In many geographic regions, both in developing and in developed countries, the number of health workers is insufficient to achieve population health goals. The 2004 Joint Learning Initiative (JLI) report Human Resources for Health estimated that "Sub-Saharan countries must nearly triple their current numbers of workers by adding the equivalent of one million workers through retention, recruitment, and training if they are to come close to approaching the MDGs [Millennium Development Goals] for health" 1; the 2006 World Health Report concluded that " [t]he severity of the health workforce crisis in some of the world's poorest countries is illustrated by WHO estimates that 57 of them (36 of which are in Africa) have a deficit of 2.4 million doctors, nurses and midwives" 2. In many developed countries, there are areas (commonly rural or poor communities) that are considered to be underserved with health workers (for instance, because the number of workers is insufficient to provide primary health care to all residents in an area) 345.¹

Interventions intended to alleviate health worker shortages include selective recruitment and training for practice in underserved areas, improvements in working or living conditions, compulsory service, or incentives 6. In this article, we systematically review the evidence on one specific set of interventions: financial incentives for return of service. These interventions work as follows. A health worker in training or a fully trained health worker enters into a contract to work for a number of years in an underserved area in exchange for a financial pay-off. Financial incentives can increase the numbers of health workers in underserved areas by a number of mechanisms. First, they can redirect the flow of those health workers who would have been educated without any financial incentive from well-served to underserved areas, for instance by decreasing the net emigration flow of nurses and physicians from developing to developed countries 789 or by increasing the net flow of physicians from urban tertiary care to rural primary care in developed countries 1011. This first mechanism can take hold if there are (future) health workers who normally would not work in an underserved area, but who are willing to do so in return for a financial incentive. Second, financial-incentive programs can add health workers to the pool of workers who would have been educated in the absence of such programs and place them in underserved areas. The second mechanism can take hold if, on the one hand, there are qualified candidates who would not have the means to finance a health care education without a financial incentive and, on the other hand, a country's health care education system can absorb additional students. Third, financial-incentive programs can decrease the outflow of health workers from underserved areas, if they prolong the retention times in underserved areas of those workers who participate in a financial-incentive program, but who would have worked in an underserved area even if they had not received a financial incentive. Improved retention in this group of health workers can be a direct result of the contractual obligation to remain for a certain number of years in an underserved area or can be caused by a program's additional efforts to increase retention (e.g., by increasing health workers' satisfaction with their work environment and career progression, or by increasing the satisfaction of health workers' families with their integration into the community) 12. Fourth, the programs can decrease the outflow of participating and non-participating health workers from underserved areas by increasing the number of health workers in those areas through any of the three mechanisms described above. Such positive feedback may occur because increasing the number of health workers can diminish reasons for non-retention in rural and remote areas, such as high workload 131415, lack of contact with colleagues 14, lack of support from medical specialists 16, or social isolation 15.

We have recently shown that a specific type of financial-incentive program, scholarships in return for a commitment to deliver antiretroviral treatment in Sub-Saharan Africa, is highly cost-beneficial under a wide range of assumptions 17. In the following, we will first systematically review studies on financial incentives for return of service. Then, we will critically summarize the findings from existing studies and draw implications for policy and future research. One previous study has systematically reviewed the evidence on financial-incentive programs for return of service. Sempowski (2004) reviewed 10 studies of financial-incentive programs published between January 1966 and July 2002 18. The author concluded that "ROS [return-of-service] programs to rural and underserviced areas have achieved their primary goal of short-term recruitment but have had less success with long-term retention" 18. Prima facie, an update of this systematic review is useful because more than six years have passed since the end of the period of publication of articles considered therein. In addition to the update of evidence, our review differs from the previous one in two aspects. First, the previous review was restricted to studies of physicians, while we consider studies of all types of health workers. Second, the previous review focused on program results (i.e., descriptions of outcomes among program participants without comparison to outcomes in non-participants) 18, while our review includes program results, program effects (i.e., analysis of program effectiveness at the individual-level through comparison of outcomes among participants and non-participants), and program impacts (i.e., analysis of program effectiveness at the population level, such as changes in physicians density or population mortality) (Table 1).

