System Dynamics Modeling of Health Workforce Planning to Address Future Challenges of Thailand’s Universal Health Coverage

Background: System dynamics modeling can inform policy decisions of healthcare reforms under Thailand’s Universal Health Coverage. We report on this thinking approach to Thailand’s strategic health workforce planning for the next 20 years. Methods: A series of group model building sessions involving 110 participants from multi-sectors of Thailand’s health systems was conducted in 2017 and 2018. Policymakers, healthcare administrators, and practitioners were facilitated to co-create a causal loop diagram representing a shared understanding of why the demands and supplies of the health workforce in Thailand can be mismatched and a stock and ow diagrams for testing the consequences of policy options. Results: Our simulation modeling found hospital utilizations created a vicious cycle of constantly increasing demands for hospital care, and hence a constant shortage of healthcare providers. Moreover, hospital care was not designed for effectively dealing with the future demands of aging populations and prevalent chronic illness. Hence, shifting emphasis to professions that can provide primary care, intermediate care, long-term care, palliative care, and end-of-life care can be more effective. Conclusions: The system dynamics modeling conrmed that shifting the care models to address the changing health demands can be a high-leverage policy of health workforce planning, although very dicult to implement in the short term.


Background
Thailand achieved Universal Health Coverage (UHC) in 2002 after decades of healthcare infrastructure development and experimenting with several nancial risk protection schemes [1]. Since then, every Thai citizen was covered under one of the three major health nancing schemes. Even before implementing Thai UHC, the planning of Thailand's health workforce has been incorporated into the National Economic and Social Development Plans. Over the decades, public healthcare facilities have been expanded nationwide. Thailand successfully built provincial hospitals in every province of Thailand by 1976, followed by the development of community hospitals in every district by 1991, and the modernization of primary care centers at the sub-district level during the 1990s [2]. Hence, historically, the national plans of workforce production and development also have focused on the providers in the public sector.
Health workforce or human resources for health (HRH) is one the building blocks of health systems, and the types and the number of healthcare providers needed in each health system are closely linked to how health care is organized in each country [3,4]. Like many other countries, healthcare systems in Thailand have been organized since the last century to make them more responsive to acute illness. Hospitals, which are historically designed for acute care, are currently the dominant providers for the UHC bene ciaries. Policymakers of the three publicly-nanced health funds have allocated most resources to providers in public hospital settings, but not in others providing in primary care, intermediate care, long-term care, palliative care, and end-of-life care. As a result, hospitals have been the major employers of healthcare providers in Thailand thus far.
More recently, a rapidly increased prevalence of non-communicable diseases and aging populations, as well as insu cient facilities speci cally designed for chronic and elderly care, have limited the effectiveness of Thailand's rst decade of the Universal Coverage Scheme (UCS) [5], the largest public healthcare nancing scheme under Thailand's UHC. This rapid change of the population's health demands could also aggravate the complex problem of inadequate health workforce domestically, which eventually can lead to equitable access to quality health care under Thai UHC. The expansion of private healthcare facilities in the private sector since the early 2000s also created a domestic "brain drain" of the health workforce, especially physicians, despite the innovative policies to retain them in the public sector.
Although Thailand has produced more healthcare workers every year, with the number of physicians or nurses per capita has been rapidly increased over the decades, many Thai UHC bene ciaries still have limited access to quality health care. Research has shown that public hospitals in Thailand, given xed inputs, have produced services relatively close to their capacity [6]. The long waiting time of patients at the outpatient department of every public hospital nationwide is self-evident for the current mismatches between supplies and demands of the health workforce in Thailand. Therefore, increasing the number of health workforce produced each year will only get us thus far. Therefore, a more comprehensive strategic planning that includes the reforms of healthcare delivery itself is needed to address this complex problem. We aimed to analyze what causes the chronic mismatches of supply and demands for the health workforce in Thailand and to synthesize more sustainable solutions to supply population health demands in Thailand in the future adequately. Using a systems thinking approach and a structured process of group model building (GMB) [7], we engaged with stakeholders who are embedded in a system to examine the nature of these complex problems, the pattern of system behaviors over time, and also to highlight the feedbacks within the systems. In the present study, we report on developing a whole-systems perspective of problems related to the health workforce in Thailand in the next 20 years, what causes them, and how potential systems interventions can be identi ed and tested by using system dynamics simulation modeling.

