From midlife emergence to old-age divergence: an empirical study on structural gendered ageism in health service utilization in Tibet Authors: Changli Jia, Weizhuo Chen, Xiang Zhang Published: 2025 ABSTRACT AbstractBackgroundPopulation aging on the Tibetan Plateau may marginalize older women through combined gender and age bias. We measured sex-specific gaps in medical spending and service use across the life course. We tested whether gendered ageism appears in midlife and grows thereafter.MethodsWe analyzed 1,274,491 person-year health insurance reimbursement claims data in three major Tibetan cities (2021–2023). We examined four outcomes: annual total medical expenditure, outpatient visits, inpatient admissions, and tertiary-hospital admissions. The key exposure was sex. Age was divided into seventeen five-year groups. We applied LASSO to prevent multicollinearity. We then estimated two-way fixed-effects models. These models controlled for Andersen’s predisposing, enabling, and need covariates. We clustered errors at the person level.ResultsWomen’s spending gap appeared at ages 36–40 (β = − 299; 95% CI: − 425 to − 123). This gap represented roughly an 8% lower annual expenditure compared with men. The maximum deficit occurred at ages 71–75 (β = − 1,006; 95% CI: − 1,401 to − 610), with women spending 28% less than men. Women’s outpatient advantage peaked at ages 31–35 (β = 0.58; 95% CI: 0.54 to 0.61). This advantage faded after age 70. In the inpatient admissions model, women first showed a significant advantage at age 31–35 (β = 0.02 admissions). From age 46–50 through 61–65, this advantage ranged from 0.03 to 0.05 admissions. From age 36–40 onward, women had fewer tertiary-hospital admissions than men (β = − 0.01). This deficit persisted across all later age bands.ConclusionsGendered ageism emerges 30 years earlier than the conventional “65-plus” threshold. It accelerates through midlife and peaks in late old age among Tibetan women. Early policy action could interrupt this cumulative inequity. We recommend higher URRBMI reimbursement, “Plateau-Women-First” telemedicine channels, and flexible delayed retirement with pension top-ups. These measures would support gender-sensitive healthy aging.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12939-025-02733-7.Keywords: Gendered ageism, Life-course inequality, Medical expenditure, Health-service utilisation, High-altitude regions Introduction Population ageing creates major equity challenges for health systems worldwide. By 2030, more than 1.4 billion people will be aged 60 years or older, and women will account for over 55% of this group [1]. This shift is likely to intensify pressures on already unequal systems, widening gaps in access to care and in the allocation of resources. A central concern is gendered ageism, the intersection of gender and age that systematically marginalizes older women within health systems [2, 3]. Compared with ageism or sexism considered separately, gendered ageism is often more subtle and persistent. In health-care settings it manifests as the underrecognition and devaluation of older women’s health needs within families and communities [4], and as implicit gender- and age-related biases in service delivery that diminish attention and resources devoted to older women [5, 6]. The cumulative consequence is reduced opportunities for examinations, diagnostic evaluation, pharmacotherapy, and hospitalization among older women [7, 8]. This bias impedes progress toward healthy ageing and warrants targeted policy and practice responses. The life-course cumulative inequality framework, also called cumulative disadvantage theory, provides a basis for understanding the widening gender gap in health care use with advancing age. The theory posits that initial advantages or disadvantages associated with one’s structural position compound over time across the life course, producing systematic differences in outcomes between groups [9]. Applied to gendered ageism, it implies that disadvantages experienced by women accumulate with age and ultimately manifest as pronounced gender gaps in midlife and older adulthood [10–12]. In adolescence, girls often fare worse on nutritional and other health indicators and have lower uptake of nonroutine vaccinations that fall outside national immunization programmes, which can translate into avoidable morbidity and cost burdens later in life [13, 14]. In midlife, international evidence shows that women are more likely to shoulder dual caregiving responsibilities, leading to downward trajectories in mental and physical health [15], and they occupy a less protected position within employer-sponsored health insurance arrangements [16, 17]. The cumulative effect is that older women experience more years lived with morbidity and higher years lived with disability [18]. Empirical