Adherence to multiple medications prescribed for a chronic disease: a methodological investigation

Many patients receive multiple medications for the treatment of a disease. While monitoring adherence is important, a composite measure of adherence is useful for estimating adherence to multiple medications in these patients. There are multiple ways to compute composite estimates of adherence to multiple medications, including (a) 80% of days covered by at least 1 medication ("at least 1"); (b) 80% of days covered by both medications ("both"); (c) 80% of days covered by each medication measured separately ("all"); and (d) computing an average of the individual medication adherence estimates ("average"). Comparison of adherence rates to individual medications and that of composite estimates are important for intervention decisions and effective disease management.

To (a) examine adherence to multiple medications prescribed for a disease; (b) estimate composite adherence to multiple medications prescribed for a disease; and (c) determine the rate of differential classification of a patient being adherent as is estimated by different available algorithms.

A retrospective cohort study was designed using 2002-2003 MarketScan Commercial Claims and Encounters data. To be included in the cohort, patients had to be less than aged 65 years and had to have separate prescriptions filled for 2 classes of diabetes medications (i.e., any sulfonylurea [SU] and any thiazolidinedione [TZD]) at least once; patients taking other diabetic medications over the observation period were excluded. Adherence was measured by proportion of days covered (PDC) over periods of 90 days (8 quarters total) and cumulatively over the 2-year study period. For some composite adherence estimates, patients were considered adherent if PDC ≥ 80%. Survival curves using the life-table method were constructed to compare the time until PDC became less than  80% as estimated by the 3 different categorical composite measures.

A total of 6,043 patients were included in the analysis. Across the 8 quarters under consideration, the average PDC estimates ranged between 69.8%-84.2% for SUs and 70.3%-85.6% for TZDs. The mean composite PDC based on the average algorithm varied between 69.4% and 84.9% when measured over each quarter or cumulatively. Similarly, the rates of composite adherence ranged from 74.5% to 88.2%, 46.4% to 61.2%, and 47.7% to 62.9% for the "at least 1," "both," and "all" methods, respectively. Many subjects were classified as adherent by 1 composite dichotomous measure but not by all 3 dichotomous measures (i.e., "all," "at least 1," and "both"); of these patients, 30.6%-38.2% were classified differently as to their adherence status over different quarters by different measures. Survival curves of categorical composite measures were different (P  less than  0.05) from one another. "At least 1" identified more patients as persistent and showed a much slower decline than did the "all" or "both" approaches.

Subjects were found to have a level of adherence-as estimated by individual medication adherence and composite adherence metrics-for multiple medications prescribed for a disease that, while not optimal from the perspective of patient care, was not entirely poor. In addition, composite estimates of adherence considerably varied depending on algorithms used. Most importantly, a large number of patients appeared to be subject to inconsistent classification based on adherence measurement algorithm. Adherence estimates produced by different composite measurement approaches give rise to difficulty in consistent interpretation, which may be detrimental to appropriate patient care decision making.