Medical costs and out-of-pocket expenditures associated with multimorbidity in China: quantile regression analysis.

OBJECTIVE: Multimorbidity is a growing challenge in low-income and middle-income countries. This study investigates the effects of multimorbidity on annual medical costs and the out-of-pocket expenditures (OOPEs) along the cost distribution.
METHODS: Data from the nationally representative China Health and Retirement Longitudinal Study (CHARLS 2015), including 10 592 participants aged ≥45 years and 15 physical and mental chronic diseases, were used for this nationally representative cross-sectional study. Quantile multivariable regressions were employed to understand variations in the association of chronic disease multimorbidity with medical cost and OOPE.
RESULTS: Overall, 69.5% of middle-aged and elderly Chinese had multimorbidity in 2015. Increased number of chronic diseases was significantly associated with greater health expenditures across every cost quantile groups. The effect of chronic diseases on total medical cost was found to be larger among the upper tail than those in the lower tail of the cost distributions (coefficients 12, 95% CI 6 to 17 for 10th percentile; coefficients 296, 95% CI 71 to 522 for 90th percentile). Annual OOPE also increased with chronic diseases from the 10th percentile to the 90th percentile. Multimorbidity had larger effects on OOPE and was more pronounced at the upper tail of the health expenditure distribution (regression coefficients of 8 and 84 at the 10th percentile and 75th percentile, respectively).
CONCLUSION: Multimorbidity is associated with escalating healthcare costs in China. Further research is required to understand the impact of multimorbidity across different population groups. © Author(s) (or their employer(s)) 2021. Re-use permitted under CC BY-NC. No commercial re-use. See rights and permissions. Published by BMJ.

The prevalence of chronic disease multimorbidity is common in both developed and developing countries.

People with multimorbidity have higher health service use compared with persons with a single disease.

Chronic disease multimorbidity has potential impacts on individuals and households.

The prevalence of chronic disease multimorbidity is high among Chinese adults.

Increased number of chronic diseases is significantly associated with greater medical expenditures.

Multimorbidity has larger effects on out-of-pocket expenditures and is more pronounced at the upper tail of the health expenditure distribution.

More focus is needed for primary and secondary prevention of multimorbidity.

There is a need to identify cost-effective strategies and payment systems for management of multimorbidity.

Development of specific guidelines for multimorbidity and its risk factors is essential.

Introduction

Non-communicable diseases (NCDs) have been a leading cause of morbidity and mortality.1 The burden of multimorbidity, defined as two or more coexisting NCDs in one person, is rising rapidly in the low-income and middle-income countries (LMICs), primarily due to increased longevity and increased exposure to risk factors.2–4 A study in Scotland found a profound difference in the proportion of multimorbidity in the study population, where 59% of those aged 65–74 years had multimorbidity, compared with 8.1% of those aged between 25 and 34 years.5 Similarly, a recent study in China, based on 11 physical NCDs, found that the prevalence of multimorbidity increased from 51% for those aged 50–54 years to 71% for those aged 75 years and above.6 Patients with multimorbidity are costly to healthcare systems due to the complex needs and greater health service utilisation.7 8 Evidence from high-income countries (HICs) suggest that the economic consequences of a health condition may differ significantly among the highest users of health services, compared with those at the bottom or middle of treatment cost distributions.8 However, there is no current systematic review of the association between total medical costs and multimorbidity. It has public health and policy implications to identify significant gaps in the existing evidence about healthcare delivery strategies for multimorbidity, economic burdens and financing systems in both HICs and LMICs.

By 2011, China reached near-universal health insurance coverage by establishing a social health insurance system. However, levels of insurance coverage and service benefit vary significantly across social health insurance schemes and locations, and patients were required to pay a substantial amount of user fees for their medical treatment costs. Recent statistics by the WHO suggests that private expenditures constitute approximately 40% of the total health expenditure in China, a level considerably higher than those of HICs with levels around 25%.8 The heavy reliance of out-of-pocket expenditure (OOPE) to fund healthcare systems for chronic conditions is concerning as evidence suggests that user fees can have negative effects on health outcomes among patients with multimorbidity. User fees in the presence of rising levels of multimorbidity can be especially harmful for the poorest population in China as they likely have higher underlying risks for several NCDs,2 6 while also having fewer financial resources to pay for their healthcare expenses.

