Abstract
Objective: Recent health spending growth has outpaced the growth of Hong Kong’s gross domestic product (GDP). Despite this, Hong Kong’s government has maintained its historical 50% share in financing health spending, with out-of-pocket and insurance funds paying for 30% and 20%, respectively. This paper quantifies the future impact of population aging on the future total costs of health care from 2023 until 2040. We decompose predicted health spending growth by inpatient, outpatient, day curative care, medical goods, and long-term care.
Methods: The past 23 years of health spending per capita data was modeled as a time series depending on the share of the population over 60, GDP per capita, and medical inflation. The best-fitting autoregressive, integrated, moving average (ARIMA) models were used to predict total and sub-components of health spending until 2040.
Results: Health spending began to grow faster than GDP in 2010 and is currently growing twice as fast as GDP. If immigration successfully slows population aging, health spending growth will return to parity with GDP growth by 2028. In all scenarios, spending’s share will rise from its current share of 7% to as much as 9% of GDP. Long-term care will be the fastest-growing component of health spending.
Discussion: Health spending in Hong Kong will continue to grow faster than GDP for the next half-decade. Proactive choices for the government include improving efficiency in the curative services of the Hospital Authority, improving the Health Department’s ability to prevent disease through better population-level public health programs, keeping up with demand by increasing fiscal space through new sources of public revenue and enabling Hong Kong’s health insurance markets to be a more effective pressure valve for the public system. The longer Hong Kong waits for a course correction, the harder reform will be. It is urgent for Hong Kong to deliberate on the best options to cope with population aging and rising health spending without suffering degradation of the publicly provided health services.
Introduction
Between 1990 and 2020, total health spending in Hong Kong rose by an average annual rate of 5.6%, while the corresponding GDP growth rate was 3.4%. Furthermore, health spending growth rates are accelerating. Health spending in the pre-Covid period grew at 6%, 6.5%, and 6.9% from 2017 to 2019. The fundamental reasons for rising health spending are population aging, rising prices for health commodities, increased sickness, and increased treatment intensity. In Hong Kong, aging is the most significant factor that will drive up health spending. Much more could be done in Hong Kong to address the price growth of new health commodities, prevent disease, and improve the cost-worthiness of treatments. Planning ways to finance the growing costs was urgent 20 years ago, and now in the throes of population aging, the problem is becoming larger and harder to fix.
In 2022, Hong Kong had 1.59 million seniors 65 or over, composing one-fifth of the population. By 2036, it will have 2.41 million seniors and they will make up one-third of the population (Hong Kong Census and Statistics Department 2023); (Hong Kong Hospital Authority 2022). Inevitably, the rising number of seniors will need an increasing supply of health services. Although this population is currently 20% of the population, they already account for half of all bed days and admissions (Hong Kong Hospital Authority 2022). Past trends in the health care utilization of seniors predict a future growing need to pay for health care. In addition, medical care prices are likely to rise as new drugs and treatments become available. New technology in health care is rarely cost-saving (Chernew and Newhouse 2011). Changes in treatment intensity can also raise costs as doctors do more things for sick patients because of an expanding range of diseases for which there are treatment options.
Health spending growth exceeds GDP growth in most countries (Farag, NandaKumar et al. 2012). This growth puts pressure on governments whose revenue typically rises only as fast as GDP. In Hong Kong total health spending was $189 billion HKD in 2019/2020 accounting for 6.8% of GDP. The government funded 54% of total health spending coming to $102 billion government spending on health out of the $731 billion HKD of total government spending in 2019/2020 (Hong Kong Government 2021). Remarkably, the Hong Kong government has been able to keep pace with its share of rising total health spending, holding it between 48% and 54% of the total since 2005 (Hong Kong Health Bureau 2023) (See Figure 1). Even though over 1 million Hong Kongers have purchased voluntary health insurance policies in recent years, Figure 1 shows that these schemes have not succeeded in channeling a greater share of total financing.
