Multidimensional health poverty among diabetic adults in China: evidence from the China Kadoorie Biobank

Abstract

Background

Despite the rising burden of diabetes in low- and middle-income countries, conventional economic metrics often fail to capture the multidimensional nature of health deprivation. This study aims to bridge this gap by assessing multidimensional health poverty among Chinese adults with diabetes using a survival-anchored multidimensional framework.

Methods

This study used data from 19,236 adults with diabetes enrolled in the China Kadoorie Biobank. Individual 5-year survival probabilities were estimated using stratified Cox proportional hazards models. Subsequently, we constructed Multidimensional Health Poverty Indices using the Alkire-Foster dual-cutoff method, integrating three dimensions: subjective health, objective health, and healthcare access capability. Scenario analyses were further employed to provide illustrative simulations of behavioral interventions on poverty reduction.

Results

The overall adjusted headcount ratio (M₀) of multidimensional health poverty was 26.92%. We observed significant demographic and socioeconomic gradients, with poverty intensity being most pronounced among older women, rural residents, and individuals with lower socioeconomic status. According to the decomposition results, reducing deprivation in healthcare access capability may warrant prioritization for lower socioeconomic strata, while counterfactual scenarios revealed that behavioral interventions were associated with larger simulated reductions for higher socioeconomic strata.

Conclusion

Diabetic adults in China face substantial multidimensional health poverty, driven principally by survival risks and structural barriers to healthcare. The divergence in intervention outcomes suggests that achieving health equity requires a targeted strategy, prioritizing financial protection mechanisms for vulnerable populations while promoting lifestyle modifications among higher socioeconomic groups.

Supplementary Information

The online version contains supplementary material available at 10.1186/s13690-026-01887-2.

Text box 1. Contributions to literature
• This study introduces a survival-based framework to assess multidimensional health poverty, capturing the multidimensional burden of chronic diseases beyond economic metrics.
• Highlights that deprivation drivers diverge, as vulnerable groups face access barriers, whereas wealthier populations primarily struggle with clinical disease burdens.
• Offers evidence supporting stratified policies, prioritizing financial protection for the underserved while promoting lifestyle modifications for higher socioeconomic groups.

Introduction

Health inequality constitutes a central dimension of broader social inequality, manifested through persistent and substantial disparities in access to health resources, health status, and health outcomes across social groups [1]. Such disparities compromise individual well-being while threatening social cohesion, equity, and long-term societal stability [2, 3]. In low- and middle-income countries, health inequalities are particularly pronounced among chronic disease patients, with diabetes being one of the most pressing public health challenges. Currently, low- and middle-income countries account for approximately 80% of the global adult diabetic population, highlighting an urgent need for targeted interventions [4].

China epitomizes this crisis, accounting for nearly one-quarter of global diabetes cases [5]. Rapid urbanization, population aging, and lifestyle shifts have driven an unprecedented increase in diabetes prevalence, making China the world’s second-largest spender on diabetes-related healthcare after the United States [6, 7]. In response, since 2009, the Chinese government has implemented extensive health system reforms, including the incorporation of diabetes management into essential public health services, in an effort to improve healthcare equity [8]. However, significant gaps persist, particularly among rural residents and individuals of lower socioeconomic status (SES), where barriers to healthcare access exacerbate outcome disparities [5]. These inequalities suggest the limitations of existing approaches, highlighting the critical need for precision management tailored to the multifaceted nature of diabetes-related deprivation.

Over recent decades, poverty research has increasingly recognized that economic indicators alone fail to fully capture the complex deprivation experienced by individuals with chronic conditions [9, 10]. Although income and consumption remain central to conventional poverty measurement, individuals with similar economic resources may experience different health trajectories. Recent literature has argued for distinguishing health poverty from economic poverty and treating low health status as a deprivation in its own right rather than merely a consequence of low income [11–13]. Within this perspective, health poverty is defined as a condition in which an individual’s health level falls below a socially or clinically acceptable threshold, regardless of socioeconomic status [14]. This threshold-based approach shifts attention from financial insufficiency to the lower tail of the health distribution and focuses on identifying individuals who remain in poor health.

However, a single health indicator is rarely sufficient to characterize health poverty. To address this, the Alkire-Foster (AF) dual-cutoff method has become a standard approach for measuring multidimensional health poverty [15–17]. Drawing on the Capability Approach, recent studies argue that health poverty stems from the complex interaction between deteriorating physical health and structural disadvantages. Integrating these intertwined health outcomes and institutional barriers into a unified index helps capture the compounded vulnerability experienced by patients [16–19]. Current empirical applications remain predominantly population- or household-level, with limited attention to disease-specific subgroups [20, 21]. The escalating health and economic burdens of diabetes render targeted assessment of multidimensional health poverty in this population both methodologically imperative and policy-relevant, enabling precise vulnerability identification, tailored interventions, and equitable health system development.

