Projection of Future Medical Expenses Based on Medical Needs and Physician Availability

Abstract

Background

Accurate scientific projections of future healthcare expenditures and workforce capacity are vital in South Korea for addressing concerns about the sustainability of the national health insurance system. This study aims to analyze projected changes in healthcare expenditures due to demographic shifts and identify appropriate healthcare workforce to meet future demands.

Methods

Data from Statistics Korea, the National Health Insurance Service, the Bank of Korea, and the Korea Development Institute were used. The Stepwise Auto Regression Model projected healthcare costs and insurance rates, considering future population estimates, the proportion of older people in the population, life expectancy, changes in medical cost rates, nominal Gross National Income, and the ratio of current medical expenses on Gross Domestic Product (GDP). The analysis applied two scenarios: maintaining the current medical school admission quota and increasing it by 1,509 students.

Results

The study anticipates a rise in future medical insurance rates alongside a gradual decline in the rate of change in medical costs. The demand for medical services is forecasted to grow by over 4% annually for the next 30 years due to an aging population and low birth rates. The ratio of current medical expenses on GDP is projected to increase significantly, reaching approximately 20.0% in 2060 from 9.7% in 2024. In two scenarios: if 3,058 medical students are added to the existing medical license holders, medical costs per active physician will increase by 2.8 billion won; if 4,567 medical students are added, the costs will increase by 2.3 billion won by 2060. Despite 1,509 new medical students annually, the number of active physicians will increase by only 1% per year, starting a decade later. Consequently, the medical market will continue to expand, and the demand for medical services per physician will not decrease. Health insurance rates are expected to rise steadily from 7.09% in 2024 to 14.39% by 2060.

Conclusion

This underlines the imperative to prioritize enhancing the sustainability of the healthcare system over solely augmenting medical student numbers. We should scientifically and precisely predict future medical costs and consider deeply whether it is right to shift the burden of intergenerational medical care to future generations at this point.

Graphical Abstract

INTRODUCTION

Governments worldwide recognize the importance of healthcare systems. A stable healthcare system is a crucial prerequisite for sustainable economic development in the long term, with investments in healthcare seen as investments in future societies.1 Consequently, healthcare policy ranks among the foremost challenges faced by nations globally. The quality of a healthcare system hinges not only on treatment outcomes but also on the efficient allocation of medical resources and the effective management of healthcare finances.

Understanding the appropriate supply of healthcare resources to meet future needs is key to the sustainability of the healthcare system. Healthcare demand changes in response to socio-economic shifts such as demographic changes and economic growth. Globally, healthcare expenditures are increasing annually, now comprising approximately 10% of the global Gross Domestic Product (GDP).2 The rapid increase in healthcare costs will consequently raises the burden on healthcare finances, threatening the system’s sustainability.

In South Korea (hereafter Korea), where there is rapid aging and one of the world’s lowest birth rates, the increase in healthcare expenditures is particularly steep. The decline in the economically active population due to aging and low birth rates is expected to reduce the National Health Insurance fund. Conversely, an increase in healthcare demand and expenditure is expected, resulting in persistent concerns about the sustainability of the healthcare insurance system. Korea is a prominent nation that has rapidly achieved and successfully maintained a universal healthcare system.3,4 However, the future outlook remains increasingly uncertain in population structure and rising costs. Accordingly, the Korean government is planning for the future supply and demand of medical personnel and examining various scenarios for maintaining the insurance system.5,6,7

To ensure the sustainability of Korea's healthcare system, realistic policies need to be formulated to address the increasing healthcare costs due to population aging and low birth rates, as well as to manage future healthcare demand through an adequate healthcare workforce supply. Prioritizing scientific estimation of anticipated healthcare expenditures and doctors’ capacity to handle them is crucial before formulating policies, to accurately understand the current situation.

Therefore, the purpose of this study is to examine changes in future healthcare expenditure levels in response to demographic shifts and to identify appropriate healthcare workforce supply aligned with future healthcare demand. Additionally, the study aims to predict future health insurance rates to achieve universal healthcare coverage for the entire population.