Figure 1. Share of financing sources for health 2000-2021.
The projected aging of Hong Kong’s population in the coming decades will put unprecedented fiscal stress on Hong Kong’s health system. This paper offers new forecasts of health spending in Hong Kong from now until 2040 to help policymakers plan ways to sustain high-quality services and financial protection. The data for our analyses came from time-series from the Health Bureau and Census and Statistics Department spanning the years 2000 to 2022 (Hong Kong Census and Statistics Department 2023, Hong Kong Health Bureau 2023).
Methods
Autoregressive, integrated, moving average (ARIMA) models are widely used to forecast health spending (Getzen and Poullier 1992, Klazoglou and Dritsakis 2018, Zheng, Fang et al. 2020). They have the virtue of requiring a minimal number of assumptions because they just use the information about aggregated spending from the past to predict future aggregated spending[1]. We use a baseline unadjusted model as well as models adjusted for population aging, medical price inflation, and GDP growth.
Baseline Model
Our baseline ARIMA model aggregates prices, diseases and population aging together, and so assumes the future overall patterns will look like past overall patterns. As discussed in the appendix we first assessed the degree to which health spending in a given year depended on spending and random shocks in prior years[2]. Then we used data from 2000 to 2022 to fit a simple model as follows
[1]
Where Ht is the total health spending in time t and T is time in years. The terms in parentheses are inserted to correct for how much health spending in one year depends on health spending in the past and were determined using the Akaike information criteria. Once b is determined, the forecast estimate is simply Ht = C + for T from 2023 to 2040.
Aging Adjusted Model
The aging adjusted model used equation [1] but also included past data on the percent of the population over 60 (PP60). The forecast estimate is Ht+ = C + and this required the census’s forecast of PP60t+ into the future. The Hong Kong Census bureau projects that population over 60 will stop rising by 2030 if Hong Kong succeeds in achieving net immigration of about 50,000 younger residents each year in coming decades.
GDP and Medical Inflation Adjusted Model
Other adjusted models used equation [1] but also included past data on the percent of the population over 60 (PP60) as well as past GDP and past medical price deflators. The forecast estimates in these models locked in GDP growth rate of 2.41% annually which equaled the average GDP growth rate of Hong Kong between 2000 and 2022. It locked in a medical inflation estimate of 4.67% based on the past medical inflation of 2000 to 2020.
Results
Summary of Data Used
Table 1. Description of Data Used
| Observations | Mean(SD)/Median (IQR) | |
| Year (years) | 23 | 2021 (11) |
| Percentage of population over 60 (%) | 23 | 20% (12.56) |
| GDP per capita ($) | 23 | $35,868 (9159.91) |
| HE per capita ($) | 23 | $2,068 (851.84) |
| Medical inflation (%) | 20 | 4.67 (2.05) |
Data are presented as mean (SD) or median (IQR). HE, health expenditure; GDP, gross domestic product. All spending predictions are presented in current dollars.
Main results
Figure 2 compares the current dollar unadjusted forecast of health spending from the baseline model [1] as a blue dashed line and current dollar health spending adjusted for the population over 60 years old as a blue solid line. For reference, the trend in GDP in Hong Kong is predicted by locking in 2.41% GDP growth and is shown as a solid red line. Unadjusted HE is a linear trend from 2022 onwards, increasing from 3 times its baseline in 2020 to over 5 times its baseline in 2040. Adjusting for the census’s predicted slowing of population aging, the adjusted health spending will only grow to 4 times its baseline in 2040.
Compared to either estimate of health spending, GDP will only approach 3 times its year 2000 baseline by 2040 if it keeps growing at 2.41%. Figure 3 compares the slopes of the curves in Figure 2 and shows that the HE growth rate will decelerate after 2028 as aging slows down. The demographic future depends on a Census Bureau projection where Hong Kong allows net immigration of roughly 50,000 younger people per year in coming decades and this softens the increases in the share of the population over 60[3].