This study seeks to address this research gap by systematically assessing the incidence, intensity, and inequality of multidimensional health poverty among Chinese adults with diabetes. Leveraging nationally representative data from the China Kadoorie Biobank (CKB), we construct a composite multidimensional health poverty measure that integrates subjective well-being, physiological burden, and service accessibility. Our analysis identifies the demographic, behavioral, and socioeconomic determinants while employing scenario analysis to explore potential improvement pathways that can inform targeted health policy interventions. The findings aim to inform precision-targeted, equity-oriented health policies while advancing theoretical understanding of diabetes-related multidimensional health poverty.

Methods

This study adopted a two-stage analytical framework to assess multidimensional health poverty among China’s diabetic population. First, we estimated individual 5-year all-cause survival probabilities using stratified Cox proportional hazards models as a key component of the objective health status dimension. Second, we constructed multidimensional health poverty indices using the AF dual-cutoff method, integrating survival outcomes with indicators of physiological burden, subjective health, and healthcare access capability. Furthermore, we implemented scenario analyses that provide illustrative, non-causal simulations of behavioral change and its association with multidimensional health poverty.

Data and study population

Individuals with diabetes were identified from the CKB, a prospective cohort of approximately 500,000 adults aged 30–79 years recruited between 2004 and 2008 from ten geographically diverse regions in China. Individuals with diabetes were identified at baseline through self-reported physician diagnosis or screen-detected criteria, including: (1) random plasma glucose ≥ 7.0 mmol/L with ≥ 8 h fasting, (2) random plasma glucose ≥ 11.1 mmol/L with < 8 h fasting, or (3) fasting plasma glucose ≥ 7.0 mmol/L [20]. Individuals newly identified with diabetes at the 2018 follow-up assessment were also included as diabetic participants in the analytical cohort. After excluding individuals lost to follow-up prior to 2016, the final analytical sample consisted of 19,236 individuals with diabetes.

The CKB provided detailed data on demographic, clinical, and lifestyle characteristics through standardized questionnaires, physical measurements, and blood sampling [22–24]. Covariates derived from the baseline survey included demographic variables (age, sex, health insurance), SES variables and health-related variables (smoking status, alcohol consumption, BMI, physical activity, diabetes-related comorbidities, and family diabetes history). SES was assessed using four indicators: annual household income (< 10,000; 10,000–19,999; 20,000–34,999; ≥ 35,000 Chinese yuan), highest education level (no formal school, primary school, middle or high school, college or above), urban or rural residence, and geographic areas. Geographic areas of China were categorized according to the National Bureau of Statistics of China and ranked from high to low per capita disposable income of households as Eastern, Northeastern, Central and Western, which reflected different levels of economic development in China [25]. BMI was categorized using Chinese-specific cut-offs [26], and physical activity was calculated by summing MET-hours per day across occupational, transportation, household, and recreational domains [27]. Assessed comorbidities include coronary heart disease (CHD), stroke or transient ischemic attack (TIA), and nephropathy.

Table 1 presents the sex-stratified characteristics of the sample. Females comprised 61% of the cohort, and 63% of all participants were aged between 45 and 64 years. Males showed higher education levels and income, whereas females reported healthier lifestyle profiles.

Table 1.

Summary statistics

Variable	Females		Males		Total
	Mean	SD	Mean	SD	Mean	SD
DM Duration (years)	9.47	1.99	9.13	2.36	9.34	2.15
BMI (kg/m2)	25.42	3.75	24.87	3.52	25.21	3.67
METs (hours/day)	16.25	11.4	17.7	14.15	16.82	12.58

Variable	Females		Males		Total
	N	%	N	%	N	%
Regular smoking	395	3.39%	3,653	48.17%	4,048	21.04%
Alcohol consumption	72	0.62%	1,986	26.19%	2,058	10.70%
Family diabetes history	950	8.15%	721	9.51%	1,671	8.69%
Health insurance	9,595	82.34%	6,795	89.61%	16,390	85.20%
Age group (year)
[30–35)	43	0.37%	33	0.44%	76	0.40%
[35–40)	383	3.29%	363	4.79%	746	3.88%
[40–45)	991	8.50%	778	10.26%	1,769	9.20%
[45–50)	1,288	11.05%	966	12.74%	2,254	11.72%
[50–55)	2,224	19.09%	1,327	17.50%	3,551	18.46%
[55–60)	2,258	19.38%	1,249	16.47%	3,507	18.23%
[60–65)	1,736	14.90%	999	13.17%	2,735	14.22%
[65–70)	1,591	13.65%	951	12.54%	2,542	13.21%
[70–75)	1,022	8.77%	811	10.69%	1,833	9.53%
[75–80)	117	1.00%	106	1.40%	223	1.16%
Household income
< 10,000	3,287	28.21%	1,586	20.92%	4,873	25.33%
10,000–19,999	3,596	30.86%	2,143	28.26%	5,739	29.83%
20,000–34,999	2,834	24.32%	2,062	27.19%	4,896	25.45%
≥ 35,000	1,936	16.61%	1,792	23.63%	3,728	19.38%
Educational level
No formal school	3,607	30.95%	629	8.29%	4,236	22.02%
Primary school	3,990	34.24%	2,355	31.06%	6,345	32.99%
Middle or high school	3,726	31.97%	3,866	50.98%	7,592	39.47%
College or above	330	2.83%	733	9.67%	1,063	5.53%
Region
Urban residence	6,210	53.29%	4,648	61.30%	10,858	56.45%
Rural residence	5,443	46.71%	2,935	38.70%	8,378	43.55%
Geographic area
Eastern	4,295	36.86%	2,849	37.57%	7,144	37.14%
Central	2,258	19.38%	1,158	15.27%	3,416	17.76%
Western	3,363	28.86%	2,090	27.56%	5,453	28.35%
Northeastern	1,737	14.91%	1,486	19.60%	3,223	16.76%
Comorbidities
Nephropathy	180	1.54%	121	1.60%	301	1.56%
CHD	735	6.31%	360	4.75%	1,095	5.69%
Stroke or TIA	392	3.36%	317	4.18%	709	3.69%
Sample size	11,653		7,583		19,236

