There is no accepted definition of what constitutes onset of antidepressant effect, and this limits evaluation of the validity of many comparative reports. In order to clarify the nature of the comparative methodology required, it is proposed that two distinct research questions are separated conceptually: 1) when do antidepressant drugs begin to clinically act? and 2) do all antidepressant drugs produce relief of symptoms in patients within a similar time period? A methodological distinction may also be made in terms of whether to assess the onset of effect of an antidepressant as having an absolute quality (“fast”) or as being fast relative to other treatments (“faster”). The choice of aim will help to define the methodological approach, the applicability of the findings and the difficulties to be addressed. Describing an antidepressant as showing a “fast” or “faster” onset of response requires that any description attempts to define (in terms that may be generalised) the nature of the comparison (faster than what?), the nature of the effect (faster to do what?) and the population in whom this effect may be expected (faster in whom?). Some details of methodological approaches are reviewed, and suggestions for study designs are made.

The issue of early onset of action (EOA) of an antidepressant was addressed by several authors. Unfortunately so far there is neither consensus nor convention on the definitions of EOA, or on measures and methods of assessments. There are several quite different approaches to the statistical analysis of the data. Our objective was to compare the results concerning EOA obtained by different statistical techniques applied on a data set which was generated by several independent conclusive double-blind, placebo controlled randomised trials with two antidepressants. The following statistical techniques were used: 1) statistically significant difference for the first time; 2) statistically significant and clinically relevant difference for the first time; 3) pattern analysis; 4) classical survival analysis without ‘sustained response’; 5) survival analysis with sustained response. The advantages and drawbacks of different methods are discussed.

In a meta-analysis of the selective serotonin reuptake inhibitor fluoxetine (n = 440) and the reversible and selective monoamine oxidase-A inhibitor moclobemide (n = 437) we investigated the time course of improvement over 6 weeks for these two classes of anti-depressants. Two different methods of approach were applied: (1) repeated measurement analysis of variance in combination with regression analysis, and (2) a survival-analytical approach that allowed us to determine onset of improvement for the individual patient, to distinguish between early and late improvers, to assess the predictive value of early improvement, and to adequately process premature withdrawals. The two antidepressants of large biochemical and pharmacological differences yielded virtually identical response- and drop-out rates, and exhibited no difference with respect to the time course of improvement. Onset of improvement occurred in more than half of the patients within the first 2 weeks of treatment, and early improvement was highly predictive of later outcome. In particular, there was no indication for a delayed onset of action of antidepressants. We suggest that future studies should be standardised with respect to washout period and sufficiently dense assessments during the first 2 weeks of trial. Thus, patient samples can be subdivided reproducibly into early-, late-, partial- and non-remitters. This, in turn, may help to identify “true” drug responders, and to discover why in some patients recovery becomes “stuck”. Data already held by the pharmaceutical companies should be reanalysed in this respect.

The onset of activity of antidepressants is a clinically and theoretically important criterion of efficacy. We estimated the onset of activity by an intraindividual regression analytical approach. It is assumed that efficient treatment causes a fast shift of depression scores from a baseline level to an end level. From this assumption follows a limited number of opportunities for the onset of activity to occur. These opportunities correspond to different regression models which are estimated for each patient's depression course. The day of onset is concluded from the regression model which produces the best fit to the empirical depression curve. Results of a double-blind clinical trial comparing mianserin and amitriptyline are reported (n = 28). The methodological approach confirms that observer ratings by physicists show a significantly earlier onset of activity than self ratings by the patients. The D-S also shows a significantly earlier onset of activity of mianserin compared to amitriptyline. The discussion emphasises methodological aspects.

In order to identify clinical and demographic variables that predict response to antidepressants and to analyse prediction of outcome as a function of definition of outcome we analysed pooled data of two independent, multicentre, double blind parallel group studies. Study I compared the efficacy of mianserin with that of fluoxetine in 65, and study II compared mianserin with fluvoxamine in 60 patients with depression. Improvement was defined as at least 20% decrease in MADRS by day 14. Patients were considered as responders if they had greater than 50% decrease and non-responders if they had ≤ 50% decrease from baseline in the MADRS at day 56. Complete remission was defined as MADRS score ≤ 6 at day 56. Patients' characteristics did not differ between mianserin and SSRI groups. Early improvement predicted response in 92% and complete remission in 55% of the patients improved at day 14. Multivariate forward stepwise logistic regression analysis showed that response to treatments at day 56 was significantly (P = 0.0003) associated with early improvement, age (responders had higher age than non-responders) and weight (responders weighted more than non-responders). Complete remission was only predicted by early improvement. Treatments could not be differentiated when data were analysed according to responder/nonresponder status or complete remission/no complete remission. However, when the same data were analysed by analysis of variance a significant treatment effect (P = 0.02, mianserin > SSRIs) and a quadratic type treatment by time interaction (P = 0.023) were found. The robustness of the analysis was further improved by inclusion of covariates (age, weight). Early clinical improvement seems to be the best predictor of 2 month response to antidepressants (mianserin, SSRIs). Younger age and lower weight may predict non-response. Quantitative analysis differentiates treatments better than analysis of responder status. As obtenation of complete remission is a realistic objective with current antidepressants, studies longer than 2 months are needed to assess effectiveness of these drugs in the obtenation of complete remission.