Method Setting
The study was carried out as a research project by Mahidol University's Faculty of Medicine Ramathibodi Hospital, collaborating with Thailand's Ministry of Public Health. A series of group model building (GMB) sessions were conducted in our workshops held in Bangkok and Nonthaburi, Thailand, during 2017 and

Study design and participants
The study employed systems thinking and modeling methodology based on the system dynamics approach [8]. We adopted ve major phases of the systems thinking and modeling methodology put forth by Maani & Cavana (2007), including 1) problem structuring, 2) casual loop modeling, 3) dynamic modeling, 4) scenario planning and modeling, 5) implementation and organizational learnings. The ndings from both our models were presented to high-level executives in the Ministry of Public Health for eliciting comments and feedbacks.
One hundred ten stakeholders, who are policymakers, healthcare administrators, and practitioners from multi-sectors in Thai health systems, participated in a series of our workshop. They were facilitated to cocreate a causal model that can explain the mismatches between demands and supplies of the health workforce in Thailand, which progressed from connecting relevant concepts to constructing qualitative causal loop diagrams (CLDs) and quantitative stock and ow diagrams (SFDs).

Group Model Building
Using scripts from system dynamics literature [10,11], we facilitated the stakeholders by using a structured group model building [7,12] to engage with relevant stakeholders. First, we discussed and agreed upon the expected outcomes in the next 20 years of the national planning of the health workforce, and drawn the reference mode of such outcomes. Second, we developed a CLD with stakeholders to gain a mutual understanding of what factors caused undesirable consequences, particularly mismatches of supply and demands for the health workforce in Thailand over the decades. Third, we continued working with stakeholders to turn our insights from the CLD into an SFD and simulated the selected health systems outcomes for the next two decades (2018-2037). Lastly, we analyze the consequences of such policy options by simulating them in our system dynamic modeling.
Our GMB sessions produced a CLD that represents a common undersetting among participating stakeholders. The critical variables discussed in the GMB sessions include the population structure of aging society, the unmet health needs of the population, utilization of healthcare services in hospital settings, utilization of healthcare services in non-hospital settings, size of the labor market of hospital care, size of the labor market in non-hospital care, the effectiveness of population health interventions and population's health literacy. The revised and nal CLD contains multiple interacting feedback loops that can be categorized into the four domains: 1) utilization of hospital care; 2) utilization of hospital care; 3) healthcare labor market for hospital care; 4) labor market for non-hospital care; 5) healthcare infrastructure; 6) self-care; and 7) drivers of population health.
As shown in Fig. 1, the balancing and reinforcing loops constitute the dynamic hypotheses of how health system components interact and result in a steady level of unmet health needs and rising demands for utilization of hospital care. Also, increasing demands for the workforce in hospital settings leads to decreasing supplies for the workforce in non-hospital settings, medical errors, rising healthcare expenditures, and an undesirable level of population health status over time.

Model Structure
The dynamic hypotheses, as depicted on CLD, formed a basis for our development of SFD and the structure of the system dynamics model. We constructed three modules to represent our insights from the CLD, which include factors and relationships that can lead to mismatches of supplies and demands for the health workforce in Thailand's health systems, including 1) population module; 2) healthcare delivery module; 3) Education and labor market module.

1) Population module
Considering how su ciency of the health workforce can impact the population health status, we considered each person can occupy a health state by the levels of severity of their illness. The status is re ected in Fig. 1 by the stocks: 1) healthy population, 2) population with simple illnesses, and 3) population with complex illnesses. Each health state corresponds to the nature of patient care teams and healthcare models that would be expected to inhibit progression into or regression from more severe health states, as represented by the in ows and out ows. Each person can also progress in terms of aging. Still, we categorized the population to only three groups by ages (0-14, 15-49, 50, and above). It also corresponds to the nature of patient care teams and healthcare models usually needed in that age group. The structure of the population is depicted in Fig. 2.

2) Healthcare delivery module
In the healthcare delivery module, we displayed the population health demands by health needs as professionally de ned [13]. Hence, on the demand side of the healthcare market, each of the health states (healthy population: HP, population with simple illnesses: SP, and population with complex illnesses: CP) creates speci c health demands for the health workforce and patient care teams in healthcare models on Fig. 2. The accessibility and utilization of each healthcare model on the population model are also described in Table 1. palliative care and end-of-life care (a.k.a. hospice care) and 8) oral healthcare usually organized in nonhospital settings. Besides, a workforce who do population health practices such as community-based disease prevention and health promotion are considered within 9) population health teams.

3) Healthcare education and labor market module
The structure of the health labor market and its relationship with health workforce education and training are shown in Fig. 3. The composition of health professions that forms a typical membership of each healthcare model is shown in Fig. 4. The supply side of the healthcare market is also the demand side of this healthcare labor market. Hence, the demands for hiring the health workforce in each profession are also determined by the capacity of the health workforce in the patient care teams, which can be categorized into nine types of healthcare teams or healthcare models.

