Long-term care insurance pilots and a severe care-gap proxy among older adults in China: evidence from CHARLS, 2011–2020

Abstract

Background

China’s long-term care insurance (LTCI) pilots aim to support older adults, but evidence on whether policy rollout is followed by reduced care gaps among functionally limited older adults remains limited.

Methods

Using CHARLS 2011–2020, we included respondents aged 60 years or older at first observation and estimated a staggered difference-in-differences model comparing early-adopter pilot cities with never-pilot cities. The primary outcome was a severe care-gap proxy, defined as ADL/IADL difficulty with no help. This estimand describes care-gap trajectories among older adults with functional limitations, not effects among confirmed LTCI beneficiaries or formally eligible disabled older adults. We also examined whether estimates differed according to whether local policy texts explicitly included in-kind home- and community-based services (HCBS), including home-visit provision.

Results

In early-adopter pilot areas, adoption was associated with a higher severe care-gap proxy by 2020 (ATT 0.023; 95% CI 0.001–0.045). The proxy showed high specificity and positive predictive value but limited sensitivity against 2020 activity-level unmet-need items. Estimates were directionally larger in cities without explicit in-kind HCBS wording, and formal/community home-based care did not measurably expand over the same period.

Conclusions

The observed 2020 pattern should be interpreted cautiously because pandemic disruption, measurement differences, attrition, and compositional shifts may have contributed. The findings support nursing-centered monitoring of severe care-gap indicators during policy rollout and more explicit attention to service activation pathways.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s12877-026-07960-3.

Background

Disability and functional decline create sustained needs for assistance with activities of daily living (ADL) and instrumental activities of daily living (IADL), and population-level disability trajectories shape the potential scale of long-term care demand [1]. In practice, care is often provided by family members, supplemented by community services and paid care when available. Inadequate care access is not only a social concern but also a nursing-relevant patient outcome: lack of assistance can elevate risk of falls, pressure injuries, malnutrition and delayed response to acute events, and has been associated with adverse outcomes and preventable health service use [2–5].

Unmet need for basic assistance has been linked to worse outcomes, including mortality among older adults with disability, underscoring its relevance as a patient-safety endpoint [6].

China has introduced long-term care insurance (LTCI) pilots to strengthen financial protection and organize long-term care delivery in the context of a rapidly evolving long-term care system [7]. In practice, these pilots were implemented with substantial local variation. Benefit design can differ in whether support is delivered as cash, institutional coverage, or in-kind services that enable home-visit and community-based care, and fragmentation across pilot designs has been highlighted as a core implementation challenge [8, 9]. For community and long-term-care nursing, the key question is whether insurance expansion translates into service entry and continuity of care, especially for those already experiencing functional limitations.

China launched LTCI pilots in 2016 and expanded pilots in 2020, providing a natural setting to evaluate whether system reforms are followed by reduced severe care-gap indicators and whether design features that support service delivery matter [10, 11].

We framed LTCI as a system-level intervention that changes the structure of financing and eligibility, which should in turn shape care processes such as timely service entry, home-visit/community care availability, and shared-care arrangements, and ultimately influence patient-centered outcomes [12]. In this study we focus on a severe care-gap proxy: functional limitation accompanied by no reported help. The term severe refers to the absence of any help despite reported ADL/IADL difficulty, not to formal certification of severe disability.

Prior evaluations frequently emphasize health outcomes or caregiving burden [13–15]. More recent work has widened the lens to wellbeing, care arrangements, and economic consequences [16–23]. Even so, relatively few studies treat care gaps themselves as the primary endpoint or ask whether policy design features shape access to patient-centered care.

Studies in China and East Asia have documented unmet assistance needs and care poverty among older adults with disability [24–26]. Related work has described unmet needs among older adults with multimorbidity and functional impairment [27], gaps in enabling resources such as bathing and toileting equipment [28], and broader concerns about unmet social care needs and need triage in home care contexts [29, 30].

We address this gap by using longitudinal survey data and a staggered DiD design to examine changes in a severe care-gap proxy and to test whether explicit in-kind HCBS inclusion is associated with more favorable trajectories. Specifically, we asked whether LTCI pilot adoption was associated with changes in the severe care-gap proxy over time, whether policies that explicitly included in-kind HCBS home-visit services were associated with directionally smaller increases, and whether care receipt patterns supported a cautious care-transition-gap interpretation.

