The value of survival analyses for evidence-based rural medical workforce planning

Health workforce undersupply in rural areas is a persistent global problem, which contributes to inequitable health outcomes for rural populations in high-, middle-, and low-income countries alike[1‐3]. Rural health workforce supply reflects the balance between current stocks and subsequent inflows (recruitment) and outflows (turnover) of workers. Considerable research has been undertaken into the complex range of issues that influence health workers’ decisions to take up, stay in, and leave rural practice (including economic, professional, personal, and community factors)[4‐7].

Unfortunately, however, substantial gaps remain in our knowledge of the flows of health workers into and out of rural areas. Much of the existing research has focussed on health worker’s job satisfaction or intentions to leave rural practice, rather than on actual observed behaviour, though within the Australian rural context there are several exemplary studies[7, 8]. Little is known, for example, about what length of stay might reasonably be predicted for a family physician practising in rural or remote locations[9]. This lack of empirical data on health worker flows and behaviours continues to hinder rural health workforce planning and decision making[10‐12].

The research reported in this paper is designed to add to the existing evidence-base. The research aims first to identify and quantify the most important factors associated with the risk of rural and remote Australian family physicians leaving a practice, and secondly, to demonstrate the value of evidence generated by rigorous survival analysis of longitudinal health workforce data to inform rural health workforce planning and retention strategies. Although this paper examines the retention of family physicians within a single jurisdiction in one high-income country, there are important parallels with other similar geographically large developed countries, including Canada, United States of America, and Germany, and the analytical method demonstrated is one that can readily be adapted to a range of settings.

Improving our knowledge of what constitutes effective workforce retention strategies is dependent on several pre-requisites. In the first instance, it is important to understand what the most appropriate metrics are for measuring rural health worker turnover and retention. A second critical requirement for strengthening the evidence base is the availability of appropriate data and sufficient capacity to analyse and report selected indicators. Thirdly, the ability to make valid comparisons between different groups, and quantify differences in workforce retention is important for policy, as it assists targeting policy to specific groups of interest more effectively. A final requirement is familiarity with the types of interventions that might be used to improve retention, together with knowledge of their effectiveness and how much they might cost[13, 14]. Within Australia, key current rural workforce strategies include the scaling of retention incentives according to location (based essentially on the degree of geographical remoteness) and restricting provider access to Medicare (the Australian universal health insurance scheme) for international doctors to designated ‘districts of workforce shortage’ and ‘areas of need’[15]. However, despite the significant financial commitment to such programs, little is known about their effectiveness or their impact on patterns of turnover and retention. This paper seeks to show how empirically derived evidence can assist to inform policy development in this area.

A recent review of the utility of different metrics for measuring health workforce turnover and retention in rural and remote contexts indicates that metrics derived using survival analysis methods have significant strengths to inform health workforce planning[16]. Survival analysis measures the time until an event occurs. In the case of this health workforce turnover and retention study, the event of interest is the time between take-up of a position until a health worker leaves that appointment. Hence, the data required include accurate commencement and exit dates for individual practitioners working in rural areas.

Between January 1^st 2003 and December 31^st 2012 there were 3,354 family physician appointments in rural and remote NSW, representing 2,783 individual family physicians (83% of appointments were first rural appointments for that physician, 13% were second appointments, 3% were third appointments, and less than 1% were fourth or subsequent appointments). Over the 10-year period of this study, a total of 14,992 family physician-years of observation time were analysed, and 1,646 (49%) appointments ended – that is, on 49% of all occasions the doctor moved a distance greater than 15 km, or left the practice for a period of at least three months to undertake other activities. Of the 2,783 rural family physicians, 1,864 (67%) were male and 1,533 (55%) were known to be Australian graduates (Table 1).

Of the 3,354 family physician appointments in rural and remote NSW, 2,237 (67%) were known to be held by family physicians who were not undertaking procedural activities in any of anaesthetics, obstetrics, or operative surgery; 492 appointments (15%) were known to be associated with ‘conditional’ registration of the family physician (conditional registration in its various forms enables overseas trained doctors who are yet to gain Australian Medical Council accreditation to work in supervised practice in designated ‘districts of workforce shortage’ and ‘areas of need’[15, 25]); and 1,741 (52%) were known to be associated with the family physician having Visiting Medical Officer (VMO) rights (rights to provide medical services in a public hospital) (Table 2).

Univariate analyses revealed significant (α = 0.05) differences in the risk of family physicians leaving an appointment according to geographic location and population size, birth year, country in which the family physician obtained their medical degree, procedural skills, registration status, age at graduation, spousal rural location prior to the family physician’s first rural posting, and VMO rights, though not according to gender. Figure 2 illustrates retention patterns for family physicians working in small towns (population size <5,000) in inner regional, outer regional, and remote/very remote locations. Retention is higher in inner regional small towns compared with outer regional and remote/very remote small towns. Unadjusted estimates of the increased risk of leaving outer regional and remote/very remote small town family practices are 1.50 (1.25, 1.79) and 2.03 (1.61, 2.56) times the risk for inner regional small towns.