Survival modeling

This study estimated individual survival probabilities using a stratified Cox proportional hazards model, adjusting for relevant demographic, socioeconomic, behavioral, and clinical covariates. Stratification was based on sex and ten 5-year age bands (30–79 years), allowing for heterogeneity in baseline hazards across sex-age strata and accommodating potential violations of the proportional hazard assumption. The outcome was defined as all-cause mortality during follow-up among adults with diabetes rather than diabetes-attributable mortality.

1

The hazard function indicates the risk at time t for individuals with specific covariates X. These covariates included SES variables and health-related variables. In this specification, is the baseline hazard, and represents the log-relative hazard associated with the i-th covariate. A hazard ratio (HR) greater than 1 implies a higher risk of mortality and thus a lower survival probability [28]. The 5-year survival probability S(t = 5) was derived for each individual and compared to corresponding national life table values by age and sex. This modeling framework accommodates both confounding control and population heterogeneity in survival expectations.

Construction of multidimensional health poverty indices

The conceptualization of multidimensional health poverty in this study adopts the AF dual-cutoff counting method [29]. Following a three-step approach commonly used in poverty research, the definition process involves three steps: (1) identification of relevant dimensions and indicators; (2) setting deprivation cut-offs for each indicator; and (3) establishing a cross-dimensional poverty cut-off to identify the multidimensionally poor [30, 31].

Dimensions, indicators, and deprivation cutoffs

We constructed a three-dimensional framework reflecting subjective health status, objective health status, and healthcare access capability. The indicators and deprivation thresholds were defined as follows:

• Subjective Health Status: Measured by self-rated health, reflecting the individual’s perceived quality of life and internal health capability. Assessed on a 4-point scale. Individuals reporting their health as “fair” (grade 3) or “poor” (grade 4) were defined as deprived.

• Objective Health Status: Measured by the count of chronic non-communicable diseases (NCDs), representing the physiological disease burden, and 5-year survival probability, offering an objective, clinically relevant projection of health stock [32, 33]. An individual was considered deprived if their predicted 5-year survival probability was more than 5% lower than the corresponding survival rate for the general population based on national life tables. Individuals with two or more NCDs were considered deprived.

• Healthcare Access Capability: This dimension captured the structural capability to access healthcare, constructed as a composite ordinal variable (graded 1–4) combining health insurance status and residential locality (Urban/Rural). The grading hierarchy was: 4 (No insurance & Rural), 3 (No insurance & Urban), 2 (Insurance & Rural), and 1 (Insurance & Urban). A score of 3 or higher was set as the deprivation cutoff. This scoring hierarchy is grounded in the context of China’s healthcare system, where the financial protection provided by basic medical insurance is paramount in mitigating the risk of catastrophic health expenditures for lifelong chronic diseases like diabetes [34]. The financial vulnerability of being uninsured in a high-cost urban environment exposes patients to severe out-of-pocket medical costs, placing them at a higher risk of health-related impoverishment compared to rural residents who possess basic health insurance coverage [35, 36].

The use of survival probability aligns with Sen’s capability approach, framing reduced survival as a deprivation of the opportunity to achieve acceptable health functioning [14, 37]. From this perspective, reduced survival reflects a severe form of capability deprivation, which encompasses cumulative disadvantages in social determinants. In conjunction with multidimensional indicators such as functional limitation and self-rated health, survival-based metrics enhance the overall validity of the assessment by quantifying irreversible outcomes and enabling robust inter-group comparison across socioeconomic strata.

Measurement and aggregation of multidimensional health poverty

To construct the multidimensional health poverty index, this study employs the AF dual-cutoff counting method [29, 38], which allows for a multidimensional assessment of prevalence, depth, and inequality [39].