METHODS

Data source

We utilized population data and projected future population data from Statistics Korea for the years 2004 to 2060, as well as the proportion of older people (over 65-years-old) in the population, life expectancy, annual medical costs data, current medical costs on GDP, nominal Gross National Income (GNI), and the number of active physicians obtained from the National Health Insurance Service. Additionally, we incorporated nominal and real GDP data provided by the Bank of Korea and the Korea Development Institute for the study period (Supplementary Table 1). In Korea, 3,058 students graduate from medical schools every year. However, the government has recently decided to increase the number of medical school students by 1,509 every year starting from 2025. We created two scenarios to account for this change: 1) adding 3,058 medical students to the existing medical license holders, and 2) adding 4,567 medical students, including the additional 1,509 students, to the existing medical license holders. It is important to note that in both scenarios, we assume the number of active physicians is 90% of the total medical license holders. Medical costs per active physician and per patient were calculated by dividing the annual population. Then, the annual medical costs were used to calculate yearly rates of change in medical costs.

Statistical analysis

The study harnessed the practicality of the Stepwise Auto Regression Model (SARM), an effective statistical model for projecting healthcare costs and insurance rates. This model is adept at selecting the most important predictors and eliminating those with minimal impact on predictions. It operates in three key steps as follows. First, we applied the SARM approach to predict the target variable, annual medical costs, and health insurance rates based on external variables. Second, we managed the persistence of the remaining unpredicted variation using an autoregressive model (AR). Third, if the error term from the first step is independent and identically distributed, the residual term can be calculated with usual AR models and the forecasted value of the target variables, including annual medical costs and health insurance rates. Given the observations and times, we constructed the final SARM for forecasting annual medical costs and health insurance rates as follows:

$$y_{i,t+h|t}\mathrm{ }=\mathrm{ }\underset{j=1}{\overset{p_i}{\sum }}{\widehat{\phi }}_{i,j}{y}_{i,t+h-j|t}\mathrm{ }+\mathrm{ }\sum _{j=0}^{p_i}{X}_{t+h-j}^{\mathrm{\text{'}}}{\widehat{\gamma }}_{i,j}$$

For more details on the above steps, please refer to the Supplementary Method 1.

This paper used the final SARM to project annual medical costs and health insurance rates from 2023 to 2060, using data from 2004 through 2019. We excluded three years (2020, 2021, and 2022) due to unusual issues such as the coronavirus pandemic. In projecting annual medical costs and health insurance rates, we finally considered the estimated future population, changes in medical cost rates, life expectancy, changes in population rates, nominal GNI, and two scenarios of the number of active physicians. The annual medical costs per capita were calculated by dividing the estimated future population.

In projecting health insurance rates, we predicted the number of employee-insured individuals based on the estimated future population and calculated their costs. Subsequently, health insurance rates were forecasted using economic indicators such as nominal GDP and GNI, along with the number of active physicians engaged in patient care or other professional activities.

Tables 1, 2, 3 present the results of polynomial regression for predicting future annual medical costs and health insurance rates of the middle-level population using the SARM model.

Table 1. Polynomial regression by the Stepwise Auto Regression Model for estimating the medical expenses and health insurance rate (middle level).