Table 1 describes the outcomes that were investigated by the 43 studies included in the review and the number of studies investigating each outcome (in parentheses). Twenty-five studies investigated 1 outcome; nine studies investigated 2 outcomes; seven studies investigated 3 outcomes; and one study investigated 4 outcomes. Two published articles report the same outcomes from the same study 4142; these study outcomes are counted only once in Table 1.

Additional file 1 shows descriptions of each of the programs that were evaluated in at least one of the included studies. When information on some program characteristics was not available in the reviewed study itself, we extracted the information from other sources (shown in the column "Other sources" in the table). All monetary values in the column "Financial incentive and obligation" are shown both as they are provided in the reviewed study and – for ease of comparison – in year-2000 United States dollars (USD). We used the purchasing power parity index from the World Bank Development Indicators 43 in order to translate the values of a non-US currency into US dollars and the consumer price index from the US Department of Labor Bureau of Labor Statistics 44 to adjust for differences in the real value of one USD over time.

The programs evaluated in the studies included in this review started between 1930 and 1998. With the exception of five programs that accepted a range of health professionals (the North Carolina Rural Loan Program, the National Health Service Corps (NHSC), the West Virginia Recruitment and Retention Community Project, the West Virginia State Loan Repayment Program in the US, and the Friends of Mosvold Program in South Africa), the financial incentives of the evaluated programs were targeted only at future or current physicians (Additional file 1).

With the exception of three programs that cover, respectively, "tuition, entrance and equipment fees and living expenses" 45,“tuition, fees” and “a living stipend" 46, and "funds for university tuition, books, residence fees and food" 47, the precise monetary values of the financial incentives of all programs included in this review were available in published articles or on web pages. The financial incentives per year of service ranged from year-2000 USD 1,358 to 28,470. One study compared the average award amount across five types of programs – service-requiring scholarship programs, service-option educational loans, loan repayment programs, direct financial-incentive programs, and resident support (in the form of service-requiring scholarships, loan repayment, or direct financial incentives). The study did not find significant differences in the size of the financial incentives 48 (Additional file 1).

We identified 43 studies that met all our inclusion and exclusion criteria. The previous systematic review of financial-incentive programs for return of service by Sempowski 18 identified only 10 articles, three of which were not included in our review. Two articles were not included because they evaluated a program that "tried to increase the number of health workers in underserved areas primarily through non-financial means" 4950 (Figure 1); one study was not included because it did not report "data from a quantitative study of results, effects, or impacts of financial incentives for return of service" 51 (see "Selection criteria" above). Of the 36 articles covered in our study but not included in the review by Sempowski, 17 were published after the end of the review period of the previous study (i.e., after 2002) 1148525354555657585960616263646566. The remaining articles were not included because the previous study used different inclusion and exclusion criteria. In particular, our review considers programmatic outcomes and health worker types that were not covered in the previous study (see "Introduction" above).

Of the 43 reviewed studies, 34 investigated financial-incentive programs in the US, 24 of which evaluated the NHSC; 8 evaluated programs in specific US states or communities; 1 evaluated both the NHSC and state-based programs; and 1 evaluated the national Commonwealth Fund Medical Undergraduate Scholarship Program. Five studies investigated the Jichi Medical University (JMU) program in Japan, two assessed the Ontario Under-serviced Area Program (UAP) in Canada, and one study each evaluated the New South Wales Department of Health Rural Resident Medical Officer Program (Cadetship Program) in New Zealand and the Friends of Mosvold Scholarship Scheme (FOMSS) in South Africa.² Additional file 2 describes the study outcomes, study designs, sample criteria, sample sizes, data sources, outcome measures, effect sizes, conclusions, and methodological limitations of all studies included in the review. Sample sizes across the thirty-seven studies in which individuals were the unit of observation ranged from 24 to 493,142. Eighteen studies had sample sizes greater than 1,000, and four had sample sizes greater than 50,000.