evidence documenting gendered ageism in health service utilization continues to accumulate. Multicountry studies show that, even after adjusting for health needs, older women have persistently lower hospitalization rates and use higher-level services less frequently than men [4, 19, 20]. Evidence from China based on the China Health and Retirement Longitudinal Study likewise indicates that women face greater barriers to care and report poorer self-rated health [21]. However, much of the literature juxtaposes “old age” with sexism and typically sets the age threshold at 65 years, an approach that obscures the earlier emergence of disparities in midlife and their cumulative amplification across the life course [22]. To address this gap, we provide large-scale empirical evidence from the Tibet Autonomous Region of China. In Tibet, access to health services faces multiple barriers; high altitude, dispersed pastoral settlements, and distinctive cultural practices further constrain the availability and use of medical resources [23]. Existing studies on Tibetan populations largely rely on small-sample surveys, focus on urban–rural or socioeconomic disparities, and typically include sex only as a control variable. Few adopt a structural perspective on gendered ageism to examine how intersecting biases shape women’s health care use across different age groups, and analyses based on large-scale medical insurance claims remain limited [24, 25]. China’s basic health insurance comprises two schemes: Urban Employee Basic Medical Insurance (UEBMI) and Urban–Rural Resident Basic Medical Insurance (URRBMI). Coverage rose from 13.4% in 2003 to 94.6% in 2023. UEBMI is mandatory and payroll-financed; URRBMI is government-subsidized and was formed in 2016 by merging URBMI with the New Rural Cooperative Medical Scheme. UEBMI operates a pooled fund plus individual accounts, whereas URRBMI relies on fiscal transfers; pooling is mainly prefecture-level for UEBMI and county or prefecture-level for URRBMI. Both cover medically necessary inpatient and outpatient care, essential medicines, and selected diagnostics and rehabilitation; most population-wide preventive services are funded via public-health programs. Maternity: routine antenatal care and uncomplicated delivery are financed under UEBMI’s maternity benefit or local URRBMI subsidies, while complications are covered by basic insurance. In-network inpatient care is settled at discharge with deductibles and coinsurance for beneficiaries; outpatient care typically has lower cost sharing; catastrophic coverage applies above thresholds; cross-county use often entails higher cost sharing. Benefit generosity varies by county and year [26–28]. Commercial health insurance remains small, about RMB 360 billion in payouts in 2022, roughly 4% of national health expenditure [29]. These features justify a claims-based approach: claims data are representative, longitudinal, and standardized, capturing utilization patterns and structural inequities across residence, sex, and employment. Guided by the life-course cumulative inequality framework, we posit that gender and age shape realized utilization through differences in health need. We therefore characterize the sex-by-age profile of utilization within China’s basic medical insurance system and assess whether gaps emerge in midlife and widen with age. Using claims for beneficiaries from three major cities in the Tibet Autonomous Region during 2021–2023, the study addresses an evidence gap for ethnic-minority and high-altitude populations. Methods Data sources and study population This study uses de-identified health insurance reimbursement data from three major cities in the Tibet Autonomous Region, covering the period from January 1, 2021 to December 31, 2023. Tibet is the principal homeland of the Tibetan ethnic group, the 2020 national census reported that 90% of residents are Tibetan, 8% Han, and 2% other groups [30]. The three study cities account for 55.4% of Tibet’s total population (2022). The database combines records from two national insurance schemes. These schemes are the UEBMI and the URRBMI. Therefore, it is the most comprehensive administrative health dataset for China’s high-altitude minority regions. Each claim record contains an encrypted personal identifier. It also records sex, birth year, ethnicity, and residence type. The database notes service date and outpatient/inpatient status. It captures the principal diagnosis, medical expenditure, and surgical code. It also includes employment status, civilservant status, poverty status, flexible employment status, insurance type and facility location. Pharmacy reimbursement data are not included. We extracted all outpatient and inpatient claims from 2021 to 2023. The dataset included 6,673,256 encounters. We kept only records for Tibetan