It is apparent that there are only few trials evaluating the effect of primary care physician education on health outcomes of patients. Especially during the last six years no results of education based interventions have been published. It seems that the research focus has shifted to more complex interventions encompassing collaborative and stepped care approaches by introducing mental health specialists to the primary care setting [46–48]. Regarding the results of the reviewed studies, this approach seems more than justified - it has yet to be shown that training practitioners alone yields significant symptom changes; however, this conclusion is only based on three relevant studies that themselves are highly diverse. While the study by Gask et al. (2004) struggles with high attrition rates, King et al. (2002) used a rather high cut-off and included chronically depressed patients, possibly leading to a conservative bias and therefore underestimating the treatment effect [38, 41]. The authors argue that the applied kind of brief cognitive behavioural therapy might have been treatment approach not sufficiently intense for highly depressed patients. Bosmans et al. (2006) find that including less severely affected patients might have led to an underestimation of the efficacy of anti-depressant treatment .
Sample selection plays a major role in assessing treatment effects in general. One could assume that severely affected patients benefit more from treatment in clinical studies (as they can show a higher reduction in quantity of symptoms). In line with this, a categorical diagnostic approach for patient inclusion by applying diagnostic categories as provided by the Diagnostic and Statistical Manual of Mental Disorders (DSM) might lead to a sample of more severely affected patients [33, 36, 37].
Furthermore, the kind of treatment has an effect. In the context of stepped care, this issue is addressed by providing more-intense treatment options to higher affected patients .
Both argumentations can provide explanations for the positive results found by studies implementing additional guideline usage by general practitioners. Small effect sizes were shown by those studies including patients with new-onset depression, rather than chronically depressed patients (as done by Kendrick et al., 2001 not resulting in positive symptom change). Obviously, the effect of mere attention to the trained practitioners as well as to the patients themselves (referred to as Hawthorne effect) has to be considered a possible moderating variable in study designs. This would lead to better outcomes and performances of the control groups even though they received no active intervention; thus, the differences found may possibly be even higher.
The justification for more complex interventions to improve primary care depression treatment is replicated in the analysis of included studies and basically goes in line with a previous review , however, we did find more evidence in newer studies that support guideline implementation to be effective in symptom reduction. One trial applying more complex strategies both yielded significant changes in symptom outcomes; however, it remains unclear how much of the effect can be attributed to the physicians' education. Bower et al. (2006) conducted a meta-regression to evaluate active ingredients of collaborative care interventions . In this analysis, primary care physician training is not associated with a positive change in depressive symptoms nor with a change in antidepressant use even in univariate calculations. Rather than provider education, systematic identification of patients, professional background of staff and supervision proved to be significant predictors of symptom change. It becomes clear that researchers should not assume an additive effect of treatment modules; especially in view of economic considerations, collaborative care cannot mean implementing as many treatment options as possible. This analysis of one specific potentially effective part of collaborative care is leading the way to a more thorough understanding of complex interventions and has to be pursued without neglecting the fact that more simple interventions can also lead to significant changes in patient outcomes as shown in this review.
However, it may not be appropriate to solely focus on outcomes of symptomatology as enhanced primary care supply may not be directly associated with such. Even the included studies show a rise in adherence to medication treatment  and medication treatment in general [34, 44, 50]. It has been shown that effectiveness of antidepressant treatment increases with depression severity ; an effect of increased antidepressant treatment in a sample of mildly depressed patients will therefore be small [as seen in the studies by [33, 39]].
Strengths and Limitations
This review only included randomised controlled trials, and therefore neglected observational and non-randomised studies. RCTs often adhere to strict exclusion criteria, thus making generalisability to primary care patients difficult. This also applies to the current review since studies with specialised co-morbid patient groups were excluded; however, regarding the heterogeneity of primary care patients, an adequate representation of studies seems hard to achieve in any case. The reported studies differ substantially in content, duration, intensity and frequency of the intervention programmes, making comparisons difficult. However, we were able to conduct a meta-analysis, quantifying the results of the studies. Meta-regression that could help determine the influence of these factors was not applicable due to the limited number of studies.
This partly results from the relatively narrow search strategy; only when education or training efforts of GPs were mentioned within title and abstract, the article was found with the applied search strategy. Earlier publications (before 1999) were not searched. Gilbody et al. mention one previous trial that showed positive effects of provider training but equally emphasise methodological weakness of this trial [21, 52], so we did include relevant trials that live up to methodological requirements.".
Furthermore, a possible publication bias cannot be ruled out or determined with a qualitative review as this, especially under the regard of not searching grey literature. Regarding the fact that almost only studies from anglophone countries were found might be an indicator for unpublished studies with negative outcomes from other countries.