The process involves two fundamental cutoffs. First, a deprivation cutoff is applied to each indicator. To objectively determine the relative importance of each dimension, we applied Principal Component Analysis (PCA). Weights were derived from the factor loadings of the first principal components that cumulatively explained over 80% of the variance. For each individual , a deprivation score is calculated as the weighted sum of their deprivations. Second, a poverty cutoff of was established; individuals with a score were identified as multidimensionally health poor [40, 41].

Following this identification, we computed three aggregate indices based on the Foster-Greer-Thorbecke (FGT) class of poverty measures to characterize multidimensional health poverty:

• Adjusted Headcount Ratio (): This primary index reflects both the incidence (, the proportion of the population that is health poor) and the intensity (, the average deprivation shares among the poor), calculated as .

• Adjusted Poverty Gap (): Captures the depth of deprivation beyond the binary cutoff.

• Adjusted Squared Poverty Gap (): Captures the severity and inequality among the poor.

To explore socioeconomic disparities, the three indices were further decomposed by subgroups (e.g., income, education, residence) and by dimensions to identify the main contributors to multidimensional health poverty [42]. Detailed axiomatic properties of the FGT measures and the mathematical formulas for subgroup and dimensional decomposition are provided in Appendix.

Sensitivity analysis

Sensitivity analysis about the survival dimension

To address potential circularity arising from using predicted survival both as an input to the index and as an outcome associated with SES, we conducted additional sensitivity analyses. First, we excluded the survival dimension from the multidimensional health poverty index and recalculated all indices using only the remaining indicators (NCDs, self-rated health, and healthcare access capability), with weights re-normalized accordingly. Second, we constructed an event-based alternative in which the predicted survival component was replaced by an observed follow-up outcome. Specifically, we defined deprivation in the survival indicator using a binary event (death within five years) and recalculated the multidimensional health poverty indices. This event-based operationalization is intentionally more conservative and coarser than a risk-based threshold because it collapses a continuous risk gradient into a realized outcome.

Sensitivity analysis about the dimension weights

Although PCA-derived weights provide a data-driven approach to summarizing multidimensional health poverty, they are not normatively grounded in explicit policy or theoretical priorities. To address this concern and assess the robustness of our findings, we conducted a supplementary analysis using an equal-weighting scheme in which each deprivation dimension was assigned identical weight. Under this normative weighting structure, we recalculated the adjusted headcount ratio (M₀) and repeated all subgroup analyses. The equal-weighted specification serves as a theory-neutral benchmark commonly adopted in multidimensional poverty measurement. By comparing results across weighting structures, we evaluate whether the observed socioeconomic disparities are sensitive to the choice of weighting approach.

Scenario analysis

To decompose the determinants of multidimensional health poverty and inform policy interventions, we conducted scenario analysis stratified by income, education, and geographic area. We designed three counterfactual scenarios to evaluate the simulated reduction in multidimensional health poverty indices compared to the baseline. In the baseline cohort, 21.04% of participants were regular smokers and 10.70% reported alcohol consumption (Table 1), so only a minority of individuals were current smokers or drinkers. In order to create more pronounced contrasts and to make changes in the multidimensional health poverty indices easier to detect, the counterfactual scenarios assume complete cessation of smoking and/or drinking among these subgroups:

• Scenario 1: Simulated a situation in which all participants who were regular drinkers at baseline are assumed to abstain from alcohol consumption.

• Scenario 2: Simulated a situation in which all participants who were regular smokers at baseline are assumed to become non-smokers.

• Scenario 3: Combined the conditions of S1 and S2 to simulate the adoption of a healthy lifestyle free from both smoking and alcohol consumption.

For these behavioral interventions (S1, S2, and S3), we employed a predictive modeling approach. The relevant smoking and drinking covariates in the dataset for baseline regular smokers and/or drinkers were set to zero, and individual 5-year survival probabilities were re-estimated using the fitted Cox proportional hazards model. These updated survival probabilities were then fed back into the AF framework to update the objective health status dimension and recalculate the poverty indices. These scenarios are therefore illustrative, non-causal simulations that quantify the sensitivity of multidimensional health poverty to stylized changes in behavioral risk factors, rather than evaluations of real-world policy effects. In this study we focused on smoking and alcohol consumption as key modifiable behaviors; additional behaviors such as physical activity were not incorporated into scenario analyses to avoid introducing further modeling assumptions and to maintain interpretability.

Results

Five-year survival rates

This study compared the 5-year all-cause survival probabilities of 19,236 individuals with diabetes to those of the general Chinese population aged 30–79 in 2010 (See Table A1, Additional File 1). Across all age groups, diabetic patients showed substantially lower survival probabilities, with male patients facing a more pronounced disadvantage than females. These results highlight the enduring survival disadvantage among individuals with diabetes, compounded by sex-based disparities.

Socioeconomic status significantly influenced survival among people with diabetes. People living in urban areas had a 30% lower risk of all-cause mortality compared to those in rural areas (HR = 0.70, 95% CI: 0.63–0.77). Higher household income (HR = 0.62, 95% CI: 0.54–0.70) and greater educational levels (HR = 0.56, 95% CI: 0.45–0.69) were also strongly associated with reduced mortality (See Fig. 1). Conversely, patients with conditions like stroke or nephropathy, as well as smokers, faced a noticeably higher risk of death.

