Objectives: We studied the independent association of adher- ence to oral hypoglycemic medications with poor glycemic con- trol among a population of adults with type 2 diabetes (T2D), adjusting for demographics, health behaviors, and clinical and treatment characteristics.
Study Design: This was a retrospective cohort design.
Methods: We studied a population of Kaiser Permanente Northwest (KPNW) members with T2D who either had: 1) good glycemic control (glycated hemoglobin [A1C] <8.0%; n = 15,891) or 2) poor glycemic control (A1C >9.0%; n = 3709). The primary independent variable was medication adherence to 1 or more oral hypoglycemic medications. High medication adherence was defined as at least 80% of days covered in the 12 months prior to the A1C test date (yes vs no). Multiple logistic regression was used to analyze the independent association of medication adherence with poor glycemic control, adjusting for demographics, health behaviors, medical comorbidities, healthcare utilization, receipt of diabetes care management services, and intensity of diabetes treatments. All measures were constructed via KPNW’s electronic heath record.
Results: Increased adherence to oral hypoglycemic medications was associated with a lower likelihood (OR, 0.54; 95% CI, 0.50-0.59; P <.0001) of having poor glycemic control, after adjusting for demographics, health behaviors, comorbidities, healthcare utilization, receipt of diabetes care management services, and intensity of diabetes treatments.
Conclusions: Higher adherence to oral hypoglycemic medica- tions is independently associated with a lower likelihood of having poor glycemic control among an adult population with T2D. Studies of the effects of measures to improve medication adherence on population-level glycemic control are needed.
Am J Pharm Benefits. 2017;9(3):82-88
Type 2 diabetes (T2D) is a chronic, progressive condition characterized by the failure of insulin secretion to compensate for resistance to insulin’s actions, resulting in hyperglycemia. The long-term complications of diabetes contribute to its status as a leading cause of premature illness and mortality.1 In 2012, 29.1 million individuals in the United States had diabetes (approximately 90% of the patients had T2D), and the prevalence appears to have increased in the past 2 decades.2 In 2012, diabetes resulted in $245 billion in costs, including $176 billion in direct medical costs and $69 billion in reduced productivity.3
The long-term control of hyperglycemia is essential in order to improve health outcomes for populations with diabetes. Poor glycemic control has been associated with microvascular complications4-6 and mortality,7,8 whereas good control reduces the development of microvascular and long-term cardiovascular complications.9 Current clinical guidelines to manage glycemic control utilize glycated hemoglobin (A1C) target levels, with adjustment for age and length of illness10; however, A1C levels greater than 75 mmol/mol (9.0%) are universally considered to reflect poor control.11 Of particular concern, according to data from a recent National Health and Nutrition Examination Survey, the proportion of adults with diabetes with poor glycemic control (A1C >9.0%) increased from 17.9% in the 2005-to-2008 period to 21.0% in the 2009-to-2012 period.12
Despite clinical evidence of the efficacy of several glucose- lowering medications, adherence levels often remain sub- optimal.13-17 Moreover, several studies have found that lower medication adherence can be associated with poorer glycemic control outcomes.14,18-23 However, these studies have several limitations, including small sample sizes and analyses not conducted on a population level.24,25
With this background, the primary objective of this study was to analyze the independent association of adherence to oral hypoglycemic medications with poor glycemic control among adults with T2D on a population level. Such research has significant clinical implications—given that medication adherence is a potentially modifiable factor—that may be influenced by quality improvement initiatives. A secondary objective was to identify demographics, health behaviors, comorbidities, healthcare utilization, diabetes care management services, and treatment intensity mea- sures associated with poor glycemic control, with an emphasis on factors that were modifiable.
The institutional review board of Kaiser Permanente Northwest (KPNW) in Portland, Oregon, approved the study protocol and waived the need for patient consent for data use, as is the protocol for data-only studies.
Study Setting and Data Sources
ThestudywasconductedatKPNW,anonprofit,group-model healthcare system in northwest Oregon and southwest Washington. The system consists of 15 medical clinics and 2 hospitals that serve more than 500,000 members. KPNW membership demographics (age, sex, race/ethnicity) are similar to the larger population in the geographic area. We used the electronic health record (EHR) (Epic Systems Corporation; Madison, Wisconsin) and administrative data systems to ascertain information on patient membership; demographics; clinical data, including height and weight; laboratory results; and medical healthcare utilization.