ethnicity. This reduced the sample to 4,599,849 encounters. We excluded records missing outcome, exposure, or covariate information. For each calendar year, we aggregated encounters by person ID. We summed numeric variables: total cost, outpatient visits, inpatient admissions, and tertiary-hospital admissions. We kept the first recorded value for categorical variables, such as sex, employment status, and residence type. The final analytic dataset contains 1,274,491 person-years. Study variables Outcome measures We analyzed four annual outcomes.The primary outcome was total medical expenditure (TME). TME was the sum of all inpatient and outpatient costs incurred within a calendar year [31]. We trimmed values above the 99th percentile. We then converted 2021 and 2022 amounts to 2023 constant yuan using the consumer-price index. We used raw yuan values rather than logarithmic values in regression. First, the large sample size ensures that the central-limit theorem provides consistent OLS estimates despite right skew. Second, log-transforms can introduce re-transformation bias and reverse coefficient signs [32]. Third, coefficients expressed in original currency units are easier to interpret. The second outcome was outpatient visits. Outpatient visits counted the number of clinic encounters per year. This count served as an indicator of service demand. The third outcome was inpatient admissions. Inpatient admissions counted the number of hospital stays per year. This count served as an indicator of service utilization. The fourth outcome was tertiary-hospital admissions. Tertiary hospitals in China are located in larger cities and provide specialized care for complex conditions. Tertiary-hospital admissions counted the number of admissions to level-III hospitals per year. This measure reflected access to high-technology care and served as a potential mediator of cost disparities. Exposure The primary explanatory variable is sex, coded as 0 for male and 1 for female. Consequently, the regression coefficients represent the difference between females and males. Categorical age variable Age was calculated using each individual’s birth year and service year. We divided age into seventeen five-year intervals: [0–5], (5–10], …, (75–80], and (80–110]. This division allowed us to capture non-linear gender disparities across the life course. We estimated the effect of sex within each interval. We pooled all individuals aged 80 years or older into a single category. This pooling avoided sparse samples in the oldest age group. Covariates Following Andersen’s behavioural model [33, 34], control variables were organized into three groups. First, predisposing factors included employment status (employed vs. not employed), civil-servant status (yes vs. no), poverty status (yes vs. no), and flexible employment status (yes vs. no). These variables served as proxies for socioeconomic position and health consciousness. Second, enabling factors comprised insurance type (UEBMI vs. URRBMI) and residence type (urban vs. rural). These variables represented participants’ ability to pay and geographic access to care. Third, need factors consisted of three measures: the annual count of unique ICD-10 diagnosis codes per beneficiary (repeated encounters with the same code counted once; this measure spans all clinical chapters recorded in the claims, including obstetric and maternity care such as childbirth and related complications), a chronic-disease indicator (based on outpatient special‐disease recognition or chronic‐disease diagnosis codes), and a surgery indicator (presence of any surgical procedure code). We further performed sensitivity analyses that adjust for disease composition as captured by the Elixhauser and Charlson comorbidity indices (ECI/CCI) (Supplement Table S1, Figure S1-3). Together, these variables measured objective health needs. Numeric covariates were summed or counted at the person-year level. Categorical covariates retained the first observed status within each year. As a result, the dataset contains one complete observation per individual per year for subsequent statistical analysis. Statistical analyses All analyses were conducted in R version 4.4.3 (R Foundation for Statistical Computing). The main packages used were glmnet for LASSO regression [35], sandwich for clustered standard errors [36], and ggplot2 alongside tidyplots for data visualization [37, 38]. All P-values were two-sided, with a significance threshold set at P < 0.05. First, we generated descriptive statistics for the person-year dataset. Continuous variables are reported as means standard deviations and medians with interquartile ranges. Categorical variables are presented as counts and percentages. We evaluated sex differences using Student’s t-test for continuous variables and the test for categorical