Improving on a Coin Toss To Predict Patient Adherence to Medications

• Patient compliance
• Human immunodeficiency virus infections
• Viral load
• Interviews
• Antiretroviral therapy, highly active

According to the father of medicine, Hippocrates, “The physician must not only be prepared to do what is right himself, but also to make the patient … cooperate” (1). Although Hippocrates' paternalistic tone might affront modern physicians and patients, the challenge of promoting patient adherence to effective treatments still confounds providers 23 centuries later. Adherence has become an increasingly central aspect of patient care because of the burgeoning array of effective treatments for many chronic diseases. In the face of remarkable medical advances, it is especially distressing that evaluation of patient adherence (also called compliance or pill taking) remains an imperfect science at best.

The article by Liu and colleagues in this issue [2] contributes to the field of adherence measurement by demonstrating an improvement in predictive validity after incorporation of data from several adherence measures when data from an electronic monitor were missing. In this study of 108 community-based patients starting highly active antiretroviral therapy, three measures were used: 1) electronic monitoring by a special pill bottle cap with a microchip that records each time the bottle is opened (Medication Event Monitoring System [MEMS]], 2) pill count, and 3) patient interview. The authors used data from these three measures to create yet another measure that they termed a composite adherence score (CAS). This additional measure should really have been called “supplemented MEMS.” When MEMS data were missing or inaccurate, the investigators replaced the data points first with pill count and then, if needed, with interview data on adherence. In an innovative move, the authors did not directly use the values from the pill count and interview measures, which seemed to overestimate adherence, to supplement the MEMS data. Instead, they calibrated pill count and interview data to adjust for this overestimate before incorporating these values when MEMS data were …