Similar to many LMICs and neighbouring countries in Asia, China’s healthcare delivery remains fragmented and hospital centred, with limited coordination between the different levels of healthcare providers within the health system.9 Strong primary care consisting of multidisciplinary teams and lead by a general practitioner is pertinent to improve multimorbidity prevention and treatment.10 11 Evidence from HICs suggested that people-centred integrated care for multimorbidity patients can sometimes be cost-effective. It is also worth noting that many LMICs may face the double burden of NCDs and infectious diseases. Therefore, the health service delivery model for multimorbidity care in LMICs also needs to pay attention to the management of NCDs with infectious disease. Rigorous evaluation of these new healthcare delivery models is warranted to ensure effectiveness, efficiency and quality of care.

Despite the increasing burden of multimorbidity in China, there has been minimal research investigating the effect of multimorbidity on economic outcomes of individuals, households and whole health systems.12 13 Of these studies, multimorbidity was found to be associated with higher healthcare utilisation and OOPE.12 13 Previous studies that estimate the average effect of the multimorbidity ignore the fact that the effect of multimorbidity might be very different in those with high medical costs than those have low medical costs.12–15 Unlike traditional regression methods, such as the ordinary least squares (OLS) regression or the generalised linear model that focus on population average/mean effects, quantile regression models look at the effect of multimorbidity on healthcare costs across the outcome distribution. The quantile regression analyses provide more detail information and a deep understanding of the financial burden of multimorbidity.16 Alternative estimation strategies using quantile regression analysis has been increasingly adopted in health systems research to investigate the associations between outcomes of interest and the explanatory variables across the distribution of a given dependent variable.16 Based on a nationally representative survey data, we present the first study that investigates the impact of multimorbidity on annual medical costs and OOPE across cost distributions.

Methods

Data sources

We used the 2015 China Health and Retirement Longitudinal Study (CHARLS) dataset, which is a nationally representative survey that collects high-quality data from respondents aged 45 years and older. The CHARLS was conducted using multistage stratified probability-proportionate-to-size sampling with ongoing follow-up surveys conducted every 2 years. The baseline sample size was 17 708 individuals. The details of the objectives, design and methods of the CHARLS are available elsewhere.17 For this study, we identified 13 354 respondents with blood test and biomarker information. After removing respondents with missing values for the dependent or independent variables, the final sample consists of 10 592 respondents. (The flow chart of subjects’ selection can be found in appendix online supplemental figure S1).

Measures

For this study, NCD multimorbidity was defined as the presence of two or more chronic NCDs.18–20 A total of 15 chronic diseases were included in the survey and were used to calculate the number of NCDs for each respondent. The chronic diseases included were hypertension, diabetes and dyslipidaemia, which was measured based on biomarkers or blood test and 10 self-reported chronic diseases (heart disease, stroke, chronic lung disease, kidney disease, digestive diseases, liver disease, cancer, memory-related illness, asthma and arthritis).

For mental illness, measured depression and self-reported diagnosed psychological disease were included in the study. Symptoms of depression were assessed using the 10-item Centre for Epidemiologic Studies Depression Scale (CESD-10),21 which has been validated and is a reliable tool for mental health assessment among China older population.20 The details of CESD-19 have been described elsewhere.22 There were four possible responses for the CESD-10: (1) rarely; (2) some days (1–2 days per week); (3) occasionally (3–4 days per week); and (4) most of the time (5–7 days per week). Respondents’ answers were coded using a range between 0 (rarely) to 3 (most of the time) for the negative question. For two positive questions included in this study, we reversed the coding as 3 (rarely) to 0 (most of the time).23 CESD-10 scoring ranged from 0 to 30. In this study, respondents with a CESD-10 score of at least 10 was defined as having depression symptoms. We also constructed a binary variable for depression.