Multivariate Cox proportional hazards modelling revealed that family physicians practising in remote/very remote and outer regional towns with fewer than 5,000 inhabitants have a significantly increased risk of leaving town compared with family physicians located in inner regional NSW towns (Table 3). Family physician age was also significantly associated with risk of leaving town: the youngest family physicians (those born in 1970 or later) and the oldest family physicians (those born before 1945 and likely to be approaching retirement age) were at increased risk of leaving compared to family physicians born between 1945 and 1970; the increased risk for these groups was 1.54 and 1.45 times, respectively. Additional factors significantly associated with risk of family physicians leaving town include country of medical school graduation, procedural practice (not practising any of operative surgery, anaesthetics, or normal obstetric deliveries), having VMO rights, and holding conditional medical registration at any time during an appointment. The largest hazard ratio was for family physicians practising in small towns (population size <5,000) in remote/very remote Australia, and was associated with a 2.65 times greater risk of leaving compared to family physicians working in inner regional NSW. Graduating from medical schools in countries other than Australia, UK, Ireland, Canada, US, or New Zealand was associated with a 45% increased risk of leaving compared to the risk for Australian graduates. Family physician age upon graduation, however, was not significantly associated with turnover risk.

Translating these ratios into predictions of median survival (the length of time until half the workforce has left), revealed, for example, a difference in length of stay of 8.1 years on the basis of geography and population size alone for Australian-trained family physicians with VMO rights and not undertaking procedural activities. Predicted median survival for those working in small towns in coastal inner regional NSW was 11.1 years compared with 3.0 years in small towns in remote/very remote NSW (Table 4).

This innovative study breaks new ground in medical workforce research in Australia. For the first time, this study applies rigorous quantitative methods to Australian longitudinal medical workforce data to identify important correlates of the risk of family physicians leaving a rural or remote location. The use of survival (time to event) analysis enables important comparisons to be made on the basis of sentinel variables such as geographical location, population size, age, and professional status, and the statistical significance, magnitude, and direction of associations to be measured and reported. For the purpose of developing effective medical workforce policies and planning, these analyses provide several key insights.

First, our research shows that over the past ten years, the risk of family physicians leaving an appointment is strongly and significantly associated with geographical location and population size. For family physicians working in small towns with a population less than 5,000 a gradient of risk was found, whereby the risk of leaving was lowest in closely-settled coastal locations, intermediate in areas of moderate population density, and highest in the most sparsely-settled locations. For Australian trained family physicians who are non-procedural and have VMO rights, these findings translate into a predicted length of service of 11 years in small coastal towns in closely-settled locations. This compares with 6 years for family physicians in small inland towns with moderate population density and 3 years for family physicians in small inland towns in sparsely settled locations. Periods of service less than this might be interpreted as indicating ‘premature’ or ‘avoidable’ turnovers, and health authorities and workforce planning agencies could monitor any ‘hot spot’ locations to see whether any specific additional interventions are required in order to extend the length of practice of family physicians.

This significant differentiation in the risk of leaving is not surprising given the demonstration by Humphreys et al. of significant associations between professional indicators known to be related to family physician retention and geographical location and population size[23]. The extent to which shorter retention in small, more sparsely settled locations is ‘optimal’ (that is, all that might be expected in these locations) or ‘sub-optimal’ (that is, illustrates premature or avoidable turnover that could be adjusted through workforce incentives or interventions), remains a moot point. It is nevertheless important to interpret these observations in the context of substantial and increasing spending by the Australian federal government on direct financial incentives paid over this period to rural and remote family physicians in an attempt to improve retention. In particular, given that the Australian government is ‘scaling’ incentives according to geographical remoteness, these results provide, for the first time, empirical evidence to guide such differentiation in the allocation of retention incentives[25].

In Australia, expenditure on specific rural family physician workforce incentives has escalated almost six-fold from $19.9 million over the eight year period between 2004–2005 and 2012–2013[25, 26]. Whilst these incentives are scaled according to remoteness, community population size is not taken into account. In the absence of any definitive evidence about the effectiveness of medical workforce retention grants, our findings suggest that existing workforce retention interventions are insufficiently effective to ensure equality of continuity of family physician care for residents of remote and very remote areas. Given that numbers are small (92 appointments or less than 3% of total appointments) in remote and very remote NSW, opportunities exist to significantly strengthen retention strategies for this group of family physicians without necessarily having a large impact on the overall program budget. Improved targeting of retention strategies to family physicians in sparsely settled locations is especially pertinent given the recent finding that the major growth in family physician rural retention payments since 2010 has been in closely-settled areas, where retention is already relatively high[25].

