We planned a three-arm parallel-group RCT with a sample of representative abstracts of RCTs of pharmaceutical treatment indexed in PubMed, each reported with 1) no mention of the funding source or authors’ financial CoI, 2) the funding source only or 3) the funding source and CoI.

We obtained ethics approval from the Institutional Review Board of Paris Descartes University, Paris, France (no. 2012-A00032-41), and the protocol is registered at ClinicalTrials.gov (NCT01679873).

The study was funded by a grant from the Fondation pour la RechercheMédicale (EquipeEspoir de la Recherche, 2010). The funders were not involved in the study design; data collection, analysis, or interpretation; the writing of the article; or the decision to submit for publication.

We selected a sample of RCTs indexed in PubMed and published from 1 January 2010 to 31 December 2010 in the Core Clinical Journals (a subset of 119 widely read journals published in English, covering all specialties of clinical medicine and public-health sciences and including all major medical journals). The search strategy is detailed in Additional file 1. One researcher screened all retrieved citations on the title, abstract and the full text when necessary. The eligibility criteria were an industry-funded RCT with at least one author having a financial CoI, testing superiority, assessing pharmaceutical treatment (drugs) prescribed in a primary-care setting (defined as drugs that may be prescribed by a general practitioner (GP) or prescribed for diseases managed jointly by a specialist and a GP), with a conclusion in favor of the beneficial effect of the experimental treatment.

Exclusion criteria were investigation of nonpharmaceutical treatments (that is, medical devices, patient education), equivalence or noninferiority trials, safety trials, trials assessing different pharmacological procedures, and abstracts reporting a negative or neutral conclusion. One researcher (CB), who is an assistant professor in primary care, screened all retrieved abstracts and selected abstracts following these inclusion criteria. In cases of doubt regarding the inclusion of an abstract, a second researcher (IB) evaluated the abstract to achieve consensus.

From the selected reports, the reviewer systematically extracted the industry funding source and all authors’ financial CoI. If these data were not reported in the abstract or full-text article, the abstract was excluded.

Of the 2,797 citations screened, reports of 75 RCTs were selected [see Additional files 2 and 3]. In 29 reports (37%), all authors had financial CoI. In 62 (81%), more than 50% of the authors had financial CoI. In 58 (77%), authors who had financial CoI were industry employees and were the first, second or last author in 34 (45%) reports.

The abstracts for all selected RCTs were standardized and modified. The journal name, date and registration number at ClinicalTrials.gov were deleted. Authors' names were substituted by names randomly selected among the 200 names most common in the United Kingdom [see Additional file 4]. Treatments were referred to as ‘experimental treatment A’ or ‘experimental treatment B’. Acronyms used for the study were deleted and dates in the text were modified to avoid trial recognition. Furthermore, each abstract was translated into French. As a quality procedure, three GPs read all abstracts to ensure the quality of the translation and the relevance of the abstracts for GPs.

The abstract for each RCT was reported in three formats as follows:

– Abstract with no mention of the funding source or CoI. If the funding was reported in the original abstract, it was removed.

Abstract reporting the funding source only. A heading ‘FUNDING’ was added after the abstract conclusions. Under this heading, we reported the name of the pharmaceutical industry funding the trial.

– Abstract reporting the funding source and CoI. Two headings were added after the abstract conclusions. Under the first heading ‘FUNDING’, we reported the name of the pharmaceutical industry funding the trial. Under the second heading ‘CONFLICTS OF INTEREST,’ we reported the initials of authors who 1) were employed by the industry and 2) had a financial interest and/or other relationship with the industry (for example, fees, travel costs, stock options, link with a family member employed by the industry). The number of authors with CoI varied in the RCT reports (for 37% of abstracts, all authors had financial CoI; for 81%, more than 50% of authors had financial CoI; for 45%, the first, second or last author was employed by the pharmaceutical industry) (see examples in Table 1).

CoI, conflicts of interest.

Consequently, we had 225 versions of the 75 abstracts selected: 75 with no mention of the funding source or CoI, 75 reporting the funding source only, and 75 reporting the funding source and CoI.

Eligible participants were French GP members of a network of GPs involved in clinical research. GPs were from all over France and had part-time activity in a university general practice department as a teacher or student supervisor.

GPs were invited by email to participate in a study evaluating the interpretation of abstracts of RCTs (with three reminders) or by an investigator during the national annual medical congress for general practice. We used two different sources of recruitment for pragmatic reasons. Such a strategy is very often used in most RCTs to increase the sample size and the generalizability of the study results. The GPs were informed that their participation would involve reading only one abstract of an RCT and answering some questions about their interpretation of the abstract. They were informed that the collected data would remain anonymous and that we would inform them of the study results when available. However, they did not know that abstracts were modified and that several formats were compared. As approved by the ethics committee, participant consent was considered obtained as soon as they logged onto the survey site. If GPs agreed to participate, they had to complete a questionnaire related to their general characteristics. They were instructed to carefully read one abstract that was randomly selected for them and they had to answer questions related to their interpretation of this abstract (see Outcomes and Additional files 5 and 6).