In the CHARLS, respondents’ systolic blood pressure (SBP) and diastolic blood pressure (DBP) were measured three times by a trained nurse using the HEM-7112 electronic monitor. Hypertension diagnosis was defined as SBP ≥140 mm Hg and/or DBP ≥90 mm Hg and/or receiving medication for hypertension.24 25 Diabetes diagnosis was defined as having one or combination of the following: (1) fasting blood plasma glucose level of ≥126 mg/dL; (2) Glycated hemoglobin (HbA1c) concentration of ≥6.5%; and (3) receiving insulin treatment and/or medication for high blood sugar level.26 27 Dyslipidaemia was defined based on: (1) total cholesterol ≥240 mg/dL; and/or (2) low-density lipoprotein cholesterol ≥ 160 mg/dL; and/or (3) high-density lipoprotein cholesterol <40 mg/dL; and/or (4) triglyceride ≥ 200 mg/dL; and/or (5) receiving antidyslipidaemia medication.28 29

Our primary outcome variables include the total treatment costs (defined as total annual health expenditure) and the annual OOPE, defined as the total direct payments for outpatient and inpatient visits occurring 1 year prior to the survey, after reimbursement from health insurance. As the total expenditure and OOPE for outpatient care was measured for 1 month at a time, we multiplied the costs by 12 months to obtain the annual outpatient care costs.2 30

This study also included the following covariates in the regression analyses: age, gender, marital status (married and partnered, unmarried and others), education attainment (illiterate, primary school, secondary school, college and above), place of residency (rural and urban), region of residency (east, central and west), household economic status quartiles (annual per capita household consumption expenditure) and social health insurance coverage (yes or no).

Statistical analysis

We summarised the mean of total treatment cost and annual OOPE by the number of chronic diseases. The number of NCDs was considered as the independent variable in the regression models. For individuals with positive total health expenditure and OOPE (expenditures >0 US$), we applied linear regression models to assess the overall effects of NCDs and quantile regression analysis to estimate the effect of multimorbidity on health expenditures at the 10th, 25th, 50th, 75th and 90th percentiles. The quantile regression is similar to OLS regression, that is, estimating the median or other quantiles for the outcome variable associated with a set of independent variables and covariates without assuming normality or homoscedasticity of the underlying distribution.31–33 The quantile regression is robust to outliers as it allows for assessing the full distribution of the outcome variable and is suitable for modelling outcomes that are not normally distributed or are highly skewed.16 34

We performed the Shapiro-Wilk W test for total medical cost and OOPE in our sample, and the results illustrated the skewed nature of both OOPE and total medical spending (p<0.05). It is becoming more important to understand the potential differential effect of multimorbidity across population group; therefore, this method has recently been widely used in health service research and policy evaluation, including the following studies that assessed costs outcomes (including OOPE or total medical costs).20 21

The coefficients at lower percentiles (10th and 25th percentiles) present the association of multimorbidity with total treatment cost and OOPE in those individuals with low health expenditures, while upper percentiles (75th and 90th percentiles) reflect the association on those with higher health expenditures. The statistical analyses in this study were conducted using Stata software V.16.0. P values <0.05 were considered statistically significant.

Results

We analysed data from 10 592 respondents. The median age of participants was 61.0 years (IQR 53.5–67.0) in 2015. There was a slightly higher percentage of female (53.1%) than male respondents. Most of the respondents were married (86.7%) and resided in rural areas (64.1%). Only 32.3% of the respondents had attained a level of education higher than primary school, and 91.5% of the respondents were enrolled in social health insurance in China. The overall prevalence of multimorbidity was 69.5%, with a high proportion in the older populations and unmarried individuals (table 1).

