Variation in adherence measures as a function of calculation methods.

Abstract

We aim to compare different operational definitions of medication adherence as well as examine the within-patient variability among these measures among patients treated for multiple comorbid conditions. Electronically monitored adherence data from a study on comorbid conditions were examined using three different calculation methods. DAILY adherence calculated the number of administrations divided by the number prescribed, without considering inter-dose interval. TIMING used predefined inter-dose intervals. Measures were aggregated to six 30-day periods. A PILLCOUNT approach counted the total administrations divided by the expected number in each 30-day period. Within-patient variability was computed based on DAILY and TIMING results for each 30-day period. Results varied by adherence calculation method. PILLCOUNT demonstrated the largest adherence rates (89%-92%); DAILY rates were lower (79%-85%); and TIMING was the lowest (62%-68%) over the 6-month period. TIMING within-patient variability (29%-35%) was larger than DAILY (20%-25%). Differences among the three methods confirm the importance of the adherence definition. TIMING may underestimate medicinal effects because patients may take medication as instructed (e.g., with meals) rather than at fixed intervals. PILLCOUNT may overestimate adherence by not accounting for inconsistent use. DAILY may best provide daily estimates of correct administration. Higher variability for TIMING may indicate patients are more likely to vary time between doses. Adherence calculation methods are important in interpreting results. Variability measures provide a more complete picture of adherence and may raise the likelihood of effects on biological outcomes. We propose studies of adherence include calculation method in the definition of adherence.