variables. We also conducted t-tests within each of the seventeen five-year age groups. These tests compared men and women on each of the four outcomes. This approach provides an intuitive depiction of how sex disparities evolve across the life course. To minimize multicollinearity, we applied LASSO regression to select covariates. We then fitted two-way fixed-effects panel models. We estimate models with region fixed effects (county or urban district) and year fixed effects. This specification absorbs time-invariant institutional and supply differences across regions and common shocks across years, while preserving between-person variation in the time-invariant regressor of interest, namely biological sex. For each age band, we estimated separate regressions for each outcome. In total, we fitted 68 models (4 outcomes × 17 age groups). Standard errors were clustered at the individual level. The estimating equation was: where 𝑖 indexes insured persons, 𝑡 the calendar year, and 𝑗 the county of the treating facility. The dependent variable, Outcome𝑖,𝑡 is one of four annual measures (TME, outpatient visits, inpatient admissions, or tertiary-hospital admissions). The key exposure, Female𝑖,𝑡 equals 1 for women and 0 for men. The vector contains predisposing, enabling and need covariates as specified by Andersen’s model, and denote year and county fixed effects, respectively. The coefficient represents the adjusted female-versus-male difference for the outcome in that age band. To quantify the age gradient in sex differences on a single scale, we additionally estimated a linear interaction specification for each outcome using the full cohort of person-years: The coefficient on the interaction term captures the age-related gradient in sex differences. Results Cohort characteristics Sex composition varied notably across age groups. Women outnumbered men in every five-year group from 21 to 25 to 36–40 years, peaking at ages 26–30 (11.0% vs. 7.8%). After age 60, the gradient reversed, and men constituted a growing share of the oldest-old. However, the absolute gap at ages ≥ 80 years was less than 1%. Table 1 presents the baseline characteristics of the 1,274,491 person-year observations and unadjusted comparisons by sex. Women incurred lower annual medical costs than men (mean ¥3,226 vs. ¥3,553; median ¥391 vs. ¥510). Women registered more outpatient visits (mean 3.25 vs. 3.09) and slightly more hospitalizations per year (0.29 vs. 0.25). Both sexes had the same mean for tertiary-level admissions (0.15), but the distribution exhibited a longer right tail among women. Table 1Correlation structure and age-stratified univariate analysesCharacteristicNo. (%) of patientsP valueTotalMaleFemaleSex1,274,491567,821 (44.6)706,670 (55.4)TME, RMB< 0.001 Mean (SD)3,371.83 (13,322.17)3,553.05 (15,120.64)3,226.22 (11,676.04) Median (IQR)454.29 (121.96 to 2,058.02)509.645 (140.00 to 2,313.00)391.840 (101.53 to 1,776.48)Outpatient visits< 0.001 Mean (SD)3.18 (4.04)3.09 (4.04)3.25 (4.03) Median (IQR)2 (1 to 4)2 (1 to 4)2 (1 to 4)Inpatient visits< 0.001 Mean (SD)0.27 (0.73)0.25 (0.73)0.29 (0.74) Median (IQR)0 (0 to 0)0 (0 to 0)0 (0 to 0)Tertiary visits< 0.001 Mean (SD)0.15 (0.56)0.15 (0.57)0.15 (0.55) Median (IQR)0 (0 to 0)0 (0 to 0)0 (0 to 0)Age group< 0.001 (0,5]64,814 (5.1)34,643 (6.1)30,171 (4.3) (5,10]58,676 (4.6)30,881 (5.4)27,795 (3.9) (10,15]47,397 (3.7)25,049 (4.4)22,348 (3.2) (15,20]44,230 (3.5)19,375 (3.4)24,855 (3.5) (20,25]73,826 (5.8)24,532 (4.3)49,294 (7.0) (25,30]121,521 (9.5)44,027 (7.8)77,494 (11.0) (30,35]146,613 (11.5)61,875 (10.9)84,738 (12.0) (35,40]132,036 (10.4)58,666 (10.3)73,370 (10.4) (40,45]110,961 (8.7)50,350 (8.9)60,611 (8.6) (45,50]103,422 (8.1)47,177 (8.3)56,245 (8.0) (50,55]104,339 (8.2)47,057 (8.3)57,282 (8.1) (55,60]95,775 (7.5)44,423 (7.8)51,352 (7.3) (60,65]67,987 (5.3)33,823 (6.0)34,164 (4.8) (65,70]42,869 (3.4)20,221 (3.6)22,648 (3.2) (70,75]30,341 (2.4)13,693 (2.4)16,648 (2.4) (75,80]18,656 (1.5)7,824 (1.4)10,832 (1.5) (80,110]11,028 (0.9)4,205 (0.7)6,823 (1.0)Employment status< 0.001 No1,036,262 (81.3)455,105 (80.1)581,157 (82.2) Yes238,229 (18.7)112,716 (19.9)125,513 (17.8)Civil servant status< 0.001 No1,166,813 (91.6)509,921 (89.8)656,892 (93.0) Yes107,678 (8.4)57,900 (10.2)49,778 (7.0)Poverty status< 0.001 No1,267,738 (99.5)565,404 (99.6)702,334 (99.4) Yes6,753 (0.5)2,417 (0.4)4,336 (0.6)Flex employment status0.098 No1,274,366 (100.0)567,775 (100.0)706,591 (100.0) Yes125 (0.0)46 (0.0)79 (0.0)Insurance type< 0.001 UEBMI298,998 (23.5)146,898 (25.9)152,100 (21.5) URRBMI975,493 (76.5)420,923 (74.1)554,570 (78.5)Residence type< 0.001 Urban335,420 (26.3)159,635 (28.1)175,785 (24.9) Rural939,071 (73.7)408,186 (71.9)530,885 (75.1)Number of diseases< 0.001 Mean (SD)1.66 (1.33)1.63 (1.33)1.68 (1.34) Median (IQR)1 (1 to 2)1 (1 to 2)1 (1 to 2)Chronic disease< 0.001 No1,179,606 (92.6)518,677 (91.3)660,929 (93.5) Yes94,885 (7.4)49,144 (8.7)45,741 (6.5)Surgery0.007 No1,233,348 (96.8)549,759 (96.8)683,589 (96.7) Yes41,143 (3.2)18,062 (3.2)23,081 (3.3)SD, Standard Deviation; IQR, Interquartile Range; UEBMI, Urban Employee Basic Medical Insurance; URRBMI, Urban and Rural Residents Basic Medical Insurance Correlation structure and age-stratified univariate analyses SD, Standard Deviation; IQR, Interquartile Range; UEBMI, Urban Employee Basic Medical Insurance; URRBMI, Urban and Rural Residents Basic Medical Insurance Correlation structure and age-stratified univariate analyses Figure 1 summarizes the preparatory analyses. Panel A shows the Pearson correlation matrix linking all control variable to each outcomes. Significant associations across the four outcomes confirm the relevance of the Andersen-based covariate set. However, inter-correlations among controls suggest a risk of multicollinearity. Panel B addresses this risk using LASSO shrinkage on the full covariate list. The coefficient path and ten-fold cross-validation show that no variable was eliminated at the optimal penalty . Each variable retained an independent contribution. Consequently, the full covariate set was carried forward into multivariable modeling without further pruning. Panel C presents box-and-whisker plots of sex differences within each five-year age group. From age 35 onward, women consistently exhibited lower annual expenditures but higher outpatient visit counts. This disparity widened and peaked in the 70–75 age group. Correlation structure and age-stratified univariate sex differences. Panel A: colour-coded correlation matrix; connecting curves highlight covariates with ≥ 0.20 (p < 0.05) for any outcome, with line thickness proportional to effect size. Panel B: LASSO coefficient paths and cross-validated mean-squared error identifying the optimal . Panel C: box-plots for TME (), outpatient visits, inpatient admissions, and tertiary admissions across 17 age bands; blue = female, orange = male. Stars above boxes denote two-sample t-tests (* p < 0.05; ** p < 0.005; *** p < 0.001) Multivariable regression results Figure 2 presents fully adjusted sex differences across age groups for each outcome. In the TME model, no significant sex difference appeared before age 35. From age 35–40, women spent 8% () less annually than men (β = − 299; 95% CI: − 425 to − 123). The gap increased with age. It reached ¥489 less for women in the 41–45 age group. In the 56–60 age group, women spent ¥919 less. The largest deficit occurred at age 71–75, when women spent 28% () less than men (β = − 1,006; 95% CI: − 1,401 to − 611). After age 80, the negative difference narrowed slightly. It remained statistically significant. For outpatient visits, women made 0.13 more visits per year than men at age 16–20 (β = 0.35). This advantage peaked at age 26–30 (β = 0.58; 95% CI: 0.54–0.61). The gap then declined. At age 61–65, women still made 0.06 more visits. By age 66–70, this difference was no longer significant. Adjusted female-minus-male differences in health-service utilisation across 17 age bands. Coefficients to the right of zero indicate higher utilization among women. Coefficients to the left indicate lower utilization. The figure includes panels for four outcomes: annual medical expenditure, outpatient visits, inpatient admissions, and tertiary-hospital admissions. All models adjust for predisposing factors (employment status, civil-servant status, poverty status, and flexible employment), enabling factors (insurance type and residence), and need factors (diagnosis count, chronic-disease indicator, and surgery indicator) under Andersen’s framework. Models also include fixed effects for calendar year and county. Standard errors are clustered at the individual level In the inpatient admissions model, women first showed a significant advantage at age 31–35 (β = 0.02 admissions). From age 46–50 through 61–65, this disadvantage ranged from 0.03 to 0.05 admissions. It decreased slightly to 0.04 admissions at age 71–80. After age 80, no significant sex difference remained. Sex differences in inpatient admissions were strongly age-patterned. Females had more admissions than males from ages 15–35, peaking at 20–25 (β = 0.15, 95% CI 0.14 to 0.16) and 25–30 (β = 0.12, 95% CI 0.11 to 0.13), and narrowing by 35–40 (β = 0.02, 95% CI 0.01 to 0.02). Beginning at 40–45, the sign reversed: females had fewer admissions than males, with the deficit growing through 65–70 (β = −0.06, 95% CI − 0.07 to − 0.04) and remaining negative at older ages. In the tertiary-hospital admissions model, sex differences were negligible before age 35. From age 36–40 onward, women had fewer high-level admissions than men (β =–0.01). This deficit persisted across all later age bands. It ranged from − 0.03 to − 0.05 admissions. In the 