Table 1

Characteristics of participants and prevalence of chronic disease*

Characteristic	Overall		No condition		One condition		Multimorbidity (n=7369)
	(n=10 592)		(n=947)		(n=2276)
	N	%	%	95% CI	%	95% CI	%	95% CI
Total	10 592	100	9	8.4 to 9.7	21.5	20.5 to 22.5	69.5	68.3 to 70.6
Age (year)
45–54	2869	27.1	11.7	10.4 to 13.2	30.1	27.9 to 32.5	58.2	55.7 to 60.6
55–64	4036	38.1	9.8		19.2	17.7 to 20.7	71	69.3 to 72.8
65–74	2727	25.8	6.1	5.1 to 7.2	17.4	15.9 to 19.1	76.5	74.6 to 78.3
75 and above	960	9.1	6.5		16.9	14.2 to 19.8	76.6	73.2 to 79.7
Gender
Male	4963	46.9	9	8.2 to 10	22.6	21.1 to 24.2	68.4	66.6 to 70
Female	5629	53.1	9.1	8.2 to 10	20.5	19.3 to 21.8	70.4	69 to 71.9
Marital status
Married and partnered	9188	86.7	9.5	8.8 to 10.3	21.9	20.8 to 23	68.6	67.4 to 69.8
Unmarried and other	1404	13.3	6	4.7 to 7.7	19	16.7 to 21.5	75	72.2 to 77.5
Education status
Illiterate	4766	45	7.5	6.6 to 8.4	19.2	18 to 20.5	73.3	71.8 to 74.7
Primary school	2404	22.7	7.7	6.5 to 9.2	21.1	19.3 to 23.1	71.2	68.9 to 73.3
Secondary school	2299	21.7	11.6	10.1 to 13.3	23.7	21.6 to 26	64.7	62.1 to 67.2
College and above	1123	10.6	12	10 to 14.5	25.7	21.9 to 29.9	62.3	57.9 to 66.5
Residence place
Urban	3799	35.9	9.9	8.7 to 11.2	21.6	19.8 to 23.5	68.5	66.4 to 70.6
Rural	6793	64.1	8.4	7.7 to 9.2	21.4	20.4 to 22.5	70.2	69 to 71.3
Region
East	4013	37.9	10.5	9.4 to 11.8	23.2	21.4 to 25.1	66.3	64.2 to 68.4
Central	4029	38	8.5	7.5 to 9.6	20.1	18.7 to 21.5	71.4	69.8 to 73
West	2550	24.1	7.4	6.3 to 8.5	20.8	19 to 22.6	71.9	69.9 to 73.8
PCE, quartile
Q1, the lowest	2648	25	8.1	6.9 to 9.4	19.5	17.8 to 21.3	72.4	70.3 to 74.4
Q2	2648	25	9.3	8 to 10.6	23.6	21.4 to 25.9	67.2	64.8 to 69.4
Q3	2649	25	9.6	8.4 to 11	21.2	19.5 to 23	69.2	67.1 to 71.2
Q4 (the highest)	2647	25	9.2	7.9 to 10.7	21.6	19.6 to 23.7	69.3	66.8 to 71.6
Social health insurance
No	902	8.5	8.9	6.9 to 11.4	24.5	20.9 to 28.5	66.6	62.5 to 70.4
Yes	9690	91.5	9.1	8.4 to 9.8	21.2	20.2 to 22.2	69.7	68.6 to 70.9

The values are weighted percentages unless otherwise indicated.

PCE, per capita household consumption expenditure.

Table 2 shows the proportion and total treatment cost and OOPE across the number of chronic diseases. Of the total participants, 28.5% and 27.4% of total participants experienced an occurrence of total treatment cost and OOPE during the last year. Overall, the mean total treatment cost and annual OOPE was US$740 and US$474 in our sample, respectively. Having multimorbidity was associated with a higher frequency of OOPE (two chronic diseases: 23.9%; three conditions: 27.7%; four or more conditions 40.6%) compared with having no NCDs or only one chronic disease (17.7% and 18.0%, respectively). Having four chronic diseases or above was associated with a substantially greater OOPE of US$717 compared with US$236 for those without any chronic diseases. Similarly, for annual medical cost, individuals with more chronic diseases had greater healthcare costs than those with a single disease.