Several well-known observer scales, including the Hamilton Depression Scale (HAM-D), Montgomery-Åsberg Scale (MADRS), Major Depression Rating Scale (MDS), Melancholia Scale (MES), and Inventory for Depressive Symptomatology (IDS) used for measuring severity of depressive states have been compared by their responsiveness in an open trial including patients treated with a combination of citalopram and mianserin. The patients fulfilled the Diagnostic and Statistical Manual (DSM)-IV criteria for major depressive episode, and all scored 18 or more on the HAM-D before treatment. Onset of antidepressant action was defined as an improvement of rating scale scores of 25% or more of pre-treatment scores. A response to treatment was defined as a reduction of 50% or more on the pre-treatment scores. The results showed that the number of treatment days until improvement was 11 to 13 with no difference between the scales. The days until response were between 18 and 21 with no difference between the scales. In conclusion, the depression scales were found to be equal in their ability to detect changes in depressive symptoms during treatment. The mean of days to response was 19 for the combination of citalopram and mianserin. This is similar to the response for the combination of fluoxetine and pinolol.

Assessments of early response to antidepressant therapy based on condition severity have been contradictory due to the poor measurement properties of the instruments employed and the inappropriateness of condition severity as a criterion of improvement. It is argued that a more valid approach would involve the assessment of improvements in quality of life (QoL) associated with effective therapy. QoL instruments should reflect the views of the patients themselves, the only group able to comment authoritatively on whether they feel benefit from treatment. Research shows that it is the perceptions of the patients that determine whether treatment is sought, how far clinical advice is followed and whether treatment is discontinued. It is argued that the Quality of Life in Depression Scale is the only instrument available that has the required conceptual basis and psychometric properties to investigate the onset of response to antidepressant therapy.

Assessing the effect of novel pharmaceutical treatments on the quality of life (QoL) of a patient, or group of patients, has been approached in numerous ways over the last 20 years. Techniques as diverse as single questions to multidimensional scales requiring trained assessors to devote several hours to each assessment; from generic questions about how life might have changed to specific issues such as the ability to use a toothbrush. In the pharmaceutical industry, the emphasis is on the registration of a product with national licensing bodies. Each body has tended to see the issue from a different perspective, which has driven study designs to be different in different countries; even different over time within one country. This paper emphasises the basic statistical steps necessary to ensure that a measure of QoL is appropriately recorded, while retaining sufficient flexibility to support the registration in several countries. Aspects about possible study design are included to assist with developing some simple concepts about analysing and then interpreting the results. It is not the intention of the paper to provide the answer, merely to provide the tools to develop the answer robustly. Put briefly, with the right approach generic solutions are feasible and these solutions will have greater utility. The challenge is to recognise exactly what QoL is, and not to deviate from it.

Our aim was to determine the effect of risperidone monotherapy treatment on disability and on the quality of life of 318 schizophrenic outpatients who had been previously treated with other neuroleptics. Patients were assessed at baseline 2, 4 and 8 months using the BPRS, CGI, WHO/DDS and SF-36. BPRS scores showed a significant decrease at month 2, both in the total score and in each of the three clusters, negative, positive and depression/anxiety. WHO/DDS scores significantly decreased both in the overall score (from 51.8 to 37.4, P < 0.0001) and in the four dimensions. SF-36 scale scores and summary measures showed a significant improvement after 8 months of risperidone treatment. Risperidone long-term treatment (8 months) has a positive effect on clinical and on psychosocial outcomes. Females, paranoids patients and patients without history of use-abuse of substances showed greater improvements in quality of life.

The main feature of observational studies is the representation of naturalistic treatment conditions. In contrast to clinical trials, they allow the evaluation and quantification of adverse event profiles of drugs under “real life” conditions. The price for this unquestionable chance is the proneness to distorting factors, which may aggravate the interpretation of the study results. Analysis of observational study results therefore has to control for potentially influential factors and reconsider possible alternatives explaining observed associations. The most important distorting factors, which should be taken into account during analysis and interpretation are under-reporting, event selection, bias, confounding and misusage? Authors and readers of such study results should be aware of this possible sources of error, in order to derive optimal benefit from this study approach.