61–75 band, the estimate was − 0.04 admissions. All tertiary-hospital gaps were significant at p < 0.001. Age-increasing female disadvantage Table 2 presents estimates from linear sex-by-age interaction models for four outcomes. The Sex×Age coefficient is negative and statistically significant for medical fees (β = −12.496; SE = 1.397; p < 0.001), outpatient visits (β = −0.002; SE = 0.000; p < 0.001), inpatient visits (β = −0.002; SE = 0.000; p < 0.001), and tertiary visits (β = −0.001; SE = 0.000; p < 0.001). The negative and significant Sex×Age coefficients indicate that the female–male gap widens with age, consistent with cumulative inequality. Table 2Age gradient in sex differencesMedical feesOutpatient visitsInpatient visitsTertiary visitsSex ×Age-12.496***-0.002***-0.002***-0.001***(1.397)(0.000)(0.000)(0.000)Sex (ref = male)46.4200.298***0.089***0.002(48.043)(0.011)(0.003)(0.002)Age44.759***0.010***0.003***0.001***(2.341)(0.001)(0.000)(0.000)ControlsYesYesYesYesCounty fixed effectsYesYesYesYesYear fixed effectsYesYesYesYesObservations1,274,4911,274,4911,274,4911,274,491R20.0830.4820.1010.080Notes: are clustered at the individual level. Models include the same set of control variables as in Fig. 2, as well as county and year fixed effects. p < 0.10; *p < 0.05; **p < 0.01 Age gradient in sex differences Notes: are clustered at the individual level. Models include the same set of control variables as in Fig. 2, as well as county and year fixed effects. p < 0.10; *p < 0.05; **p < 0.01 Discussion Using 1.27 million beneficiary-years of claims from three Tibetan cities in 2021–2023, we found that gendered age disparities in health spending and high-level care begin by midlife and widen with age. Women had persistently lower total medical expenditures and lower probabilities of inpatient and tertiary admissions beginning around age 36–40, while maintaining higher outpatient frequency; gaps peaked in the early 70s at approximately ¥1,000 per beneficiary-year and a sustained 0.03–0.05 shortfall in tertiary-hospital admissions. These patterns remained after adjustment for need covariates, individual fixed effects, and county-year controls, which suggests structural rather than purely preference-driven differences. Our findings move the apparent onset of gendered ageism forward to ages 35–49, in contrast to prior work that typically begins at 65 years. Around age 35 many Chinese women enter or deepen the “sandwich generation,” simultaneously providing continuous and contingency care for older parents and young children. Women devote approximately 2.5 times as many hours to unpaid caregiving as men; this time poverty intensifies in midlife as family responsibilities accumulate, compressing the time and attention required for long-distance referrals, elective surgery, and inpatient courses of care. Consequently, women are more likely to select outpatient services with lower time and financial costs, which is consistent with the relative shortfall we observe in admissions and tertiary utilization [39, 40]. The absence of disadvantage among younger women is partly explained by reproductive-age protections that underpin access, especially maternity-oriented benefits for pregnant women [41]. Evidence indicates that the UEBMI maternity benefit improves perinatal outcomes and, by implication, lowers barriers to related outpatient, emergency, and inpatient services during the childbearing years. As childbearing concludes [42], this floor recedes, structural disadvantages become more salient, and gaps widen after about age 35. This is consistent with our sensitivity analysis (Supplement Figure S4). After excluding pregnancy-related and maternity encounters, the onset of divergence shifted to younger ages. Gendered ageism intensifies at ages 51–65 and peaks after age 65, with the sex-by-age-group interactions suggesting, to some extent, a cumulative pattern. This trajectory is consistent with prior work showing lower spending and lower use of higher-level care among older women relative to men. First, underuse of high-intensity or tertiary services among midlife and older women, such as coronary angiography and revascularization and joint replacement, has been repeatedly documented in multicountry and Chinese clinical cohorts [43, 44]. Second, large-scale evidence on hospitalization spending from India shows that sex gaps are largest in older age groups, indicating that as care becomes more expensive, particularly for inpatient and surgical services, women’s relative under-spending becomes more pronounced [20]. Moreover, the intensification of gendered ageism at this stage exhibits a cumulative pattern, consistent with the widening TME gap observed at ages 51–65. Older women face greater financial and access barriers and are more likely to forgo needed care when ill [45]. In addition, although