. This is not surprising given that older physicians have had more exposure to risk of lawsuits because they have been practicing for longer periods of time. Whereas 8.2 percent of physicians under the age of 40 have been sued, almost half of physicians over the age of 54 have been. The probability of being sued two or more times and the average number of claims filed against physicians increase with age as well. Although fewer than 2 percent of physicians under 40 have been sued at least twice, this fraction rises to 28.0 percent for physicians ages 55 and over. The final column shows that while 10 claims per 100 physicians had been filed against physicians under 40, this number increases to 109 claims for those over 54. Finally, there is less variation by age in the probability of having been sued recently. Fewer than 3 percent of physicians were sued in the last year regardless of their age.

There is also variation in claim frequency by gender. Female physicians were less likely to be sued than their male counterparts. Almost 40 percent of male physicians have been sued over the course of their careers, compared to 22.8 percent of women. While 20.4 percent of male physicians had at least two claims filed against them, 9.7 percent of female physicians did. On average, women had about half the number of claims filed against them (41 per 100) than did male physicians (82 per 100). As with age, however, there was little variation by gender in the likelihood of getting sued recently. The percentage of both men and women who were sued in the last year was just above 2 percent.

There are a number of apparent reasons why women are less likely to get sued than male physicians, such as differences in age and specialty.6 In separate analyses, we find that women physicians tend to be younger than their male counterparts so they have been in practice for a shorter period of time and thus have had less exposure to liability risk. 7 Thus, lower age is one explanation for part of the gender differential in claim frequency. We also find that with one exception (OB/GYN), women tend to practice in lower risk specialties than their male counterparts. In short, just controlling for age and specialty largely reduces the gender differential in claim incidence.

Table 2 also shows wide variation in claim frequency by specialty. Psychiatrists and pediatricians were the least likely among the specialties to be sued. Sixteen percent of psychiatrists and 17.8 percent of pediatricians have been sued, and about 6 percent of physicians in those specialties have been sued two or more times. At the other end of the distribution, general surgeons and OB/GYNs

5 This last statistic is simply the average number of claims per physician multiplied by 100. 6 See the AMA's Physician Characteristics and Distribution in the US, 2015 Edition, which reports physician distribution by gender, age and specialty. 7 Results not shown. 4 were the most likely to be sued. Over 63 percent of general surgeons and OB/GYNs had a claim filed against them. Half of general surgeons and 44.1 percent of OB/GYNs were sued two or more limes. Among OB/GYNs, there was an average of 162 claims filed per every 100 physicians, and the number among general surgeons surpasses 200. Although in general there was low incidence and little variation by specialty in getting sued recently, general surgeons and OB/GYNs also stood out, with 8.0 percent and 6.7 percent, respectively, getting sued in the last year.

Table 3 reports percentages of physicians who have been sued by both age and specialty. This analysis also reveals wide variation in claim frequency. One notable finding is that before they turn 55, over half of general surgeons and OB/GYNs have already been sued. Also noteworthy is that of those over 54, over three quarters of physicians in those specialties had at least one claim filed against them. Even older family practitioners had a high claim incidence, with more than half of those over 54 getting sued at some point in their careers.

In contrast to the findings by age, gender and specialty, there is much less variation in claim frequency by type of practice and physician ownership status (Table 4). In general, these results suggest that solo practitioners have a somewhat higher claim frequency than physicians in other practice types. About 30 percent of physicians employed in hospitals and those who are in multi­ specialty groups have been sued, compared to 40.4 percent of solo practitioners. Solo practitioners also had a higher average number of claims filed against them (89 per 100) than physicians in other practice types.

Employed physicians had a somewhat lower claim frequency than owners and independent contractors. Thirty-seven percent of both owners and independent contractors had a claim filed against them. In contrast, the share among employees was 30.1 percent.