Of the 43 studies included in the review, 34 evaluated financial-incentive programs located in the US. The remainder examined programs in other developed countries (Japan, Canada, New Zealand), with but one exception that described a program in South Africa. The US financial-incentive programs have placed substantial numbers of health workers in underserved areas. For instance, between 1972 and 2009, the NHSC – the largest financial-incentive program in the US – placed approximately 30,000 primary care clinicians in underserved areas 46. At the same time, the US programs have met only a small proportion of national unmet health care need. In 2008, 4,600 clinicians were serving in the NHSC, but according to NHSC estimates 27,000 additional primary care professionals were required to provide care to the 50 million people who still lacked access to primary health care in the United States 91.

While most of the evaluated programs were located in the US, the US market for health care education is unusual in comparison to many other countries in that students pay high tuition for their education. Countries where students of health care do not usually incur large debt, such as many Western European countries, may not be as successful as the US in recruiting students and health professionals into programs that provide scholarships or repayment of educational loans in return for service in underserved areas. In many developing countries, by contrast, education for a health profession can be quite costly because of tuition and school fees as well as costs of housing and living. Some of the experiences from the US may thus be more applicable to health care education markets in developing countries than to other developed countries. On the other hand, (future) health workers in the US have many options for funding their education, while funding opportunities for education may be few in some developing countries. Thus, the generalizability of US findings to other countries where students have substantial financing need for health care education may be limited because the selection into financial-incentive programs for underserved service may depend on the availability of funding alternatives. Numerous other differences, such as in the capacity to enforce and monitor obligated service (compare 92), may limit the generalizability of the studies included in this review to other settings. One study from South Africa suggests that scholarship programs for health care education can be a successful instrument to recruit health workers for practice in rural Africa 66. Future studies should evaluate outcomes of financial-incentive programs from other developing countries where such programs have been offered in the past or are currently offered, such as Swaziland 93, Ghana 94, and Mexico 95.

Notwithstanding the above caveats about generalizability, it is useful to summarize some of the key findings from our review. First, most of the financial-incentive programs experienced substantial losses to recruitment before the start of the service obligation. Across the 25 programs included in the meta-analysis in our review, on average about 3 in 10 participants did not fulfill their commitment to work in an underserved area. However, there was wide variation in loss to recruitment. As reported previously by Pathman and colleagues 48 and Jackson and colleagues 58, state programs in the US that committed students to service (service-requiring scholarships and educational loans with service requirements) had significantly lower recruitment proportions than state programs that committed health workers after their training (direct financial incentives and loan repayment programs). This finding is not surprising, because preferences change over time. For instance, students who found careers in primary care appealing at entry into medical school may develop a strong interest in highly specialized health care during their training, which depends on technology that is usually not available in underserved areas.

Furthermore, we find that the recruitment proportion did not differ significantly between programs that offered a buy-out option and those that did not. While this result suggests that participants who have decided not to serve their obligation will do so independent of the conditions of the program they are enrolled in, it is important to note that the proportion of participants who would have taken up work in underserved areas had they not enrolled in a specific financial-incentive program is unknown. Thus, it is impossible to infer from such comparison the relative recruitment effectiveness of different types of programs.

Second, participants in financial-incentive programs were significantly more likely to leave their sites of first practice after completion of their obligation than non-obligated health workers in comparable sites of first practice after service of similar length of time. There may be several reasons for this finding. For one, some of those health workers who – without financial incentive – find practice in any underserved area less attractive than practice in sites that are not underserved decide to enroll in financial-incentive programs and to complete their obligations. These health workers are likely to leave the underserved area after having served the obligated term. On the other hand, even those obligated health workers who find practice in underserved areas to be the most attractive career path in general may be more likely to leave their sites of initial practice than their non-obligated colleagues in the same underserved areas. Obligated health workers have less choice over the particular underserved area in which they first practice than their non-obligated peers and are thus less likely to be satisfied with their work and personal life in the underserved area. For instance, one study of the NHSC concludes that NHSC enrollees "placed in rural sites in the late 1980s experienced a site-matching process that they felt offered few acceptable sites" and "offered little opportunity to locate the bestsuited site among those offered" 12. Financial-incentive programs aiming to achieve high retention of health workers in the sites where they fulfill their obligated service should attempt to accommodate health workers' wishes to practice in a particular underserved area to the greatest extent possible.

