Health Services Accessibility and Self-Reported Health Among Chinese Middle-Aged and Older Adults: Propensity Score Matching and Mixed-Effects Ordinal Logit Analysis

Abstract

This study evaluates the impact of unmet healthcare needs on self-reported health among middle-aged and older Chinese adults (≥45 years), with a focus on urban-rural disparities. Using 2011, 2013, and 2015 CHARLS data, we applied propensity score matching and mixed-effects ordinal logit models, with interaction effects regression for urban-rural differences. We analyzed the impact of unmet healthcare needs—defined as instances where individuals perceived a need for medical care but did not receive it—on self-reported health, which was measured using a 5-point Likert scale. Unmet outpatient care alone did not significantly affect self-reported health but had a negative impact when financial barriers were present (AOR = 0.44, 95% CI = [0.24, 0.82]). Unmet inpatient care significantly decreased self-reported health (AOR = 0.65, 95% CI = [0.57, 0.74]), with financial barriers worsening the effect (AOR = 0.25, 95% CI = [0.15, 0.42]). The negative impact of unmet inpatient care was significant only for rural residents (AOR = 0.67, 95% CI = [0.50, 0.89]). Unmet healthcare needs, particularly due to financial barriers, significantly harm self-reported health, with rural populations more affected, highlighting the need for targeted reforms in both financial protection and healthcare system delivery.

Keywords

health services accessibility self-reported health financial barriers urban-rural health disparities propensity score matching

Background

Access to healthcare services is crucial for enhancing public health outcomes, mitigating disparities in healthcare access, and advancing the agenda of Universal Health Coverage.¹ Although China has made significant progress toward achieving universal health coverage through social health insurance reform,² Unmet Healthcare Needs (UHN), defined as instances where an individual does not receive an available and effective treatment that could have improved their health,³ persist^4-7and represent an ongoing challenge for health equity and system performance.

Research utilizing data from the Chinese Longitudinal Healthy Longevity Survey (CLHLS) and the China Health and Retirement Longitudinal Study (CHARLS) has consistently highlighted the negative effects of poor healthcare accessibility on various health outcomes, including self-reported health, functional limitations in activities of daily living, cognition decline, depression symptoms, all-cause mortality and frailty.^5,6,8-10 These findings indicate that UHN is a multifaceted issue, with potential implications for population health, particularly among middle-aged and older adults in China.

Among these diverse health outcomes, Self-Reported Health (SRH) stands out as a particularly comprehensive and predictive measure—not only because of its strong association with mortality and other health conditions,¹¹ but also due to its practicality and inclusiveness in health research. It is widely regarded as one of the most feasible, inclusive, and informative measures of health status.¹² Moreover, SRH has been extensively used in research examining the relationship between UHN and health outcomes, providing valuable insights into how healthcare access influences perceived health status across different population groups.^13-15

Existing studies have revealed the complexity of the health impacts of UHN, demonstrating that different types of UHN can affect various populations in distinct ways. For instance, inadequate access to healthcare was found to be significantly higher among older adults in rural areas compared to their urban counterparts in China, indicating disparities based on geographical location.⁹ Similarly, Wu et al reported that the adverse effects of UHN vary across different populations among older adults.¹⁵ A related study in South Korea found that the relationship between UHN and health outcomes varied depending on the reasons for the unmet needs and the age group of the individuals.¹³

Despite these insights, most studies on UHN in China focus on its general impact,^5,6,8 often without distinguishing between types of UHN, such as Unmet Outpatient Healthcare Needs (UOHN) versus Unmet Inpatient Healthcare Needs (UIHN), or between financial and non-financial barriers. However, China’s social health insurance system is characterized by a distinct urban–rural dual structure. Urban employees and rural residents are enrolled in different insurance schemes and are subject to significantly different reimbursement policies.¹⁶ These policies also vary between outpatient and inpatient services within the same scheme.¹⁷ This structural complexity underscores the need to differentiate types and causes of UHN, as well as population characteristics, to better understand how unmet needs translate into diverse health outcomes and to inform more targeted policy responses.

Study Hypothesis

Building on the findings of prior research, we now outline the specific hypotheses that guide our study.

Hypothesis 1: We hypothesize that UIHN will have a significant negative association with SRH among middle-aged and older adults in China, whereas UOHN will have a comparatively weaker or non-significant association. This hypothesis is informed by prior research indicating that inpatient and outpatient services differ in their health implications¹⁸ and that unmet inpatient needs are more closely linked to severe health deterioration¹⁵

Hypothesis 2: We hypothesize that UHN due to financial barriers—whether UOHN or UIHN—will have a significant negative impact on SRH among middle-aged and older adults in China. Financial constraints remain a major contributor to UHN in this population group.⁷ While international studies have demonstrated that cost-related barriers to healthcare are associated with poorer self-reported health, particularly among older and economically disadvantaged populations—as seen in South Korea¹³—this raises the question of whether such associations also hold in the Chinese context, where the health insurance system and population aging present distinct structural characteristics.

Hypothesis 3: We hypothesize that the negative impact of UHN on SRH will be more pronounced among rural residents compared to their urban counterparts. While the structural divide between urban and rural healthcare systems in China is well-recognized,^9,19 less is known about how these systemic differences influence the health consequences of UHN. We therefore expect rural residents to experience disproportionately stronger negative effects of UHN on their self-reported health.

The overarching purpose of this study is to examine the effects of UHN on SRH among middle-aged and older adults in China. Specifically, this study aims to differentiate between the impacts of UOHN and UIHN and assess how financial barriers and urban-rural disparities influence SRH.

Methodology

Data and Population Sample

This study utilized data from the 2011, 2013, and 2015 waves of CHARLS, the first nationally representative longitudinal survey of middle-aged and older adults in China, which provides high-quality data on aging, healthcare utilization, and health. Although data collection continued in later waves (2018 and 2020), those waves do not include variables on unmet healthcare needs (UHN). In contrast, the 2011 to 2015 waves offer comprehensive data on unmet outpatient and inpatient needs (UOHN and UIHN), along with rich demographic, socioeconomic, and lifestyle variables. These waves include large sample sizes—17 708 in 2011, 18 605 in 2013, and 21 095 in 2015^20,21—and remain the most suitable and comprehensive dataset for our analysis, providing valuable insights on the relationship between UHN and SRH among middle-aged and older adults in China.

As shown in Figure 1, several exclusion steps were applied to construct the final analytic samples.

Figure 1.

Flow chart of sample selection procedure.

We began by excluding individuals under age 45. Next, respondents with missing values on covariate variables were removed. For most variables, observations with missing values were excluded from the analysis through listwise deletion. However, for household per capita expenditure—a key socioeconomic control variable with relatively high missingness—values were imputed using the average expenditure level within the same community. This approach minimized sample loss while preserving contextual accuracy.