An independent statistician created a computerized randomization list with a 1:1:1 ratio using random block sizes. The sequence was generated to have the same number of assessments for each of the three formats of a given abstract. A computer engineer uploaded the randomization list to a secure Internet system to assure allocation concealment. Participants logged onto this secure Internet system and were randomized to assess one of the 75 abstracts presented in one of the three formats.

Participants were informed that the aim of the study was to assess their interpretation of RCT abstracts, but they did not know that abstracts were modified and that several formats were compared.

The primary outcome was GPs’ confidence in the abstract’s conclusions. GPs answered the following question: ‘On a scale ranging from 0 to 10, indicate your confidence in the conclusion reported,’ with 0, not at all confident, to 10, completely confident. The secondary endpoints were the perceived methodological quality of the study and the interpretation of the treatment benefit in terms of safety and efficacy. For these outcomes, GPs were asked ‘On a scale ranging from 0 to 10, what is the methodological quality of the study?’ with 0, very poor quality, to 10, excellent quality, and ‘On a scale ranging from 0 to 10, is the experimental treatment beneficial in terms of safety and efficacy?’ with 0, not at all beneficial, to 10, totally beneficial.

Details of the questionnaire are in Additional file 6.

With a significance level of 1.67% fixed for each of the three 2 × 2 comparisons (Bonferroni correction to maintain an overall significance level of 5%), we needed 118 evaluations for each format to demonstrate an effect size of 0.5 on the numeric scale with a power of 90% for each 2 × 2 comparison. Because we expected that each participant would read a single abstract, we needed to include 118 participants per arm (354 in total).

Descriptive results are reported with means and standard deviations (SDs), median and interquartile ranges (Q1 to Q3) for quantitative variables and frequencies and percentages by modality for qualitative variables.

We had three main comparisons: abstracts with no mention of the funding source or CoI versus reporting the funding source only; abstracts with no mention of the funding source or CoI versus reporting the funding source and CoI; and abstracts reporting the funding source only versus reporting the funding source and CoI. Differences in primary outcome between groups were estimated by a linear model and were compared by Tukey’s honestly significant difference test to adjust for multiple testing 26 . Differences in secondary outcomes were compared in the same manner as for the primary outcome. Post hoc sensitivity analysis adjusted on the mode of recruitment were performed. All analyses involved use of R software v3.0.1 27 .

The flow of participants is shown in Figure 1. Between October 2012 and June 2013, among 605 GPs contacted, 354 (58%) agreed to participate, and 118 were allocated to each arm. The characteristics of GPs are given in Table 2. In all, 65% were men; the median (Q1 to Q3) age was 51 (36 to 57). Less than half (43%) had participated in a pharmaceutical industry–funded clinical trial. Only 22% had received some fees from the pharmaceutical industry for speaking, consulting or enrolling patients in trials. Half had not received any visits from pharmaceutical industry representatives.

CoI, conflicts of interest.

Abstracts were read by a median of six GPs (Q1 to Q3, 3 to 6). In all, 34 abstracts were read three times (once in each group), 39 were read six times (twice in each group) and two were read nine times (three times in each group).

Data on the mean (SD) GP confidence in abstract conclusions are reported in Table 3. The mean difference (95% confidence interval (CI)) in confidence with conclusions was 0.2 (-0.6; 1.0) (P = 0.84) for abstracts reporting the funding source only versus no funding source or CoI; -0.4 (-1.3; 0.4) (P = 0.39) for abstracts reporting the funding source and CoI versus no funding source or CoI; and -0.6 (-1.5; 0.2) (P = 0.15) for abstracts reporting the funding source and CoI versus the funding source only (Table 3). The adjusted post-hoc analysis did not obtain exactly the same results as the analysis that was not adjusted on the mode of recruitment. However the adjusted results were very consistent and confirm that the results are robust (see Additional file 7).

The confidence in abstract’s conclusions is evaluated on a 0, (not at all), to 10, (completely confident) scale.

The methodological quality of the study is evaluated on a 0, (very poor), to 10, (excellent quality) scale.

Treatment benefit is evaluated on a 0, (not at all), to 10, (completely confident) scale.

Differences in outcomes between groups were estimated using a linear model and were compared with a Tukey’s HSD test 26 . CI, confidence interval; CoI, conflicts of interest; GP, general practitioner; HSD, honestly significant difference; SD, standard deviation.

The mean difference (95% CI) in perception of quality was 0.1 (-0.7; 0.9) (P = 0.97) for abstracts reporting the funding source only versus no funding source or CoI; -0.4 (-1.2; 0.4) (P = 0.41) for abstracts reporting the funding source and CoI versus no funding source or CoI; and -0.5 (-1.3; 0.3) (P = 0.30) for abstracts reporting the funding source and CoI versus the funding source only (Table 3).

The mean difference (95% CI) in assessment of treatment benefit was -0.1 (-1.0; 0.7) (P = 0.93) for abstracts reporting the funding source only versus no funding source or CoI; -0.8 (-1.7; 0.02) (P = 0.06) for abstracts reporting the funding source and CoI versus no funding source or CoI; and -0.7 (-1.5; 0.1) (P = 0.13) for abstracts reporting the funding source and CoI versus the funding source only (Table 3).