Third, while participants in financial-incentive programs were less likely than non-participants to remain in the particular underserved area of first practice, the reviewed studies suggest that participants were more likely to practice in some underserved area or to work with an underserved population than their peers who did not participate in a financial-incentive program. This summary finding from our systematic review is in contrast to the conclusion of the one previous review of financial incentives for return of service, which concluded that incentive programs "have achieved their primary goal of short-term recruitment but have had less success with long-term retention" 18.

Many of the analyses of retention in studies in this review compared the behavior of participants in financial-incentive programs to that of non-participants, controlling for a few observed health worker characteristics, such as sex, age, ethnicity, or marital status. However, since participants self-selected into programs, it is difficult to identify whether any difference in behavior between participants and non-participants was due to unobserved characteristics distinguishing participants from non-participants or due to program effects. It is possible that those health workers with the strongest preferences to serve underserved populations chose to participate in financial-incentive programs and that these unobserved preferences fully explain the different work and retention patterns in participants and non-participants, i.e., participants would have worked for exactly the same lengths of time in underserved areas without the incentives they received.

An ideal strategy to identify causal effects of financial-incentive programs is randomized controlled trials. However, since program participation is an individual choice, it will be impossible to randomize individuals into program participation and control arms. While it would theoretically be possible to randomize cohorts of medical students (e.g., by year of graduation or by medical school) to financial-incentive offers of different sizes, such a randomization strategy may not be politically or administratively feasible. An alternative strategy to identify causal effects involves the use of statistical models that control for selection into financial-incentive programs on unobserved individual characteristics. Two studies in this review (one of program effect 53 and one of program impact 64) implemented selection models of program participation. The two studies used medical school characteristics (e.g., the "historical proportion" of graduates pursuing careers in primary care) to identify program effect. However, the type of medical school that students choose is likely to be related not only to the decision to enroll in financial-incentive programs, but also – independent of program participation – to the decision to work in underserved areas. For instance, students with strong preferences to work in underserved areas may be more likely than their peers with weaker preferences for such care to select medical schools with a high "historical proportion" of graduates pursuing careers in primary care, because such schools are likely to focus on medical education relevant for underserved practice. This selection may determine work location decisions, independent of any effect the medical school characteristic may have on participation in financial-incentive programs. Thus the characteristic may not be a valid variable to identify program effects. Despite the difficulty in finding variables to identify program effects in selection models, future studies using already-existing data should emphasize control of biases due to selection effects. In the absence of a valid method to control for selection into program participation on unobserved variables, studies of retention and provision of care should attempt to control for variables capturing health workers' preferences to work in underserved areas before financial-incentive programs could have influenced those preferences. Studies in this review controlled for intention to work in an underserved area prior to the decision to enroll in a financial-incentive program 8487, for the intention to practice in an underserved area for a time period that is longer than the service obligation in a financial-incentive program prior to such practice 12, or for variables that are likely to be closely related to the preference to work in an underserved area in the absence of financial incentives, such as having grown up in an underserved area 8487 or a predilection for living in small communities 81. Two of these four studies found that physicians participating in the NHSC were significantly less likely than non-NHSC physicians to remain in the same underserved area where they initially took up work 1281, while the other two found that NHSC participants are significantly more likely to provide care in an underserved area than non-NHSC physicians 8487. These four studies suggest more strongly than the other studies in this review that the finding that program participants are more likely than non-participants to work in underserved areas in the long run (even though they are less likely to remain at their site of original placement) is indeed causal. Nonetheless, they cannot rule out that the observed effects are due to selection on unobserved variables.