Table 2

Treatment costs and out-of-pocket health expenditure by the number of chronic conditions

Variable	Any total treatment cost (%)		Total annual treatment cost (US$)		Any OOPE (%)		Total annual OOPE (US$)
	Proportion	95% CI	Mean	95% CI	Proportion	95% CI	Mean	95% CI
Total participants	28.5	27.3 to 29.6	740	658 to 822	27.4	26.3 to 28.6	474	413 to 534
No conditions	18	15.2 to 20.8	387	158 to 616	17.7	15 to 20.5	236	76 to 396
One condition	19.1	17.2 to 21	497	342 to 652	18	16.2 to 19.8	312	220 to 405
Two conditions	24.8	22.1 to 27.6	635	463 to 807	23.9	21.1 to 26.6	417	289 to 545
Three conditions	28.4	25.9 to 30.9	735	517 to 954	27.7	25.2 to 30.1	482	300 to 663
Four conditions and above	42.1	39.8 to 44.4	1131	974 to 1289	40.6	38.3 to 42.9	717	599 to 835

The values are weighted percentages and means unless otherwise indicated.

OOPE, out-of-pocket expenditure.

Results of quantile regression analysis

Table 3 presents the effects of the number of chronic diseases on total treatment cost by the treatment cost quantile. An increase in the number of chronic diseases was significantly associated with greater health expenditure across every quantile group. The effect of chronic diseases on total treatment cost was found to be larger among the upper tail than those in the lower tail of cost distributions (coefficients 12, 95% CI 6 to 17 for 10th percentile; coefficients 296, 95% CI 71 to 522 for 90th percentile). Table 4 shows that annual OOPE increased with chronic diseases from the 10th percentile to the 90th percentile. Multimorbidity had larger effects on OOPE and more pronounced at the upper tail of the outcome distribution (coefficients of 20 at the 25th percentile and 84 at the 75th percentile, respectively). Regarding sociodemographic covariates, the results only showed that household wealth has significant positive associations with health total medical cost and OOPE.

Table 3

Quantile regression on total treatment cost associated with multimorbidity (total treatment cost >0, n=3021)

Variable	Overall		10th percentile		25th percentile		50th percentile		75th percentile		90th percentile
	Coeff	95% CI	Coeff	95% CI	Coeff	95% CI	Coeff	95% CI	Coeff	95% CI	Coeff	95% CI
No. of NCD	70	42 to 97	12	six to 17	35	23 to 47	70	41 to 99	144	70 to 217	296	71 to 522
Age (years) (45–54)
55–64	25	−88 to 137	24	−1 to 49	52	−4 to 107	25	−98 to 147	90	−247 to 426	653	−529 to 1835
65–74	70	−66 to 206	23	−4 to 51	11	−49 to 72	70	−64 to 205	176	−257 to 610	539	−741 to 1819
75 and above	179	−23 to 380	56	13 to 99	67	−23 to 157	179	−30 to 388	478	−143 to 1099	1112	−782 to 3005
Gender (male)	−35	−146 to 76	24	3 to 46	−6	−52 to 40	−35	−138 to 69	94	−201 to 390	−223	−1176 to 730
Marital status (married)	−133	−247 to −19	−16	−49 to 17	−37	−92 to 18	−133	−243 to −23	−236	−590 to 118	−261	−1594 to 1073
Education status (illiterate)
Primary school	−23	−145 to 100	17	−5 to 40	4	−48 to 56	−23	−152 to 106	368	−46 to 783	494	−556 to 1544
Secondary school	35	−115 to 185	−2	−31 to 27	7	−50 to 63	35	−113 to 182	328	−69 to 725	1022	−534 to 2579
College and above	121	−72 to 313	22	−20 to 64	91	−24 to 206	121	−91 to 333	428	−178 to 1035	914	−1021 to 2848
Residence place (urban)	−101	−215 to 12	−20	−43 to 3	−61	−104 to −17	−101	−206 to 4	−293	−674 to 88	−407	−1436 to 622
Region (east)
Central	−136	−258 to −14	13	−13 to 40	−4	−57 to 48	−136	−261 to −11	−218	−595 to 158	−310	−1592 to 971
West	−247	−363 to −132	−3	−28 to 23	−51	−102 to 0	−247	−378 to −117	−466	−809 to −122	−953	−2142 to 237
Social health insurance (no)	−158	−385 to 70	3	−39 to 45	−3	−91 to 84	−158	−376 to 61	−611	−1511 to 289	−2016	−4651 to 618
PCE, quartile (Q1, the lowest)
Q2	−14	−115 to 88	18	−4 to 39	25	−20 to 70	−14	−113 to 85	19	−289 to 327	−211	−1161 to 740
Q3	185	71 to 300	46	15 to 76	89	32 to 146	185	65 to 305	572	154 to 990	967	−308 to 2242
Q4 (the highest)	499	312 to 687	61	34 to 88	167	106 to 227	499	311 to 687	1800	1234 to 2366	5303	3765 to 6841