Methods

Study design and reporting

This was a longitudinal observational policy evaluation using repeated individual observations across five CHARLS waves. We report the study in line with STROBE principles for observational analyses.

Data source and study population

We used CHARLS data collected in 2011, 2013, 2015, 2018 and 2020. CHARLS is a nationally representative longitudinal survey of middle-aged and older adults in China that uses multistage stratified probability sampling with periodic follow-up [31]. We analyzed repeated observations at the individual level and allowed an unbalanced panel. Variable harmonization and cross-wave comparability were supported by the Harmonized CHARLS resources developed by the Gateway to Global Aging Data [32].

We restricted the analytic population to respondents aged 60 years or older at their first observed wave.

Policy exposure

LTCI pilot exposure was assigned based on respondents’ city and the pilot adoption timing. Because CHARLS is fielded in discrete survey waves, we treated a city as exposed starting in the first wave conducted after adoption and defined event time in survey-wave increments. Cities becoming operational after 2020 were treated as unexposed within the 2011–2020 analysis window, and late-2020 starts were also coded unexposed when the respondent’s interview occurred before benefit activation. Some large cities implemented partial pilots, for example at district level; our primary analysis excluded partial pilots, with sensitivity analyses reported in Additional file 1.

Outcomes

Primary outcome: severe care-gap proxy

We defined the severe care-gap proxy as having ADL/IADL difficulty with no help, using a harmonized measure available across waves. We use the term severe because the proxy requires no reported help despite functional difficulty; it should not be read as formal severe-disability certification or confirmed LTCI eligibility. The definition prioritizes specificity for clear care gaps and supports comparable estimation across waves (Fig. 1; Additional file 2). The resulting care-needs population is broader than formal LTCI eligibility, which varies across cities and often requires more severe disability than can be identified from a wave-consistent survey proxy.

Fig. 1.

Dynamic event-study estimates of the association between LTCI pilot adoption and the severe care-gap proxy among older adults (aged 60 years or older at their first observed wave; early-adopter cohort plus never-treated controls; n = 13,774 unique respondents contributing 47,487 person-wave observations). Estimates are group-time ATT aggregated by event time using a staggered-adoption DiD approach (Callaway and Sant’Anna) with baseline covariate adjustment (age, sex, education, hukou) and city-clustered bootstrap standard errors. Partial pilots excluded. Shaded bands are 95% confidence intervals. Source data in Additional file 2 are restricted to the early-adopter cohort used for primary inference. (Additional file 2)

To clarify the estimand, the primary outcome describes severe care-gap-proxy trajectories among older adults aged 60 years or older with functional limitations in adopting cities, not treatment effects among confirmed LTCI beneficiaries. The estimates therefore indicate whether LTCI pilot adoption coincided with changes in severe care gaps among older adults with care needs, rather than whether LTCI succeeded or failed for those who passed formal assessment.

Care-receipt pathway outcomes

Among respondents with care needs, we assessed changes in receipt of any help, any informal care, any formal institutional care, paid/community home-based care, and an indicator for any formal-or-community care. For clarity, Fig. 4 summarizes these care-receipt pathway estimates for the early-adopter cohort (Additional files 4 and 5).

Fig. 4.

Care-receipt pathway evidence among respondents with care needs (early-adopter cohort): event-study estimates for care receipt pathways. Panels A-E show any help received, any informal care, any institutional formal care, any paid/community home-based care, and any formal-or-community care. Same approach as Fig. 1, restricted to the care-needs subgroup (ADL/IADL difficulty; n = 7,082 unique respondents), with partial pilots excluded. (Additional files 4 and 5)

2020 cross-validation outcome

In 2020, CHARLS provides activity-level unmet-need items. We used these to cross-validate the severe care-gap proxy (Additional file 7).

Policy design heterogeneity: explicit in-kind HCBS inclusion

We coded a binary indicator for whether local policy documents explicitly stated in-kind HCBS home-visit service provision. Coding followed a prespecified rule set covering source-priority hierarchy, explicit inclusion/exclusion criteria, and ambiguity handling, documented in Appendix A in Additional file 1 and supported by clause-level evidence and a city registry (Table S5; Additional file 13). Local policy documents used for coding and exposure timing are listed in Table S5.