Fig. 1.

Hazard ratios for 5-year mortality among individuals with diabetes. Note: Significance levels are denoted as follows: ***P < 0.001, **P < 0.01, *P < 0.05

Weights and dimensional contributions

PCA was employed to objectively assign weights to the three dimensions of multidimensional health poverty. As shown in Table 2, the derived weights were relatively balanced but indicated that the survival indicator within the objective health dimension had the largest weight. This contribution suggests that while survival remains the most critical component, healthcare access capability and subjective health are influential in defining multidimensional health poverty among diabetic patients.

Table 2.

Weights and contributions of dimensions in multidimensional health poverty indices

Dimension	Objective Health Status		Subjective Health Status	Healthcare Access Capability
	5-year survival rates	NCDs	Self-rated health
Weight (%)	26.34%	23.81%	25.54%	24.31%
Contributions (%)	41.08%	6.51%	40.27%	12.14%

Multidimensional health poverty

Using a multidimensional poverty cutoff of , we calculated that the adjusted headcount ratio () was 26.92%, the adjusted poverty gap () was 0.1271, and the adjusted squared poverty gap () was 0.0923 for the overall population.

Table 3 summarizes the age- and sex-stratified poverty indices. The incidence and intensity of multidimensional health poverty increased progressively with age, peaking in the 60–69 age group. Sex disparities were pronounced; males consistently exhibited higher deprivation scores compared to females across most age cohorts. This gender difference indicates that male participants experienced higher multidimensional health poverty in middle age, likely driven by the intersection of lower survival rates and higher prevalence of behavioral risk factors. Notably, however, poverty severity among women rose sharply in older age groups, indicating a late-life crossover pattern in which older women exhibit higher intensity of deprivation than men.

Table 3.

Multidimensional health poverty in China, by age group and sex

Group	M0		M1		M2
	Male	Female	Male	Female	Male	Female
By age
[30–35)	29.56%	19.07%	0.1309	0.0871	0.0937	0.066
[35–40)	14.37%	15.75%	0.0691	0.0840	0.0503	0.0649
[40–45)	19.09%	12.52%	0.0873	0.0680	0.0641	0.0548
[45–50)	21.84%	18.90%	0.0983	0.0931	0.0722	0.0725
[50–55)	24.26%	23.91%	0.1073	0.1128	0.0761	0.0862
[55–60)	30.85%	26.70%	0.1407	0.1238	0.0991	0.0936
[60–65)	31.94%	34.96%	0.1524	0.1588	0.1071	0.1149
[65–70)	30.56%	37.26%	0.1467	0.1774	0.1024	0.1237
[70–75)	27.32%	37.57%	0.1342	0.1849	0.0933	0.1293
[75–80)	25.33%	32.45%	0.1298	0.1742	0.0901	0.1222
Overall	26.20%	27.39%	0.1220	0.1305	0.0864	0.0962

Table 4 presents the distribution of multidimensional health poverty stratified by socioeconomic characteristics and gender. A steep income gradient was observed across all three poverty indices. For individuals in the lowest income bracket, was alarmingly high (Male: 39.74%, Female: 40.79%), compared to approximately 12–15% among those earning ≥ 35,000 yuan (Male: 15.01%, Female: 12.62%). The corresponding severity of poverty () for the lowest income group (Male: 0.1609, Female: 0.1569) was nearly four times that of the highest income group, indicating deep inequality. Educational disparities were similarly prominent. Those with no formal schooling exhibited an exceeding 32% and an above 0.16 for both sexes, whereas college-educated individuals showed significantly lower poverty levels, particularly females (: 12.54%; : 0.0331).

Table 4.

Multidimensional health poverty in China, by SES and year of diagnosis

Group	M0		M1		M2
	Male	Female	Male	Female	Male	Female
Household income
< 10,000	39.74%	40.79%	0.2156	0.2085	0.1609	0.1569
10,000–19,999	29.68%	29.52%	0.1353	0.1380	0.0939	0.1007
20,000–34,999	21.89%	19.25%	0.0904	0.0837	0.0613	0.0594
≥ 35,000	15.01%	12.62%	0.0595	0.0528	0.0405	0.0387
Educational level
No formal school	32.79%	33.09%	0.1762	0.1643	0.1295	0.1207
Primary school	30.75%	29.82%	0.1500	0.1404	0.1074	0.1042
Middle or high school	24.27%	20.60%	0.1074	0.0947	0.0750	0.0696
College or above	16.11%	12.54%	0.0621	0.0466	0.0421	0.0331
Region
Rural residence	31.90%	31.44%	0.1628	0.1576	0.1203	0.1222
Urban residence	22.60%	23.85%	0.0962	0.1068	0.0650	0.0734
Geographic area
Western	35.22%	35.65%	0.1786	0.1822	0.1307	0.1386
Central	33.05%	34.20%	0.1619	0.1646	0.1174	0.1248
Northeastern	26.38%	31.43%	0.1102	0.1367	0.0747	0.0938
Eastern	16.70%	15.72%	0.0703	0.0696	0.0474	0.0489
Year of diagnosis
2004	39.85%	43.56%	0.1987	0.2156	0.1464	0.1629
2005	32.06%	34.03%	0.1579	0.1697	0.1149	0.1289
2006	23.57%	25.82%	0.1056	0.1191	0.0738	0.0858
2007	21.03%	20.50%	0.0904	0.0917	0.0604	0.0645
2008	17.82%	14.82%	0.0771	0.0668	0.0530	0.0477

