Clinical Outcomes Associated With Rates of Sulfonylurea Use Among Physicians
Published Online: July 08, 2013
Katalin Bognar, PhD; Kelly Fee Bell, PharmD, MS Phr; Darius Lakdawalla, PhD; Anshu Shrestha, PhD; Julia Thornton Snider, PhD; Nina Thomas, MPH; and Dana Goldman, PhD
Over the past 2 decades, the prevalence of diabetes mellitus has rapidly increased in the United States1 and globally.2 Estimates indicate that in 2010, nearly 19 million Americans had diagnosed diabetes, with 7 million more presumably undiagnosed.3 Type 2 diabetes mellitus (T2DM) accounts for over 90% of new cases of diabetes in adults.3 The wide range of available treatments has mitigated the impact of this growth4 but has increased the complexity of T2DM management.
Metformin, a biguanide, is widely accepted and used as first-line treatment for T2DM.4-6 Unlike many other agents,metformin causes neither acute hypoglycemia nor significant weight gain. Metformin use is usually only limited by gastrointestinal adverse events and is contraindicated in patients with impaired renal clearance. Although T2DM management guidelines do not name a preferred second-line agent,4 sulfonylureas are prescribed as if they were the favored choice. After metformin, sulfonylureas are the second-most popular T2DM medication, prescribed to approximately one-third of patients.7,8 Although sulfonylureas achieve glycemic control effectively,9 their long-term effects ondiabetes-related complications are not well established.10 Widespread sulfonylurea use continues despite the availability of several newer T2DM agents, presumably because sulfonylureas are inexpensive and well established.11-14
With the passage of the 2010 Affordable Care Act (ACA), the Centers for Medicare & Medicaid Services (CMS) is required to report performance data on physicians billing Medicare through the Physician Quality Reporting System.15 Currently, physicians voluntarily report performance data to CMS, but by 2015, the ACA will make it mandatory.15 This system is intended to reward quality of care and reflects a movement toward physician assessments such as doctor report cards.16 Because physician performance will be a key metric in determining US healthcare quality, it seems reasonable that physicians and stakeholders should have greater access to quality “physician-level” data to inform healthcare decisions. Specific performance measures are still being defined; the study presented here offers 1 possible measure.
In this retrospective, commercial, claims-based study, we examined the association between T2DM medication prescription patterns and physician performance. We measured physician performance using the occurrence of T2DMrelated complications, including hypoglycemic and cardiovascular events, neuropathy, and lower-extremity, vision, and renal complications. We then determined whether T2DM medication prescription patterns were associated with physician performance rank. Our study takes the first steps toward assessing physician prescribing patterns as one simple and easily measurable tool for predicting physician performance. Moreover, we provide payers with a broader context to evaluate sulfonylureas and other T2DM therapies.
Humana is a large provider of commercial and Medicare Advantage health insurance plans. We examined patient claims data aggregated at the physician level using the 2007-2011 Humana database. To acquire the physician-level data, we first identified the patient cohorts that would be aggregated to the physician level by examining all commercially insured patient claims data and extracting incident and prevalent cohorts of T2DM patients. To reduce potential bias,17 the incident cohort was used for base case analyses, while the prevalent cohort and 2 incident subcohorts were reserved for sensitivity analyses (reported in eAppendix A,www.ajmc.com). T2DM patients were identified as those with at least 1 claim with the International Classification of Diseases, Ninth Revision (ICD-9) code 250. x0 or 250.x2 and at least 1 claim for an anti-diabetic medication following the first observed claim with a T2DM diagnosis. Based on published results, these search criteria have a specificity range of 0.93 to 0.99 and sensitivity of 0.44 to 0.91.18,19 Patients with claims for pregnancy (ICD-9 codes 630-79, V22.x–V24.x, V27.x, V29.x, V61.6, V61.7) were excluded during pregnancy and for 6 months thereafter. The cohort includes only working-age patients (aged 18-64 years).
