To retrieve the level of evidence for the prescription of ADs in non-psychiatric PC settings, an initial search was performed using the MeSH terms for the 24 medical conditions collected in our previous qualitative study and the terms “antidepressants” and “therapeutic use” in PubMed 21 . During this process, 30 additional conditions possibly leading to AD prescription were found resulting in a list of 54 conditions or medical circumstances for which ADs could possibly be prescribed. Ten conditions - psychiatric disorders, non-primary care and non-adult conditions - were then excluded: five purely psychiatric disorders, narcolepsy and myotonia because of their low prevalence in primary care, nocturnal enuresis as a paediatric condition and amphetamine withdrawal and cocain dependence being conditions to be managed in multidisciplinary and specialised environment. This process resulted in a list of 44 non-psychiatric disorders manageable in PC and potentially leading to AD prescription in PC (Table 1). This list was used for the data collection search strategy.

The terms matching with the 44 conditions and medical circumstances for possible AD prescription were listed in French and English. The MeSH validated translation and synonyms were searched for in the terminology section of the catalogue and index of the French-language Health Resources Website (CISMeF). When there was no MeSH term available for the medical condition or circumstance, the researchers used a broader MeSH term, on a higher level of hierarchy, including the description (e.g., cancer AND antidepressants for “cancer related fatigue, musculoskeletal diseases AND antidepressants for musculoskeletal symptoms). Using this list, searches were performed to find the available guidelines (GL) in the databases of the following government agencies: the French “Haute Autorité de Santé” (HAS), the English National Institute for Clinical Excellence (NICE) and the National Guideline Clearinghouse (NGC), the last one including most international guidelines. The searches were performed from March 2011 to August 2011. The search covered all of the documents made available by the three agencies.

During the same period, we performed simultaneously, a search in the Cochrane database of systematic reviews (CDSR) and in Medline on the same basis, using the same criteria and process. For the Medline search, the following limits were used: “clinical trial, meta-analysis, clinical guidelines, randomized controlled trial, and adults, published after 2004”. The aim was to find scientific data published after the launch of the guidelines. Those studies might change the evidence and possibly lead to a change in prescription attitudes. All detailed searches are available in the full version of all guidelines. Individual studies (RCTs in Pubmed) included in this review did not constitute the bulk of the results.

All guidelines before 1997 were excluded, as were those regarding paediatrics, nursing practices, patient information or education, medical records, continuing medical education for care providers, medical imaging, biology and surgical techniques. Full text of guidelines have been downloaded from the agencies databases. All non-psychiatric conditions were included and all standalone psychiatric conditions were excluded. Remaining documents on anxiety, depression and side effects of ADs were also excluded after reading. The documents were independently read and assessed by two researchers (AM & IAA) and disagreements were resolved by discussion. Duplicates were removed. The documents in the databases were crosschecked using a search with the generic term “antidepressants” in order to retrieve any documents that may have been missed by previous searches. In the full text documents, the terms “antidepressant”, “Tricyclic agents”, “TCA”, “SNRI”, “serotonin”, “SSRI”, “tricyclic”, “imipramine”, “monoamine”, “duloxetine”, “venlafaxine” and their French translations were searched for. All guidelines containing the keywords were selected. In order to find the latest evidence, the most recent documents available on each topic in each guideline database were retained when available, and the most up-to-date version of the selected guidelines was kept. The overall process ended with a list of 78 documents.

The final stage of the study aimed to retrieve guidelines on recommended therapeutic use of ADs from these documents. Sections of specific interest were extracted and analysed. Information on pharmaceutical classes, drugs, recommended therapeutic use and level of evidence was collected when available. Agreements and discordances were noted, and inconsistencies between the various guidelines were highlighted. Recommendations from the guidelines were graded according to the EFNS scheme 22 (Additional file 1).

All results for pain conditions are described in Table 3.

(1) Neuropathic pain is related to different treatment strategies and different conditions detailed in this table.

(2) Diffuse pain, refractory or recurrent pain, central pain, pain connected with multiple sclerosis, dysesthesia after stroke or paraplegia.

(3) See also comments in plain text: General or non-specific conditions and general symptoms.

The guidelines, reviews and RCTs found prescription of sertraline, citalopram and trazodone to be potentially beneficial in treating behavioural perturbations, mood disorders and agitation in patients with dementia, though no level of evidence was available 40 41 . Potential side effects and difficulty in managing the prescription were emphasised 42 . ADs were found to have no specific effects in treating Parkinson’s disease, apart from those in treating its associated psychiatric indications 43 .

