As is the case when attempting to measure patient behaviour in many other contexts, it is difficult to derive accurate estimates of patient adherence to medication for depression. Across studies, several techniques have been used including clinician estimation or patient self-report, pill-counting, estimation of blood levels of drug, metabolite or tracer substance, and the use of electronic monitoring systems that record pill dispensing. Two studies directly compared methods of measurement. In 1990 Kroll et al., using a small sample of patients with mixed diagnoses, demonstrated that levels of medication in the blood correlated with clinical outcome, and that many patients who claimed to be taking a medication regularly had low levels of it in their blood (3). In 2000, George et al. compared four methods of assessment in depressed patients treated by primary care practitioners, and were able to show that an event monitoring system (EMS) that electronically counted the amount of medication dispensed from its container was the most sensitive method of measuring adherence, although the specificity of a patient report of nonadherence was also high (4). Estimations of plasma levels of drugs and their metabolites were less useful. Although these types of measure overcome some of the bias associated with either physician observation or patient self-report, they still lack some of the features required of a "gold-standard" measure (i.e. being direct, objective and unobtrusive).
The second important methodological issue is the nature of the patient samples studied. Much research has been conducted on hospital outpatients or inpatients, or patients recruited into randomized trials to test the efficacy of medications. This pre-selection bias makes it very unlikely that the patients in these studies represent the true population of depressed patients receiving treatment in primary care settings. This makes it hard to generalize from the results of these studies.