For the prevalent cohort, the first observed claim with a T2DM diagnosis provided the index date. Patients with less than 1 year of follow-up from the index date were excluded. The incident sample included all patients from the prevalent sample with at least 1 year of continuous enrollment prior to the index date, with no T2DM-related medical or pharmacy claims. We considered all available follow-up months of these patients. We also identified 2 incident subcohorts for sensitivity analyses: the second-line subcohort and the longrun therapy subcohort. In the secondline subcohort, we excluded patients from the incident sample if they didnot require a second-line agent during the study period, and we considered all available follow-up months of the remaining patients. A first-line therapy was defined as the first anti-diabetic drug used after diagnosis. A secondline therapy was added to or replaced the first agent. For the long-run therapy subcohort, we selected follow-up months for incident patients in whichpatients used at least 1 drug class they also had used continuously for at least 6 months prior. The study selection criteria are detailed in Figure 1.
We collected patient characteristics including age and gender, and tracked monthly comorbidities, medication use, and complications. The Elixhauser Comorbidity Index (ECI) was calculated each month for each patient using comorbid diagnoses over the previous 12 months.20,21
For each patient, we identified antidiabetic drug use using the National Drug Code (NDC) Directory. We grouped T2DM medications by mechanistic class; specifically, biguanides (metformin), sulfonylureas, thiazolidinediones (TZDs), dipeptidyl peptidase-4 (DPP-4) inhibitors, glucagon-like peptide-1 (GLP-1) agonists, insulins, and “other” (including alpha-glucosidase inhibitors, amylinomimetics, and meglitinides). For each patient, we identified T2DM-related adverse events (cardiovascular, lower-extremity, renal, ophthalmic, neurological, and severe hypoglycemic events) using ICD-9 and Current Procedural Terminology (CPT) codes (listed in eAppendix B, www.ajmc.com), consistent with previous literature.22-27
For each patient cohort (incident, prevalent, second-line, or long-run), we used physician identifiers from the pharmacy claims to link physicians prescribing T2DM medicines to each patient and compile an associated cohort of physicians. Once a patient received T2DM prescriptions from a provider, the patient was considered that provider’s patient for the next 6 months. A patient receiving prescriptions from multiple providers was considered the patient of each. For each physician in each month, we calculated complication rates, average patient characteristics (age, sex, ECI), and average T2DM drug usage (fraction of patients using each class) among the physician’s T2DM patients.
After tabulating drug use and complication rates for the patient cohort, we created for each month in the study period a case-mix-adjusted measure of physician performance. Specifically, we calculated the fraction of a physician’s patients experiencing any of the study complications in a given month. We then used a linear regression model to predict the monthly rate of any T2DMrelated complication as a function of average age, gender, and ECI among patients in the practice in a given month, as well as a monthly time trend. This model provided a measure of physicians’ performance as relative success at avoiding T2DM-related complications, after adjusting for their patients’ age, sex, and comorbidities.
After deriving this measure of physician performance, we sought to determine whether it was related to T2DM prescription patterns. To do this, we compared each physician’s actual performance to the “risk-adjusted” performance predicted by patient characteristics alone (Figure 2). Those doctors with lower complication rates than predicted based on patient characteristics were considered “high performers,” whereas those with higher complication rates than predicted were considered “low performers.” We ranked physicians in each month based on their performance that month, and then sorted doctor-months into 10 ordered and equally sized groups (deciles). After establishing the 10-group ranking of physician performance, we analyzed whether the prescribing patterns in the month prior of highest performing doctors were different than those of the lowest performing doctors for each of the drug classes.
Finally, to quantify the impact of physicians’ prescribing decisions, we expanded the previous regression model of physicians’ T2DM complication rates on practice characteristics to include rates of use for each of the T2DM drug classes in the month prior. We used this model to predict the change in T2DM complications when switching from “low performer” (bottom decile) prescribing patterns to
those of “high performers” (top decile). We performed sensitivity analyses on the prevalent, second-line, and longrun samples.
We identified an incident T2DM cohort of 7905 patients. Demographic descriptions are provided in Table 1. A majority of patients were men (n = 4418; 55.90%), and the most frequent age category was 46-55 years (n = 3155, 39.91%). The average age was 50.1 years.
PDF is available on the last page.