Figure 4 shows the trends of HE as a percentage of GDP over a 42-year period from 2000 to 2042. HE share of the GDP is projected to grow from 7% in 2020 to over 9% in 2040 under the baseline unadjusted trend model. However, adjusting for immigration-based demographic changes, despite peaking at nearly 8% around 2032, the share of HE in the economy is expected to be lower by 2040, at around 7.5%. This suggests that factors related to population aging have a moderating effect on the growth of economic resources allocated to health spending.
Additional models forecasting the sub-components of health spending are shown in Appendix 2. As can be seen from the Appendix Figure 2.1, the largest health spending components are outpatient curative care, inpatient curative care, day curative care, medical goods, and long-term care in 2000 and 2020. Their predicted relative contributions will be unchanged in 2040. Within these five primary components, as shown in Appendix Figure 2.2, long-term care will be the fastest-growing component of health spending. Long-term care will rise from over 6 times its baseline in 2020 to around 12 times its baseline in 2040. Day curative care will be the second-fastest growing component, growing to nearly 9 times its baseline in 2040. It is worth mentioning that the preventive care component peaked at around 10 times its baseline from 2020 to 2022, which can be attributed to the COVID-19 outbreak when preventive care became the fastest-rising component of health spending. For robustness, we found that the ARIMA models adjusting for GDP or medical inflation into the base case model did not substantially alter our results. This occurs because the effects of medical inflation and GDP on health spending were already embedded in the growth pattern of the HE variables from 2000 to 2020. (Appendix 1)
Figure 2. Trend of HE&GDP over the 42-year period from 2000 to 2042 in Hong Kong normalized against their value in the year 2000. Blue dashed line, trend for HE based on the baseline model; Blue solid line, trend for HE based on the aging adjusted model; Red solid line, trend for GDP. HE, health expenditure; GDP, gross domestic product.
Figure 3. Predicted trends of growth rates for HE&GDP over the 20-year period from 2023 to 2042. Blue dashed line, predicted trend for HE growth rates based on aging adjusted model; Red dashed line, predicted trend for GDP growth rates. HE, health expenditure; GDP, gross domestic product.
Figure 4. Trends of HE as a percentage of GDP over 42-year period from 2000 to 2042. Red dashed line, HE/GDP based on the baseline model; Blue dashed line, HE/GDP based on the aging adjusted model. HE, health expenditure; GDP, gross domestic product.
Discussion
The prospect of growth in Hong Kong’s health spending will put an increasing strain on government finances. Health spending growth will outpace GDP growth at least until 2028. If Hong Kong can achieve net immigration of about 50,000 younger people per year in the coming decade, it could bend the curve and restrain health spending’s share of the GDP to 7% by 2040. However, without the benefit of immigration to soften population aging, Hong Kong’s health spending will rise to 9% of GDP by 2040.
Policy options to improve Hong Kong’s health system have been extensively discussed, but major reform has been elusive. Solutions need to be multi-faceted. Principal options are 1) Invest more in keeping people out of the hospital through prevention, health promotion, and comprehensive primary health care; 2) Improve efficiency in the use of current health care resources through demand and supply side utilization controls; 3) Manage the entry and pricing of new drugs and technologies; 4) Help citizens plan and pay for culturally appropriate long term care; 5)Help Hong Kong’s health insurance markets play a larger role in financing health care[4]; 6) Create new government funding by hypothecated revenues or social insurance (Leung and Bacon-Stone-J 2006).
We have shown that health spending growth will continue to outpace the government’s ability to pay. Because Hong Kong’s health spending growth is driven mostly by aging, its lived experience in the public sector will be an increase in admissions, bed days and outpatient visits demanded by more seniors who need more services from bedraggled public sector health workers. Waiting times for services in the public sector that are already long, will get longer. Hong Kong’s celebrated achievements in population health indicators (e.g. low infant mortality, long life expectancy) will become harder to maintain. It is beyond the scope of this paper to recommend which of the above-mentioned reforms are best suited for Hong Kong’s population. Our results call attention to a financial sustainability problem that has been growing and will continue to grow unless more steps are taken.