Regression Analysis of Medical Liability Claim Frequency

The analysis in the previous section showed that having liability claims filed against physicians is not rare. It also revealed that the chance of getting sued varies by age, gender and specialty, and to a much lesser extent, by practice type and ownership status. However, those descriptive analyses do not allow one to control for correlations among the specific characteristics themselves. For example, while they show that men are more likely to be sued than women, we also know that male physicians tend to practice in higher risk specialties. Therefore, in this section we complement those analyses with an analysis that looks at the correlation between a specific characteristic (e.g. gender), while controlling for other important factors. To illustrate, the following more systematic regression analysis allows us to assess whether women are still less likely to be sued than men after controlling for their age, specialty and practice arrangements.

Regression results of the probability of being sued, being sued two or more times and the average number of claims per 100 physicians are presented in columns 1 through 3 of Table 5, respectively. The first results we discuss are the estimates by age. For ease of presentation, we compare the claim frequency of older (age 40-54 and 55+) to younger physicians (under 40). These results corroborate the results from the descriptive analysis above. Age is still strongly positively correlated with claim frequency, and the sizes of the differences do not change much when controlling for other factors. Again this is not surprising given that older physicians have had more exposure to liability 5 risk. Compared to physicians under 40, those 40-54 were 19.5 percentage points and those 55 and over were 39.4 percentage points more likely to have been sued. The probabilities of being sued two or more times and the average number of claims are also higher for older physicians.

Consistent with the results from the descriptive analysis, women were still less likely to have claims filed against them than their male counterparts. Interestingly, however, the size of those differences shrinks substantially, suggesting that other factors-in particular age and specialty-are responsible for a significant part of the gender differentials reported in Table 1. Controlling for age and specialty reduces the gender differential in getting sued by about 6 percentage points, or roughly 40 percent. 8 Table 5 shows that women were 8.7 percentage points less likely to have been sued and had about 23 fewer claims per 100 physicians than male physicians.

We still find wide variation in claim frequency by specialty after controlling for other factors. We compare claim incidence in different specialties to internists. General surgeons, OB/GYNs, emergency medicine physicians, surgical subspecialists and, to a lesser extent, radiologists had a higher incidence of claims than internists. In contrast, pediatricians and psychiatrists have lower claim frequency. General surgeons and OB/GYNs have the highest incidence of claims. For example, OB/GYNs were 34 percentage points more likely to have been sued and 32.5 percentage points more likely to have been sued two or more times than internists.

Turning to the results by type of practice, in contrast to the findings from the simple analysis above which suggested a somewhat higher claim frequency among solo practitioners, here we find no evidence of variation. Compared to single-specialty physicians, those in other practice types (including solo practitioners) are no more or less likely to have been sued. Solo practitioners may have had more claims filed against them than those in single-specialty groups, but the estimate is not statistically significant. This suggests that the differences reported in Table 4 were due to other factors for which we are now controlling.

Finally, we turn to the results by physician ownership status. The interesting finding here is that, in contrast to the results from the simple analysis in Table 4 which show that employed physicians had a somewhat lower incidence of claims than owners, here we find some opposing evidence. First, employed physicians were no more or less likely than owners to have been sued. Second, employees were actually a bit more likely (3 percentage points) than owners to have had two or more claims filed against them, and they had 12 more claims per 100 physicians than did owners. It is not clear why employees would have more claims than owners. A separate analysis shows that these findings are driven by employed physicians in single specialty groups-in particular, a few physicians with a relatively large number of claims.9

8 This is from a comparison of unweighted unadjusted and adjusted (regression) estimates. Note that the unadjusted estimates in Table 1 are weighted, whereas the regression estimates (Table 5) are not so they are not directly comparable. Not weighting the unadjusted estimates reduces the 16.59 percentage point gender difference in Table 1 by about 1.67 percentage points, to 14.92, which is what we compare to the 8.7 percentage point difference estimated by the regression. 9 Results not shown. 6

Conclusion

Using data from the 2016 Benchmark Survey, this paper presents population estimates of medical liability claim frequency for all physicians as well as by age, gender, specialty, practice type and physician ownership status. Those data are complemented by a regression analysis that examines the correlation between a given characteristic (e.g. gender) and liability risk, while controlling for other important factors. We note that getting sued is not necessarily indicative of medical error.