We conducted a retrospective cohort of KPNW members who met all the following inclusion
fasting glucose >125 mg/dL, nonfasting glucose >200 mg/ dL), or 3) 1 or more elevated A1C test results (>6.5%). In addition, those with T2D had an International Classification of Diseases, Ninth Revision (ICD-9)-related encounter (in any medical setting) as far back as membership existed. T2D ICD-9 codes include any of the following: 250.00, 250.02, 250.12, 250.20, 250.22, 250.30, 250.32, 250.40, 250.42, 250.52,250.60,250.62,250.70,250.72,250.80,250.82, 250.90, and 250.02.
Uncontrolled and Good Glycemic Control Groups
The process flow describing the study population is shown in the FIGURE. We identified 37,514 members who met the initial inclusion criteria. Of this population, 4759 members had an A1C value greater than 9.0% and formed the poor glycemic control group, while 27,514 had an A1C value less than 8.0% and made up the good glycemic control group. Another 5250 members with A1C values between 8.0% and 9.0% were not included in the analysis. The final analytic population (n = 19,600) included poor (3709) and good (15,891) glycemic control populations that received 1 or
criteria: 1) had an A1C test between July 1, 2014, and June 30, 2015; 2) were 18 years or older on the A1C test date; 3) were on the KPNW diabetes registry with a T2D indication at the time of the A1C test; and 4) had continuous health plan enrollment for 12 or more months prior to the A1C test date, which was defined as the index date. If the results of multiple A1C tests were available between July 1, 2014, and June 30, 2015, the latest value was used in the analysis.
KPNW members enter the diabetes registry through any of 3 ways: 1) receipt of treatment with oral hypo- glycemic medications, 2) 2 or more abnormal fasting or nonfasting glucose test results in the previous 3 years (eg, more oral hypoglycemic medications in the 12 months prior to the A1C test. Generally, adherence to oral hypoglycemic medications and other study covariates was measured in the year prior to the A1C test date, for a complete study period between July 1, 2013, and June 30, 2015.
Dependent variable. The primary dependent variable was poor glycemic control (A1C >9.0%) versus good glycemic control (A1C <8.0%).
Medication adherence. Adherence to oral hypoglycemic medications was the primary independent variable in the analysis. We used the proportion of days covered (PDC) to construct medication adherence—a measure well described in the literature26 and considered to be more consistent than another adherence measure commonly used with electronic pharmacy information, such as the medication possession ratio.27 The denominator included the total number of days between the first hypoglycemic medication fill and the end of the 365-day observation period, prior to the A1C test date. The numerator included total days covered by prescription fills during the denominator period. High medication adher- ence was defined as 80% or more days covered during the denominator period. If multiple medications were prescribed, a measure of total average days covered was constructed. No insulin use was included in the adherence measure.
Covariate measures. We identified 6 groups of covariate measures: demographics, health behaviors, comorbidities, healthcare utilization, use of diabetes care management services, and diabetes treatment-intensity measures. The rationale for selecting these measures was 2-fold. First, they are important factors previously associated with poor glycemic control in the literature. Second, adjusting for these measures on a population level increased the ability to better assess the independent association of medication adherence with poor glycemic control.
Demographics. Using KPNW’s EHR, we measured age (continuous: assessed upon A1C test date), sex (female vs male), race/ethnicity, and primary language status (non–English speaker: yes vs no).
Health behaviors. Two health behavior measures were included. The first, current tobacco use (yes vs no), was assessed at the time closest to the A1C test, with a 1-year look-back period. Second, current physical activity status—whether an individual exercised 4 or more times per week (yes vs no)—was the most recent measurement assessed closest to the A1C test date. This information, systematically assessed at routine KPNW office visits and asked in person by clinic staff, was the most recent update in the member’s EHR.
Comorbidities. Four comorbidity measures were analyzed. The first was obesity, as reflected by a body mass index (BMI) of 30 or higher (yes vs no) if height and weight were available, using the most recent assessment with a look-back period of 60 months (5 years). The Charlson Comorbidity Index (CCI) score was used to assess patient- level comorbidities. CCI score (continuous), an established risk of mortality score in the medical literature,28,29 was constructed using utilization of inpatient and outpatient services in the year prior to the A1C test date. The duration of diabetes was assessed by measuring months on the diabetes registry (continuous) as far back as membership existed.