Explanatory variables			Coefficients	Standard error	t-statistics	P value
Medical expenses
	Step 1. Included all explanatory variables
		Proportion of older people	5.08	1.02	0.50	0.623
		Life expectancy	−1.54	2.70	−5.69	< 0.001
		Nominal GNI	4.31	8.23	5.23	< 0.001
		Changes in medical costs	−7.51	3.32	−2.26	0.037
		Changes in population	2.27	9.02	2.51	0.024
		Current medical costs on GDP	−4.04	5.44	−0.74	0.465
		Changes in medical costs by 3,058 medical students	5.37	4.25	12.65	< 0.001
		Changes in medical costs by 4,567 medical students	−1.60	3.25	−4.92	< 0.001
	Step 2. Excepted insignificant variables from step 1
		Life expectancy	−1.66	1.86	−8.92	< 0.001
		Nominal GNI	4.32	4.99	8.65	< 0.001
		Changes in medical costs	−7.98	3.20	−2.50	0.023
		Changes in population	2.39	8.75	2.73	0.015
		Changes in medical costs by 3,058 medical students	5.10	1.15	44.33	< 0.001
		Changes in medical costs by 4,567 medical students	−1.42	2.02	−7.02	< 0.001
Health insurance rate
	Step 1. Included all explanatory variables
		Portion of Workers in the Population	−2.55	5.58	−4.56	< 0.001
		Current medical costs on GDP	−3.73	9.21	−4.05	0.001
		Life expectancy	3.88	8.08	4.80	< 0.001
		Nominal GNI	−2.42	1.15	−2.10	0.041
		Changes in medical costs by 3,058 medical students	3.37	3.38	9.98	< 0.001
		Changes in medical costs by 4,567 medical students	−1.11	7.96	−1.39	0.170
	Step 2. Excepted insignificant variables from step 1
		Portion of workers in the population	−1.79	1.34	−13.36	< 0.001
		Current medical costs on GDP	−4.09	8.92	−4.59	< 0.001
		Life expectancy	2.79	2.07	13.52	< 0.001
		Nominal GNI	−2.02	1.13	−1.79	0.078
		Changes in medical costs by 3,058 medical students	3.04	2.43	12.49	< 0.001

R² = 0.999.

GNI = Gross National Income, GDP = Gross Domestic Product.

Table 2. Comparison of the forecasting medical expenses (middle-level) from the SARM model in 2004–2019.

Years	Medical expenses by per population		SARM
	Predicted (1,000 won)	Actual (1,000 won)	ARIMA (1,2,0)
2004	463.97	464.18
2005	511.19	510.68
2006	556.73	581.14	MAE
2007	650.49	659.25	2126933763
2008	736.13	705.78
2009	752.14	792.22	MASE
2010	878.29	874.67	0.48
2011	956.22	922.70
2012	970.25	950.22	RMSE
2013	977.88	1,007.26	2652715996
2014	1,064.02	1,082.75
2015	1,157.99	1,153.78	MAPE
2016	1,224.93	1,275.30	3.65
2017	1,397.25	1,381.06
2018	1,487.70	1,518.70
2019	1,657.04	1,677.35

The exchange rate in 2024 is approximately 1,000 won to 0.77 dollars.

SARM = Stepwise Auto Regression Model, ARIMA = autoregressive integrated moving average, MAE = mean absolute error, MASE = mean absolute scaled error, RMSE = root mean square error, MAPE = mean absolute percentage error.

Table 3. Comparison of the forecasting health insurance rate (middle-level) from the SARM model in 2004–2019.

Years	Health insurance rate (%)		SARM
	Predicted	Actual	ARIMA (0,1,0)
2004	4.27	4.21
2005	4.46	4.31
2006	4.43	4.48	MAE
2007	4.53	4.77	0.001
2008	4.87	5.08
2009	5.35	5.08	MASE
2010	5.27	5.33	0.68
2011	5.47	5.64
2012	5.94	5.80	RMSE
2013	6.16	5.89	0.002
2014	6.20	5.99
2015	6.30	6.07	MAPE
2016	6.30	6.12	2.22
2017	6.26	6.12
2018	6.10	6.24
2019	6.21	6.46

SARM = Stepwise Auto Regression Model, ARIMA = autoregressive integrated moving average, MAE = mean absolute error, MASE = mean absolute scaled error, RMSE = root mean square error, MAPE = mean absolute percentage error.