Coefficients estimated after adjusting for study variables, including age, gender, marital status, level of education, residence place, region, household economic level and health insurance status.

NCD, non-communicable disease; PCE, per capita household consumption expenditure.

Table 4

Quantile regression on OOPE associated with multimorbidity (OOPE >0, n=2926)

Variable	Overall		10th percentile		25th percentile		50th percentile		75th percentile		90th percentile
	Coeff	95% CI	Coeff	95% CI	Coeff	95% CI	Coeff	95% CI	Coeff	95% CI	Coeff	95% CI
No. of NCD	45	28 to 62	8	5 to 11	20	13 to 28	45	29 to 61	84	36 to 133	156	0 to 312
Age (years) (45–54)
55–64	−31	−122 to 60	8	−8 to 25	18	−21 to 58	−31	−124 to 61	−149	−434 to 135	173	−723 to 1070
65–74	−72	−172 to 29	2	−15 to 18	−36	−78 to 6	−72	−173 to 30	−74	−392 to 244	193	−580 to 965
75 and above	−121	−235 to −6	15	−7 to 37	−33	−88 to 22	−121	−246 to 5	−12	−381 to 357	138	−1164 to 1440
Gender (male)	8	−62 to 78	7	−6 to 21	8	−24 to 40	8	−65 to 81	113	−96 to 322	520	−153 to 1193
Marital status (married)	−82	−154 to −9	−4	−21 to 13	−33	−68 to 1	−82	−161 to −3	−247	−514 to 20	−135	−1463 to 1193
Education status (illiterate)
Primary school	−13	−98 to 73	11	−4 to 26	−1	−38 to 36	−13	−98 to 72	124	−127 to 376	330	−409 to 1069
Secondary school	−20	−118 to 78	5	−11 to 21	−27	−68 to 15	−20	−115 to 75	−6	−326 to 315	473	−658 to 1604
College and above	−19	−138 to 101	10	−17 to 37	−4	−64 to 56	−19	−143 to 106	325	−334 to 984	670	−453 to 1793
Residence place (urban)	−6	−71 to 60	−13	−25 to 0	−21	−55 to 13	−6	−71 to 60	70	−151 to 290	126	−620 to 873
Region (east)
Central	−15	−101 to 70	−2	−18 to 15	8	−28 to 44	−15	−103 to 72	−84	−327 to 158	−433	−1186 to 321
West	−136	−214 to −59	−5	−19 to 9	−39	−79 to 2	−136	−216 to −57	−300	−549 to −50	−647	−1412 to 118
Social health insurance (no)	−122	−274 to 31	−8	−38 to 23	−65	−135 to 4	−122	−288 to 45	−697	−1363 to −31	−2349	−3985 to −713
PCE, quartile (Q1, the lowest)
Q2	−21	−90 to 48	9	−5 to 23	4	−29 to 37	−21	−91 to 49	115	−112 to 342	293	−453 to 1039
Q3	166	85 to 248	11	−6 to 28	57	11 to 102	166	84 to 248	431	188 to 674	672	−244 to 1588
Q4 (the highest)	294	192 to 396	43	23 to 64	112	65 to 159	294	195 to 394	1201	812 to 1590	3464	2106 to 4823

Coefficients estimated after adjusting for study variables, including age, gender, marital status, level of education, residence place, region, household economic level and health insurance status.

