Pill count measures of compliance in a drug trial: variability and suitability

To evaluate pill counts as a compliance measure for drug trials, we followed 121 ambulatory hypertensives selected for good compliance over less than or equal to 12 months. The medication regimens consisted of either pinacidil or hydralazine as monotherapy or with propranolol and/or hydrochlorothiazide. Pill counts for the two primary drugs were obtained at each of the 20 return visits. The population was characterized by chronic uncomplicated hypertension and sociodemographic diversity; mean age was 53 years. Despite excellent average weekly pill counts (overall mean compliance rate [+/- SD] = 96.0 +/- 13.2%), we observed large intersubject and intrasubject variance in weekly pill count assessment: individuals' mean standard deviation = 13.7% (range = 0%-86%) and mean coefficient of variation = 0.138 (range = 0.001-0.410). By pill count, 35% of individuals exhibited greater than 110% compliance on at least 1 visit. We conclude that (a) pill count variability is large, even among highly selected subjects, (b) traditional reports of overall pill counts are suboptimal, and (c) pill counts may unreliably measure medication-taking behavior because "supranormal" compliance by this method is improbable but common.