We conducted a two-round modified online Delphi survey. The Delphi method is a systematic approach to obtain opinions of a group of experts by means of a series of short self-administered questionnaires [4].
Ethics and informed consent
No formal ethics approval was necessary for this study. Participants received a full written explanation about the scope and goals of the study. All experts participated voluntarily and provided written consent by completing the two Delphi rounds.
Identification of panel members
One researcher generated a list of potential participants of the expert consensus panel. Candidates were identified by checking the author lists of primary studies summarized in a recent systematic review [5]. Contact details of the first, second and last author of each publication were retrieved from the original publication. We complemented the list with experts known to the authors. Current e-mail addresses were verified via a general web search. Fifty experts were invited to participate. Every expert received an e-mail containing a description of the purpose of the study and an invitation to participate. Anonymous participation was granted to ensure freedom of expression. If an e-mail was rejected, we contacted the last known host institution of the expert. A reminder letter was sent out two weeks later.
Survey set-up
The questionnaire of the first round was designed by one research fellow. Following the approval from all authors the questionnaire was implemented into the survey software. The survey was conducted using the SurveyMonkey online tool [6]. The online tool allows to implement an introduction page providing information about the survey methods. Also, the system dynamically adapts its user interface offering sufficient space for the number of answers given. Moreover, it is possible to set input requirements allowing only a specific format of the answer e.g. range of the attributed weights must be a number between one and ten.
First round questionnaire
Each consenting expert received a personalized e-mail containing a description of the study method and a web-link to access the survey. Before starting to complete the questionnaire, each participant provided information about his or her specific expertise and the current affiliation. In the first round, experts were asked to name possible relevant predictors besides the well-established parameters of FRAX for the occurrence of osteoporotic fractures. They could specify these possible predictors for one or more specific subgroups of patients (e.g. patients with secondary osteoporosis).
Second round questionnaire
The feedback from the first round was extracted, examined and prepared for the second round. Obvious misspellings were corrected and plurals were replaced by singulars (e.g. fall vs. falls). Terms with the same root word were collapsed into one term if the intended meaning remained unchanged. If appropriate, a general concept was used to summarize a specific condition (i.e. kidney transplantation, lung transplantation, heart transplantation was collapsed into one term “history of transplantation”). In the questionnaire, the parameters were then grouped into the seven different subgroups. Every participant received after round one a compiled inventory of the answers. In the second round, participants weighed for each subgroup every predictor by assigning a number between one (not important) and ten (very important).
Attaining a consensus on relevant parameters
We calculated the median of the attributed weights and the corresponding interquartile ranges (IQR) for each of the parameters within each of the clinical subgroups. Using the thresholds of a previous Delphi study, we defined an expert consensus as the IQR of a predictor ≤2 [7]. The cut-off value of the median attributed weights for a relevant predictor was set at ≥7. Statistical analyses were performed using the Stata 14.2 statistical software package (StataCorp. 2015. Stata Statistical Software: Release 14. College Station, TX: StataCorp LP.). At the end of the Delphi study, every participant received the results of the Delphi including median and IQR of all items.