Groups were chosen using a purposeful sampling method. This study population was drawn from two different sources. Firstly, four focus groups (FG) were held with GPs recruited from existing "quality circles". These GPs already knew each other from previous professional training sessions based on the French "peer group" technique. In this technique one of the participants presents to the group of GPs a structured analysis of patient files with a specific medical condition (e.g. diabetes, medications for high blood pressure). The group analyses all of the issues, decisions or potential improvements to the care of the patient and provides feedback to the participant presenting the case. The dynamic interaction amongst these group members enabled us to collect reliable data on their actual practices. Secondly, we organized two focus groups with locum GPs. Locum physicians could have a different, more critical point of view on the diagnostic and therapeutic options of the doctors they stand in for: They were taught about the importance of the guidelines during their recent initial medical training. All of the locums were replacing either rural or urban GPs and not only private-practice GPs (GP principals) involved in the peer groups. The option of running FG with locums was chosen in order to disclose attitudes or facts that the initial prescriber may have concealed. The first group of locums were invited using purposeful sampling and the second by a local association of locums. Four of the locum participants were unable to take part in the focus group and were invited for a semi-structured interview using the same topic guide. All participants were sent an invitation by the group leader, and also received a telephone call informing them about the general subject of the focus group session. The participants gave their informed consent before participating. According to the rules of the research centre, ethical approval was not necessary for this study. Five focus groups were held in Normandy (two in rural areas), and the other took place in a northern suburb of Paris. Overall, six focus groups were held, using a continuous process of collection and analysis. We ensured that the focus groups included younger and older, more experienced male and female private-practice GPs and locums from both urban and rural areas. Data saturation was reached after 4 focus groups with GP principals, and the second focus group with the locums did not bring any additional material. Information on participant profiles is detailed in the "results" section.

Qualitative data were collected through focus groups and interviews with GPs. The focus groups were conducted between November 2008 and April 2009. A topic guide was created based on the existing literature 17 and the researchers' previous experiences 18. The interviews focused on the GPs' experiences, circumstances, influences, justifications, and explanations, concerning both the psychiatric and non-psychiatric conditions for which the antidepressants were described. Two trained moderators briefly explained the aim of the study and led the focus groups, using a detailed topic guide (table 1). The sessions began by assessing the difficulty of taking care of certain patients based on guidelines. The relevance of the GPs' prescriptions was never challenged. The participants were asked to share their personal experiences in dealing with antidepressant drugs. The moderator ensured that all issues were covered and that everyone participated. One researcher observed the group members, without intervening, and gathered information on nonverbal communication and the interaction between the participants. Audio recordings of all of the focus groups were made, and these recordings were later transcribed.

A phenomenological approach was used. The first aim of this approach was to gather material coming from GPs personal experiences in their own real situations. Secondly, the analysis focused on what doctors said about themselves or their peers regarding their experiences with prescribing ADs. We gathered information on what the GPs said regarding the social, family and professional situations of their patients and on how they saw the prescription of ADs helping them to solve their patients' problems. The final aim was to identify the background and disease conditions that influenced the doctors' decision to prescribe. QSR Nvivo 8.0 Software was used to perform the analysis.

In practice, information on non-psychiatric diagnoses and situations other than those based on guideline criteria that lead to prescription of an antidepressant was brought together. Firstly, the researchers listened to the audio recordings and noted any emerging themes. Secondly, working on the written transcription, each researcher created a code list without any pre-conceived framework (open coding). These codes were shared and discussed within the research team. Units of information (codes on situations, e.g. diseases and social conditions) and concepts (e.g. the doctor's role, the doctor as a medication and placebos) were labelled. This information was examined and cross-analyzed with semantic indicators of doubt, insistence and certainty. The codebook was continuously revised, with the researchers comparing all of the codes in the event of disagreement and attempting to clarify their meaning, going back to the context until mutual consent was reached. Following an axial coding process, a matrix was developed in order to highlight the GPs' opinions about ADs in these various situations and to focus on the decision-making process.