Healthcare utilization. Four dichotomous (yes vs no) variables assessed healthcare utilization in the year prior to the A1C assessment date: 1) primary care visits (1 or more visits vs none), 2) specialty care visits (excluding visits to internal medicine, family practices or pediatrics; 1 or more visits vs none), 3) hospital admission (1 or more admissions vs none), and 4) emergency department (ED) utilization (1 or more visits vs none).
Diabetes care management services. Enrollment in diabetes care management services was assessed by any contact with a diabetes care management program (yes vs no) in the year prior to the A1C assessment date.
Diabetes treatment intensity measures. Two diabetes treatment measures were assessed in the year prior to the A1C test date: 1) use of oral hypoglycemic medications (1 medication vs 2 or more medications) and 2) use of insulin (any use vs no use).
We first examined the association of adherence to oral hypoglycemic medications and other covariate measures with poor glycemic control using χ2 analysis. Variables that were significant in the bivariate analysis (P <.05) were included in a final logistic regression model. The model examined the independent association of adherence to oral hypoglycemic medications and other covariate measures with poor glycemic control. Odds ratios (ORs) and 95% confidence intervals (CIs) were constructed for medication adherence and each covariate measure. The final logistic regression model included all covariate measures with the exception of: 1) ED utilization, which was not significant in the bivariate analysis, and 2) proteinuria. The proteinuria measure, although significant in the bivariate analysis, was dropped from the final logistic regression model because of a high level of missing information. The final model included the following variables: age (continuous), sex (female vs male [reference group]), and race/ethnicity (Hispanic, African American, Asian American/Pacific Islander, other race [Native American/Aleutian/Eskimo] vs white [reference group]).
Bivariate results are presented in TABLE 1 and multivariate results are in TABLE 2. Significant multivariate results are presented below.
Adherence to Oral Hypoglycemic Medications
Those with high adherence to oral hypo- glycemic medications were less likely to have poor glycemic control (OR, 0.54; 95% CI, 0.50-0.59) compared with those with low medication adherence, adjusting for other study covariates.
Association of Covariate Measures With Poor Glycemic Control
Demographics. Of the demographic factors, age, sex, and race/ethnicity were significantly associated with poor glycemic control. Specifically, younger age (OR, 0.96; 95% CI, 0.95-0.96) was associated with poorer glycemic con- trol, while women (OR, 0.88; 95% CI, 0.81-0.96) were less likely to have poor glycemic control than men. Lastly, African Americans (OR, 1.40; 95% CI, 1.16-1.69), Hispanics (OR, 1.62; 95% CI, 1.40-1.87), Asian Americans/Pacific Islanders (OR, 1.31; 95% CI, 1.11-1.55), and those of other race (OR, 1.44; 95% CI, 1.17-1.77) were more likely to have poor glycemic control compared with non-Hispanic whites.
Health behaviors. One of 2 health behaviors, exercise status was signifi- cantly associated with glycemic control in the multivariate analysis. Those who exercised 4 or more times per week were less likely to have poor glycemic control (OR, 0.75; 95% CI, 0.68-0.82) compared with those who exercised 3 or fewer times per week.
Comorbidities. Two comorbidity measures were associ- ated with poor glycemic control. Specifically, those with a BMI of 30 or greater (OR, 1.18; 95% CI, 1.07-1.30) were more likely to have poor glycemic control compared with those with a BMI of less than 30. Moreover, longer duration on the diabetes care registry (OR, 1.06; 95% CI, 1.05-1.07) was associated with increased likelihood of having poor glycemic control.
Healthcare utilization. Three of 4 utilization variables were significantly associated with glycemic control. Having a primary care visit (OR, 0.95; 95% CI, 0.94-0.97) or specialty care visit (OR, 0.98; 95% CI, 0.98-0.99) was associated with a lower likelihood of poor glycemic control compared with no primary or specialty care utilization. Interestingly, those with 1 or more hospital admissions were less likely to have poor glycemic control (OR, 0.85; 95% CI, 0.79-0.92) compared with those with no hospital admissions, adjusting for patient-level covariates.
