Current treatment for depression in primary care and other out-patient settings demonstrates a pattern that is incongruous with the magnitude of the burden of depression suggested by its associated disability.

To review important considerations in current depression treatment with a focus on antidepressant use.

Factors influencing the under treatment of depression in real-world settings are examined.

Patient and clinician behaviour as well as the incentives created by the health care system affect the likelihood of realising effective antidepressant therapy in practice.

Given the complexities of clinical practice, selection criteria for an antidepressant should include safety, efficacy and tolerability, as well as the ability of the antidepressant to deliver real-world efficacy while balancing health care costs in the long term.

Depression, which only a few decades ago was considered to be a short-term illness requiring short-term treatment, is now recognised as a recurrent, sometimes chronic, long-term illness.

To highlight the clinical importance of long-term antidepressant therapy in the treatment of depression.

The current literature was reviewed to examine the relationship between duration of antidepressant therapy and efficacy.

Approximately one-third to a half of patients successfully stabilised in acute-phase treatment will relapse if medication is not sustained throughout the continuation period. Only 10–15% will relapse if medication is continued. For maintenance-phase therapy, approximately 60% of patients at risk will experience a recurrent episode of depression within I year if untreated, whereas those who continue in treatment will have a recurrence rate of between 10% and 30%.

Risk of relapse and recurrence of depression can be significantly reduced if adequate continuation and maintenance therapy durations are achieved.

Although the efficacy of antidepressants has been demonstrated in randomised, controlled clinical trials, it is how an antidepressant is used in clinical practice that determines its clinical effectiveness, or real-world efficacy.

To explore the frequency with which antidepressants are used at adequate dose and duration to obtain remission of symptoms and prevent relapse in clinical practice and discuss potential implications for clinical outcomes.

Studies of antidepressant prescribing were reviewed and comparisons made between antidepressant classes and individual compounds within those classes.

Naturalistic studies show that patients who begin therapy on tricyclic antidepressants often receive sub-therapeutic doses for inadequate duration; conversely patients who begin therapy on selective serotonin reuptake inhibitors more often receive an adequate dose of therapy for a longer duration.

How antidepressants are used in clinical practice can determine the clinical outcomes that are achieved. Antidepressants that are more forgiving of sub-optimal prescribing and use patterns by providers and patients, respectively, may help to improve real-world efficacy.

Economic considerations increasingly play a role in the selection of antidepressant drugs and are often based on analyses from prospective and retrospective studies. However, the non-randomisation found in retrospective studies may result in significant selection bias.

To highlight the use of statistical methods in non-randomised studies and the application of those methods to economic analyses.

The literature on the observational studies of economic outcomes with alternative antidepressants is reviewed and several statistical methodologies to control for biases that can occur in non-randomised study designs are described.

In comparisons of antidepressant drugs, differences in acquisition costs are consistently found to be at least offset by other components of care when broad measures of health care resource utilisation are considered.

Economic evaluations of antidepressants should be based on broad measures of health care expenditure and can rely on data generated in real-world settings if appropriate statistical methods are used to control for the potential biases of non-randomisation.