Ten years ago research in this field was limited, but considerable progress has since been made. Although by no means complete, we now have data for estimating the extent of the problem, and there is an increasing awareness of its clinical and social impact and of the fact that high levels of patient adherence to treatment and physician adherence to best-practice protocols are important co-determinants of treatment outcome. There is broader recognition that, at least for those patients with severe and recurrent illness, a chronic disease model should be adopted.

Furthermore, practitioners treating patients with depression can be guided by several recent findings that are summarized below.

• If the problem of poor adherence is not addressed, 30 - 40% of patients will discontinue their medication early (after 12 weeks), regardless of perceived benefits or side-effects.

• Simple to follow advice and education such as that tested by Lin et al. (14) is beneficial, and such advice should be given both in the early phase of treatment (5) and repeated at later stages (28).

• If patients admit to poor adherence, then it is highly likely that they are not taking their medication as prescribed; if they report good adherence, but lack of clinical progress suggests that adherence may nevertheless be a problem, the most sensitive method of detection is electronic monitoring.

• There is at best only weak evidence that treatment with the newer antidepressants leads directly to better rates of adherence and this is therefore probably not a material factor in choice of medication.

• Improved patient outcomes in primary care are probably best achieved through complex interventions such as those used by Katon et al. comprising improved assessment and monitoring of patients and relapse prevention counselling, together with specific advice targeted at the needs and concerns of individual patients.

A considerable research agenda still remains. More accurate estimations of the prevalence of adherence are needed in addition to research to address and measure the different forms that poor adherence may take, e.g. patients missing doses, taking "drug holidays", substituting agents, changing dosing, not filling prescriptions or discontinuing treatment early. The ways in which primary care physicians assess depression and deliver treatment should be further explored to identify determinants that explain adherence (and nonadherence) behaviours. Electronic event monitoring systems offer a useful approach to measuring some forms of adherence. An improved understanding of the relationships between health beliefs and medication-taking behaviour should lead to more robust theoretical frameworks, and to more effective methods of improving adherence, that can be added to existing techniques. Depression management programmes of the type pioneered by Katon and others in the United States require evaluation in other health care systems to ascertain whether their apparent benefits are transferable to other situations.