Individuals who had not reported illness in the past month or had not been advised to be hospitalized in the past year were excluded from the analysis, as they were considered to have no healthcare needs.^10,22 The final sample was then divided into 2 groups: Those ill in the month before the survey (2011: n = 4343; 2013: n = 4874; 2015: n = 4990) and those advised for hospitalization in the year before the survey (2011: n = 2093; 2013: n = 2596; 2015: n = 3,163). Since individuals without healthcare needs were purposefully excluded at this stage, many respondents appeared in only one survey wave. Therefore, the data are treated as repeated cross-sectional rather than panel data.

Variables

Dependent Variables: SRH was assessed using 1 item from the CHARLS questionnaire that asked participants, “Would you say your health is very good, good, fair, poor or very poor?” The response option was a 5-point Likert scale, ranging from 5 (very good) to 1 (very poor).²³

Independent Variables

Two primary independent variables were established: UOHN and UIHN.

Unmet Outpatient Healthcare Need (UOHN) in this study—defined as reporting illness in the past month but not receiving any outpatient care—aligns with the concept of “subjective unmet need” widely used in international health services research. This construct reflects perceived healthcare needs that remain unmet due to barriers such as limited accessibility, high costs, or long wait times.^24,25 UOHN derived from 2 questions in the CHARLS survey: “(1) Have you been ill in the last month? (2) In the last month, have you visited a public hospital, private hospital, public health center, clinic, health worker’s or doctor’s practice, or been visited by a health worker or doctor for outpatient care?” Those who answered “yes” to the first question but “no” to the second were classified as having HOHN (UOHN = 1), while those who did visit were classified as not having UOHN (UOHN = 0).

This study defines Unmet Inpatient Healthcare Need (UIHN) as a situation in which an individual was advised by a physician to be hospitalized within the past year but did not receive inpatient care. This operational definition aligns closely with the concept of “subjective, clinician-validated unmet need” proposed by Allin et al,³ which refers to cases where a person perceives a need, accesses care, but does not receive the treatment that a clinician would judge as appropriate. UIHN was identified based on the survey question: “In the past year, did a doctor suggest that you needed inpatient care but you did not get hospitalized?” Respondents who answered “yes” were classified as having UIHN (UIHN = 1), while those who had been hospitalized as recommended were coded as not having UIHN (UIHN = 0).

Furthermore, 2 additional independent variables—UOHN/UIHN for different reasons—were established. For the UOHN for different reasons variable, when a respondent reported experiencing UOHN and answered “no money” to the question “What’s the main reason for not seeking medical treatment?,” they were categorized as experiencing UOHN due to financial barriers. If the response was “No time,” “Inconvenient traffic,” or other reasons, the respondent was categorized as experiencing UOHN due to non-financial barriers.

For the UIHN for different reasons variable, when a respondent reported experiencing UIHN and answered “Not enough money” to the question “What’s the main reason for not seeking inpatient care?,” they were categorized as experiencing UIHN due to financial barriers. If the response was “No ward available,” “Not willing to go to the hospital,” or other reasons, the respondent was categorized as experiencing UIHN due to non-financial barriers. (UOHN/UIHN for different reasons = 0).²²

Covariate Variables

Based on their well-established associations with health outcomes and consistent inclusion in prior research on UHN and SRH,^9,13,26,27 we included 3 categories of covariates—demographic, socioeconomic, and lifestyle variables—to control for potential confounders.²⁸

Demographic Characteristics: These variables include gender, age, number of living children, educational level, marital status, living arrangement, Hukou, and number of chronic diseases. Age and number of living children were treated as continuous variables, with age scaled in decades (ie, age divided by 10). Educational level was categorized into 4 groups. Marital status was categorized as married (including “Married with Spouse Present,” “Married but Not Living with Spouse Temporarily,” and “Separated”) or single (including “Divorced,” “Widowed,” “Never Married,” and “Cohabiting”). Living arrangement was classified as living alone or living with family members. The “Hukou” variable reflects a registration system that distinguishes between urban and rural residence status. The number of chronic diseases was categorized based on 14 chronic conditions from the CHARLS questionnaire and grouped into 4 categories. More details on variable coding are provided in the descriptive statistics section (Table 1).

Table 1.

Descriptive Characteristics of the Sample by UOHN and UIHN.

Variables	UOHN			UIHN
Variables	UOHN = 0N = 9710	UOHN = 1N = 4497	P-value	UIHN = 0N = 5309	UIHN = 1N = 2543	P-value
SRH (0 = Very poor)	875 (9.0)	436 (9.7)	<.001	562 (10.6)	433 (17.0)	<.001
1 = Poor	3151 (32.5)	1326 (29.5)		1848 (34.8)	986 (38.8)
2 = Fair	4585 (47.2)	2205 (49.0)		2310 (43.5)	937 (36.8)
3 = Good	752 (7.7)	403 (9.0)		403 (7.6)	131 (5.2)
4 = Very good	347 (3.6)	127 (2.8)		186 (3.5)	56 (2.2)
Age	6 (5.3-6.7)	5.9 (5.2-6.7)	.010	6.2 (5.5-7)	6 (5.3-6.7)	<.001
Gender(0 = Female)	3940 (40.6)	1937 (43.1)	.005	2469 (46.5)	1052 (41.4)	<.001
Education level (0 = Illiterate/Semi-literate)	4647 (47.9)	2249 (50.0)	.013	2569 (48.4)	1270 (49.9)	.56
1 = Primary school	2141 (22.0)	1009 (22.4)		1204 (22.7)	556 (21.9)
2 = Middle school	1832 (18.9)	795 (17.7)		980 (18.5)	467 (18.4)
3 = High school and above	1090 (11.2)	444 (9.9)		556 (10.5)	250 (9.8)
Marital status(0 = Single)	8326 (85.7)	3823 (85.0)	.25	4468 (84.2)	2159 (84.9)	.40
Number of children	2 (2-4)	2 (2-3)	.006	3 (2-4)	2 (2-4)	<.001
Living arrangement (0 = Living with family)	497 (5.1)	243 (5.4)	.48	320 (6.0)	143 (5.6)	.48
Hukou(0 = Urban,1=Rural)	7650 (78.8)	3574 (79.5)	.35	3916 (73.8)	1991 (78.3)	<.001
Number of chronic diseases (0 = No chronic disease)	2115 (21.8)	1138 (25.3)	<.001	975 (18.4)	328 (12.9)	<.001
1 = 1 chronic disease	2831 (29.2)	1323 (29.4)		1333 (25.1)	575 (22.6)
2 = 2-3 chronic diseases	3446 (35.5)	1463 (32.5)		2040 (38.4)	1032 (40.6)
3 = 4 or more chronic diseases	1318 (13.6)	573 (12.7)		961 (18.1)	608 (23.9)
Region (0 = West)	3175 (32.7)	1960 (43.6)	<.001	1938 (36.5)	1018 (40.0)	<.001
1 = Middle	3098 (31.9)	1346 (29.9)		1717 (32.3)	869 (34.2)
2 = East	3437 (35.4)	1191 (26.5)		1654 (31.2)	656 (25.8)
Medical insurance (0 = No insurance)	426 (4.4)	263 (5.8)	<.001	224 (4.2)	132 (5.2)	<.001
1 = UEBMI	986 (10.2)	398 (8.9)		686 (12.9)	230 (9.0)
2 = URRBMI /URBMI/NRCMI	7684 (79.1)	3589 (79.8)		4018 (75.7)	2026 (79.7)
3 = other insurance	614 (6.3)	247 (5.5)		381 (7.2)	155 (6.1)
Per capita household expenditure	8.86 (8.28-9.45)	8.80 (8.23-9.39)	<.001	9.04 (8.42-9.61)	8.86 (8.30-9.44)	<.001
Smoking status (0 = not current smoker)	2726 (28.1)	1530 (34.0)	<.001	1384 (26.1)	726 (28.5)	.020
Drinking status (0 = not current drinker)	2225 (22.9)	1246 (27.7)	<.001	1184 (22.3)	661 (26.0)	<.001
Social activity	1 (0-1)	1 (0-1)	.34	0 (0-1)	1 (0-1)	.007

