In a review of methodological and conceptual issues relevant to measuring adherence in patients with diabetes, Johnson (127) suggested that the prevalence of adherence may vary across the different components of the diabetes regimen and the patient's lifespan, during the course of the disease, as well as between populations of patients with diabetes (i.e. type 1 and type 2). Johnson also noted the conceptual problems encountered in defining and measuring adherence including:

- the absence of explicit adherence standards against which a patient's behaviour can be compared;

- inadvertent nonadherence attributable to miscommunication between patient and provider and deficits in the knowledge or skills of the patient;

- the behavioural complexity of the diabetes regimen; and

- the confounding of compliance with diabetes control.

Furthermore, the multiplicity of measurements used to assess adherence (i.e. health status indicators; provider ratings; behavioural observations; permanent products, and patient self-reports, including behaviour ratings, diaries and 24-hour recall interviews) also makes comparison of studies troublesome. Johnson concluded that a measurement method should be selected on the basis of reliability, validity, non-reactivity, sensitivity to the complexity of the diabetes regimen behaviours and measurement-independence from the indicators of health status. Glasgow et al. (30) also noted the methodological shortcomings of studies on diabetes self-care correlates, the lack of clear conceptualizations and the failure to differentiate between regimen adherence, self-care behaviour and metabolic control, as well as the empirical - atheoretical nature of many studies that lacked a comprehensive model or theory.

The present review of studies reported from 1980 to 2001, has revealed that research on adherence to treatment for diabetes yields some inconsistent findings. These inconsistent results may have several causes including variability in:

- research designs (e.g. longitudinal as opposed to cross-sectional studies) and study instruments;

- sampling frames employed for study recruitment;

- the use of general measures (e.g. general stress) as opposed to more specific ones (e.g. diabetes-specific stress);

- sample sizes (in some studies the small samples used decreased the likelihood of detecting significant associations between the variables); and

- lack of control of potentially confounding variables.