Ethics statement

All data used in this study were collected and publicly released by the national government, thus ethical approval was not required. The authors assert that all procedures contributing to this work comply with the ethical standards of the relevant national and institutional committees and with the Helsinki Declaration. The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

RESULTS

Fig. 1 provides essential information about the future of healthcare. It shows the projected medical costs as a percentage of GDP and the annual medical costs for the estimated future population from 2023 to 2060, excluding the period between 2020 and 2022 due to the coronavirus pandemic. Despite a decreasing population, the projected current medical costs as a percentage of GDP are expected to increase from 9.7% in 2024 to 20.0% in 2060.

Fig. 1. Current medical cost projection on GDP (middle-level estimated, 2023–2060). The confidence intervals (UCI and LCI) would depend on the distribution of error terms in the Stepwise Auto Regression Model. The red line represents the current projection of medical costs on GDP, while the blue and yellow dotted lines indicate the upper and lower confidence intervals, respectively.

GDP = Gross Domestic Product, LCI = lower confidence interval, UCI = upper confidence interval.

In Fig. 2A, annual medical costs per capita are predicted to increase rapidly, reaching approximately 10 million won by 2060 based on the estimated middle-level future population. Two scenarios were considered: adding 3,058 medical students to the existing medical license holders would lead to a 2.8 billion won increase in medical costs per active physician, while adding 4,567 medical students would result in a 2.3 billion won cost increase by 2060.

Fig. 2. Medical cost projection (middle-level and low-level estimated, 2023–2060). The confidence intervals (UCI and LCI) would depend on the distribution of error terms in the Stepwise Auto Regression Model. The purple and green lines show the projected per capita allocated medical costs for 3,058 and 4,567 students, respectively. The blue line represents the overall per capita projection of medical costs, while the green and yellow dotted lines indicate the upper and lower confidence intervals, respectively.

LCI = lower confidence interval, UCI = upper confidence interval.

Fig. 2B shows projected medical costs based on the estimated low-level future population from 2023 to 2060, except between 2020 and 2022. The results indicate an approximate 2 million won increase in annual medical costs per capita, reaching approximately 12 million won by 2060. As shown in Fig. 2, annual medical costs are projected to continue increasing until 2060. However, increasing the number of doctors does not reduce annual medical costs in the estimated middle—and low-level populations.

Fig. 3 presents projected health insurance rates based on estimated middle-level medical costs from 2023 to 2060. In 2024, Korea’s health insurance rate for employee-insured individuals is 7.09%. According to Figs. 1 and 2, as annual medical costs increase, the health insurance rate will also increase linearly to 14.39% by 2060. Based on the estimated low-level medical cost projection for the same period, the health insurance rate for employee-insured individuals will rise to 15.85% by 2060, as shown in Fig. 4. Compared to the middle-level medical cost projection, health insurance rates show a 1.46% difference. Figs. 3 and 4 indicate that medical health insurance rates are likely to increase, while the rate of change in medical costs is expected to decrease over time. Table 1 presents estimated health insurance rates for the future middle-level population using polynomial regression by the SARM model. The results encompass 99% of the portion and average medical costs for employee-insured individuals in the population.

Fig. 3. Rate of medical health insurance based on medical cost projection (middle-level estimates, 2023–2060). The confidence intervals (UCI and LCI) would depend on the distribution of error terms in the Stepwise Auto Regression Model. The purple and green bar histograms represent the rates of total medical costs and rates of medical costs of the insured employee, respectively.

The orange line represents the overall per capita projection of the medical insurance rate, while the blue and green dotted lines indicate the upper and lower confidence intervals, respectively.

LCI = lower confidence interval, UCI = upper confidence interval.

Fig. 4. Rate of medical health insurance based on medical cost projection (low-level estimates, 2023–2060). The confidence intervals (UCI and LCI) would depend on the distribution of error terms in the Stepwise Auto Regression Model. The purple and green bar histograms represent the rates of total medical costs and rates of medical costs of the insured employee, respectively.

The orange line represents the overall per capita projection of the medical insurance rate, while the blue and green dotted lines indicate the upper and lower confidence intervals, respectively.