Fourth, financial-incentive programs varied substantially in the level of participant satisfaction. Participants in some programs were more satisfied than non-participants with their work and personal life in underserved areas, while the converse was true for participants in other programs. Health workers' satisfaction with work and personal life in underserved areas is important for several reasons. For one, health worker satisfaction influences retention, as has been shown in several studies 969798, including in studies of financial-incentive programs for return of service 124163. Moreover, health worker satisfaction is associated with patient satisfaction 99 and quality of care 100101. Health workers are also likely to share their experiences with colleagues and may thus influence the supply of health workers to underserved areas as well as participation in financial-incentive programs. The reviewed studies offer some insight into the mechanism through which individual programs affect participant satisfaction. This evidence, based on case reports and participants' accounts, suggests that programs that achieved high participant satisfaction successfully interacted with participants during different stages of program enrolment, viz. participant selection, the matching of underserved areas to the preferences of individual participants, preparation of the participants and their families before the start of the obligated service, as well as career guidance, mentoring, monitoring of problems, and ongoing support during the service 124858667077. Detailed case studies of relatively successful and unsuccessful programs could further improve our understanding of management skills, organizational processes, and program features that increase participant satisfaction and retention in underserved areas.

Fifth, there is no clear evidence that financial-incentive programs had any significant impact on the supply of health workers to underserved areas. The results of three studies suggest that certain programs led to an increase in health worker numbers or densities, while two other studies did not find such program impacts. This discrepancy could be due to actual differences in impact between programs or over time; or they could be caused by methodological limitations of the studies. The impact of financial-incentive programs on health worker numbers and densities is not only a function of program scale and program effect on participating individuals, but depends also on the effect of the programs on non-participating health workers. It is plausible that participating health workers will deter non-participants from practice in underserved communities because the former will compete with the latter for patients and practice personnel. Conversely, it also seems plausible that the inflow of program participants into underserved communities attracts non-participating health workers to the same communities because the former decrease the overall work load per health worker (which may be perceived as too high) and increase opportunities for referral and exchange among colleagues. A study by Pathman and colleagues is significant insofar as it suggests "that the NHSC contributed positively to the non-NHSC primary care physician workforce in the rural underserved counties where its clinicians worked during the 1980s and 1990s" 55. In the above discussion of summary findings from our review, we caution that the existing evidence regarding program results, effects, and impacts does not allow (strong) causal inferences. It is further important to keep in mind that the summaries are across five countries, five types of programs, programs of different geographic reach ranging from community to country, seven types of health workers, and study publication dates ranging from 1963 to 2008. Program recruitment, retention, and satisfaction outcomes differed widely, even within some of the strata defined by program location, type, geographical reach, health worker type, and time period. Health planners can use our review to gain an overview of the existing evidence. In designing future programs, however, they need to carefully consider the applicability of findings from the studies in the review to the market for health care education in their country, the specific health worker group they intend to target with a program, and the type of underserved areas they aim to supply with program participants.

Financial-incentive programs for return of service are one of the few health policy interventions to improve the distribution of human resources for health on which substantial evidence exists. However, the majority of studies to date are from the US and only one study reports findings from a developing country. The existing studies show that financial-incentive programs placed substantial numbers of health workers in underserved areas and that program participants were more likely than non-participants to work in underserved areas in the long run, even though they were less likely to remain at their site of original placement. As none of the existing studies can fully rule out that the observed differences between participants and non-participants are due to selection effects, the evidence to date does not allow the inference that the programs have caused increases in the supply of health workers to underserved areas. In order to improve the scope of evidence on financial-incentive programs for return of service in underserved areas, future studies should evaluate programs from a more diverse set of countries, in particular in the developing world. In these studies, researchers should attempt to control selection biases as rigorously as possible, using selection models in observational studies and randomized controlled trials where funders and policy makers are willing to support such experiments.

The authors declare that they have no competing interests.

TB and DB jointly formulated the study design, obtained and analyzed the data, interpreted the findings, and wrote the article. Both authors read and approved the final manuscript.