In a first step, we identified orthopaedic clinical research questions that could be evaluated with IPD MAs. Then, using personalized emails, we systematically contacted corresponding authors of randomized controlled trials (RCTs) related to those clinical questions and asked them whether they would agree to share the IPD from their trials for an MA.
Identification of clinical research questions in orthopaedic surgery
To identify relevant clinical research questions in orthopaedic surgery that could justify an IPD MA, we relied on questions assessed in recently published systematic reviews with MAs of aggregated data.
Search of systematic reviews with MAs of aggregated data assessing orthopaedic surgical procedures
On 6 April 2014, we searched MEDLINE via PubMed for systematic reviews published between 1 January 2013 and 31 December 2013, using the search equation reported in Additional file 1. This search equation combined MESH terms related to: orthopaedics and free-text words corresponding to the main orthopaedic surgical procedures; publication type and free-text words corresponding tosystematic reviews and MAs; and a modified version of the Cochrane Highly Sensitive Search Strategy to identify RCTs. In addition, we performed a search of the Cochrane Database of Systematic Reviews (Orthopaedics and Trauma and Rheumatology) to identify Cochrane systematic reviews assessing an orthopaedic surgical procedure and published in 2013. The search results were pooled and duplicate records were removed.
Selection of systematic reviews
The title, abstract and full text, when necessary, of all identified references were screened by two authors (BV and PB). We included systematic reviews written in English or French and published in 2013 that assessed an orthopaedic surgical procedure, with no restriction on comparator (usual care, placebo, conservative intervention, pharmacological treatment or other surgical implant or procedure), and including an MA of aggregated data based on two or more RCTs. Systematic reviews and MAs that were withdrawn were excluded, as were those for which the full text was not available.
Extraction of clinical research questions from the systematic reviews
Two reviewers (BV and PB) independently extracted all elements of the clinical research question assessed from the full text of the systematic reviews by using the Population, Intervention tested, Comparator and primary Outcomes (PICO) acronym . Then, all clinical questions were classified by anatomical region (e.g., shoulder) and surgical procedure tested (e.g., arthroplasty) to identify any potentially redundant research questions.
Development of protocol synopsis of IPD MA for each clinical research question
For each clinical research question, we developed a standardized protocol synopsis for an IPD MA, with the background and objectives sections derived from the corresponding systematic review with MA of aggregated data and the methods section based on the Cochrane Handbook for Systematic Reviews of Interventions . An example of a protocol synopsis is presented in Additional file 2.
Identification of RCTs relevant to the clinical research questions
We identified all RCTs included in the systematic reviews with MA of aggregated data for the clinical research questions. All RCTs were selected provided they were written in English or French and results were published since 2000, because for trials with results published before 2000, we anticipated difficulties in contacting corresponding authors and obtaining IPD from these trials. Non-randomized studies and quasi-RCTs were excluded, as were RCTs not indexed in MEDLINE. If corresponding authors were involved in several RCTs corresponding to different clinical research questions, we contacted them for only one clinical question chosen at random. Only the RCTs corresponding to this clinical question were selected.
Extraction of characteristics of the selected RCTs
Using a pre-tested standardized form, we extracted the following characteristics from the full text and online Additional files, as well as journal websites:
Corresponding author: name and email address of the corresponding author. When the email address was not available in the article or not functional, the name was entered in PubMed to search for another study published by the author for which the email address was reported. We also screened the website of the author’s institution to search for an email address.
Characteristics of the trial: publication date, whether the trial was a single-centre or multicentre trial, location of the study (Africa, Asia, Australia and New Zealand, Europe, North America, South America) and funding source (public, private, both public and private, not reported). The geographic location of the corresponding author’s affiliated institution was used to define the location of the study when the study location was not reported or when the RCT was an international multicentre study. We also extracted the number of patients randomized in each RCT.
Characteristics of the journal in which the RCT was published: name of the journal, whether it was specialized or general and its impact factor. We classified journals according to whether they were in the 10 highest impact factors for a medical condition according to the Journal Citation Reports. We also recorded whether the journal had a data sharing policy and, if so, whether data sharing was a mandatory condition to publish.
Contacting the corresponding authors of identified RCTs
Corresponding authors of each RCT were contacted by personalized email to participate in a specific IPD MA project and to provide IPD from their RCT. The email stated that the French Cochrane Centre aimed to initiate collaboration among trialists to perform IPD MAs on important orthopaedic topics and that the first project had as an objective the clinical research question for which the RCT was eligible. We guaranteed protection of patient data and secure storage of datasets. We also systematically asked whether the corresponding author or another colleague wanted to be a co-author of the published IPD MA and whether we should cover costs related to data extraction. The emails were personalized for each trialist and clinical research question, with inclusion of the name of the corresponding author, title, year of publication and journal in which the author’s trial was published, as well as the objective of the IPD MA and a link to the protocol synopsis of the IPD MA corresponding to the research question assessed. The subject of the email was ‘Your article ‘Trial title’, published in ‘Journal’ on ‘Year of publication”. Emails were sent using a dedicated Cochrane address and were signed by an academic orthopaedic surgeon on behalf of the French Cochrane Centre, the French Equator Centre and the Inserm Research Centre U1153. An example of the email sent is available in Additional file 3. Two similar reminders were sent to the authors 15 and 30 days after the first email if we did not receive a response to the initial request.
It has to be noted that our hypothesis was that the rate of positive answers would be low but we expected to be able to perform several IPD meta-analyses as part of a PhD program for the clinical questions for which we would have received a sufficient number of positive answers.
Characteristics of systematic reviews and RCTs were described along with the number and percentage for categorical variables. We compared characteristics of trials and journals for trials for which we had a positive response and a negative or no response to our request for data sharing by two-sided chi-square or Fisher’s exact test, as appropriate, with a type I error level of 0.05. We excluded from this comparison authors who responded that our request was under consideration. Data were analyzed by using R version 2.13.1 .