Diabetes care management services and treatment. Perhaps serving as a proxy for disease severity, those with 1 or more diabetes care management visits (OR, 2.80; 95% CI, 2.55-3.07) were more likely to have poor glycemic control compared with those with no contact. Similarly, those using more than 1 oral hypoglycemic medication (OR, 2.09; 95% CI, 1.93-2.27) and any insulin (OR, 2.31; 95% CI, 2.09-2.55) were more likely to have poor glycemic control compared with those using 1 oral hypoglycemic medication and no insulin, respectively.
Improving glycemic control is a central goal of contemporary diabetes management. Nevertheless, many patients with diabetes do not succeed in reaching goal A1C levels. This can be a source of significant frustration for patients and clinicians and portends a greater disease burden upon health systems. Although a range of demographic, behavioral, and clinical factors were associated with poorer glycemic control, we found that medication adherence, as reflected by PDC of at least 80%, was the most significant factor associated with satisfactory glycemic control. In our study, we found that those with high adherence to oral hypoglycemic medications were 46% less likely to have poor glycemic control compared with those with low adherence. Our study results are unique, given that we found this relationship among a population of adults with T2D receiving care in an integrated system (serving more than 500,000 individuals) after adjusting for a comprehensive set of patient-level covariate measures.
Surprisingly, receipt of diabetes care management services was associated with poor glycemic control. Rather than showing the potential ineffectiveness of care management services, the receipt of these services is likely a marker for higher disease severity. Similar results were found for insulin use and higher use of oral hypoglycemic medications.
Our findings are consistent with other recent research findings that examined medication adherence and glycemic control. A study by Feldman and colleagues30 of the EHRs of more than 200,000 adults with diabetes in Israel found that lower medication adherence was associated with a greater likelihood of having poor glycemic control (A1C >9%). Similarly, a retrospective cohort study by Farmer and colleagues31 using the Clinical Practice Research Datalink found that reduced medication adherence was associated with smaller reductions in A1C values compared with greater medication adherence.
Our results have direct applicability to clinical practice, given that medication adherence is a modifiable factor that could improve. A recent study by de Vries McClintock and colleagues32 found that a brief in-person and telephone-based intervention program improved adherence to diabetes medi- cations, which was associated with a subsequent improve- ment in glycemic control. Our results, coupled with previous research findings, provide a focus for quality improvement opportunities to improve medication adherence.
We also found interesting relationships between several covariate measures and glycemic control that have direct applicability to clinical care. First, we found that African Americans and other ethnic minorities were more likely to have poor glycemic control compared with non-Hispanic whites, even after adjusting for medication adherence and other patient-level covariates. These results are similar to research findings from Lafata and colleagues33 that found that African American–Caucasian differences in glycemic control could not be explained by medication adherence. These results give the larger KPNW delivery system an opportunity to develop quality improvement interventions to further address racial disparities in glycemic control.
In addition, 2 potentially modifiable factors were found to be associated with glycemic control. First, we found that increased exercise (4 or more times per week vs 3 or fewer times) was independently associated with a lower likelihood of having poor glycemic control after adjusting for medication adherence and other study covariates. Second, a BMI of 30 or higher (vs BMI <30) was associated with a greater likelihood of poor glycemic control. These results reinforce the importance of lifestyle interventions to improve glycemic control and identify another important quality improvement opportunity, specifically, interventions that target weight loss and improvement in medication adherence for adults with T2D.
Our study has some important limitations. First, the T2D population was identified using electronic databases; thus, they may not be completely accurate and may misclassify some patients. Similarly, medication adherence was constructed from pharmacy dispensing information and may not reflect the medications that were actually taken by the patients observed in the study. Third, given that the study occurred in a group-model health maintenance organization (HMO) setting in the Pacific Northwest, the results cannot be generalized beyond this setting.
Future research should examine which system-level interven- tions are most successful in improving medication adherence and whether such improvement is associated with subsequent improvements in glycemic control. Such interventions should be pragmatic clinical trials or rigorous observational studies that can be implemented in real-world delivery systems with high levels of generalizability to clinical practice.
Among a population of patients with T2D in a group- model HMO care delivery system, we found that higher adherence to oral hypoglycemic medications was indepen- dently associated with a lower likelihood of having poor glycemic control. This finding remained significant, even after adjusting for patient-level demographics, health behaviors, comorbidities, healthcare utilization, use of diabetes care management services, and diabetes treatment measures.
