Exploration of claims-based utilization measures for detecting potential nonmedical use of prescription drugs.

Abstract

Nonmedical use of prescription medication is a significant and growing public health concern in the United States. Drug utilization measures that can reliably quantify the extent of nonmedical use of a given medication or medication class would greatly facilitate efforts to identify, monitor, and constrain nonmedical prescription drug use. Measures making use of prescription claims data would be especially valuable given the ready availability of these data among health care plans and payers. To explore the extent to which claims-based utilization measures can provide information that aids in identifying and quantifying nonmedical drug use. Prescription claims from a large employer-based administrative claims database (MarketScan) were used to evaluate drug utilization during the first year after an index prescription for 6 classes of drugs with a known abuse potential and 3 classes without. Traditional population-level measures of adherence (i.e., medication possession ratio [MPR] and proportion of days covered [PDC]) and a novel measure of overlapping days supply (MPR/PDC ratio) were calculated for all medications. Measures of asymmetrical use within a population were evaluated with the Lorenz curve, representing the total drug supply used by the heaviest 1%, 5%, and 50% of all users. All of the measures across compounds were compared with Spearman nonparametric rank correlations, and the Friedman's test was performed to determine whether the rankings were consistent in pairwise analysis. The ability of each measure to discriminate between abusable and nonabusable compounds was evaluated using the c-statistic. The study cohort included 6,291,810 patients, mean age 52 years, 57.9% female. The mean MPR and mean PDC for drugs with known abuse potential were both lower than for drugs without known abuse potential. The MPR/PDC ratio (MPR:PDC) ranged from 1.02-1.09. Highest values for the Lorenz-1 curve were seen for acetaminophen with codeine, acetaminophen with oxycodone, oxycodone, and acetaminophen with hydrocodone. The individual MPR and PDC were strongly correlated to each other (r = 0.99; P less than 0.001 for both), moderately correlated with the MPR:PDC (r = 0.62 and 0.57, respectively) and moderately inversely correlated with Lorenz 1 (r = -0.83 and -0.86, respectively, P less than 0.001 for both). The MPR:PDC was inversely correlated with the Lorenz-1 (r = -0.39; P = 0.02). After rank ordering individual drugs by each measure from highest to lowest value, the MPR:PDC resulted in consistent distributions with the individual MPR (P = 0.0097) and PDC (P = 0.0018), but not the Lorenz-1 (P = 0.0835), nor Lorenz-50 (P = 0.1343). The MPR, PDC, Lorenz-1, and Lorenz-50 were able to discriminate between the drugs with known abuse potential and those without (c-statistic 0.979 to 1), while the MPR:PDC was not (c-statistic 0.592). When comparing classes of drugs with a known abuse potential with classes of drugs not prone to such use, using an array of drug utilization measures, significantly different patterns emerge, although the ability of these measures to serve as reliable indicators of the extent of nonmedical prescribing may be limited. There is a need for valid and reliable algorithms to detect the extent of nonmedical use at a population level to help target public health interventions aimed at constraining illicit use. Further work is warranted on the development of novel measures that make use of individual patient-level use patterns.