Statistical analysis

We used a staggered-adoption difference-in-differences approach to estimate cohort- and time-specific effects and to aggregate them into dynamic event-study effects [33]. This approach is recommended under staggered adoption and heterogeneous effects, where conventional two-way fixed effects event-study specifications can be biased [34, 35]. Never-treated units served as controls within each cohort-specific comparison.

Treatment timing is defined at the city level. We aligned city adoption years with the CHARLS interview waves (2011, 2013, 2015, 2018, 2020) and defined exposure from the first post-adoption wave observed in CHARLS. In the analytic city pool after excluding partial pilots, 19 treated cities were first observed post-adoption in 2018 and one treated city was first observed post-adoption in 2020 (Guangzhou) (Table S8; Additional file 16). Because the 2020 cohort is sparse in CHARLS, our primary inference focuses on the early-adopter cohort; the late-cohort estimate is presented as descriptive.

Uncertainty was quantified with city-clustered bootstrap standard errors (300 iterations) and pointwise 95% confidence intervals. Baseline covariates were handled with prespecified rules, including assignment of missing binary indicators to the reference category and sample-median imputation for education. Estimates are reported as average treatment effects on the treated. We also report robustness checks including negative control outcomes and formal tests of HCBS effect modification [36].

Bias and missing data

Potential biases include wave-to-wave measurement non-equivalence, particularly in 2020, pandemic-related disruption to home-visit and community services, attrition or compositional change over time, and exposure misclassification due to within-wave policy start months. We addressed these risks through wave-consistent outcome definitions, pre-trend checks, sensitivity analyses (Tables S1-S7 in Additional file 1), and a 2020 cross-validation against activity-level unmet-need items. Baseline covariate missingness was low for age, sex and hukou in the analysis sample, but education missingness was substantial at approximately 20% overall, and we therefore used prespecified median imputation for education and report robustness to alternative covariate sets (Table S1 in Additional file 1).

Results

Sample and descriptive context

Starting from the full CHARLS panel (25,941 unique respondents across all ages and waves), we retained 14,085 unique older adults aged 60 years or older at their first observed wave, contributing 48,428 person-wave observations across five survey waves. The primary event-study specification excluded partial pilots and focused on the early-adopter cohort plus never-treated controls, leaving 13,774 unique older adults and 47,487 person-wave observations across 123 cities (19 treated; 104 never treated); the corresponding care-needs subgroup (ADL/IADL difficulty) comprised 7,082 unique respondents. The mean baseline age in this primary analytic sample was 67.7 years (SD 7.0). Table 1 summarizes the wave-specific profile of this primary analytic sample.

Table 1.

Wave-specific profile of the primary analytic sample (adults aged 60 years or older; early-adopter cohort plus never-treated controls; partial pilots excluded

Characteristic	Overall	2011	2013	2015	2018	2020
Unique respondents in panel, n	13,774
Person-wave observations, n	47,487	9,142	9,143	9,630	9,110	10,462
Cities represented, n	123	123	123	123	123	122
Respondents with care needs, n	7,082	1,319	3,303	3,545	3,355	2,692
Care need present in full sample, %	29.9	14.4	36.1	36.8	36.8	25.7
Severe care-gap proxy in full sample, %	7.6	5.4	9.2	10.1	9.7	3.9
Any help among those with care needs, %	76.9	91.3	74.5	72.6	73.6	84.8
Any informal care among those with care needs, %	59.9	44.5	91.9	43.4	56.7	53.4
Any paid/community home-based care among those with care needs, %	2.6	1.1	4.6	1.3	4.2	0.3
Any formal-or-community care among those with care needs, %	3.0	1.1	4.8	2.0	5.2	0.4

Percentages in the last four rows use respondents with care needs as the denominator

As a negative-control check for differential compositional change, a rural residence indicator showed a small but statistically detectable shift (ATT 0.014; 95% CI 0.004 to 0.024; Table S4 in Additional file 1; Additional file 12). This pattern is compatible with some compositional change between waves and reinforces the need for cautious interpretation of 2020 estimates; the primary models adjust for baseline hukou type.