Geographic differences were also evident. Rural residents consistently faced a heavier poverty burden (: 31–32%) across all measures compared to urban residents (: 22–24%). Regionally, a distinct West–East gradient was evident. The Western and Central areas exhibited the highest deprivation, with values in the Western region surpassing 35% and reaching levels above 0.13. Conversely, the Eastern region reported the lowest indices (: ~ 16%, : ~ 0.05), consistent with persistent structural disadvantages in inland provinces. Table 4 also examines multidimensional health poverty by year of diagnosis. Patients diagnosed more recently demonstrated markedly better outcomes, with dropping from over 40% in the 2004 cohort to under 18% in the 2008 cohort.

Decomposition of multidimensional health poverty

Figure 2 illustrates the dimensional decomposition of the Multidimensional Health Poverty Index () across varying demographic and socioeconomic subgroups. Across almost all strata, the objective health dimension (survival and NCD burden) and subjective self-rated health consistently emerged as the predominant contributing to deprivation, collectively accounting for approximately 80% of the total poverty index.

Fig. 2.

Relative contribution of each indicator at k = 0.3. Notes: By showing which dimensions (survival, NCDs, self‑rated health, and healthcare access capability) dominate the index for each subgroup, this figure helps identify whether policy responses should prioritize improving clinical management, strengthening psychosocial support, or removing access barriers for specific populations

However, patterns of dimensional contributions differed systematically across socioeconomic and demographic gradients. The contribution of healthcare access barriers was most pronounced among vulnerable subgroups specifically females, rural residents, and those in the lowest income and education brackets. For instance, lack of healthcare access contributed 16.09% to total deprivation for the lowest-income group, nearly three times higher than that observed in the highest-income group (5.45%). Conversely, the relative contribution of the NCD burden followed a positive gradient with age and socioeconomic status. This pattern indicates that a larger share of measured deprivation among lower socioeconomic groups is attributable to healthcare access capability, whereas among higher socioeconomic and older groups, a larger share arises from the clinical accumulation of comorbidities. Detailed decomposition percentages for all subgroups and dimensions are visually presented in Fig. 2.

Sensitivity analysis

As a robustness check, we re-estimated all multidimensional health poverty indices after excluding the survival indicator from the index (retaining NCD burden, self-rated health, and healthcare access capability with re-normalized weights). The subgroup rankings and socioeconomic gradients remained broadly consistent with the baseline results (Appendix Table A4-A6). In a further sensitivity analysis, we used observed survival events as the survival indicator. Observed survival proportions were generally close to, but slightly lower than, the corresponding predicted probabilities from the Cox model. Because deprivation in this specification was defined using a binary indicator of survival status rather than the 5% shortfall relative to national life-table survival used in the baseline index, the resulting multidimensional health poverty levels were somewhat lower overall; however, the demographic and socioeconomic subgroup patterns and the main conclusions remained broadly similar to those in the main analysis (Appendix Table A7-A10).

Using the equal-weighting scheme, the recalculated adjusted headcount ratio (M₀) yielded highly consistent patterns compared with the PCA-based results. The relative ranking of demographic and socioeconomic subgroups remained unchanged, and the magnitude and direction of deprivation gradients across income, education, and regional groups were substantively similar. Detailed results from the equal-weighting robustness analysis are presented in Appendix Table A11-A13.

Scenario analysis

Figure 3 and Appendix Table A14-A15 detail the changes in multidimensional health poverty indices and under three counterfactual scenarios.

Fig. 3.

Multidimensional health poverty index across socioeconomic groups under counterfactual scenarios. Note: Panel A presents the projected Adjusted Headcount Ratio () under each hypothetical scenario. Panel B illustrates the relative simulated reductions under these scenarios, expressed as the percentage reduction in compared to the Baseline scenario. The simulation scenarios are defined based on the following hypothetical assumptions: S1 (hypothetical complete abstinence from alcohol among participants who were regular drinkers at baseline); S2 (hypothetical complete smoking cessation among participants who were regular smokers at baseline); and S3 (simultaneous cessation of both smoking and alcohol among participants who were regular smokers and/or drinkers at baseline). These scenarios are illustrative simulations based on the fitted survival model and AF index, and are intended to show the sensitivity of multidimensional health poverty to stylized changes in behavior rather than to provide causal estimates of real‑world policy effects