NCD, non-communicable disease; OOPE, out-of-pocket expenditure; PCE, per capita household consumption expenditure.

The result of quantile regression analysis among those incurring any healthcare costs suggests that an increased number of chronic diseases was significantly associated with greater total medical cost and OOPE at higher health expenditure quantiles compared with lower health expenditure quantiles (figure 1A, B). The total medical cost and OOPE attributable to multimorbidity over the distribution of these costs are also depicted in online supplemental figure S2a, b. Using the binary variable of multimorbidity, the total treatment cost and OOPE associated with multimorbidity were significant in almost in all quantiles, except for the 90th quantile of OOPE. Variations in OOPE associated with multimorbidity became more pronounced as expenditure approached the upper percentiles.

Figure 1

Distribution of total treatment cost (above) and annual OOPE (below) associated with the number of chronic conditions by medical cost quantiles. OOPE, out-of-pocket expenditure.

Discussion

Principal findings

Our results showed that the number of NCD is associated with greater annual medical costs and OOPE. Importantly, our results suggest that the effect of multimorbidity on healthcare spending and OOPE was not constant across the quantiles of health expenditures. The quantile regression method adopted in our study accounts for the important association of number of the NCD with health expenditures that differs along the costs distribution. As a whole, our study identified important heterogeneity among key multimorbidity patient groups at different levels of medical costs and OOPE. Our results reveal a more nuanced measure of the association between NCDs and healthcare costs, and OOPE. These estimates can be used to forecast the future burden of multimorbidity and further evaluate the cost-effectiveness of prevention strategies for multimorbidity across population groups. Our results highlight the need for stronger and clearer analysis on the economic impact of multimorbidity in cost-effectiveness analysis on multimorbidity treatment and prevention strategies.

Compare with existing literature

In this study, the prevalence of multimorbidity (69.5%) was much higher than previous research in China.35 36 Using the 2015 CHARLS, a study by Zhang et al35 found that NCD multimorbidity was only present in 43.6% of people aged ≥60 years. The difference in the prevalence could be due to the inclusion of objective biomarker and scale measurements for chronic diseases (hypertension, diabetes, dyslipidaemia and depression) in this study, which may have led to an increased diagnosis of multimorbidity compared with the results from the self-report chronic diseases.

The higher treatment cost and financial burden for individuals with multimorbidity are in line with earlier published studies.6 13 37 Chen et al13 found that the expenditure on patients with multiple chronic diseases was over three times higher than those with a single disease. This positive relationship between the number of NCDs and OOPE was also consistent with earlier studies.38 39 Based on an earlier study in China, the mean OOPE for outpatient care was around 1.5 times greater for those with multimorbidity compared with those without any NCDs.38 Similar results have been documented in earlier studies in both LMICs (eg, India and Mexico) and HICs.39–42 The presence of mental illness is associated with greater health service use and increased OOPE.43–45 A study by Hsieh and Qin44 showed that people with depression were more than 10% more likely to use health services in a year. The economic burden attributable to depressive symptoms and depression accounted for 142 Renminbi (RMB) and 126 RMB per person/year.

Our findings on the higher treatment costs and financial burden may be explained by more prevalent polypharmacy among those with chronic conditions, leading to higher medicine expenditures. The challenges in polypharmacy may be due to the application of single-disease guidelines on patients with multimorbidity, which were designed based on frameworks that exclude multimorbidity conditions.38 46 Some evidence also suggested that certain combinations of chronic diseases yielded higher medicine expenditure due to certain diseases requiring treatment using more expensive drugs.7 Multimorbidity patients with higher OOPE on medicines had to allocate fewer resources to other healthcare services, and thus, it is more likely to increase inequality in healthcare and patient outcomes at the health system level when the social health insurance schemes for people with multimorbidity have not been well established.