In addition, the evidence generated by our study may help guide the relative lengths of service that might be required in the form of ‘return of service obligation’ for medical practitioners mandated or bonded to work in non-metropolitan areas. A range of Australian government programs currently scale return of service obligations, once again only according to geographic remoteness (ASGC-RA), but not on the basis of any empirical evidence. For example, the scaling ratio for reducing return of service obligations for the Medical Rural Bonded Scholarship Scheme is inner regional 1.0: outer regional 1.3: remote 1.5: very remote 1.8. Our work suggests that population size should also be taken into account and that the ratios for remote and very remote locations compared to inner regional locations could be higher.

A further important finding is that graduates from medical schools in countries without an Australian Medical Council-designated competent authority (countries other than Australia, UK, Canada, US, New Zealand and Ireland, which we term ‘other’ countries) had a substantially (1.45 times) increased risk of leaving a family physician appointment in rural NSW compared with Australian graduates. In terms of predicted median length of stay, this translates to Australian trained graduates staying for almost a year longer in small remote towns and for almost 2 years longer in small inland towns in regions of moderate population density. These differences are after the modelling adjusts for the lower risk of leaving an appointment for family physicians with conditional registration. Our findings are consistent with existing evidence that physicians obliged to work in a location not of their choosing are at increased risk of leaving that location in the longer term compared to non-obliged physicians[5, 14]. These findings are particularly important for rural and remote workforce policy development since such a large proportion (33%) of family physicians in rural and remote NSW during the past 10 years are graduates of ‘other’ medical schools. Indeed, in 2009–2010 almost 50% of all family physicians in rural and remote Australia were international medical graduates[25]. Given this heavy reliance on internationally trained family physicians, it becomes critical to identify the root causes of their high turnover and address them as a matter of urgency. Recent work by McGrail et al. indicates a much higher relative dissatisfaction of overseas-trained family physicians (especially those with restrictions on where they can practise) compared with local graduates[27]. Their study also pinpoints various professional and non-professional aspects associated with dissatisfaction, some of which may be responsive to policy intervention.

Finally, our research has identified VMO status and procedural activities in the area of obstetrics, anaesthetics, and operative surgery as important correlates of reduced risk of family physician turnover, consistent with previous cross-sectional analyses[7]. Not only are these professional activities likely to be associated with a higher overall income, but also with a greater sense of autonomy, a wider variety of work, increased opportunities to use an extended skill set, and a heightened sense of responsibility. In other words, VMO status and procedural activity are associated with important indicators of family physician professional satisfaction[28], which may in turn be associated with reduced turnover. These findings have important implications for future investment in rural training pathways that develop advanced skills needed for rural and remote hospital work, as for instance with the successful generalist model being promoted in Queensland, Australia[25]. Furthermore, provision and maintenance of infrastructure to foster hospital-based activities of rural and remote family physicians is also important because of its association with relatively higher family physician retention.

A number of limitations of this study are acknowledged. First, despite family physician vocational trainees making a substantial and important contribution to the rural and remote workforce in NSW, they were excluded from analysis since the factors driving their relocation decisions are likely to be different from those of family physicians and because their training program often requires rotations at various times regardless of their satisfaction with any practice location. Second, some variables of interest were not included in the final multivariate model because of the extent of missing data. These included rural origin of spouse and recipient status for some specific rural scholarships. Some variables of current policy relevance, such as Medical Rural Bonded Scholarships, also had insufficient numbers of recipients to permit reliable estimation of the effect. This is due to the long lag time between receipt of financial support and commencement of rural or remote practice as a family physician. Third, the retention profile used in this study coded a ‘failure’ as a location move of more than 15 km or break in service provision of more than 3 months[29], in recognition of the importance of continuity of care as a key dimension of primary health care[30]. While realistic and relevant for our research in non-metropolitan Australia, this retention profile may not be appropriate for all workforce-planning purposes. Pathman’s pioneering work on physician retention described how physicians may variously be considered as being successfully retained when they stay with an initial practice, when they stay within the initial community, when they stay within any rural location within the jurisdiction, or even when they remain actively practising clinical medicine[29]. Indeed, a strength of survival analysis is that it can successfully be applied to each of these different policy problems simply by coding a ‘failure’ in different ways – for example, as leaving a particular ASGC-RA, or as leaving rural NSW or even as leaving the medical profession – depending on the availability of requisite data and the particular policy question being asked.

This study highlights how survival analyses can be used to generate rigorous evidence to inform policy development in the area of health workforce planning, particularly, for example, in the strengthening and improved targeting of retention strategies in rural and remote areas. In this instance, survival analyses identified strong associations between geographical location and population size, country of primary medical degree, procedural activity, and VMO status, and the risk of NSW rural and remote family physicians leaving a community. Such quantitative empirical evidence establishes a better baseline against which to monitor the effectiveness of workforce strategies and guide workforce planning.

DR, JH, and MM are members of the Centre of Research Excellence in Rural and Remote Primary Health Care (CRERRPHC), conducting research in accessible and equitable primary health service provision in rural and remote Australia. IC is the Chief Executive Officer and PW is the information manager at NSW RDN and IC is also a member of the National Advisory Committee of the CRERRPHC.