Concerning stroke, three kinds of problems were assessed: treatment of emotionalism, prevention of depression and the benefit of prescribing an AD in the acute stage to facilitate recovery of motor skills. ADs were recommended for emotional instability (Level B) 44 45 . They reduced the frequency and severity of crying and laughing episodes. The effect did not seem specific to one drug or class of drugs. Early prescription prevented depression, but no improvement in its severity was found when depression was actually occurring 46 . Early prescription of fluoxetine with physiotherapy found that patients with ischemic stroke and moderate to severe motor deficit could enhance motor recovery after 3 months 47 .

Antidepressants were found to provide no benefit in treating isolated sleeping disorders and primary insomnia even though they were found to be potentially beneficial in the event of psychiatric comorbidity 48 .

Antidepressants were not recommended for use in cases of restless legs syndrome, which was in fact presented as a side effect of ADs 49 .

As well, there was no evidence of a benefit in prescribing ADs for cases of sialorrhea related to neurological conditions (Amyotrophic lateral sclerosis (ALS), Parkinson’s disease), although prescription was sometimes recommended 50 . Antidepressants were found to have no proof of benefit for cases of tinnitus 51 .

Duloxetine was found to have potential benefits as a second-line (Level C) treatment for patients with stress urinary incontinence. Duloxetine significantly improved quality of life but TCAs did not. Anticholinergic agents, such as TCAs, were found to have potential benefits for patients with overactive bladder syndrome 52 53 54 . Proof of benefit was observed for incontinence caused by other urological conditions. Antidepressants were cited as a potential source of side effects for these conditions 55 .

One Cochrane review attested that all SSRIs were highly effective in reducing symptoms related to severe premenstrual syndrome, (also called pre-menstrual dysphoric disorder or luteal phase dysphoric disorder) (SMD -0.53, 95% CI: 0.68 to -0.39; P < 0.00001) with no level of evidence available 56 . Another Cochrane review reported that SSRIs and SNRIs had a mild to moderate effect in reducing hot flashes during menopause in women with a history of breast cancer, as well as in men with a history of prostate cancer (Level B) 57 .

SSRIs were not recommended for erectile dysfunction. They were, however, recommended as a first-line (Level A) treatment for premature ejaculation 58 59 .

Nortriptyline was shown to be effective for tobacco withdrawal (OR 2.79 (95% CI: 1.70-4.59) whereas moclobemide, venlafaxine and SSRIs did not show any effectiveness (no level of evidence available) 60 . All guidelines and reviews agreed that ADs were not indicated for alcohol misuse or dependence 61 . It is clearly stated that depression is a direct consequence of alcohol abuse, and that AD prescription is useless if the patient does not stop drinking.

The term “fatigue” pooled together a wide variety of health problems. No data was retrieved for isolated fatigue. In the NICE guideline, ADs were not found to be beneficial in treating chronic fatigue syndrome. ADs were considered as useless for cancer-related fatigue but beneficial in treating depression related to purely physical conditions, with no effect on the physical conditions themselves 62 .

Prescribing an AD provided no benefit for musculoskeletal symptoms, except for fibromyalgia, which is assessed in the “pain conditions” section of the results (see also Table 3) 63 . No data was retrieved regarding prescription for unexplained complaints, somatoform disorders, treatment refusal, chronic pruritus, helping type 2 diabetes patients to lose weight, and improving medical adherence or patient compliance.

According to the CDSR, TCAs could be used as a second-line treatment and SSRIs as a third-line treatment, for irritable bowel syndrome (IBS) (no level of evidence available) 64 65 .

Antidepressants were found to be potentially useful with a high level of evidence for 15 conditions and potentially useful with a lower level of evidence for 5 others. Prescribing an AD was found to be potentially beneficial for patients with many pain conditions as well as urological, gastroenterological and gynaecological conditions. Nevertheless, while prescribing, the GP has to refer to on-label use for his own country, because unfortunately, proof of benefit is not related to “on label use”. The evidence reviewed provided insufficient support for prescribing ADs for 24 other conditions. Restless legs syndrome, non-specific urinary incontinence and erectile dysfunction exposed in the “proof of no benefit” section were side effects of ADs. For rheumatologic conditions, (musculoskeletal symptoms and non-specific low back pain) the “proof of no benefit” should be interpreted with caution, due to the difficulty in assessing these patients in a primary care setting. Pain symptoms or somatoform complaints may be similar to those encountered in fibromyalgia, a clinical syndrome that lacks a clear definition and is still a subject of debate 66 . For the other 15 conditions listed in the “No proof of benefit” section, this should be understood as “having no specific effect on the non-psychiatric condition mentioned”. For example, no guidelines found ADs to have a specific effect in treating Parkinson’s disease comparable to their effect in treating dementia-related agitation. Nevertheless, ADs should be prescribed in the event of a major depressive episode related to Parkinson. The same can be said for treating depression in physically ill people or those with cancer 62 . Thus, an AD considered as having no specific benefit for a given condition may prove beneficial for certain specific PC patients. As a result, it is difficult for prescribing physicians to assess the potential benefit of using them in clinical practice in PC settings where several conditions are involved.