References
Chernew, M. E. and J. P. Newhouse (2011). Health care spending growth. Handbook of health economics, Elsevier. 2: 1-43.
Farag, M., A. NandaKumar, S. Wallack, D. Hodgkin, G. Gaumer and C. Erbil (2012). “The income elasticity of health care spending in developing and developed countries.” International journal of health care finance and economics 12(2): 145-162.
Getzen, T. E. and J.-P. Poullier (1992). “International health spending forecasts: Concepts and evaluation.” Social Science & Medicine 34(9): 1057-1068.
Hong Kong Census and Statistics Department. (2023). “Hong Kong population projections for 2022-2046.” Retrieved November 23 ,2023, 2023.
Hong Kong Government (2021). 2020-2021 Budget. Hong Kong.
Hong Kong Health Bureau. (2023). “Domestic Health Accounts.” from https://www.healthbureau.gov.hk/statistics/en/dha/dha_summary_report.htm.
Hong Kong Hospital Authority (2022). Strategic Plan 2022-2027. Hong Kong.
Klazoglou, P. and N. Dritsakis (2018). Modeling and Forecasting of US Health Expenditures Using ARIMA Models, Cham, Springer International Publishing.
Leung, G. M. and Bacon-Stone-J (2006). Health Financing Reform. Hong Kong’ Health System. Hong Kong, Hong Kong University Press: 339-396.
Leung, G. M., K. Y. Tin and W.-S. Chan (2007). “Hong Kong’s health spending projections through 2033.” Health Policy 81(1): 93-101.
Wanless, D. (2002). Securing our future health: taking a long-term view. C. o. t. Exchequer. London.
Zheng, A., Q. Fang, Y. Zhu, C. Jiang, F. Jin and X. Wang (2020). “An application of ARIMA model for predicting total health expenditure in China from 1978-2022.” J Glob Health 10(1): 010803.
APPENDIX 1 Sensitivity analyses results
Appendix Figure 1.1. Trend of HE&GDP over 42-year period from 2000 to 2042 (GDP adjusted model). Blue solid line, trend for HE; Red solid line, trend for GDP. HE, health expenditure; GDP, gross domestic product.
Appendix Figure 1.2. Predicted trends of growth rates for HE&GDP over 20-year period from 2023 to 2042 (GDP adjusted model). Blue dash line, predicted trend for HE growth rates; Red dash line, predicted trend for GDP growth rates. HE, health expenditure; GDP, gross domestic product
Appendix Figure 1.3. Trend of HE&GDP over the 42-year period from 2000 to 2042 (GDP and medical inflation adjusted model). Blue solid line, trend for HE; Red solid line, trend for GDP. HE, health expenditure; GDP, gross domestic product.
Appendix Figure 1.4. Predicted trends of growth rates for HE&GDP over 20-year period from 2023 to 2042 (GDP and medical inflation adjusted model). Blue dash line, predicted trend for HE growth rates; Red dash line, predicted trend for GDP growth rates. HE, health expenditure; GDP, gross domestic product.
Appendix 2 Shares of HE Past Present Future
Appendix Figure 2.1. Share of components of HE in 2000, 2020, and 2040. HE, health expenditure.
Appendix Figure 2.2. Trend of HE components & GDP over 42-year period from 2000 to 2042. Dash lines, trend for HE components; Black solid line, trend for GDP. HE, health expenditure; GDP, gross domestic product. The light blue spike in preventive spending due to COVID-19 pandemic is unlikely to lead to continued large increases in prevention.
Appendix Figure 2.3. Predicted trends of growth rates for HE components & GDP over 20-year period from 2023 to 2042. Dash lines, predicted trend for HE components growth rates; Black solid line, predicted trend for GDP growth rates. HE, health expenditure; GDP, gross domestic product. The upper panel is adjusted just for Population over 60. The lower panel is adjusted for Population over 60 and GDP and medical inflation.