We find that over a third of all physicians have been sued at some point in their careers. Due to the narrower time frame, there was a low (2.3 percent) chance of getting sued in the last year.

Older physicians had a much higher incidence of claims than younger physicians. Controlling for other factors, physicians 55 and older were about 40 percentage points more likely to have been sued than physicians under 40.

Women had lower claim frequency than male physicians. Controlling for other factors shrinks the differentials substantially, though women were still 8.7 percentage points less likely to have been sued than their male counterparts.

General surgeons, OB/GYNs and, to a lesser extent, emergency medicine physicians and surgical subspecialists had a relatively high incidence of claims. General surgeons and OB/GYNs had an over 30 percentage point higher probability of being sued and getting sued two or more times than did internists. Psychiatry and pediatrics were the relatively lower risk specialties.

It is not rare for physicians to have liability claims filed against them. Physicians' liability risk is significant. We also find variation in claim frequency by certain characteristics. Gender, and especially age and specialty are strong predictors of claim incidence. There are plausible reasons why we would expect at least some of these findings. Women physicians are relatively younger and tend to work in lower risk specialties; older workers have had more exposure to liability claims; and some specialties are inherently more risky than others. More research is needed to better understand the extent of these correlations and why they arise.

AMA Economic and Health Policy Research, December 2017 2017-5 7

References

Kane, CK. Updated data on physician practice arrangements: physician ownership drops below 50 percent. Chicago (IL): American Medical Association; 2017 [cited 2017 Nov 14]. (Policy Research Perspectives 2017-2). Available from https://www.ama-assn.org/sites/defaull/files/media­ browser/public/health-policy/PRP-2016-physician-benchmark-survey.pdf.

Schaffer, AC. Jena, AB. Seabury, SA. Singh, H. Chalasani, V. Kachalia, A. Rates and characteristics of paid malpractice claims among US physicians by specialty, 1992-2014. JAMA Internal Medicine. 2017; 177(5): 710-718. Available from https ://jamanetwork.com/journals/jamainternalmedicine/fullarticle/2612118.

Kane, CK. Medical liability claim frequency: a 2007-2008 snapshot of physicians. Chicago (IL): American Medical Association; 2010 [cited 2017 Nov 14]. (Policy Research Perspectives 2010-1). 8

Table 1. Medical Liability Claim Frequency by Physician Age and Gender, 2016

Percentage of Physicians Number of Claims Sued Sued in Last 12 per 100 Ever Sued 2+ Times Months Physicians (1) (2) (3) (4)

All Physicians 34.0% 16.8% 2.3% 68

Under aae 40 8.2% 1.8% 1.2% 10 Age 40-54 28.7% 11.6% 2.6% 50 Age 55 and over 49.2% 28.0% 2.6% 109

Men 39.4% 20.4% 2.5% 82 Women 22.8% 9.7% 2.1% 41

Observations 3211 3145 3147 3145

Source: Author's tabulation of data from the AMA's 2016 Benchmark Survey.

Table 2. Medical Liability Claim Frequency by Physician Specialty, 2016

Percentage of Physicians Number of Sued Sued in Last Claims per 100 Specialty Ever Sued 2+ Times 12 Months Physicians (1) (2) (3) (4)

Anesthesiolonv 36.3% 17.9% 1.3% 64 Emeraencv medicine 51.7% 25.7% 3.0% 108 Family practice 33.4% 13.8% 1.1% 55 General surqery 63.2% 50.1% 8.0% 205 Internal medicine 31.7% 14.8% 3.1% 57 Internal medicine sub-specialties 25.5% 11.0% 1.0% 44 Obstetrics/Gvnecoloav 63.6% 44.1% 6.7% 162 Pediatrics 17.8% 6.0% 1.0% 28 Psychiatry 16.1% 5.9% 1.9% 25 Radioloav 37.6% 21.4% 0.4% 82 Suraical sub-specialties 47.4% 25.0% 3.3% 110 Other specialties 19.5% 5.8% 2.5% 29

