Alignment of diagnosis and pharmacy claims data coding of medication adherence among patients with diabetes or hypertension.

ICD-10-CM codes exist that facilitate provider designation of patients as "nonadherent to therapy"; however, it is unclear whether this label accurately reflects patient behavior according to widely accepted medication adherence metrics using pharmacy claims data. To determine the extent to which patients are accurately coded for and have calculated rates of nonadherence using ICD-10-CM codes and claims, respectively. This was a retrospective cohort study using commercial insurance and Medicare Advantage claims data from 2015 to 2016. The analysis focused on adults aged 18 years and older who had been diagnosed with and were being treated for hypertension and/or diabetes and had been coded as nonadherent by a provider during an outpatient encounter. Adherence (proportion of days covered [PDC]) to oral antihypertensive and/or antidiabetic therapy was calculated 6 months before and after the first nonadherence diagnosis identified in outpatient encounters, using 2 distinct calculation methods. Inferential statistics and multivariable logistic regression were used to determine predictors of coding agreement and changes in adherence after the nonadherence diagnosis controlling for available patient characteristics. A total of 1,142 patients who had been coded as nonadherent were identified, of which between 5.3% and 22.0% (depending on metric and condition) had PDCs before the nonadherence code deeming them adherent according to claims, conflicting with nonadherence diagnosis codes documented by their providers. Mean PDCs increased significantly (20.5%-24.3%, all < 0.001) among both conditions following the nonadherent code, as did the proportion adherent (PDC > 80%), irrespective of disease (all < 0.01). The odds of being correctly labeled nonadherent according to claims decreased with age (diabetes odds ratio [OR]: 0.82, 95% CI = 0.694-0.976; hypertension OR: 0.86, 95% CI = 0.773-0.944) but were higher among those taking more medications (diabetes OR: 2.97, 95% CI = 1.658-5.326; hypertension OR: 3.0, 95% CI = 2.095-4.305). Following the nonadherence coding, the odds of being adherent increased with age in both models (diabetes OR: 1.17, 95% CI = 1.012-1.363; hypertension OR: 1.13, 95% CI = 1.048-1.223) yet decreased with increasing medications (diabetes OR: 0.25, 95% CI = 0.138-0.468; hypertension OR: 0.47, 95% CI = 0.368-0.592) and were lower if the patient was observed to be nonadherent before the index encounter (diabetes OR: 0.33, 95% CI = 0.146-0.760; hypertension OR: 0.25, 95% CI = 0.152-0.423). In general, providers are properly classifying patients as nonadherent using ICD-10-CM codes, but additional assessment is needed to determine the reasons for the remaining mismatch between claims- and diagnosis-based nonadherence. In addition, the correct claims-based metric needs to be established to improve alignment with provider interpretation of patient medication use. No outside funding supported this study. Gatwood reports grants from GlaxoSmithKline, Merck & Co., and AstraZeneca, outside the submitted work. Kovesdy reports consulting fees from Amgen, Sanofi, Fresenius Medical Care, Keryx, Bayer, Abbott, Abbvie, Dr. Schar, Astra-Zeneca, Takeda, Tricida, and Reata and grants from Shire, outside the submitted work. The other authors have nothing to disclose. Findings described in this article were presented as a poster at the American College of Clinical Pharmacy Annual Meeting in New York City, October 2019.