Note. Data are presented as median (Interquartile Range, IQR) for continuous variables and as n (%) for categorical variables, Percentages (%) have been omitted in the table but are represented within parentheses. The Chi-square test was used for categorical variables, while the Wilcoxon rank-sum test was applied for continuous variables. Corresponding P values are listed in the table.

Socioeconomic Variables: These variables include region, per capita household expenditure, and type of medical insurance. Regions are categorized according to the official regional classification by the National Bureau of Statistics of China.²⁹ Per capita household expenditure, chosen for its higher data quality in CHARLS,³⁰ was log-transformed for standardization, with missing values imputed by community averages. Medical insurance was classified into 4 categories: no insurance, UEBMI (Urban Employee Basic Medical Insurance), and a combined category comprising Rural Residents Basic Medical Insurance (URRBMI), Urban Residents Basic Medical Insurance (URBMI), and New Rural Cooperative Medical Insurance (NRCMI) based on their policy similarity and their integration into a unified insurance scheme,³¹ and other insurance.

Lifestyle Variables: These include smoking, drinking status, and social activity, which have been identified in previous studies as important influences on health. Information on social activity was drawn from the CHARLS questionnaire item: “Have you done any of these activities in the last month?,” and was treated as a count data variable.

Statistical Methods

Descriptive statistics were performed to describe demographic, socioeconomic, and lifestyles characteristics (Table 1). Propensity Score Matching (PSM) and Multinomial Propensity Score Matching (MNPS) were employed to balance observed covariate differences. Mixed-effects Ordinal Logit (MEOLOGIT) regression was employed to examine the relationship between UOHN or UIHN and SRH, and the relationship between UOHN or UIHN for different reasons and SRH.

Endogeneity Issues Methodology

Endogeneity poses a significant challenge in studying the impact of UHN on health outcomes due to potential inconsistencies between treatment and control groups or unobserved factors. To address this, we applied 3 strategies. First, by excluding individuals without illness from the sample, the analysis focused on those with actual healthcare needs, thereby reducing heterogeneity. Second, PSM and MNPS were employed to balance observed covariate differences across treatment groups, reducing observable selection bias,^32-34 MEOLOGIT models were applied to control for time-invariant unobserved factors, thereby mitigating potential omitted variable bias; however, time-varying unobserved factors may still pose challenges.^35,36

Propensity Score Matching

To perform PSM, 1-to-2 nearest neighbor matching with a caliper of 0.05 was chosen for its superior matching performance, effectively reducing selection bias and enhancing the comparability between treated and control groups. For multinomial treatment variables, MNPS are employed using Generalized Boosted Models, where the inverse of the estimated propensity scores is applied as weights to balance covariates across treatment groups^34,37,38 All PSM and MNPS procedures are conducted separately for each year to avoiding “self-matching” bias and accuracy of the estimates.^39,40

Mixed-Effect Ordinal Logit Regression Model (MEOLOGIT)

The primary dependent variable in this study, SRH, is ordinal. The exclusion of individuals without illness from the sample resulted in an unbalanced panel data structure, with over 50% of individuals having data for only one period. Therefore, a MEOLOGIT model was employed for the analysis.^35,36,41 This model accommodates the unbalanced nature of the data by incorporating individual random effects to control for individual heterogeneity,^42,43 while treating covariates and time as fixed effects to capture their influence on health outcomes across different periods.⁴⁴ This approach ensures that the ordinal nature of SRH is appropriately modeled and that the analysis of the unbalanced panel data is more accurate.⁴⁵ Given that the MEOLOGIT model accounts for individual heterogeneity and adjusts for potential variations across different levels of the dependent variable through random effects, the requirement to strictly test the proportional odds assumption is relaxed.⁴⁶

As the parameters of the Ordinal Logit model are not easily interpretable, the marginal effects of the MEOLOGIT model provide a clearer understanding of the relationships.^47-49 Therefore, marginal analysis is employed following MEOLOGIT regression to facilitate interpretation.

The statistical analyses were primarily conducted using Stata 18. For the MNPS, both Stata 18 and R 4.4.0 were used in combination.³⁴

Results

Descriptive Results

The sample size for UOHN is 14 207, indicating that 31.65% of middle-aged and elderly individuals who experienced illness in the past month did not seek medical care. For UIHN, the sample size is 7852, with 32.39% of those advised by a doctor to be hospitalized not following through. Complete study population characteristics are presented in Table 1.

Propensity Score Matching (PSM)

PSM for UHN

Using SRH as the outcome variable and UOHN or UIHN as the treatment variables, PSM was applied to covariates. The main results are presented in Table 2.

Table 2.

Propensity Score Matching Indicators for UHN Impact on SRH by Year.

Treatment variables	Year	Sample	Ps R²	LR χ²	P>χ²	MeanBias	MedBias	B	R	% Var
UOHN	2011	Unmatched	.027	137.8	.000	8.1	5.7	39.8*	1.06	0
	2011	Matched	.001	3.17	.999	1.6	1.6	7.2	1.01	12
	2013	Unmatched	.018	112.7	.000	7.0	6.2	32.6*	1.09	0
	2013	Matched	.001	4.45	.992	1.4	1.1	7.5	1.09	12
	2015	Unmatched	.018	117.8	.000	7.3	6.2	32.7*	.98	12
	2015	Matched	.001	4.33	.993	1.8	1.7	7.2	.97	0
UIHN	2011	Unmatched	.037	99.23	.000	9.6	8.5	46.9*	.98	25
	2011	Matched	.005	8.76	.846	3.1	3.3	16.0	1.09	12
	2013	Unmatched	.035	114.62	.000	8.6	5.8	46.2*	.84	25
	2013	Matched	.002	3.45	.998	1.6	1.4	9.2	1.04	0
	2015	Unmatched	.040	160.17	.000	11.6	11.0	48.6*	.88	25
	2015	Matched	.004	12.60	.558	4.6	4.6	15.6	.91	0

Indicates a statistically significant imbalance before matching.