Observations 3211 3145 3147 3145

Source: Author's tabulation of data from the AMA's 2016 Benchmark Survey. 9

Table 3. Percentage of Physicians Ever Sued by Age and Specialty, 2016

Specialty Under Age 55 Age 55+

Anesthesioloqy 24.5% 51.4% Emergency medicine 41.3% 72.2% Family practice 18.1% 51.4% General surqerv 54.8% 75.5% Internal medicine 18.5% 45.6% Internal medicine sub-specialties 14.7% 43.1% Obstetrics/Gvnecoloov 53.7% 76.5% Pediatrics 10.9% 27.9% Psychiatry 8.7% 23.0% Radiology 23.2% 53.5% Suraical sub-specialties 31.0% 67.4% Other specialties 12.4% 33.2%

Observations 1977 1234

Source: Author's tabulation of data from the AMA's 2016 Benchmark Survey.

Table 4. Medical Liability Claim Frequency by Type of Practice and Physician Ownership Status, 2016

Percentaqe of Physicians Number of Claims Sued per100 Ever Sued 2+ Times Physicians ( 1) (2) (3) Practice tvne Solo practice 40.4% 21.2% 89 Sinole specialty oroup 34.9% 17.5% 68 Multi-specialty Qroup 29.5% 14.1% 59 Direct hospital employee 29.3% 15.2% 62 Other' 33.5% 14.7% 62

Physician ownership status Owner 37.4% 18.6% 75 Employee 30.1% 15.2% 62 Independent contractor 37.4% 16.5% 68

Observations 3211 3145 3145

Source: Author's tabulation of data from the AMA's 2016 Benchmark Survey. Note: 1 The "other" category includes ambulatory surgical center, urgent care facility, HMO/MCO, medical school, faculty practice plan, and fill-in responses. 10

Table 5. Some Correlates of Medical Liability Claim Frequency

Sued Number of Claims Independent Variable Ever Sued 2+ Times per 100 Physicians (1) (2) (3) Age 40-54 0.195*** 0.095*** 38.70*** (0.021) (0.017) (5.83) Age 55 and over 0.394*** 0.253*** 96.36*** (0.022) (0.017) (6.15) Female -0.087*** -0.062*** -22.84*** (0.017) (0.013) (4.75) Anesthesiology 0.040 0.023 8.49 (0.037) (0.029) (10.28) Emergency medicine 0.224*** 0.110*** 54.50*** (0.043) (0.034) (11.87) Family practice 0.023 -0.002 0.194 (0.031) (0.024) (8.44) General surgery 0.305*** 0.362*** 148.0*** (0.049) {0.039) (13.83) Internal medicine sub-specialties -0.024 -0.013 -3.38 (0.030) (0.024) (8.28) Obstetrics/Gynecology 0.341 *** 0.325*** 116.0*** (0.040) (0.032) (11.36) Pediatrics -0.099*** -0.059** -18.34** (0.033) (0.025) (8.96) Psychiatry -0.166*** -0.088*** -34.37*** (0.039) (0.030) (10.66) Radiology 0.067 0.069* 28.29** {0.046) {0.037) (12.90) Surgical sub-specialties 0.150*** 0.105*** 52.29*** {0.032) (0.026) (9.02) Other specialties -0.050 -0.051 * -12.66 (0.034) (0.027) (9.46) Solo practice 0.008 0.009 10.07 (0.023) (0.018) (6.40) Multi-specialty group -0.021 -0.020 -4.07 I0.020) I0.016) (5.50) Other practice type 0.004 -0.027 -7.05 (0.030) (0.023) (8.22) Direct hospital employee -0.025 -0.011 -3.69 {0.032) {0.025) (8.86) Employee 0.014 0.029* 12.29** (0.019) (0.015) (5.17) Independent contractor 0.033 0.020 8.21 (0.035) (0.028) (9.85)

Observations 3211 3145 3145

Notes: Estimates from a regression of the dependent variable in columns 1-3 on the independent variables in the leftmost column. Omitted categories are physicians under the age of 40, in internal medicine, in single­ specialty groups, and owners. ***p < 0.01, **p < 0.05, *p < 0.10.