Appendix 3 on Methods
The alternative actuarial approach (not used in this study) breaks down a population by age, sex, utilization, and cost per visit (Wanless 2002, Leung, Tin et al. 2007). The total spending of a population is modeled as the product as follows:
[1] Total Cost = S (Njt) ×(Djkt) × (Mjkt) × (Cijkt ).
Where Njt is the number of people in a group defined by age and sex. Dkjt Number of disease episodes of type k in group j. Mijkt is the number of medical treatments of type i for disease type k in group j. Cijkt is the cost of treatment i at time t. The disadvantage of the method is that it requires dozens of separate forecasts about the future evolution of N, D, M, and C in different ages, diseases, and treatment types. The database of past costs, disease trends, and treatment trends is often limited so analysts have to introduce many assumptions that multiply the uncertainty around the forecasts.
In the ARIMA approach that we adopted the past trends in overall spending are assumed to aggregate all of the information about past trends in morbidity, utilization, and inflation.
Statistical method for model fitting For the base case analysis, the model evaluation included the yearly trend and subsequent considerations for population aging. In the sensitivity analysis, additional predictors, such as GDP and the combination of GDP and medical inflation, were incorporated. Line plots were employed to visually represent the temporal evolution of growth rates for HE and GDP from 2023 onwards and the normalized HE and GDP over the 42-year period from 2000 to 2042. We used augmented Dickey-Fuller tests to assess the stationarity of total and sub-components of HE. The best fit number of autoregressive (AR) and moving-average (MA) lags were chosen in each ARIMA model according to the reported Akaike’s information criterion (AIC). The equation for our ARIMA model is outlined below,
[2]
Where is the health expenditure at time
,
t is the error term of the health expenditure time series,
are the autoregressive parameters for the lag terms, and
are the moving average parameters for the lagged error terms,
are the coefficients for the independent variables including time, population over 60, GDP, and medical inflation, respectively.
Data The dataset to train the model included past information from 2000 to 2022 on the growth of the total population(Hong Kong Census and Statistics Department 2023), the population over 60(Hong Kong Census and Statistics Department 2023), gross domestic product (GDP) at current market prices(Hong Kong Government 2021), and total health expenditure (HE) in constant 2019 Hong Kong(Hong Kong Health Bureau 2023). We also applied the forecasting technique to sub-components of total health expenditure: inpatient, outpatient, day curative care, medical goods, and long-term care. In each case we determined the best fit ARIMA model, fit the model to population share over 60, and current GDP then forecast the future based on projections of population over 60, and GDP. For robustness we also assessed models using past measures of Hong Kong’s medical inflation. To project past trends into the future we used a quadratic interpolation of Hong Kong census bureau’s predictions for population over 60 from 2023 to 2042. We also tested the sensitivity of the model to include a simple forecast of future GDP based on extending the last 23 year’s average 2.41% annual growth rate into the next two decades.
[1] An actuarial approach (not used in this study) disaggregates health spending by age, sex, disease, inpatient/outpatient episodes per disease, and cost per episode (Wanless 2002, Leung, Tin et al. 2007). The actuarial approach uses multiple hidden assumptions about the future time course of medical prices, disease rates, hospitalization rates, and population size.
[2] Spending patterns are called “autoregressive” when they depend significantly on past spending patterns from 1, 2, 3… years earlier. This happens because of multi-year labor contracts and capital financing. They are said to have “moving averages” when shocks in a given year influence spending in subsequent years e.g. due to sudden enduring changes in epidemics like tobacco-related disease, HIV or hepatitis.
[3] The Census estimate predicts net increases of 106,000, 104,600, 89,700, and 68,300 domestic households of 2.6 to 2.7 people between 2021-26, 26-31, 31-36, and 36-41. (Hong Kong Census and Statistics Department 2023)
[4] Hong Kong’s private insurance market has grown its subscriber base without taking on a larger share of health spending (See Figure 1)