Results suggest that PSM significantly improved covariate balance between the treated and control groups. Reductions in pseudo R², Likelihood Ratio Chi-Square (LR χ²), Mean Bias, and Median Bias values post-matching demonstrate improved comparability. The B statistic also decreased, reflecting reduced standardized mean differences, while the R statistic remained close to 1, indicating balanced variance between the groups.

The Supplemental Tables show the PSM results for UOHN (Table 1A-C) and UIHN (Table 2A-C) for 2011, 2013, and 2015, respectively. They demonstrate substantial reductions in % Bias and t-values across covariates after matching, indicating improved comparability between the groups (Table 3).

Table 3.

Estimated Treatment Effects of UOHN and UIHN on SRH by Year.

Treatment variables	Year	Sample	Treated mean	Control mean	Difference	S.E.	t
UOHN	2011	Unmatched	2.599	2.584	0.015	0.029	0.53
		ATT	2.599	2.571	0.029	0.035	0.83
		ATU	2.585	2.582	−0.002	-	-
		ATE	-	-	0.006	-	-
	2013	Unmatched	2.708	2.646	0.062	0.027	2.33
		ATT	2.708	2.715	−0.007	0.032	−0.23
		ATU	2.647	2.702	0.056	-	-
		ATE	-	-	0.035	-	-
	2015	Unmatched	2.653	2.699	−0.047	0.027	−1.72
		ATT	2.653	2.686	−0.033	0.032	−1.02
		ATU	2.700	2.637	−0.064	-	-
		ATE	-	-	−0.053	-	-
UIHN	2011	Unmatched	2.294	2.538	−0.244	0.041	−6.02
		ATT	2.294	2.493	−0.199	0.050	−3.99
		ATU	2.539	2.284	−0.254	-	-
		ATE	-	-	−0.236	-	-
	2013	Unmatched	2.345	2.586	−0.241	0.039	−6.21
		ATT	2.345	2.531	−0.186	0.048	−3.87
		ATU	2.585	2.394	−0.191	-	-
		ATE	-	-	−0.190	-	-
	2015	Unmatched	2.433	2.618	−0.185	0.034	−5.39
		ATT	2.435	2.529	−0.094	0.043	−2.19
		ATU	2.611	2.550	−0.061	-	-
		ATE	-	-	−0.072	-	-

The unmatched and matched samples (ATT, ATU, and ATE) for UOHN show minimal and statistically insignificant differences in SRH between treated and control groups. For example, in 2011, the ATT for UOHN was 0.029 (t = 0.83), with similarly non-significant results in 2013 and 2015.

In contrast, even after matching, UIHN still exhibits significant differences between treated and control groups. In 2011, the unmatched sample had an SRH difference of −0.244 (t-statistic: −6.02), and the matched ATT was −0.199 (t = −3.99). This pattern continued in 2013 and 2015, with significant ATT values of −0.186 and −0.094.

These findings indicate that while UOHN has a negligible impact on SRH, UIHN significantly affects SRH, even after matching.

MNPS for UHN With Reasons

Using SRH as the outcome variable, UOHN with reasons, and UIHN with reasons as the treatment variables, MNPS was performed on covariates year by year. The main results are presented in Table 4.

Table 4.

MNPS Balance Test Results for UHN With Reasons.

Treatment variables	Year	Comparison	Max Std. Diff	Min P-value	Max KS Stat	Min KS P-value
UOHN With reasons	2011	Unweighted	0.503	<.001	0.213	<.001
		es_mean	0.166	.069	0.071	.473
		ks_mean	0.168	.065	0.072	.459
	2013	Unweighted	0.509	<.001	0.235	<.001
		es_mean	0.204	.011	0.091	.234
		ks_mean	0.208	.003	0.094	.257
	2015	Unweighted	0.444	<.001	0.189	<.001
		es_mean	0.119	.096	0.071	.264
		ks_mean	0.117	.159	0.080	.214
UIHN With reasons	2011	Unweighted	0.356	<.001	0.176	<.001
		es_mean	0.142	.081	0.073	.400
		ks_mean	0.141	.082	0.073	.393
	2013	Unweighted	0.369	<.001	0.160	<.001
		es_mean	0.076	.215	0.046	.507
		ks_mean	0.077	.209	0.046	.497
	2015	Unweighted	0.300	<.001	0.142	<.001
		es_mean	0.118	.020	0.051	.277
		ks_mean	0.118	.020	0.051	.275

The results show a notable improvement in balance after weighting, reflected by lower Maximum Standardized Differences (Max Std. Diff) and Kolmogorov-Smirnov Statistics (Max KS Stat), along with higher minimum P-values and KS P-values. This indicates that the treatment groups became more comparable, reducing potential bias in estimating the effects of UHN with reasons on SRH.

Further details are provided in Supplemental Table 3-A and Supplemental Table 3-B, which demonstrate that MNPS positively impacted covariate balance, reducing inconsistencies between groups.

The Impact of UHN on SRH

The Regression Results of UHN on SRH

To test Hypothesis 1, the MEOLOGIT regression model for UOHN and SRH, as well as UIHN and SRH were employed, the main results reported in Table 5.

Table 5.

The Impact of UOHN and UIHN on SRH.

Variable	UOHNAOR (95% CI)	UIHNAOR (95% CI)
UOHN/UIHN	1.03 (0.94, 1.14)	0.65*** (0.57, 0.74)
Gender(0 = Female)	1.13* (0.98, 1.30)	1.15 (0.97, 1.38)
Age	0.95 (0.89, 1.02)	1.06 (0.97, 1.16)
Marital status (0 = Single)	1.25** (1.04, 1.50)	1.19 (0.94, 1.50)
Number of children	0.99 (0.95, 1.04)	1.05* (1.00, 1.11)
Living arrangement (0 = Living with family)	1.05 (0.80, 1.38)	1.16 (0.82, 1.65)
Hukou (0 = Urban)	0.68*** (0.58, 0.80)	0.58*** (0.47, 0.72)
Per capita household expenditure	1.03 (0.97, 1.09)	1.05 (0.97, 1.14)
Drinking status (0 = not current drinker)	1.56*** (1.38, 1.75)	1.54*** (1.31, 1.81)
Smoking status (0 = not current smoker)	0.90 (0.78, 1.04)	1.02 (0.85, 1.22)
Social activity	1.22*** (1.17, 1.28)	1.26*** (1.19, 1.35)
Education level: (0 = Illiterate/Semi-literate)
Primary school	1.09 (0.95, 1.25)	1.10 (0.92, 1.31)
Middle school	1.20** (1.03, 1.40)	1.11 (0.91, 1.36)
High school and above	1.70*** (1.39, 2.08)	1.23 (0.95, 1.58)
Number of chronic diseases: (0 = no chronic disease)
1 chronic disease	0.47*** (0.41, 0.54)	0.43*** (0.34, 0.53)
2-3 chronic diseases	0.21*** (0.18, 0.24)	0.21*** (0.17, 0.27)
4 or more chronic diseases	0.09*** (0.07, 0.11)	0.10*** (0.07, 0.13)
Region: (0 = West)
Middle	1.03 (0.91, 1.17)	0.93 (0.80, 1.09)
East	1.48*** (1.30, 1.68)	1.23** (1.04, 1.45)
Medical insurance (0 = No insurance)
UEBMI	1.34* (1.00, 1.81)	1.36 (0.90, 2.04)
URRBMI /URBMI/NRCMI	1.14 (0.89, 1.46)	1.17 (0.84, 1.64)
Other insurance	1.43** (1.06, 1.94)	1.58** (1.06, 2.37)
Wave (0 = 2011)
2013	1.25*** (1.11, 1.40)	1.09 (0.93, 1.28)
2015	1.51*** (1.34, 1.70)	1.54*** (1.31, 1.81)
Random intercept variance	5.54*** (3.78, 8.11)	4.79*** (2.69, 8.51)
N	9476	5131
ICC	0.34	0.32