women live longer on average, they bear a higher burden of nonfatal conditions across the life course, including musculoskeletal pain, depression, and anxiety, which amplifies the cumulative effects of gendered ageism [46]. The study’s findings are best interpreted as a structural chain linking macro-institutions, healthcare supply, and household decision-making rather than individual preferences. At the macro-institutional level, women are more likely to be covered by URRBMI and to retire earlier [47], which lowers effective inpatient reimbursement relative to UEBMI and raises exposure to out-of-pocket costs [48, 49]; despite integration reforms, a substantial reimbursement gap with UEBMI persists, especially for higher-cost inpatient care [50]. These scheme features interact with Tibet’s spatial configuration, where high-level resources are agglomerated and travel times to higher-tier facilities are long, increasing the full price of referral care for rural women [51]. On the supply side, fee-for-service and revenue-weighted performance metrics can tilt intensive procedures toward patients perceived as higher return [52], while the geographic concentration of tertiary capacity in and around prefectural capitals compounds selection at the point of referral. Evidence from plateau and western regions documents limited general-practice capacity and referral bottlenecks [51, 53], consistent with the persistent shortfall we observe in women’s tertiary admissions from age 40 onward. On the patient-demand side, gendered ageism in women’s health service use is further shaped by intra-household bargaining and resource allocation. Related evidence indicates that rural households allocate more medical utilization and spending to boys, suggesting that when families face treatment trade-offs, female members, including adult women, are more likely to cede access to higher-cost, higher-tier care [54]. Analyses based on the China Health and Retirement Longitudinal Study show that increases in women’s family status or bargaining power improve the health of women themselves and their children [55], and that exogenous increases in elders’ bargaining power, for example through rural pension programs, raise doctor visits and medical expenditures [56, 57]. In light of our empirical finding from high-altitude settings that gendered ageism emerges around age 35, we propose an integrated policy response that prioritizes timely access, financial capacity over the life course, and routine accountability. First, leverage telemedicine and nurse-training initiatives operating in Tibet since 2023 to establish a “Plateau Women First” fast-track that links teleconsultation with tiered referral, creating a dedicated online window for women aged 40 years or older to complete chronic-disease follow-up and preoperative assessment; referrals through this channel should receive reserved tertiary beds and point-of-care subsidies, reducing distance and language barriers to higher-level care [58]. Second, complement access measures with financial reforms that strengthen women’s ability to pay in later life, including flexible delayed-retirement options, a one-time catch-up pension contribution for up to five previously missing years as China phases in later statutory retirement beginning in 2025, and increased credits to women’s individual medical savings accounts. Third, require routine intersectional monitoring by disaggregating administrative insurance data by sex, age, and ethnicity and publishing an annual Intersectional Indicators Bulletin that reports hospitalization rates in five-year age bands, tertiary-care utilization, and out-of-pocket shares for beneficiaries, thereby enabling prospective evaluation of policy impact. Limitations First, our outcomes are claims-based measures of realized utilization and spending observed in 2021–2023. Claims cannot capture latent clinical need, unmet need, or motives for forgone care. Accordingly, our parameter of interest is the conditional difference in realized use among beneficiaries within the insured system; we interpret these differences as patterns of revealed access consistent with structural gendered ageism and related constraints, not as proof of provider bias per se. The observation window is relatively short and straddles the pandemic and immediate recovery period; longer panels with broader geographic coverage would permit stronger tests of persistence and generalizability. Second, the data cover three prefecture-level cities that together account for 55.4% of Tibet’s population in 2022 and include both UEBMI and URRBMI beneficiaries across tertiary, secondary, and primary facilities. Enrollment exceeds 95% in these jurisdictions, so the claims capture virtually all beneficiaries and provide strong internal validity for the insured populations