Main finding: the severe care-gap proxy was higher by 2020

In early-adopter pilot areas, LTCI adoption was associated with a higher severe care-gap proxy by 2020 (ATT 0.023; 95% CI 0.001 to 0.045; Fig. 1; Additional file 2). Pre-policy estimates at 5 and 3 years before adoption were near zero and statistically compatible with no difference (event time − 5: ATT − 0.003; 95% CI -0.025 to 0.019; event time − 3: ATT 0.005; 95% CI -0.018 to 0.027; Fig. 1). Excluding cities with late-2020 operational timing concerns yielded a nearly identical point estimate (ATT 0.024; 95% CI -0.000 to 0.047; Table S6 in Additional file 1).

On an absolute scale, an ATT of 0.023 corresponds to approximately 23 additional older adults per 1,000 classified as having a severe care gap in the post-adoption period.

Descriptively among respondents with care needs, the share classified as unmet/no care increased between 2018 and 2020 (Fig. 2; Additional file 6). Figure 2 is a descriptive care-structure display based on the age-eligible care-needs sample and therefore does not apply the partial-pilot exclusion used in the primary event-study models. This descriptive shift provides context but is not, by itself, causal evidence.

Fig. 2.

Care structure composition among respondents with care needs (ADL/IADL difficulty) across CHARLS waves. Bars show the share of respondents whose primary care source is classified as institution, formal/community home-based care, children, spouse, relatives, or unmet/no care. Descriptive sample restricted to adults aged 60 years or older at their first observed wave. Unlike the primary event-study models, this descriptive figure does not exclude partial pilots. (Additional file 6)

Heterogeneity: directionally smaller increases in cities with explicit in-kind HCBS wording

The increase in the severe care-gap proxy was larger in cities without explicit in-kind HCBS home-visit inclusion (ATT 0.037; 95% CI 0.014 to 0.059) than in cities with explicit inclusion (ATT 0.013; 95% CI -0.021 to 0.046; Fig. 3; Additional file 3). Formal contrast testing did not show statistically conclusive effect modification (difference 0.024; 95% CI -0.016 to 0.064; p = 0.239, conservative approximation; Table S6 in Additional file 1), and interaction-model results were directionally consistent but non-significant (Table S7 in Additional file 1). Because this comparison is based on policy-text design features, it should be interpreted as design-relevant evidence rather than a direct measure of implementation intensity.

Fig. 3.

Heterogeneity by benefit design: dynamic event-study estimates stratified by whether local policy texts explicitly include in-kind HCBS home-visit services (0/1). Same approach as Fig. 1, with partial pilots excluded. Interpretation is design-relevant policy-text evidence rather than implementation intensity. Analytic population: adults aged 60 years or older at their first observed wave in the early-adopter cohort plus never-treated controls (n = 13,774 unique respondents). (Additional file 3)

Care-receipt pathway evidence: reduced help receipt without increased community home-based care

In the early-adopter cohort, among respondents with care needs, we did not observe a corresponding increase in paid/community home-based care (ATT − 0.002; 95% CI -0.013 to 0.008; Fig. 4; Additional file 5). Formal institutional care showed at most small changes, while help receipt estimates were directionally negative but imprecise (Fig. 4; Additional file 4). Read together, these estimates indicate that the severe care-gap proxy increased during a period in which formal/community home-based care did not measurably expand; they do not directly identify assessment delays, provider capacity, service activation, or substitution between family and formal care.

Cross-validation and high-need sensitivity: interpreting the proxy

In 2020, the severe care-gap proxy showed specificity and positive predictive value of 1.00 against activity-level unmet-need items, while sensitivity was 0.45 (Additional file 7). This pattern means that observations classified by the proxy were also classified as unmet on activity-level items, but the proxy missed many activity-level unmet-need cases. The measure is therefore best interpreted as a high-specificity severe care-gap proxy rather than a complete measure of all unmet need.