Note. Coefficients of the meologit models are presented as Adjusted Odds Ratios (AOR), while for mixed-effects models, the regression coefficients are directly presented. 95% Confidence Intervals (CI) are shown in parentheses. The Intraclass Correlation Coefficient (ICC) and the variance of the random intercepts for individual-level clustering in the meologit models are reported. The Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC) are reported. These notes also apply to Tables 6 and 7.

P < .1, ** P < .05, *** P < .01.

Results indicate that that UOHN does not have a statistically significant impact on SRH (AOR = 1.03, 95% CI = [0.94, 1.14]). Conversely, there was a significant negative impact of UIHN on SRH (AOR = 0.65, 95% CI = [0.57, 0.74]). Therefore, Hypothesis 1 is supported: while UIHN has a significant negative impact on SRH, UOHN does not.

The regression results indicate that several control variables significantly influence SRH. Current drinkers and middle-aged and older adults with more social activities are more likely to report better SRH. Conversely, rural residents and those with a greater number of chronic diseases tend to report poorer SRH. Additionally, regional differences are evident, with individuals in eastern China reporting better SRH compared to those in western regions.

The Regression Results of UHN With Reasons on SRH

To test Hypothesis 2, the MEOLOGIT regression model for UOHN with reasons and SRH, as well as UIHN with reasons and SRH were employed. Results indicate that compared to those who accessed outpatient or clinical services, individuals who were ill but did not seek outpatient care due to financial constraints were significantly less likely to report better SRH (AOR = 0.44, 95% CI = [0.24, 0.82]). Similarly, compared to those who were hospitalized, individuals who did not receive inpatient care due to financial reasons also reported lower SRH (AOR = 0.25, 95% CI = [0.15, 0.42]). Detailed results are presented in Table 6.

Table 6.

Impact of UHN on SRH: Financial and Non-Financial Reasons.

Variables	UOHN for different reasons	UIHN for different reasons
UOHN/UIHN for different reasons (0 = Already received healthcare)
UOHN/UIHN due to financial barriers	0.44*** (0.24, 0.82)	0.25*** (0.15, 0.42)
UOHN/UIHN for non-financial barriers	1.27* (0.97, 1.66)	0.78 (0.43, 1.43)
Gender (0 = Female)	1.04 (0.77, 1.42)	1.35 (0.87, 2.09)
Age	0.84** (0.72, 0.99)	1.04 (0.83, 1.29)
Marital status (0 = Single)	2.52*** (1.51, 4.19)	2.13** (1.05, 4.33)
Number of children	1.03 (0.92, 1.16)	1.15* (0.99, 1.33)
Living arrangement (0 = Living with family)	2.18** (1.17, 4.07)	2.78** (1.01, 7.64)
Hukou (0 = Urban)	0.52*** (0.34, 0.80)	0.51** (0.26, 0.99)
Per capita household expenditure	1.00 (0.85, 1.16)	0.95 (0.74, 1.22)
Drinking status (0 = not current drinker)	1.91*** (1.40, 2.62)	2.66*** (1.44, 4.89)
Smoking status (0 = not current smoker)	1.11 (0.73, 1.68)	1.13 (0.64, 2.00)
Social activity	1.38*** (1.22, 1.57)	1.50*** (1.21, 1.86)
Education level (0 = Illiterate/Semi-literate)
Primary school	1.32** (1.01, 1.73)	1.86*** (1.24, 2.78)
Middle school	1.98*** (1.47, 2.68)	1.96*** (1.26, 3.05)
High school and above	4.03*** (2.68, 6.05)	4.01*** (2.19, 7.33)
Number of chronic diseases (0 = No chronic disease)
1 Chronic disease	0.27*** (0.19, 0.40)	0.15*** (0.08, 0.28)
2-3 chronic diseases	0.05*** (0.03, 0.07)	0.04*** (0.02, 0.07)
4 or more chronic diseases	0.02*** (0.01, 0.03)	0.01*** (0.00, 0.02)
Region (0 = West)
Middle	1.27** (1.00, 1.62)	0.95 (0.68, 1.34)
East	2.79*** (2.18, 3.56)	1.98*** (1.39, 2.83)
Medical insurance (0 = No insurance)
UEBMI	1.33 (0.59, 2.98)	0.66 (0.18, 2.44)
URRBMI /URBMI/NRCMI	1.14 (0.59, 2.21)	0.60 (0.21, 1.70)
Other insurance	1.76 (0.79, 3.93)	0.90 (0.26, 3.14)
Wave: (0 = 2011)
2013	1.32* (0.97, 1.78)	1.27 (0.81, 1.99)
2015	2.13*** (1.57, 2.87)	2.37*** (1.48, 3.82)
Random intercept variance	23.11 (20.84, 25.64)	31.21 (26.07, 37.36)
N	14 207	7852
ICC	0.88	0.90

P < .1, ** P < .05, *** P < .01.

Heterogeneity Analysis

To test Hypothesis 3, a heterogeneity analysis to examine urban-rural differences in the relationship between UOHN or UIHN and health outcomes was employed. Key interaction results are presented in Table 7, while full results, including all covariates, are available in Supplemental Table 4.

Table 7.

Regression Results of UOHN and Hukou Interaction on SRH.

Variables	UOHN AOR (95% CI)	UIHN AOR (95% CI)
UOHN/ UIHN	0.95 (0.78, 1.16)	0.89 (0.69, 1.15)
Hukou	0.65*** (0.54, 0.79)	0.70*** (0.54, 0.90)
UOHN/UIHN and Hukou Interaction	1.11 (0.88, 1.40)	0.67*** (0.50, 0.89)
Random Intercept Variance	5.54*** (3.78, 8.10)	4.69*** (2.64, 8.34)
N	9476	5131
AIC	21 961.00	12 368.61
BIC	22 175.70	12 564.90
ICC	0.34	0.32

***

P < .01.