Policy implications

This study presents novel evidence on the increasing financial burden due to multimorbidity among the Chinese population. As multimorbidity is costly to individuals and the health systems, there is a need for concerted efforts and other LMICs with ageing populations to reduce the treatment burden among patients with multimorbidity and its impact on financial risk protection.6 Prevention strategies for NCDs should adopt approaches that take into account multimorbidity in healthcare management, rather than single-disease approaches, particularly in funds allocation and in designing policies on financial protection.7 Furthermore, preventing multimorbidity and developing more sustainable models of care is a key priority. Instead of focusing on the prevention and treatment of a single disease, NCD care delivery models should account for patients with multiple comorbid health conditions.47 Health service delivery models should also adapt to more personalised and structured care, with an emphasis on care cocoordination across multidisciplinary teams alongside with better patient education and self-management. There has been some evidence showing healthcare delivery supported by digital technology and information system can improve effectiveness in managing patient with chronic conditions.48 49 These insights can be used to guide China’ policymaking in NCD prevention and management.

In the context of social health insurance reforms, emphasis should be placed on strengthening the financial risk protection function by reducing OOPE among patients with multimorbidity.6 Furthermore, policy measures could introduce a special exemption from certain costs for vulnerable populations (eg, the elderly and poor population groups) including lower copayments and vital drugs subsidisation.9 Prescription drug cost sharing benefit plans and the National Basic Public Health programme should be geared towards providing broader coverage for multimorbidity, particularly for disease combinations that potentially yield higher OOPE and cause catastrophic health expenditure. In terms of clinical implications, to ensure treatment adherence and to avoid patients foregoing treatment due to high OOPE, medical practitioners should consider the risks of financial burden among patients with multimorbidity particularly relating to polypharmacy.7

Strengths and limitations

To our knowledge, this is the first Chinese study that investigated the effect of multimorbidity on treatment cost and OOPE, using nationally representative survey data. Furthermore, the application of quantile regression models allowed us to assess the variations in the impact of multimorbidity across the distribution of healthcare costs and OOPE. This approach provides clearer and more detailed information on the economic consequences of multimorbidity compared with previous studies that used basic regression models such as OLS, which only assess the mean costs under the assumptions that costs are normally distributed and that the impact of multimorbidity on utilisation and OOPE would be similar across the outcome distributions when it is often not. Furthermore, our findings also show that by using quantile regression, we were able to explore further the differences in healthcare cost and OOPE by considering the variations in the level of the multimorbidity. Finally, while quantile regression is still infrequently used in public health studies, this econometric analysis offers broader applicability in studies of healthcare utilisation and associated costs.

Our study has several limitations. First, undiagnosed and untreated chronic diseases might lead to an underestimation of medical costs due to self-reported measures for some of the chronic conditions. Second, there is a limited number of the types of chronic diseases included in the CHARLS questionnaire. Third, we did not account for the different types of combinations of diseases the construction of multimorbidity variable and due to data limitation of diseases’ severity. Fourth, because outpatient care expenditures were collected for last month in CHARLS, this study estimated the total annual health expenditure and annual OOPE by adding up the inpatient care expenditure during last year and the outpatient care expenditure multiplying 12 months. This could underestimate the total spending on outpatient services among patients with chronic diseases because some patients may have no outpatient visit within 4 weeks before the survey.

Conclusion

The effect of multimorbidity on total medical cost and OOPE increases gradually when approaching the higher percentiles of the health expenditure distribution. Policymakers must recognise the need for better equity and reducing economic burdens. To deliver more cost-effective and better care for multimorbidity patients, preventing multimorbidity and developing people-centred integrated care models is a key priority. Social health insurance reforms should emphasise reducing OOPE among patients with multimorbidity by providing more effective methods of financing and service delivery.