A variety of approaches have been taken to the measurement of treatment adherence within research settings, each with their own advantages and disadvantages. For researchers to be able to draw firm conclusions from their work, it's important that measurement tools used have been appropriately validated. This section considers evidence in this area.

Within clinical settings, measurement tools must be practical - some methods commonly used in clinical trials are too resource-heavy to be used in day-to-day practice.

The principal techniques used are patient self-report, pill counts, electronic pill-container caps such as the Medication Event Monitoring System (MEMScaps), biological assays, or occasionally, the judgement of professionals.

An early review of the general literature on treatment adherence recommended biological assays as being the most accurate, followed by pill count and finally patient interviews (Eraker 1984). Unfortunately, these findings do not apply so easily to HIV treatments because some, particularly protease inhibitors, have such short plasma half-lives that measuring blood plasma levels only identifies whether patients took their most recent dose (Hecht 1998). Nevertheless, this may be a guide to adherence levels over longer periods, and there is evidence that drug levels can be markers for poor adherence (Hugen 2002). On the downside, drug level monitoring remains a somewhat experimental test within HIV care, and is less-well suited to accurate assessment of some therapies, specifically nucleoside analogues. Further to this, it is a doctor-led intervention which may be experienced by patients as intrusive - an attempt to 'catch them out'.

Research has shown physician judgement to be an unreliable measure of adherence to HAART (Paterson 2000; Liu 2001). For instance, a clinic study found that clinicians over-estimated adherence by 9% and correctly detected poor adherence in only 24-66% of cases, depending on the definition of non-adherence used (Miller 2002). Clinicians were most likely to over-estimate adherence among young people, long-standing patients and people with higher CD4 nadirs prior to commencement of treatment.

Self-report is widely used in adherence studies, not least because of its convenience. However it has been found to overestimate adherence when compared to other measures such as pill count (Hales 1998; Heinasmaki 1998), MEMScaps (Miller 2002) or both (Bangsberg). There is clearly potential for patients to under-report non-adherence, especially if there is a professional expectation of high adherence. However pill counts can easily be undermined by the removal of pills prior to going to clinic and MEMScaps can under-estimate adherence if patients take more than one dose from the bottle at once. A self-administered computer-based survey is an alternative form of self-report which may improve disclosure of non-adherence, although one study reported that half of people incorrectly reported their level of adherence using this form of self-report (Bangsberg 2002).

Nevertheless, there are data validating self-reported adherence as an accurate predictor of virological response to HAART (Mannheimer 2002, Walsh 2002). Mannheimer and colleagues employed a questionnaire which invited self-recall of adherence over the previous seven days, whilst John Walsh and colleagues validated the accuracy of two week and one month recall. These methods are easily adopted within clinical settings, and emphasise the active role of patients in their own care; two significant advantages.

A longitudinal study of 108 people taking antiretroviral therapy compared four types of adherence measures: MEMS, pill count, interview, and a composite adherence score which relied heavily on MEMS but drew on additional data from the other measures. The composite adherence score was most strongly associated with undetectable viral load after six months of treatment, followed by MEMS, pill count and self-report by interview. The latter was found to be a very poor predictor of adherence with only 14% sensitivity. Pill count was sensitive to non-adherence in only 75% of instances. In contrast, MEMS tended to under-estimate adherence. This may be due to the use of adherence-enhancing devices such as pill boxes which mean pill bottles are not opened for every dose (Liu 2001).

Because there is such a clear relationship between low adherence and virological or immunological failure, these surrogate markers may too be used to infer low adherence. Whilst viral rebound should always prompt doctors to discuss adherence behaviour with patients, clearly there are a variety of reasons for treatment failure, and even perfect adherence may not result in the desired response for some individuals.

References are listed in Interventions to improve adherence in Anti-HIV therapy: Adherence.