The regression results in Table 7 indicate that UOHN does not have a significant impact on SRH for either urban or rural middle-aged and older adults. The Odds Ratios for UOHN are not statistically significant across all samples, aligning with earlier findings that UOHN does not significantly influence SRH in this population. Furthermore, the interaction term between UOHN and Hukou is also non-significant, suggesting no meaningful difference in the effect of UOHN on SRH between urban and rural residents.

In contrast, the significant interaction between UIHN and rural Hukou (AOR = 0.67, 95% CI = [0.50, 0.89]) in the mixed-effects ordinal logistic regression indicates a stronger negative effect of unmet inpatient need on self-rated health among rural older adults. The marginal predicted probabilities further illustrate this disparity. Among older adults with unmet inpatient needs, the probability of reporting very poor self-rated health (SRH = 1) was 17.27% for rural residents, compared to 10.25% for their urban counterparts. Conversely, the probability of reporting very good self-rated health (SRH = 5) was only 1.90% for rural residents, significantly lower than the 3.69% observed among urban residents. These differences highlight a disproportionate negative impact of unmet inpatient needs on both ends of the health spectrum for rural populations.

Additionally, the Wald test of the UIHN and Hukou interaction term in the regression model revealed statistical significance (χ² = 67.89, P = .000), confirming that the impact of unmet inpatient healthcare needs (UIHN) on SRH differs significantly between urban and rural areas.

Thus, Hypothesis 3 is partly supported. A significant difference in the impact of UIHN on SRH is observed between urban and rural middle-aged and older adults, with rural residents more adversely affected. However, no significant difference is found in the impact of UOHN on SRH between urban and rural residents.

Robustness Checks

Three regression models were employed to conduct robustness checks for the results of UOHN and UIHN on SRH by adjusting for PSM methods, MNPS weighting, and regression models. The results indicate that the regression coefficients of UOHN and UIHN on SRH remain relatively stable, confirming the consistency of the findings. Detailed results are provided in Supplementary Table 5A and B.

Similarly, 3 regression models were employed to conduct robustness checks for the results of the reasons behind UOHN and UIHN on SRH, using a mixed effects model with es_mean weights, a mixed effects model with Kolmogorov-Smirnov Mean (ks_mean) weights, and a MEOLOGIT model with ks_mean weights. The results show that the regression coefficients for the reasons behind UOHN and UIHN on SRH remain relatively stable, further supporting the reliability of the findings. Detailed results are provided in Supplemental Tables 5C and D.

Endogeneity Issues

The methods used to address endogeneity proved effective, as confirmed by the results and robustness checks. Post-matching analysis showed no significant covariate differences, indicating successful control of observable confounders. The MEOLOGIT model accounted for potential omitted variable bias by including fixed effects and time variables. High Intraclass Correlation Coefficient values and significant variance in random intercepts justified the model’s use,⁵⁰ addressing individual-level heterogeneity and reinforcing the robustness of the findings.

Discussion

This study corroborates existing research, confirming that UHN have a negative impact on health. The findings reveal that while UOHN are not significantly associated with SRH, UIHN are associated with greater odds of reporting lower SRH. Moreover, UHNs due to financial barriers, whether related to outpatient or inpatient services, are shown to lead to significantly poorer SRH outcomes. Heterogeneity analysis further reveals that rural middle-aged and older individuals are particularly vulnerable to the negative health impacts of UIHN compared to those in urban areas.

It demonstrates that not all UHN result in poorer SRH. Specifically, UOHN do not directly affect SRH among middle-aged and older adults, which contrasts with the findings of Wu et al, who reported a 3.4% decrease in SRH due to unmet outpatient needs.¹⁵ This discrepancy may stem from differences in sample selection, as Wu et al¹⁵ included all individuals without unmet needs as controls, which may have led to the inclusion of a substantial proportion of healthy individuals who were not at risk of experiencing adverse health outcomes, potentially leading to an overestimation of the impact of UOHN on SRH. This highlights the importance of sample composition when assessing the impact of unmet healthcare needs on health outcomes. In contrast, UIHN, which typically involve more serious conditions, have a significant negative effect on SRH, consistent with Ko, who noted that unmet needs for mild symptoms are less likely to impact health outcomes.¹³

The research also reveals that financial barriers have a significant negative impact on SRH, while UHN due to non-financial reasons are not associated with poorer health. This contrasts with Ko¹³, who found that financial barriers, access limitations, and mild symptoms all contributed to poorer SRH. The discrepancy may be attributed to the more profound negative effects of unmet medical needs driven by financial reasons on the SRH of middle-aged and elderly individuals in China. Additionally, the limited outpatient coverage under China’s social health insurance system⁵¹ for insured individuals may exacerbate the negative impact of financial barriers on SRH. Financial difficulties can lead to delays or avoidance of necessary care, resulting in worsened health outcomes due to individuals’ inability to afford essential treatments. As noted in the literature, financial difficulties exacerbate health disparities by limiting access to care, delaying treatment, and increasing stress.^7,52,53

It extends the findings of Zhang et al,⁹ which highlighted SRH differences between urban and rural populations due to UHN. Our research shows that only UIHN significantly lower SRH among rural middle-aged and older adults compared to those in urban areas, while UOHN does not exhibit such differences. This could be explained by the higher reliance on inpatient services in rural areas, where UIHN play more critical for maintaining health.

It is important to note that the analytic sample was deliberately restricted to middle-aged and older adults who experienced health problems in the past month or had a physician-recommended need for hospitalization in the past year. This sample design enhances the comparability between individuals with and without unmet inpatient needs by ensuring that all participants had some degree of healthcare need, thus allowing for a more meaningful assessment of the association between unmet need and self-rated health. Including healthy individuals—who have no demand for inpatient care—would introduce substantial heterogeneity and measurement dilution, thereby undermining the validity of the estimated effects. We believe that the findings of this study are not only relevant to individuals with healthcare needs but also hold broader implications for understanding health disparities among the middle-aged and older population, especially in the context of unmet healthcare needs.

Limitations of the Study

This study has some limitations, primarily due to the reliance on CHARLS data from 2011, 2013, and 2015, as later waves lack detailed information on healthcare underutilization. Additionally, while the study distinguishes between UOHN and UIHN effects on SRH, it does not fully explore the underlying mechanisms. Future research should incorporate more recent data and further investigate how unmet healthcare needs, particularly financial barriers, impact health through pathways such as delayed care or increased stress.

While the methodological strategies used in this study—such as PSM, MNPS, and MEOLOGIT—are effective in reducing selection bias and controlling for unobserved heterogeneity, they have inherent limitations. PSM and MNPS balance covariates across treatment groups to reduce observable selection bias but can only control for observed confounders, leaving room for residual bias from unobserved factors. MEOLOGIT controls for time-invariant unobserved heterogeneity, but cannot account for time-varying factors, such as changes in health behaviors or policy, which may still influence the relationship between UHN and SRH. Moreover, excluding individuals without healthcare needs helps reduce heterogeneity but limits the generalizability to those with perceived healthcare demand. Future research could consider using instrumental variable methods or natural experiments to better control for unobserved confounding and strengthen causal inferences.