Author Affiliations: Center for Health Research, Kaiser Permanente Northwest (DMM, ABS, HC), Portland, OR; Kaiser Permanente Northwest Quality Management and Systems (HG, ACF), Portland, OR.
Source of Funding: None.
Author Disclosures: The authors report no relationship or financial interest with any entity that would pose a conflict of interest with the subject matter of this article.
Authorship Information: Concept and design (DMM, HC, ACF, HG); acquisition of data (ABS, ACF); analysis and interpretation of data (ABS, DMM); drafting of the manuscript (DMM, HC, ACF, HG); critical revision of the manuscript for important intellectual content (DMM, HC, ABS, ACF, HG); statistical analysis (ABS); provision of patients or study materials (ABS).
Address Correspondence to: David M. Mosen, PhD, MPH, Kaiser Permanente Northwest, Center for Health Research, 3800 N Interstate Ave, Portland, OR 97227. E-mail: firstname.lastname@example.org.
1. IDF diabetes atlas – 7th edition. International Diabetes Federation website. www.diabetesatlas.org. Accessed March 28, 2017. 2. 2014 national diabetes statistics report. CDC website. http://www.cdc. gov/diabetes/data/statistics/2014StatisticsReport.html. Published 2014. Accessed March 28, 2017.
3. American Diabetes Association. Economic costs of diabetes in the U.S. in 2012. Diabetes Care. 2013;36(4):1033-1046. doi: 10.2337/dc12-2625.
4. Patel A, MacMahon S, Chalmers J, et al; ADVANCE Collaborative Group. Intensive blood glucose control and vascular outcomes in patients with type 2 diabetes. N Engl J Med. 2008;358(24):2560-2572. doi: 10.1056/ NEJMoa0802987.
5. Stratton IM, Adler AI, Neil HA, et al. Association of glycaemia with macro- vascular and microvascular complications of type 2 diabetes (UKPDS 35): prospective observational study. BMJ. 2000;321(7258):405-412.
6. Nathan DM, Genuth S, Lachin J, et al; The Diabetes Control and Complica- tions Trial Research Group. The effect of intensive treatment of diabetes on the development and progression of long-term complications in insulin-dependent diabetes mellitus. N Engl J Med. 1993;329(14):977-986.
7. Huang ES, Liu JY, Moffet HH, John PM, Karter AJ. Glycemic control, complica- tions, and death in older diabetic patients: the diabetes and aging study. Diabetes Care. 2011;34(6):1329-1336. doi: 10.2337/dc10-2377.
8. Ho PM, Rumsfeld JS, Masoudi FA, et al. Effect of medication nonadherence on hospitalization and mortality among patients with diabetes mellitus. Arch Intern Med. 2006;166(17):1836-1841.
9. Holman RR, Paul SK, Bethel MA, Matthews DR, Neil HA. 10-year follow-up of intensive glucose control in type 2 diabetes. N Engl J Med. 2008;359(15):1577-1589. doi:
10.1056/NEJMoa0806470. 10. American Diabetes Association. Glycemic targets. Diabetes Care. 2016;39(suppl 1):S39-S46. doi: 10.2337/dc16-S008.
11. HEDIS 2015 volume 2: technical update. National Committee for Quality Assurance website. http://www.ncqa.org/Portals/0/HEDISQM/Hedis2015/ HEDIS%202015%20Volume%202%20Technical%20Update_Final.pdf. Pub- lished October 1, 2014. Accessed February 3, 2016.
12. Office of Disease Prevention and Health Promotion. Glycemic control, adults with diagnosed diabetes, 2005-08 and 2009-12. HealthyPeople.gov website. https://www.healthypeople.gov/2020/topics-objectives/national-snapshot/ glycemic-control-adults-with-diagnosed-diabetes-2005%E2%80%9308-and. Published August 25, 2014. Accessed February 3, 2016.
13. Cramer JA. A systematic review of adherence with medications for diabetes. Diabetes Care. 2004;27(5):1218-1224.
14. Krapek K, King K, Warren SS, et al. Medication adherence and associated hemoglobin A1C in type 2 diabetes. Ann Pharmacother. 2004;38(9):1357-1362.
15. Donnan PT, MacDonald TM, Morris AD. Adherence to prescribed oral hypoglycaemic medication in a population of patients with type 2 diabetes: a retrospective cohort study. Diabet Med. 2002;19(4):279-284.