The high-need sensitivity analyses further clarify the estimand. Among respondents with two or more baseline ADL limitations, a subgroup closer to typical LTCI eligibility thresholds, the 2020 estimate was directionally negative but imprecise (ATT − 0.035; 95% CI -0.089 to 0.019; Table S10; Additional file 18). A broader baseline ADL/IADL threshold produced an imprecise positive estimate (ATT 0.025; 95% CI -0.111 to 0.161; Table S10; Additional file 18). These analyses do not overturn the main population-level estimate, but they caution against generalizing the main result to formally eligible beneficiaries or those with the highest baseline disability.

Discussion

A severe care-gap proxy is a patient-safety endpoint in community and long-term-care nursing because it reflects functional limitation without reported assistance. In this national longitudinal evaluation, LTCI pilot adoption in early-adopter areas was associated with a higher severe care-gap proxy by 2020. This population-level pattern contrasts with high-need sensitivity analyses restricted to respondents with two or more baseline ADL limitations, where the point estimate was directionally negative (Table S10; Additional file 18). Care-receipt pathway analyses showed no measurable expansion of paid or community home-based care, while help receipt estimates were directionally negative. In combination, these findings are compatible with a potential gap between financing expansion and the local delivery of care, while acknowledging that the underlying pathway cannot be observed directly in the present data.

One plausible interpretation is that financing reform altered care arrangements more quickly than local systems could expand home-visit and community-based services at meaningful scale. In that reading, coverage may have been introduced before the service infrastructure needed to translate entitlement into actual care was fully in place. That same pattern could also reflect pandemic-related interruption of home visits and community services in 2020, or informal-care rearrangement in which family members reduced input more quickly than formal services replaced it. Alternative explanations therefore remain plausible, including measurement non-equivalence in 2020 modules, compositional change in high-need survivors between 2018 and 2020, and temporary disruption around eligibility assessment and benefit activation.

This interpretation aligns with established transitional care and care-transition frameworks that emphasize continuity, shared accountability and service entry during vulnerable handoffs [37–42].

Implications for nursing and service planning

From a community and long-term-care nursing perspective, the severe care-gap proxy is a patient-safety endpoint that can reveal service shortfalls not captured by broader health indicators. The directionally smaller increase seen in cities with explicit in-kind HCBS home-visit provisions is compatible with the idea that clearer service-entry language may matter, but the subgroup contrasts were not statistically conclusive and should not be read as proof of effect modification. Even so, the pattern is consistent with evidence that outcomes can differ across home- and community-based services versus institutional pathways, that structured home-based support programs can improve function, and that the home-care workforce has not kept pace with growth in HCBS demand [43–46].

The main nursing implication is not that this study tested a specific nurse-led intervention model, but that severe care-gap monitoring may serve as a practical geriatric quality signal: an older adult with ADL/IADL difficulty and no help is a person for whom risk review, interim safety planning, and referral checks may be warranted. In community nursing workflows, this indicator could trigger reassessment of functional limitations, caregiver availability, fall and pressure-injury risk, nutrition and medication support, and referral to LTCI assessment or contracted community providers when formal activation is delayed.

The policy implications extend across financing, commissioning, and delivery. At minimum, financing reforms need service pathways that are explicit, monitorable, and locally deliverable. Local governments can monitor the interval from eligibility approval to service activation and strengthen provider networks where delays persist. Provider organizations can use unresolved care-gap indicators to identify settings where formal entitlement is not yet translating into timely assistance at home, and to prioritize outreach for older adults whose needs are visible before benefits are fully activated.

These service-planning priorities align with integrated community care guidance for older people that emphasizes early identification, tailored care plans and functional support in community settings [47]. They are also consistent with nursing outcome frameworks that link structural inputs and care processes to patient-centered outcomes [48], and with dementia care planning work emphasizing continuity across community, institutional and health-care settings [49, 50].

Strengths and limitations

Key strengths include longitudinal national survey data spanning a decade, a staggered DiD event-study design with transparent presentation of dynamic effects, policy-text coding for design heterogeneity supported by clause-level evidence, care-receipt pathway outcomes, and 2020 cross-validation against activity-level unmet-need items.