Conclusions

This study highlights the significant negative impact of unmet inpatient healthcare needs (UIHN) on self-rated health, particularly when compounded by financial barriers. While outpatient unmet needs (UOHN) alone were not significantly associated with poorer health outcomes, their adverse impact became evident when linked to affordability constraints. To address these challenges, policymakers should expand inpatient reimbursement for low-income residents, strengthen outpatient mutual-aid mechanisms (menzhen gongji), and improve pre-admission cost protection. Additionally, reforming provider payment methods could enhance the efficiency of health insurance fund allocation and reduce unnecessary hospitalizations. Efforts to reinforce primary care referral pathways and promote community-level health literacy could help mitigate treatment delays and prevent condition escalation. Targeted reforms in both financial protection and system-level delivery are critical to preventing financial barriers from turning into unmet needs and deteriorated health among China’s aging population.

Supplemental Material

sj-doc-1-inq-10.1177_00469580251397123 – Supplemental material for Health Services Accessibility and Self-Reported Health Among Chinese Middle-Aged and Older Adults: Propensity Score Matching and Mixed-Effects Ordinal Logit Analysis

Supplemental material, sj-doc-1-inq-10.1177_00469580251397123 for Health Services Accessibility and Self-Reported Health Among Chinese Middle-Aged and Older Adults: Propensity Score Matching and Mixed-Effects Ordinal Logit Analysis by Heling Ai, Ariel Shensa and Faina Linkov in INQUIRY: The Journal of Health Care Organization, Provision, and Financing

Footnotes

Acknowledgements

We thank all of the participants enrolled in the China Health and Retirement Longitudinal Study. We thank the CHARLS research team for providing the data.

List of Abbreviations

UHN: Unmet Healthcare Needs

UOHN: Unmet Outpatient Healthcare Needs

UIHN: Unmet Inpatient Healthcare Needs

MEOLOGIT: Mixed-effects Ordinal Logit regression

SRH: Self-reported health

PSM: Propensity Score Matching

MNPS: Multinomial Propensity Scores

ORCID iDs

Heling Ai

Ariel Shensa

Ethical Considerations

The CHARLS data collection process received approval from the Ethics Committee of Peking University.

Consent to Participate

Participants provided informed consent at the time of data collection.

Consent for Publication

Not applicable. No experiments were conducted, nor were patients involved in this study.

Author Contributions

Heling Ai: Conceptualization, Methodology-Statistical analysis, Writing—original draft, review & editing, project administration, visualization, data curation, Software. Ariel Shensa: Methodology—Statistical analysis, methodology, visualization, Writing—review & editing. Faina Linkov: Conceptualization, Supervision, methodology, visualization, Writing—original draft, review & editing. Both authors have read and agreed to the published version of the manuscript.

Funding

The authors received no financial support for the research, authorship, and/or publication of this article.

Declaration of Conflicting Interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Data Availability Statement

Data used in this study are available in publicly accessible repositories as follows: (1) The CHARLS datasets are publicly available from the National School of Development at Peking University (http://charls.pku.edu.cn/en) and can be accessed after submitting a data use agreement to the CHARLS team. (2) City-level healthcare resource data were primarily sourced from the 2016 China City Statistical Yearbook and 2019 China City Statistical Yearbook ().

Supplemental Material

Supplemental material for this article is available online.

References

World Health Organization. World Health Statistics 2016: Monitoring Health for the SDGs, Sustainable Development Goals. World Health Organization; 2016.

Yip

Jian

, et al Universal health coverage in China part 1: progress and gaps. Lancet Public Health. 2023;8(12):e1025-e1034. doi:10.1016/s2468-2667(23)00254-2

Allin

Grignon

Le Grand

. Subjective unmet need and utilization of health care services in Canada: what are the equity implications? Soc Sci Med. 2010;70(3):465-472. doi:10.1016/j.socscimed.2009.10.027

Zhou

Huang

Wang

. Does universal health insurance coverage reduce unmet healthcare needs in China? Evidence from the National Health Service Survey. Int J Equity Health. 2021;20(1):43. doi:10.1186/s12939-021-01385-7

Lang

. The impact of medical service accessibility on the health of the elderly. Chin J Heal Stat. 2021;38(06):912-915.

Qiu

Y-L

Ran

X-X

. Impact of medical service accessibility on the health of the elderly: analysis based on CLHLS data. Chin J Health Policy. 2019;12(07):1-10.

Gao

Prina

Y-T

Mayston

. Unmet healthcare needs among middle-aged and older adults in China. Age Ageing. 2022;51(1):1-9. doi:10.1093/ageing/afab235

Meng

Kang . The impact of medical service accessibility on disability of the elderly: from the perspective of urban and rural differences. J Ningxia Univ (Human Soc Sci Ed). 2021;43(03):186-195.

Zhang

Dupre

Qiu

Zhou

Zhao

. Urban-rural differences in the association between access to healthcare and health outcomes among older adults in China. BMC Geriatr. 2017;17(1):151. doi:10.1186/s12877-017-0538-9

10.

Chen

. Unmet healthcare needs predict frailty onset in the middle-aged and older population in China: a prospective cohort analysis. Front Public Health. 2023;11:1064846. doi:10.3389/fpubh.2023.1064846

11.

Wang

Zhao

, et al The relationship between self-rated health and objective health status: a population-based study. BMC Public Health. 2013;13(1):9.

12.

Jylhä

. What is self-rated health and why does it predict mortality? Towards a unified conceptual model. Soc Sci Med. 2009;69(3):307-316.

13.

. Unmet healthcare needs and health status: panel evidence from Korea. Health Policy. 2016;120(6):646-653. doi:10.1016/j.healthpol.2016.04.005

14.

Sohn

Che

Park

. Unmet healthcare needs, catastrophic health expenditure, and health in South Korea’s universal healthcare system: progression towards improving equity by NHI type and income level. Healthcare. 2020;8(4):1-14. doi:10.3390/healthcare8040408

15.

Liu

. Unmet healthcare needs, health outcomes, and health inequalities among older people in China. Front Public Health. 2023;11:1082517. doi:10.3389/fpubh.2023.1082517

16.

Miao

Zhang

Sandeep

. Improving the performance of social health insurance system through increasing outpatient expenditure reimbursement ratio: a quasi-experimental evaluation study from rural China. Int J Equity Health. 2018;17:89.

17.

Shensa

Linkov

. Outpatient pooling policy and healthcare satisfaction among middle-aged and older adults: evidence from China. BMC Health Serv Res. 2025;25(1):688. doi:10.1186/s12913-025-12769-8

18.

Huang

Gan

. The impacts of China's urban employee basic medical insurance on healthcare expenditures and health outcomes. Health Econ. 2017;26(2):149-163. doi:10.1002/hec.3281

19.

