Study design
In this population-based cohort study, we assessed our outcome (becoming practicing GP by 2016) with a survey and an internet search to identify non-responders. The cohort was all participants in the WHM training program between 2006 and 2015 (n = 381). We used retrospective data from the WHM administrative database, and supplemented it with an online survey administered to training program participants in 2016. More data on the exposure (GP training module) and trainee’s satisfaction with the module was prospectively collected from 2006 to 2015 (Fig. 1).
Outcome
Our primary outcome was the working status of former GP trainees (practicing GP by 2016: yes/no). Our secondary outcome was common program-related parameters that favoured or deterred trainees.
Training program
WHM GP training places trainees at a GP office, where they work under the supervision of a GP (trainer). All GP trainers attended a two-day course on how to supervise GP trainees. WHM and GP trainers pay GP trainee salaries [27]. GP trainees must declare their interest in becoming a GP. In Switzerland, trainees can change their career path during or even after their GP training.
GP training modules are now also organized at a regional level [26]; WHM often administers candidates and salaries for these programs. Thus, WHM is responsible for most GP trainees. More details about the WHM program can be found on www.whm-fmf.ch.
Participants
Between 1998, when WHM offered the first GP training module, and 2015, 809 GP trainees completed the program. We included all trainees who registered for a GP training module in GP offices with WHM in the last 10 years (from 2006 to 2015, Fig. 1). We chose this period because, throughout this period, the evaluation forms were the same. We had 494 trainee records. Exclusion criteria were 1) those who had not completed training; and, 2) those who dropped out of training in under a month. No trainee was counted more than once, because we merged the records of trainees who had completed multiple modules and summed their hours.
Data sources
WHM registers all trainings in a database that contains administrative data about the trainees and the trainers, and the evaluation forms trainees and trainers complete when they have finished a training.
In January, 2016, we sent an email invitation to all trainees in the WHM database (2006–2015) to assess our primary outcome. We invited them to complete a short online survey in either German or French (see Additional files 1 and 2). Participants were asked for their current working status. Practicing GPs were then asked when they started practicing, and how much influence their GP training module had on their decision (5-point Likert scale). The other questions were demographic. If they did not respond, they were sent two email reminders (2 weeks and 4 weeks after the initial invitation).
To assess our primary outcome for trainees who did not respond after two reminders, we first searched the National Registry of Medical Professions in Switzerland (www.medreg.admin.ch), which prospectively records all graduating medical students in Switzerland. The register is public and notes whether and when trainees were board-certified as GPs. However, in Switzerland, a board-certified GP might not be in practice as a GP (e.g., they could be doing research, positioned abroad, working for an insurance company, etc.). If we could not locate them, we searched professional directories, including www.doktor.ch, www.doctorfmh.ch, social media websites, including www.linkedin.com, www.facebook.com, and www.google.ch to find each trainees’ current working status. If we could not determine current working status, we categorized these trainees as “could not be assessed” and excluded them from the analysis. Based on previous experience with this kind of internet search, we expected that <10% of trainees would be excluded for this reason.
We used data from WHM database to capture GP trainees and GP trainer characteristics. For GP trainees, we collected data on gender, age at time of GP training, the years GP training started and ended, months of training, percent of training if part-time (e.g. 50% = 2.5 days/week), and how long after graduation the GP entered the program. We also stratified length of GP training module based on length of training (adding up part-time training so it was a percentage of full-time; for example, 2 months of 50% training were counted as 1 month of full-time training). We stratified by three periods because, in Switzerland, stakeholders want to know if length of training (<6, >6, or >9 months) has an effect on a trainee’s decision to become GPs.
For GP trainers, we collected data from WHM, including how old the trainers were when they taught the module, gender, and type of practice (e.g., solo or group practice). We used the postal code of their GP practice to categorize GP practices by population density, which we calculated with public population census data.
To collect data on satisfaction of GP trainees with the GP practice where they trained, we analysed the evaluation forms that WHM routinely collects immediately after the GP completes the training module. However, since the evaluation form is very long (79 questions) we a priori selected a sample of questions to analyse. Limiting our sample helped us avoid methodological problems caused by overfitting. If we had included all the questions, and used a significance level of 5%, one in twenty associations we identified would have been a statistical artefact. We thus invited a panel of experts to choose questions they though most likely to be associated with the exposure (longer vs. shorter GP training) and the outcome (becoming a practicing GP by 2016). We decided in advance that we would include questions that at least 5 of 6 experts had selected. Four out of the original 79 questions met that criteria: 1) overall satisfaction with the training module; 2) quality of supervision by GP trainer; 3) ability to acquire perceived competencies during GP training module; and, 4) trainee’s opinion of how well the GP trainer taught.
Statistical analyses
We described basic characteristics of GP trainees, GP training modules, GP trainer, and satisfaction with GP training module, both for GP trainees who became practicing GPs and those who did not. We used the chi square test to compare categorical data, and the t-test or non-parametric Wilcoxon ranksum test, where appropriate, to compare continuous data.
We calculated univariate time-to-event curves to become practicing GP and stratified the data by GP trainee’s gender, part-time training (yes/no), length of GP training module (<6/>6 months), and whether they were trained before or after the fourth year of GP training. The curves were constructed with the Kaplan-Meier method, and compared using a log-rank test. We defined time-to-event as starting the year GP training module was completed and ending the year the trainee went into practice as a GP. If time data for individuals was missing, we calculated the mid-point between the end of the GP training module and the time the subject completed the questionnaire. For all cofactors, we calculated hazard ratios (HRs) to become practicing GPs and corresponding 95% confidence intervals (CI). We calculated both univariate and multivariate HRs with Cox proportional hazard models. In a multivariate model, we further adjusted for gender of both GP trainee and GP trainer, year of training, part-time training, duration of training (calculated based on full-time training, stratified into <6 months, 6–9 months, or >9 months), dates of training (stratified into three groups: 2006–2008, 2009–2012, and 2013–2015), solo practice, and the four a priori questions experts had chosen from the evaluation form. We drew forest plots that used random effects models to visualize the associations of the selected cofactors using HRs.
We considered a two-sided p-value of 0.05 to be statistically significant. We performed all analyses with STATA version 14.2 (Stata Corp, College Station, TX, USA).