16. Lopez JM, Bailey RA, Rupnow MF, Annunziata K. Characterization of type 2 diabetes mellitus burden by age and ethnic groups based on a nationwide sur- vey. Clin Ther. 2014;36(4):494-506. doi: 10.1016/j.clinthera.2013.12.016. 17. Long JA, Wang A, Medvedeva EL, et al. Glucose control and medication adherence among veterans with diabetes and serious mental illness: does collocation of primary care and mental health care matter? Diabetes Care. 2014;37(8):2261-2267. doi: 10.2337/dc13-0051.
18. Egede LE, Gebregziabher M, Hunt KJ, et al. Regional, geographic, and racial/ethnic variation in glycemic control in a national sample of veterans with diabetes. Diabetes Care. 2011;34(4):938-943. doi: 10.2337/dc10-1504.
19. Rozenfeld Y, Hunt JS, Plauschinat C, Wong KS. Oral antidiabetic medication adherence and glycemic control in managed care. Am J Manag Care. 2008;14(2):71-75.
20. Lawrence DB, Ragucci KR, Long LB, Parris BS, Helfer LA. Relationship of oral antihyperglycemic (sulfonylurea or metformin) medication adherence and hemoglobin A1c goal attainment for HMO patients enrolled in a diabetes disease management program. J Manag Care Pharm. 2006;12(6):466-471.
21. Pladevall M, Williams LK, Potts LA, Divine G, Xi H, Lafata JE. Clinical outcomes and adherence to medications measured by claims data in patients with diabetes. Diabetes Care. 2004;27(12):2800-2805.
22. Guillausseau PJ. Influence of oral antidiabetic drugs compliance on meta- bolic control in type 2 diabetes. A survey in general practice. Diabetes Metab. 2003;29(1):79-81.
23. Schectman JM, Nadkarni MM, Voss JD. The association between diabetes metabolic control and drug adherence in an indigent population. Diabetes Care. 2002;25(6):1015-1021.
24. Asche C, LaFleur J, Conner C. A review of diabetes treatment adher- ence and the association with clinical and economic outcomes. Clin Ther. 2011;33(1):74-109. doi: 10.1016/j.clinthera.2011.01.019.
25. Gonzalez JS, Schneider HE. Methodological issues in the assessment of diabetes treatment adherence. Curr Diab Rep. 2011;11(6):472-479. doi: 10.1007/s11892-011-0229-4.
26. Benner JS, Glynn RJ, Mogun H, Neumann PJ, Weinstein MC, Avorn J. Long-term persistence in use of statin therapy in elderly patients. JAMA. 2002;288(4):455-461.
27. Choudhry NK, Shrank WH, Levin RL, et al. Measuring concurrent adherence to multiple related medications. Am J Manag Care. 2009;15(7):457-464.
28. Kieszak SM, Flanders WD, Kosinski AS, Shipp CC, Karp H. A comparison of the Charlson comorbidity index derived from medical record data and adminis- trative billing data. J Clin Epidemiol. 1999;52(2):137-142.
29. Deyo RA, Cherkin DC, Ciol MA. Adapting a clinical comorbidity in- dex for use with ICD-9-CM administrative databases. J Clin Epidemiol. 1992;45(6):613-619.
30. Feldman BS, Cohen-Stavi CJ, Leibowitz M, et al. Defining the role of medi- cation adherence in poor glycemic control among a general adult population with diabetes. PLoS One. 2014;9(9):e108145.
31. Farmer AJ, Rodgers LR, Lonergan M, et al; MASTERMIND Consortium. Adher- ence to oral glucose-lowering therapies and associations with 1-year HbA1c: a retrospective cohort analysis in a large primary care database. Diabetes Care. 2016;39(2):258-263. doi: 10.2337/dc15-1194.
32. de Vries McClintock HF, Morales KH, Small DS, Bogner HR. A brief adherence intervention that improved glycemic control: mediation by patterns of adherence. J Behav Med. 2015;38(1):39-47. doi: 10.1007/ s10865-014-9576-3.
33. Lafata JE, Karter AJ, O’Connor PJ, et al. Medication adherence does not explain black-white differences in cardiometabolic risk factor control among insured patients with diabetes. J Gen Intern Med. 2016;31(2):188-195. doi: 10.1007/s11606-015-3486-0.