This was a cross sectional study of Australian podiatrists using data collected through an online survey from 2017 to 2020. The CHERRIES (Checklist for Reporting Results of Internet E-Surveys) guided the reporting of collected data [18]. The Monash University Human Research Ethics Committee approved this research (19959).

Podiatrists and podiatric surgeons working in Victoria were invited to participate in Waves 1 and 2 (2017–2018) of the survey. The survey was open to all podiatrists and podiatric surgeons in Australia for Waves 3 and 4 (2019–2020). There were an estimated 5429 podiatrists and 36 podiatric surgeons registered in Australia when the final wave closed [19]. There were no restrictions on practice setting.

Data were collected as part of the Podiatrists in Australia: Investigating Graduate Employment (PAIGE) study, and the methodology is published [20]. PAIGE survey questions were based on a longstanding longitudinal medical workforce survey in Australia, the Medicine in Australia: Balancing Employment and Life (MABEL) study [21], tailored to exploring different elements of the podiatry workforce with a core bank of questions each wave and question bank elements added at each wave (Table 1). Data were collected relating to demographics (all waves), measurement of constructs impacting on labour decisions such as job satisfaction (all waves), earnings (Wave 1), impact on family (Wave 1), workplace setting (all waves), and mental health (Waves 2, 3 and 4). All four surveys are provided as Supplementary Files 1, 2, 3 and 4.

Demographic data collected from participants included information about their gender, age, pre-registration education, postcode, current work setting and employee/employer status, number of working locations, time spent working at a location, exposure to regional/rurality placement during education, leave availability and professional development availability.

All waves of the PAIGE study included the 10-item revised job satisfaction scale [22]. Participants were asked to indicate satisfaction relating to different aspects of work. The original 7-point Likert scale used within the MABEL study was reduced to 5-point item scale (1 = very dissatisfied, 2 = moderately dissatisfied, 3 = neutral, 4 = moderately satisfied, 5 = very satisfied) [23]. A not applicable response option was also provided for each item. This adaption was on suggestion from members of the MABEL team who provided advice during survey build [24].

Waves 2, 3 and 4 included the abbreviated Maslach Burnout Inventory (aMBI), a nine-item scale used for assessing burnout [25, 26]. It has three subscales including emotional exhaustion, depersonalization and personal accomplishment [27]. An additional three questions were included on job satisfaction with reference to being a health professional [28]. Items were scored on a seven-point Likert scale (1 = everyday, 2 = a few times a week, 3 = once a week, 4 = a few times a month, 5 = once a month or less, 6 = a few times a year, 7 = never). Higher scores for emotional exhaustion [10,11,12,13,14,15,16,17,18] and depersonalisation [10,11,12,13,14,15,16,17,18] and lower scores for personal accomplishment (0–9) and job satisfaction (0–9) indicated greater burnout [27].

Podiatrists and podiatric surgeons working in Victoria (Waves 1 and 2) and Australia (Waves 3 and 4) were invited to participate in the online survey every year through its promotion on social media (Facebook, Twitter, LinkedIn and Instagram), at podiatry conferences and through targeted emails from peak bodies such as the Australian Podiatry Association and Australasian College of Podiatric Surgeons. Podiatrists who had completed previous waves and left contact details were emailed directly. Podiatrists who responded were given the opportunity to enter a competition for gift cards or to receive professional development vouchers.

Qualtrics® software (Qualtrics, Provo, UT, USA) was used to collect each wave of the online survey data [29] and subsequent waves linked responses by each participant creating their own unique identifier code. Forced or requested response prompts were used to minimise missing data, and participants could withdraw at any time by closing their internet browser. All non-completed questions were treated as missing data for the remaining non-completed variables. Question logic was used to minimise question blocks if a participant indicated that they had participated in previous waves. These logics included if there were no job changes, or no changes in location, these responses were carried over at each round. Cookies were used to allow responses to be saved up to 4 hours within partial completion. Code routinely collects Internet Protocol (IP) addresses as part of the de-identified metadata in the survey response and IPs were only viewed and used as a last resort to match data where other linking variables were incomplete.

All data were initially cleaned to remove any responses that were partially completed, including where there were no core demographics (core data included gender, age, postcode, recency of practice). As podiatrists were only requested to complete some sets when their job role or living situation had changed, a final per podiatrist response was created with the most recent response, with additional completed data from prior waves inputted as required whereby the most recent response was the one applied to the current analysis (Fig. 1). In response to the coronavirus pandemic, we undertook preliminary analysis to understand if there were distinct differences between Wave 3 (2019) and Wave 4 (2020) cohorts, particularly regarding mental health scores, burnout and any impact on job satisfaction or intent to leave the profession. We did not find significant differences in our domains of interest between waves and their impact on metropolitan or rural responses, therefore all data were analysed using the combined data set from Wave 4.

Descriptive statistics of all variables were grouped and recoded where appropriate (e.g. Satisfied responses were combined with Very Satisfied). The Modified Monash Model (MMM) was used on postcode data to recode location into metro (MMM 1) or rural (MMM 2, 3, 4, 5, 6 and 7) [30]. Univariate and multivariate regression models were used to determine factors associated with location of work (as a proxy measure of recruitment) and intent to stop seeing patients (Yes versus Unknown/No) and intent to leave the profession (Yes versus Unknown/No; as a measure of retention). Univariate logistic regression was used to explore any associations between the variable of interest (location, intent to stop seeing patients, intent to leave the profession) and those known to impact recruitment and retention to build the multiple regression model. Variables known to impact recruitment and retention were chosen based on prior studies supporting their impact on recruitment and retention, including age [3], recency of practice [3], primary work setting (private or public) and business relationship (owner or partner, salaried/contract, locum/not working and other) [31], number of working locations and time spent working at a location [32], exposure to regional/rurality through life or education [6], leave availability [3], professional development availability [33] and working in a location close to family and friends [34].

Backward stepwise multivariate regression was performed where univariate analysis revealed a value of p < 0.20. This analysis took form by removing the variable with the least significant fit in a stepwise procedure until all remaining variables were significant at p < 0.05. Regression analyses were performed using Stata 15 software (StataCorp, College Station, TX, USA).