studied. However, external validity to the entire Tibet Autonomous Region may be incomplete if unobserved prefecture-level factors (e.g., governance rules, provider mix, travel time) differ. Third, the database omits roughly 5% of residents who remain uninsured. Because our estimand is defined on insurance claims, excluding the uninsured does not introduce selection in constructing the outcomes and preserves internal validity for the insured system to which our policy implications pertain. At the same time, uninsured residents are likely more socioeconomically disadvantaged, and may include a higher share of women, so population-wide sex gaps could be larger than our insured-only estimates. On balance, our estimates should be viewed as conservative lower bounds on gender disparities in utilization and spending. Finally, we model age using seventeen five-year bands. This choice enhances life-course interpretability and aligns with policy-salient thresholds, but it can introduce sensitivity to bin definitions, inflate Type I error through multiple comparisons, and yield sparser observations at the tails. To mitigate these concerns, we estimate all age effects within a single specification that includes sex-by–age-group interactions and a common set of controls, and we emphasize the shape and magnitude of the age profile (with confidence intervals) rather than isolated pointwise significance. Conclusions Using 1.27 million person-years of claims (2021–2023) from three Tibetan cities, we applied two-way fixed effects within the Andersen framework, LASSO-guided covariate selection, and individual-clustered errors to map the sex-by-age profile of medical spending and service use. Women’s spending gap emerged by age 36, widened sharply between ages 51 and 65, and peaked at ages 71–75 at roughly ¥1,000 per beneficiary-year. Women also exhibited higher outpatient frequency and persistently lower use of high-level inpatient care. These disparities remained after adjustment for macro-level and need factors, indicating a structural chain of gendered ageism spanning institutions, supply constraints, and caregiving roles in a high-altitude minority context. Policy should shift the intervention window to midlife women, raise URRBMI reimbursement, establish a “plateau-women-first” tele-referral channel, and adopt flexible delayed retirement to interrupt cumulative disadvantage and promote gender-sensitive healthy ageing. Future work should extend to province-wide, harmonized claims linked to facility characteristics and geographic accessibility to test generalizability; integrate claims with clinical registries and time-use or household surveys to measure clinical need, forgone care, and intra-household bargaining; and evaluate the proposed policies using quasi-experimental designs (e.g., staggered rollouts) to estimate causal effects on age- and sex-specific utilization and spending. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Acknowledgements We acknowledge the research support from the National Natural Science Foundation of China (No. 72074084). Abbreviations Urban Employee Basic Medical Insurance Urban Rural Resident Basic Medical Insurance Total medical expenditure Author contributions Changli Jia: Writing – original draft, Writing – review & editing, Methodology, Data curation, Conceptualization. Weizhuo Chen: Writing – original draft, Writing – review & editing, Formal analysis, Data curation, Conceptualization. Xiang Zhang: Writing – review & editing, Project administration, Funding acquisition, Supervision. Funding This study was supported by the National Natural Science Foundation of China (No. 72074084). The funders did not participate in any aspect of the study, including study design, data collection, data analysis, data interpretation, the writing of the report, or the decision to submit the paper for publication. Data availability The study data and related documents are available upon reasonable request from the corresponding author, Professor Xiang Zhang, via email at zhangchxiang@163.com. Declarations Ethics approval and consent to participate Ethics approval for this study was granted by the Medical Ethics Committee of Tongji Medical College, Huazhong University of Science and Technology (protocol numbers: [2022](S092)). All participants’ information have been de-identified. The study was performed in accordance with the Declaration of Helsinki. Consent for publication Not applicable. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Changli Jia and Weizhuo Chen contributed equally to this paper. References Associated Data Supplementary Materials Supplementary Material 1 Data Availability Statement The study data and related documents are available upon reasonable request from the corresponding author, Professor Xiang Zhang, via email at zhangchxiang@163.com.