Model Parameters
The parameters used in the model are shown in Table 2. These parameters were used in the initial steady state of our model, which represents a dynamic equilibrium and is numerically sensitive to model parameters. To test for policies, we evaluate the policies on four outcomes that concern health workforce planning at the national level. From our GMB process, the su ciency of the health workforce in Thailand can be seen by 1) population health status, 2) unmet health needs, and 3) health care expenditures.
The rst outcome is the overall population health status represented by the percentage of a healthy population in the country, which indicates an adequate health workforce in the effective healthcare models for the demands of population health. Another population health outcome is the health-related quality of life (HRQoL) of the Thai population, which captures the degree and effectiveness of long-term care and palliative care necessary for aging, disabled, and terminal stage patients who cannot be converted to a healthy state. The second outcome is unmet health needs, which re ect limited access to necessary care for their health status. An inadequate health workforce does not only compromise population health status, but can also create long-waiting time, congested patients at healthcare facilities, and equitable access to necessary care. The third outcome is the healthcare expenditure, which is the primary concern of the government and partially address the cost-effectiveness of policy interventions from the societal perspective. Policy Experimentation We ran our system dynamics simulations under four scenarios in three main model parameters were changed (i.e., service gap, out-of-pocket cost, and the number of doctors) to conduct policy experimentation and illustrate the potential impacts of each policy in the next 20 years (2017-2038) under the following scenarios: 1. Scenario I Business-As-Usual (BAU): All key policy variables were kept constant. Under this scenario, all model inputs, including the effectiveness of the available health workforce actively working in all healthcare models in Thailand, was assumed to be equal and remain unchanged over the simulation time. The model is validated using unit consistency test, structural validity test, and behavioral replication test [14]. To test for unit consistency, we used the unit test function in the Stella Architect software. We focused on two dimensions. First, the unit of each variable must have the meaning and consistent with the description of that variable. The second dimension is that the unit must be consistent throughout the model. After testing for the unit consistency, the unit of all variables represents the real meaning of those variables. Besides, Stella software shows no unit error, which indicates that the unit is consistent throughout the model. Therefore, the model passes the unit consistency test.

Scenario II
For the structural validity test, we tested the model by showing the model to the group of experts who works in the healthcare industry, research relating to healthcare service, and the government agencies who manage healthcare security and healthcare services. The experts agree that the structure of the model re ects the actual situation. Therefore, the model passes the structural validity test.
Lastly, we did a behavioral replication test. The reference model was drawn using multiple data, including the

Results
Our simulation modeling produced results, as shown in Figs. 5a, 5b, and 5c, displaying the impacts of the four scenarios on the four primary outcomes. The three policy options were compared to our baseline or the "Business-As-Usual" (BAU) scenario. We can observe the consequences of current health workforce policies that most workforce have been working in hospitals.
Under Scenario I (the BAU Scenario), the population health outcomes, both the ratio of a healthy population and health-related quality of life, gradually got worse over the next two decades. Both health systems' performance also declined, as the unmet health needs slowly increased, and health care expenditures kept rising over the whole period.
Under Scenario II (Policy#1), we considered the impacts of decentralization of primary care units from central government to local governments and limiting new recruitments of physicians into the MOPH facilities from the year 2027 on. These policy options emerged from our GMB process, but our simulation revealed that it produced almost the same patterns of systems behaviors like that of the BAU scenario. The healthy population and unmet health needs of the people got slightly worse than that of the BAU approximately after ten years of this policy implementation, or from the year 2027 (2567 BE) on.
Under Scenario III (Policy#2), we considered the impacts of expanding public nancing for private health care delivery while modernizing primary care in the public sector, especially implementing the digitalization of MoPH primary care units. From the year 2022 (2564 BE) or approximately after ve years of this policy implementation, the ratio of the healthy population and health-related quality of life rapidly improved. The unmet health also needs shapely dropped around the year 2022 and more gradually dropped furthermore after 2025 (2567 BE). The simulation of total healthcare expenditures displayed an interesting pattern of "worse before better" by immediately and sharply increased after policy implementation but decreased approximately after eight years, or from the year 2025 (2567 BE).
Lastly, under Scenario IV (Policy#3), we considered the impacts of signi cant reforms of all care delivery models by shifting the focus from only lling the health workforce in MoPH hospitals care and producing a substantial proportion of health workforce to promote non-hospital care. The ratio of a healthy population, health-related quality of life, and the unmet health need rapidly improved, similar to the pattern observed under Scenario III. However, we can observe the improvement slightly faster than that of Scenario III. The signi cant difference was on health care expenditures, which slightly increased from that of the BAU Scenario but not as highly increased as Scenario III. However, unlike Scenario III, health care expenditures never went down under Scenario IV.