The behavioral intervention scenarios, which simulated the control of risk factors via the Cox prediction model, yielded modest reductions in multidimensional health poverty, with simulated reductions disproportionately favoring higher socioeconomic groups. Scenario 1 (No Alcohol) resulted in minimal improvements across all strata, with reduction rates generally remaining below 1% (−0.03% for the lowest income group and −0.51% for the highest). Scenario 2 (No Smoking) showed more pronounced simulated reductions, particularly for better-educated individuals. For those with a college education, S2 reduced by 7.30% (from 0.15 to 0.14), whereas the reduction was negligible (−0.31%) for those with no formal schooling. Scenario 3, combining both interventions, amplified this trend. While it reduced by 8.44% and by 8.53% among the college-educated, the reduction for the lowest income group remained marginal (−0.87% for ).

Discussion

This study focuses on the multidimensional health poverty of adults with diabetes, by establishing survival‑anchored multidimensional health poverty indices and identifying the key factors that contribute to this deprivation. Our empirical results highlight that structural mechanisms of deprivation diverge [43, 44]. Barriers to healthcare access appear to disproportionately constrain vulnerable groups, particularly rural residents and those with lower socioeconomic status. Conversely, higher socioeconomic groups face deprivation in which the accumulation of clinical NCDs accounts for a larger share of the multidimensional index. Counterfactual scenario simulations also illustrate that potential policy‑relevant associations are stratified. In these illustrative models, behavioral interventions targeting lifestyle risk factors were associated with larger simulated reductions for wealthier cohorts [45].

Stratified analyses revealed demographic gradients, identifying sex and age as key determinants of multidimensional health poverty severity. A striking gender crossover effect was observed. While men exhibited higher poverty headcount () and severity in early adulthood (30–45 years), women significantly surpassed men in older age groups (> 60 years), particularly in poverty severity (). Dimensional decomposition clarifies this reversal. For young males, deprivation was predominantly accounted for by the survival dimension (accounting for ~ 43% of ), reflecting heightened biological vulnerability and behavioral risks that compromise life expectancy [46]. Conversely, the deepened poverty among older women was driven disproportionately by healthcare access dimension, which contributed nearly four times more to their deprivation compared to their male counterparts. This finding supports the feminization of poverty thesis, suggesting that older women face structural exclusion and economic dependence rather than purely physiological decline [47, 48]. Furthermore, the age-related trajectory followed a U-shaped pattern with a marked deterioration in the 65–74 age bracket, compounded by cumulative NCDs and diminishing self-rated health. Notably, lower deprivation levels in the oldest age group may be attributed to survivor bias, as healthier individuals are more likely to survive into advanced age with better self-care capacity and healthcare access, especially for men [49].

Socioeconomic factors exerted a profound influence on multidimensional health poverty. Higher income and education provided a protective buffer, consistent with established social determinant frameworks [11, 50]. Interestingly, this protective effect was more pronounced for women, suggesting higher efficiency in converting resources into health benefits among high-SES females [51]. Furthermore, while socioeconomic-related health poverty research may not fully capture the deprivation faced by the affluent [21, 49], our dimensional decomposition reveals that high-SES groups are not immune to multidimensional health poverty; rather, they face a distinct form of deprivation driven primarily by adverse health outcomes and the physiological burden of NCDs. Geographic disparities further reinforced structural inequality [52]. While deprivation in the Western region appeared more closely linked to resource constraints [53], particularly affecting women, the Northeast’s burden was driven by NCDs. This may reflect regional epidemiological differences potentially tied to diet and climate. Beyond spatial and socioeconomic factors, disease duration played independent and substantial roles in shaping multidimensional health poverty patterns. A clear disease duration gradient was evident, where patients diagnosed earlier exhibited significantly higher deprivation than those diagnosed recently, illustrating how chronic disease acts as a poverty trap that progressively erodes both physiological resilience and household assets over time [54].

Scenario simulations unravel a distinct dichotomy in policy simulated outcomes. According to the decomposition results, improving healthcare access capability may be particularly relevant for vulnerable populations, because the share of deprivation attributable to the healthcare access dimension is substantially higher among low‑SES groups than among high‑SES groups (Fig. 2). In contrast, behavioral interventions (S1-S3) exhibited a socioeconomic gradient of simulated efficacy. They produced negligible effects for the impoverished but were associated with noticeably larger simulated reductions among college‑educated participants. This pattern is consistent with a possible diminishing marginal return of financial protection for high‑SES groups, whose residual measured deprivation appears more strongly linked to survival‑related and lifestyle‑associated risks than to access barriers. These counterfactual scenarios are illustrative simulations rather than causal estimates, and their comparative magnitudes should be interpreted with caution because the model mechanically updates only the survival component while holding other deprivation dimensions constant.