Shi

Liang

Ding

. Urban-rural disparities in health care utilization among Chinese adults from 1993 to 2011. BMC Health Serv Res. 2018;18(1):102. doi:10.1186/s12913-018-2905-4

20.

Cao

Yang

Jin

, et al A machine learning-based aging measure among middle-aged and older Chinese adults: the China health and retirement longitudinal study. Front Med. 2021;8:698851. doi:10.3389/fmed.2021.698851

21.

Zhao

Wang

Nicholas

, et al Health-related quality of life and Health Service use among multimorbid middle-aged and older-aged adults in China: a cross-sectional study in Shandong province. Int J Environ Res Public Health. 2020;17(24):9261. doi:10.3390/ijerph17249261

22.

Yang

Sun

. Inequalities in unmet health care needs under universal health insurance coverage in China. Health Econ Rev. 2024;14(1):2. doi:10.1186/s13561-024-00494-7

23.

Zhao

Strauss

Xinxin Chen

, et al China Health and Retirement Longitudinal Study Wave 4 User’s Guide. National School of Development; Peking University; 2020.

24.

Smith

Connolly

. Re-thinking unmet need for health care: introducing a dynamic perspective. Health Econ Policy Law. 2020;15(4):440-457. doi:10.1017/S1744133119000161

25.

Vahedi

Torabipour

Takian

, et al Socioeconomic determinants of unmet need for outpatient healthcare services in Iran: a national cross-sectional study. BMC Public Health. 2021;21(1):457. doi:10.1186/s12889-021-10477-6

26.

Pickett

Pearl

. Multilevel analyses of neighbourhood socioeconomic context and health outcomes: a critical review. J Epidemiol Community Health. 2001;55(2):111-122. doi:10.1136/jech.55.2.111

27.

Kim

Han

, et al Association between unmet healthcare needs and health-related quality of life: a longitudinal study. Eur J Public Health. 2017;27(4):631-637. doi:10.1093/eurpub/ckw264

28.

Bombak

. Self-rated health and public health: a critical perspective. Front Public Health. 2013;1:15. doi:10.3389/fpubh.2013.00015

29.

National Bureau of Statistics. China Statistical Yearbook. National Bureau of Statistics; 2023.

30.

Gong

Wang

Zhao

. Consumption and poverty of older Chinese: 2011–2020. J Econ Ageing. 2022;23:100410. doi:10.1016/j.jeoa.2022.100410

31.

Liu

Vortherms

Hong

. China’s health reform update. Annu Rev Public Health. 2017;38:431-448. doi:10.1146/annurev-publhealth-031816-044247

32.

Heckman

Ichimura

Todd

. Matching as an econometric evaluation estimator: Evidence from evaluating a job training programme. Rev Econ Stud. 1997;64(4):605-654.

33.

Benedetto

Head

Angelini

Blackstone

. Statistical primer: propensity score matching and its alternatives. Eur J Cardiothorac Surg. 2018;53(6):1112-1117. doi:10.1093/ejcts/ezy167

34.

Burgette

Griffin

McCaffrey

. Propensity Scores for Multiple Treatments: A Tutorial for the mnps Function in the Twang Package. R package Rand Corporation; 2017.

35.

Lau

Ataguba

. Investigating the relationship between self-rated health and social capital in South Africa: a multilevel panel data analysis. BMC Public Health. 2015;15:266. doi:10.1186/s12889-015-1601-0

36.

Steele

. Multilevel models for longitudinal data. J R Stat Soc Ser Stat Soc. 2008;171(1):5-19.

37.

McCaffrey

Griffin

Almirall

Slaughter

Ramchand

Burgette

. A tutorial on propensity score estimation for multiple treatments using generalized boosted models. Stat Med. 2013;32(19):3388-3414.

38.

Shurrab

Jackevicius

, et al A review of the use of propensity score methods with multiple treatment groups in the general internal medicine literature. Pharmacoepidemiol Drug Saf. 2023;32(8):817-831.

39.

Böckerman

Ilmakunnas

. Unemployment and self-assessed health: evidence from panel data. Health Econ. 2009;18(2):161-179. doi:10.1002/hec.1361

40.

Xie

Fan

Wan

. Improvement and application of classical PSM-DID model. Stat Res. 2021;38(2):146-160.

41.

Exavery

Kirigiti

Balan

Charles

. Multivariate mixed-effects ordinal logistic regression models with difference-in-differences estimator of the impact of WORTH Yetu on household hunger and socioeconomic status among OVC caregivers in Tanzania. PLoS One. 2024;19(4):e0301578. doi:10.1371/journal.pone.0301578

42.

Antweiler

. Nested random effects estimation in unbalanced panel data. J Econom. 2001;101(2):295-313.

43.

Skrondal

Rabe-Hesketh

. Protective estimation of mixed-effects logistic regression when data are not missing at random. Biometrika. 2014;101(1):175-188. doi:10.1093/biomet/ast054

44.

Bruyndonckx

Hens

Aerts

. Simulation-based evaluation of the linear-mixed model in the presence of an increasing proportion of singletons. Biom J. 2018;60(1):49-65. doi:10.1002/bimj.201700025

45.

Hedeker

. A mixed-effects multinomial logistic regression model. Stat Med. 2003;22(9):1433-1446. doi:10.1002/sim.1522

46.

Rabe-Hesketh

, ed. Multilevel and longitudinal modeling using Stata. 4th ed. STATA Press; 2022.

47.

Hedeker

du Toit

SHC

Demirtas

Gibbons

. A note on marginalization of regression parameters from mixed models of binary outcomes. Biometrics. 2018;74(1):354-361. doi:10.1111/biom.12707

48.

Williams

. Using the margins command to estimate and interpret adjusted predictions and marginal effects. Stata J. 2012;12(2):308-331.

49.

Mize

Doan

Long

. A general framework for comparing predictions and marginal effects across models. Sociol Methodol. 2019;49(1):152-189. doi:10.1177/0081175019852763

50.

Liljequist

Elfving

Skavberg Roaldsen

. Intraclass correlation - a discussion and demonstration of basic features. PLoS One. 2019;14(7):e0219854. doi:10.1371/journal.pone.0219854

51.

Zhang

Chen

. Inequality in benefit distribution of reducing the outpatient cost-sharing: evidence from the outpatient pooling scheme in China. Front Public Health. 2024;12:1357114. doi:10.3389/fpubh.2024.1357114

52.

Khan

Socha-Dietrich

. Investing in Medication Adherence Improves Health Outcomes and Health System Efficiency: Adherence to Medicines for Diabetes, Hypertension, and Hyperlipidaemia. OECD; 2018.

53.

Osborn

Kripalani

Goggins

Wallston

. Financial strain is associated with medication nonadherence and worse self-rated health among cardiovascular patients. J Health Care Poor Underserved. 2017;28(1):499-513. doi:10.1353/hpu.2017.0036

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

1.55 MB