LCI = lower confidence interval, UCI = upper confidence interval.

DISCUSSION

Medical spending encompasses expenditures on disease prevention, treatment, and health improvement, influenced by both micro- and macroeconomic factors such as income growth, advancements in medical technology, pharmaceutical developments, and government policies. Healthcare expenses also can be affected by various factors, including population size and demographic structure.8 In contrast to the past, there is a growing interest in higher life expectancy in developed and some developing countries. This trend has resulted in an increasingly aging population over time, along with changes in social standards and personal preferences. Disease prevention programs have also contributed to increased health spending.9

In the future, Korea is expected to experience a decrease in its population due to aging and low birth rates. According to a report from Statistics Korea,10 comparing the years 2022 and 2072 reveals significant demographic shifts: the proportion of Korea’s working-age population decreased from 71.1% to 45.8%, the elderly population increased from 17.4% to 47.7%, and the youth population decreased from 11.5% to 6.6%. In addition, the proportion of older people (over 65-years-old) is expected to continue to increase, reaching 17.5% of the population by 2025 and surpassing 20.6% in 2035, leading to a super-aged society with over 30.1% in 2050. On the other hand, in 2024, the average birth rate among OECD member countries will be 1.6. However, Korea's birth rate is declining, reaching 0.7, and women’s average childbirth age is gradually increasing.

In economic terms, Korea’s current medical expenses to GDP ratio in 2022 is 9.7%, higher than the OECD average of 9.5%.11 With rapid economic growth and an increasing elderly population, managing healthcare expenditures relative to GDP is crucial. An OECD report (2015) highlighted health insurance expansion and long-term care insurance introduction as key drivers of Korea’s healthcare cost escalation, alongside increased pharmaceutical spending.12 While rising healthcare expenditures can stimulate economic activity, excessive cost increases pose sustainability risk to the healthcare system and strain the broader economy. The shifting population demographics present new societal challenges, necessitating detailed analyses for accurate healthcare spending projections.

Forecasting medical expenditures involves critical considerations such as selecting the forecasting timeframe and incorporating pertinent variables. Prior research has typically focused on medium- and long-term factors like inflation, income levels, population dynamics, disease outbreaks, severity and utilization, health insurance, physician availability, medical technology, government regulation, and health system structure.9 Population structure, economic conditions, medical demand, and physician availability relatively influence medical expenditure forecasts.13 Thus, our study examines crucial variables, including total annual medical costs, changes in population, proportion of older people, life expectancy, nominal GNI, current medical costs on GDP, and physician supply changes. Sensitivity analyses incorporate normal, good, and bad economic scenarios in terms of future GDP since 2023. The results of the sensitivity analysis are shown in Supplementary Figs. 1 and 2.

Numerous previous studies have focused on predicting total healthcare spending in their respective countries. For instance, the microsimulation model is a powerful tool that uses different scenarios to help policymakers assess the characteristics and behaviors of a large sample of individuals representing the entire population of interest.14 The component-based model considers a wide variety of factors, such as financing agents, providers, consumed health services, and individual groups in cohort-based models.15 Furthermore, the macro-level model restricts the analysis of overall health spending in long-term forecasts due to unresolved trends and the lack of structural breaks.16 However, these three models have certain limitations. Although the component-based model is generally less data-intensive than the micro-simulation model, it uses estimates divided into categories. The macro-level model has the lowest data requirements but can only forecast for the short-term period.17 Earlier research often employed exponential trend analysis, decomposition, and regression in time series modeling. However, recent time-series studies have shifted towards using forecasting models such as the Box-Jenkins methodology, the Autoregressive Integrated Moving Average (ARIMA) model, the Vector Autoregression model, and Neural Networks.18