Limitations cluster around the 2020 wave. CHARLS instruments and module structures differ from earlier waves, and some measurement non-equivalence may remain despite harmonized cross-wave definitions and cross-validation. The high specificity and positive predictive value of the 2020 proxy support interpretation as a clear care-gap indicator, but the limited sensitivity means that it undercounts broader activity-level unmet need. In addition, 2020 was the first pandemic year, so reduced availability of home visits, community services, household mobility, or informal-care arrangements could have amplified care gaps independently of LTCI design. Attrition and survival-related changes may also have altered the observed composition of functionally limited older adults between 2018 and 2020.

Other limitations concern policy alignment and observability. Treatment timing is defined at the survey-wave level and cannot fully resolve within-year operational start months for all cities; we address partial pilots and late starts through exclusion and sensitivity strategies. Our survey-based care-need definition is broader than formal LTCI eligibility, which often requires more severe disability and locally specific assessment thresholds; the estimates therefore describe care-gap trajectories among older adults with care needs rather than treatment effects among confirmed beneficiaries. We do not directly observe expectation shifts in family caregiving, assessment delays, or service activation intervals, so the service-activation interpretation is inferential rather than directly measured. Compositional change among high-need older adults between waves, including differential mortality or survival, cannot be fully ruled out; this concern is reinforced by a small but statistically detectable shift in one negative-control composition measure (rural residence). Policy documents characterize design features but do not measure implementation intensity or service quality, so heterogeneity results should be interpreted as design-relevant rather than definitive implementation evaluation; we therefore report a structured set of robustness checks in Additional file 1.

Finally, LTCI pilot adoption and its sequencing were not randomly assigned. While the pre-policy estimates are statistically compatible with no difference in our primary event-study, we cannot fully rule out residual confounding from time-varying city-level shocks correlated with adoption timing (e.g., differential service infrastructure growth or local governance capacity).

High-need sensitivity. In sensitivity analyses restricted to respondents with two or more baseline ADL limitations, a closer proxy to typical LTCI eligibility thresholds, the 2020 point estimate was directionally negative (ATT − 0.035; 95% CI -0.089 to 0.019) but imprecise due to reduced sample size (Table S10; Additional file 18). While not statistically conclusive, this pattern is compatible with more favorable trajectories among those closest to formal eligibility, alongside broader population-level care-gap dynamics captured by the main specification. This is why we interpret the main finding as a care-gap trajectory among functionally limited older adults rather than a beneficiary-specific estimate.

Conclusions

Across 2011–2020, LTCI pilot adoption in early-adopter pilot areas was associated with a higher severe care-gap proxy and directionally lower help receipt among those with care needs, without clear evidence of expanded community home-based care. The findings support nursing-centered evaluation of policy reforms using care-gap endpoints and underscore the need to align financing changes with service activation, while interpreting 2020 estimates in the context of pandemic disruption and variable local eligibility rules.

Abbreviations

ADL

Activities of daily living

ATT

Average treatment effect on the treated

CHARLS

China Health and Retirement Longitudinal Study

CS-DiD

Callaway and Sant’Anna difference-in-differences estimator

HCBS

Home- and community-based services

IADL

Instrumental activities of daily living

LTCI

Long-term care insurance

Authors' contributions

HZ and WC conceived the study. HZ curated the data, performed the formal analysis, and drafted the manuscript. WC supervised the study and critically revised the manuscript. Both authors interpreted the data, approved the final manuscript, and agree to be accountable for all aspects of the work.

Funding

This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.

Data availability

The CHARLS data used in this study are available to qualified users upon registration through the official project website. Analysis-derived source data underlying the figures and supplementary tables are provided with this submission as Additional files 2-18. Analysis code and a public reproducibility package are available via GitHub (https://github.com/Zhenghongwei11/charls-ltci-care-gap-repro) and archived on Zenodo ( https://doi.org/10.5281/zenodo.21502530). To aid verification, we also provide a claim-level effect-size ledger linking manuscript-reported estimates to their corresponding source-data files (Additional file 9).

Declarations

Ethics approval and consent to participate

CHARLS was approved by the Peking University Institutional Review Board (main survey IRB00001052-11015; biomarker collection IRB00001052-11014), and all respondents provided informed consent [31]. The study procedures were conducted in accordance with the Declaration of Helsinki and relevant institutional/national ethical standards. This study is a secondary analysis of de-identified public-use data.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.