These findings are consistent with the conclusions of recent research on health system reforms in China. Over the past decade, China has consolidated resident schemes into the Urban and Rural Resident Basic Medical Insurance, increased contribution and reimbursement levels, and built a multi-tier system that links basic insurance, catastrophic coverage and medical assistance [55, 56]. These reforms mainly ease deprivation in healthcare access capability by lowering cost sharing, extending outpatient benefits for chronic diseases and improving protection against catastrophic spending, especially for low-income and rural households [57–59]. In parallel, the National Essential Public Health Service Package and primary care-based chronic disease programs for hypertension and diabetes have expanded regular follow-up, biomarker monitoring and health education [60], thereby strengthening objective health (better control and delayed complications) and subjective health (improved symptom management and perceived support). However, existing evaluations point to persistent urban–rural and socioeconomic gaps in coverage and benefit incidence, suggesting that these reforms have not yet fully reached those who are most health‑poor. Our multidimensional framework helps identify which dimensions remain most deprived in which subgroups, and thus where incremental policy effort may yield the greatest equity gains.

Methodologically, this study contributes a transferable analytical framework. By embedding survival-based deprivation and counterfactual scenarios within the AF structure, we propose a model that may map which policy might alleviate which dimension of poverty, and for whom. This approach can help bridge the gap between deprivation diagnostics and simulated policy responses, extending multidimensional poverty measurement from merely monitoring inequality to better inform targeted intervention design, thereby providing empirical support for dimension-sensitive and equity-oriented policy formulation.

Despite its contributions, several limitations merit consideration. First, although survival probabilities represent robust proxies for health outcomes, our estimates rely on predicted rather than directly observed survival durations, potentially masking subgroup heterogeneity. Future studies may benefit from incorporating stochastic simulations or machine learning methods to refine individual-level survival prediction [61, 62]. Second, the operationalization of the healthcare access dimension relied on a composite proxy deriving from the interaction of residence and insurance status. While this captures the intersectionality of geography and financial protection, it may oversimplify other nuances of accessibility. Third, although the CKB provides valuable longitudinal data, follow-up assessments represent spot-checks of baseline participants, with a relatively modest sample of diabetic individuals, thus limiting comprehensive capture of time-dependent effects. Finally, counterfactual Scenarios serve as illustrative simulations and likely underestimate the benefits of behavioral interventions. In our model, lifestyle changes only improved survival probability without mathematically updating self-rated health or NCD dimensions [63, 64]. While counterfactual scenarios demonstrate strong associations, they do not establish causality. Future research might utilize causal inference approaches to precisely quantify how specific policies influence survival-related health poverty outcomes.

Conclusion

This study reveals substantial survival-based multidimensional health poverty among Chinese diabetic adults. While deprivation among vulnerable groups is driven by structural barriers, higher socioeconomic groups also face poverty risks stemming from lifestyle-related clinical burdens. Decomposition results indicate that the healthcare access dimension contributes disproportionately to deprivation among low-SES groups, highlighting financial protection and access barriers as priority targets for equity-oriented intervention. Illustrative behavioral scenarios (S1-S3) suggest comparatively larger simulated reductions among higher-SES cohorts, consistent with their deprivation profile being more closely linked to survival-related and lifestyle-associated risks. These findings advocate for a stratified policy approach, prioritizing financial protection for the underserved while promoting behavioral changes among higher socioeconomic groups.

Abbreviations

SES

Socioeconomic status

AF

Alkire-Foster

CKB

China Kadoorie Biobank

FGT

Foster-Greer-Thorbecke

CHD

Coronary heart disease

TIA

Transient ischemic attack

Authors’ contributions

Y.M.Z. and Y.L. contributed to the study design. Y.Y.T. and M.Y.Z. were responsible for the statistical analyses and drafted the initial version of the manuscript. Y.W.B. and F.C. critically reviewed the manuscript and ensured the accuracy of the data analysis. All authors read and approved the final manuscript.

Funding

This study was supported by the National Natural Science Foundation of China under Grant 72004231 and the 2023–2024 Open Fund of the Key Laboratory of Health Technology Assessment of the National Health and Wellness Commission.

Funding for the development of and maintenance of the CKB resource has been received from Chinese Ministry of Science and Technology, Chinese National Natural Science Foundation, the Kadoorie Charitable Foundation in Hong Kong and the Wellcome Trust. Unless specified elsewhere, these funders have not supported the research work required for the preparation of this paper.

Data availability

The data that support the findings of this study are available from the China Kadoorie Biobank but restrictions apply to the availability of these data, which were used under license for the current study, and so are not publicly available.

Declarations

Ethics approval and consent to participate

CKB received ethical approvals from the Ethical Review Committee of the Chinese Center for Disease Control and Prevention (approval number 005/2004) and the Oxford Tropical Research Ethics Committee at the University of Oxford (approval number 025–04). All participants provided written informed consent. The study was conducted in accordance with the Declaration of Helsinki. All methods were carried out in accordance with relevant guidelines and regulations.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.

This research has been conducted using the China Kadoorie Biobank resource (www.ckbiobank.org, Request No. DAR-2024-00008). Publication of results does not require or imply approval by the membership of the CKB Collaborative Group.