The characteristics of the participants and practices are summarized in table 2.

First of all, the expected usefulness of AD prescription for several medical conditions is presented. Many conditions result in prescription, which is seen as effective but generally only with a symptomatic effect. Secondly, the GPs' options for assessing patients and situations are presented. We will see how GPs cope with patient demands, struggling through difficulties and inventing a personal approach.

Several points support these findings. Firstly, all of the GPs harshly criticized the guideline criteria and agreed they were irrelevant in primary care settings, thus clearly and explicitly assuming responsibility for the nature of their actual practices. From this point of view, the study's phenomenological approach was successful: the GPs recounted their true-life experiences with prescribing ADs, and the analysis was based on these experiences, not on general opinions. This was facilitated by a comprehensive approach: the participating GPs did know that this study was being conducted by GP researchers. This option was chosen in order to ensure the GPs that they would not be judged by psychiatrists: otherwise, the GPs may not have discussed their choices and difficulties so freely. Secondly, a wide variety of diagnoses, in addition to psychiatric conditions, were assessed. This leads us to believe that the GPs talked very freely about their real problems, even though this study was based on related behaviours. Finally, and despite the fact that they were younger and less experienced, the locums mostly shared the same opinions. We clearly made this choice in order to try to discover borderline therapeutic choices among GPs who would have declined the invitation to take part in a focus group on the sensitive topic of AD prescription. None of the locums recounted unexplained prescription of ADs. The choice to use a sample composed of both urban and rural GPs was also made to try to discover behaviours and difficulties related to overbooked practices or a variety of patient situations. No clear differences in their approach or behaviour appeared during the data analysis. This was the case for both younger and older, more experienced male and female GPs. This strengthens the hypothesis that our outcomes are on the whole related to primary care situations, and not only to certain practice characteristics.

Conducting focus groups on this sensitive subject could have proven to be a limitation, hiding relevant material: the GPs could have chosen to conceal odd prescriptions or decisions not covered by validated scale criteria. Nevertheless, the options chosen for the sample could not have selected doctors very interested in this topic, or on the contrary those with a strong aversion to psychiatric conditions. Another limitation is the need for more insight on the exact usefulness of the GP's personal way of assessing a patient: feelings and knowledge are well known in primary care, but we did not collect any relevant material on the way the GPs made them part of their decision-making process. One key finding was that the GPs used a personal scale, but that no data were collected to evaluate the actual performances of these scales. No evidence was gathered concerning a threshold, determined using these skills, after which an antidepressant is prescribed. Our study was not designed to explore these aspects in detail. The outcomes enable us only to make hypothesis on GPs' possibly prescribing more than necessary, and not to focus on a particular explanation. It is already well known that the positive predictive value for routine diagnosis of depression is low 18. Could GP assessments using these personal tools increase this rate?

Another finding concerned GPs' doubts regarding the usefulness of guideline depression scales in primary care settings. In our study, we saw GPs reinventing criteria. This is consistent with other studies, showing that physicians' familiarity with a patient was an important condition in recognizing and managing depression 23. If physicians do not see these tools as useful and do not use them at all, we should perhaps draw some conclusions: instead of making every effort to generalize the use of these tools, shouldn't we assess what is currently used in real practice? Such an assessment would likely result in practices that were more often used in everyday primary care settings. The relevance of antidepressant treatment in non-psychiatric situations in primary care seems to be a key question amongst many researchers. Various studies and trials have tested the usefulness of antidepressant drugs in "non-psychiatric" situations 2425. As regards primary care, the external validity of these studies is controversial, as primary care rarely involves one clear condition, but rather combines a wide variety of physical symptoms, psychological distresses and social issues. The usefulness of antidepressant treatment in various primary care settings needs to be comprehensively assessed. Prescribing fewer drugs should be considered along with making counselling and psychotherapy more available. The next step is to try and understand decision making in real settings by collecting data on antidepressant prescriptions from patient records.