Data analysis

We conducted interviews until data saturation was reached (i.e., until interviews no longer yielded new information). Interview data were coded inductively by A.H., using a thematic analysis approach.²³ NVivo version 1.5.1 (QSR International) software was used to manage the data. Interview data were organized into codes that were identified iteratively from the data set. Codes were then compared across the data set for regularities and divergences and modified accordingly. Through this process of inductive analysis, patterned responses developed into themes, which retained strong links with the original data set.³⁰

We considered data saturation for document review to be when document collection from interviewees and collaborators and Web searches no longer yielded new information. Because the documents gathered were mostly unsuited to line-by-line coding, we read them closely and highlighted the relevant information, before they were clustered thematically following the principles of thematic analysis (A.H.).³¹

The document and interview analyses were then compared for regularities and variations (A.H.). A subset of 7 interviews and the full data set of documents were independently reviewed by 1 author (P.N.) for coding and emergent themes to ensure consistency and reliability in the analysis. Two authors (A.H. and P.N.) then discussed the coding and reached consensus on final themes. Another author (S.S.) was consulted to resolve any disagreements.