Good and poor adherence: optimal cut-point for adherence measures using administrative claims data.

OBJECTIVE: To identify the adherence value cut-off point that optimally stratifies good versus poor compliers using administratively derived adherence measures, the medication possession ratio (MPR) and the proportion of days covered (PDC) using hospitalization episode as the primary outcome among Medicaid eligible persons diagnosed with schizophrenia, diabetes, hypertension, congestive heart failure (CHF), or hyperlipidemia. RESEARCH DESIGN AND METHODS: This was a retrospective analysis of Arkansas Medicaid administrative claims data. Patients > or =18 years old had to have at least one ICD-9-CM code for the study diseases during the recruitment period July 2000 through April 2004 and be continuously eligible for 6 months prior and 24 months after their first prescription for the target condition. Adherence rates to disease-specific drug therapy were assessed during 1 year using MPR and PDC. MAIN OUTCOME MEASURE AND ANALYSIS SCHEME: The primary outcome measure was any-cause and disease-related hospitalization. Univariate logistic regression models were used to predict hospitalizations. The optimum adherence value was based on the adherence value that corresponded to the upper most left point of the ROC curve corresponding to the maximum specificity and sensitivity.
RESULTS: The optimal cut-off adherence value for the MPR and PDC in predicting any-cause hospitalization varied between 0.63 and 0.89 across the five cohorts. In predicting disease-specific hospitalization across the five cohorts, the optimal cut-off adherence values ranged from 0.58 to 0.85.
CONCLUSIONS: This study provided an initial empirical basis for selecting 0.80 as a reasonable cut-off point that stratifies adherent and non-adherent patients based on predicting subsequent hospitalization across several highly prevalent chronic diseases. This cut-off point has been widely used in previous research and our findings suggest that it may be valid in these conditions; it is based on a single outcome measure, and additional research using these methods to identify adherence thresholds using other outcome metrics such as laboratory or physiologic measures, which may be more strongly related to adherence, is warranted.