Discussion And Conclusion
Our study was among the rst to investigate plausible scenarios of the strategic health workforce planning by taken into the account of healthcare delivery reforms of either Thailand or other low-and middle-income countries (LMICs). The evidence can inform the governance of Thailand's UHC in the next decades to come. Using a GMB process, the policymakers and stakeholders gained a better understanding of causal relationships among factors in Thai healthcare systems related to the su ciency of the health workforce or the mismatch of supplies and demands of the health workforce. Moreover, and policy options were tested by our quantitative simulation modeling to compare the consequences of each policy.
Initiating signi cant reforms of all care delivery models, by shifting the focus from only lling health workforce hospitals care to promoting health workforce placements in non-hospital care settings, or creating new care delivery systems for the integration of hospital care and non-hospital, can lead to the most desirable outcome consistently with suggestions from a stream of literature on integrated care [15,16] and value-based care [17,18]. The outstanding results can be clearly observed, especially when compared to merely hiring a new health workforce in the existing care models as depicted by the business as usual. The healthcare expenditure would increase by approximately 1.3 times of the starting year of 2017. More importantly, the better ratio of health population and the lower level of unmet health needs would result in fewer demands for the health workforce in the long run. The reduced unmet health needs can affect fewer demands for new facilities in both the public and private sectors.
Overall, linking workforce planning strategy with healthcare delivery reforms would provide better outcomes in population health status and health systems performance. It would be a far superior policy option, especially when compared to implementing a set of new health workforce policies in the exiting healthcare delivery models. While the complexity of managing the health workforce can be signi cantly increased during healthcare reforms, at the same time, inadequate preparation of human resources for incoming health systems reforms also can impact the performance of health systems negatively [19].
Alternatively, policymakers can implement new health workforce policies that emphasize new nancing mechanisms for existing care delivery models. The argument would be to increase the e ciency of the health workforce, their healthcare teams, and healthcare organizations. On the downside, as demonstrated by Scenario II, the unmet health needs of people without any access to necessary care would be kept at 20% in the next two decades. Yet, the healthcare expenditure would be approximately 2fold in the rst eight years and then decreased to a similar level of the BAU Scenario. Even with the assumption of using more information and communications technologies (ICTs) to receive a greater e ciency of health workforce utilization and care delivery models. While some policymakers believe using more ICT in healthcare delivery can be more e cient than producing and managing the health workforce, our ndings are consistent with a stream of literature that suggests the limited effects of ICT without shifting resources among care models or improving the design of health care delivery systems [20][21][22]. Hence, training a new workforce or retraining the existing ones already working in the health systems would provide a much better outcome.
Beyond the healthcare expenditures, any potential policies that rely on the new workforce, payment mechanism, or ICT systems implemented upon the existing healthcare delivery but not providing an incentive for the reforms of healthcare delivery will minimally affect the health status of the populations.
Hence, health workforce policies with a focus on the reforms of healthcare delivery itself, e.g., one that promotes a more balance between hospital care and non-hospital care or a greater integration among care models care, should be preferred. However, these policy options are unlikely successful if only a limited number of healthcare providers in the market offer integrated care. To increase the supplies, the focus of health workforce policies should not limit only public providers and include both public and private providers who quali ed. For instance, primary care clinics or rehabilitation centers in the private sector are currently not a major focus of the reimbursement systems of all major public healthcare funds. In Thailand, the payer, such as UCS, can team up with public or private hospitals to establish an integrated care process for their patients. However, by this option, healthcare expenditures can increase more rapidly in the early years due to the higher unit cost of health care services in the private sector compared to that of public providers.
As put forth by Milstei, Homer & Hirsch (2010) [23], system dynamics modeling can demonstrate the consequences of policy options of healthcare reforms in a more comprehensive way. However, our study may have some limitations in predicting future outcomes if the assumptions used to construct our system dynamics modeling is too far from the complex reality. The health outcomes can be altered from the simulated ones for several reasons, including 1) the quantity and quality of health workforce in the future might be inadequate for all health demands of Thai populations, 2) the patients have a preference for speci c types of healthcare teams, or 3) their accessibility to new care models was not as high as expected. Moreover, the healthcare expenditures may increase even more than the simulated numbers if the government expands the UHC bene t packages from the existing ones. Lastly, due to the exploratory nature of our study, our model reveals the trend of population health status and systems performance outcomes as the consequences of each policy option. Still, we did not aim to precisely forecast an exact amount of healthcare expenditures or any other results. More speci cally, for simulated healthcare expenditures, we did not take into account of the in ation in our model yet.
Nonetheless, building upon the present study, policymakers of healthcare reforms can bene t from further analyses. The synthesis of additional policy options by group model building and testing such policies by simulation modeling can help not only the strategic planning health workforce at the national level but also the planning and evaluation of the ongoing UHC reforms. Our modeling process also informs policymakers and stakeholders about what data in health information systems is crucial to the strengthening of UHC governance, particularly regarding managing the health workforce and health systems performance. Hence, this iterative nature of data collection and data analysis could be a lesson learned for the UHC policy process, not only in Thailand but also in other LMICs. The CLD of insu ciency of the health workforce in the hospital care and the non-hospital care settings of Thailand from the GMB process