The first strength of the study was the meticulous methodology and careful step-by-step process used to extract the data. The analysis was not started from a theoretical framework, but from real daily practice in primary care, based on the qualitative material of our previous research 21 . No previous study has assessed in that way the appropriateness of AD prescriptions for non-psychiatric conditions. This review gives a new insight on studies emphasizing misuse as an explanation for the increasing rate of AD prescription 67 68 69 70 . A second strength was that it provided all GPs a clear overview of information on ADs not limited to the information and guidelines available in their own country. Finally, a third strength of the study was that it updated the guidelines with recent reviews, in order not to miss relevant information. This study, being an analysis of guidelines and systematic reviews did not focus on a single condition, population or intervention, but on a family of drug (ADs) and their indications. Thus, the nature of this analysis made it impossible to perform a systematic review fulfilling the “Prisma” statements. Another limitation of this study was the consistency of the available data. First of all, levels of evidence were not always available. When available, levels of evidence were not always consistent between the different guidelines, as was the case with diabetic painful neuropathy. Additionally, either a given AD, only certain ADs or an entire class of ADs could be considered as having a potential benefit in treating a specific condition. For example, among the SSRIs, only citalopram and sertraline were assessed in treating dementia-related agitation. Also, the optimum dosage to be used in treating a condition was not always determined, as in the case of using amitryptiline to treat migraine or neuropathic pain. Finally a limitation of our study could be that some guidelines in other languages than French and English were not selected in the search. However we used the big databases such as Medline and NGC, which already include guidelines, sometimes translated in English, from many guideline developers worldwide.

Performing a strict assessment and comparison of several conditions appeared arduous: although the subjects of the recommendations were similar, the guidelines did not refer to strictly comparable conditions. For example difficulties were encountered with fibromyalgia, chronic fatigue syndrome and rheumatologic pain, which were difficult to distinguish from one another. This lack of precision can lead to difficulties in assessing and summarizing information, and consequently in correctly understanding the potential benefit of ADs for patients. As well, the randomised trials were designed for well-defined conditions and homogeneous populations, which is rarely the case for PC patients. The inclusion criteria for the studies often involved secondary care and did not reflect primary care situations.

The limitations of available data and the wide variety of conditions made it difficult to formulate clear recommendations. In any case, the results of the study provide new insight to enable clinicians to prescribe ADs with more accuracy, and could potentially serve as a point of reference for health policy organizations in specifying prescription rules.

Comparing the 24 non-psychiatric health conditions for which the GPs claimed to prescribe ADs 21 with the results of this review study, 12 conditions were found to be based on evidence. Of the 12 other conditions for which GPs claimed a prescription of an AD, there was no scientific evidence. In our qualitative study, the GPs correctly assessed the benefit of ADs in treating pain conditions 21 . This seems contradictory with observational studies finding that pain management could be improved in PC settings 71 . However GPs’ proper assessment of the benefit of ADs, depends on his ability to make a fine distinction between “the scientifically proven effect of an AD for a certain condition” and “the potential benefit of an AD at the right dose for a specific patient presenting with several psychiatric or non-psychiatric conditions”. Besides the GP’s decision is based not only on biomedical observations, but also follows a bio-psychosocial model in which patient-centeredness and the environment are taken into account 67 . The major benefit of this study is to understand better and justify GPs’ AD prescription behaviour.

On the whole, the guideline agencies often stressed a lack of head-to-head comparative trials assessing the clinical effectiveness of the drugs, which would allow physicians to prioritise the choice of therapy for a given condition. This is impossible without any involvement from healthcare institutions. Another challenge is to better understand the way physicians prescribe ADs in actual PC settings. We currently know that GPs’ decisions are shaped by a combination of diseases and psychosocial influences. The potential benefit of a prescription for a patient cannot be assessed without incorporating all of the dimensions of actual prescription. This requires a strict comparison that directly assesses the prescriptions, indications and patients’ environment, based on observational cohort studies to collect data on medical conditions, psychosocial environment and prescriptions in PC settings.