In the U.S., stochastic time series models have been used to predict health spending, utilizing data on the aging population and annual health spending as a percentage of GDP between 2002 and 2075.19 In Canada, the least-squares method and the general least-squares model were used to forecast future health spending from 1965 to 2008.20 Additionally, the ARIMA model was employed to project health costs for respiratory illnesses in Shanghai, China, in 201221 and to predict China’s total health spending as a percentage of GDP.22 However, the ARIMA model has several potential limitations for forecasting. Firstly, the available period is relatively short. Secondly, it requires consideration of explanatory variables, and a multicollinearity problem may be encountered with these variables. Additionally, it assumes nonstationarity and may exhibit model error with autocorrelations.23 As a result, we explored a new approach using the SARM model with Ordinary Least Squares regression. The SARM model was proposed to address these potential issues. In the study, the SARM model first removed insignificant variables stepwise among all explanatory variables considered and ultimately predicted outcomes, including explanatory variables that affect medical costs.

Over the past 30 years, Korea’s economy has grown steadily, marking a golden age with the highest proportion of the working population. Factors such as an aging population, advanced medical technology, and increased welfare needs have prompted significant investments in medical health insurance. The Korean government has implemented policies to expand health insurance coverage and improve accessibility to medical care while reducing economic burden. However, it is crucial to ensure that this growth is sustainable. Without guaranteed rapid economic growth and improvements in population structure, past achievements could become future burdens. Furthermore, society needs to adequately discuss the necessary preparations and responses for when this structure of unlimited supply and demand collapses. We are now at a point where it is becoming increasingly challenging to manage the rapid increase in medical demand and consider its progress in our society.

In 2024, Korea plans to implement an expansion policy, increasing the annual intake of medical college students from 1,509 to meet the growing medical demand. However, attempts to rectify the imbalance between medical supply and demand by increasing medical enrollments may expedite future sustainability depletion. As shown in Figs. 1 and 2, the demand for medical services is expected to increase by more than 4% annually over the next 30 years, mainly due to the aging population and low birth rate. Despite an additional 1,509 medical students entering the field each year, the total number of active physicians will only increase by 1% per year, beginning ten years later. This implies that the growth in the number of practicing physicians will not keep pace with the rising demand for medical services despite the increased medical student enrollment. As illustrated in Table 1, the total number of active physicians correlates with annual medical costs. However, the annual influx of 1,509 medical students is still insufficient to reduce future annual medical costs. Consequently, the medical market will continue to expand, and the demand for medical health services per physician will not decrease. Moreover, the health insurance rate is estimated to increase steadily from 7.09% in 2024 to 14.39% by 2060. This shift necessitates a change in perspective, emphasizing the need to enhance the sustainability of the medical health system rather than simply increasing the number of medical students. It is important to consider deeply whether it is right to shift the burden of intergenerational medical care to future generations at this point.

Although the forecasting results relied on various scenarios—such as economic conditions, population changes, estimated future population growth, fluctuations in medical costs, life expectancy, and two scenarios regarding the number of active physicians—the study faced significant limitations when using the SARM model to predict medical spending. First, while the SARM model can reduce the impact of unknown variables, unexpected factors can still affect the predicted medical costs. Second, the analysis did not consider regional or age-related effects, which may lead to wide 95% confidence intervals in the long-term estimated medical costs. Third, we excluded data from the years 2020 to 2022 due to the unexpected coronavirus pandemic, which resulted in a somewhat erratic forecast curve for 2023. To address this, we processed only the data from 2023, using an arithmetic mean that considered the four years prior. For these reasons, it is recommended that future studies broaden the forecasting of medical costs by exploring other established methodologies that could better reduce error correction modes.

SUPPLEMENTARY MATERIALS

Supplementary Method 1

The Stepwise Auto Regression Model (SARM)

Supplementary Table 1

Inclusion variables for projecting future medical expenses

Supplementary Fig. 1

Sensitivity analyses of medical cost projection according to impact of normal, good, and bad economic scenarios (middle-level estimated, 2023–2060).

Supplementary Fig. 2

Sensitivity analyses of medical cost projection according to impact of normal, good, and bad economic scenarios (low-level estimated, 2023–2060).