Gendered experiences of cause-specific unmet medical need: An intersectional analysis using MAIHDA

Abstract

Unmet medical needs (UMN) persist even under universal health coverage and remain unequally distributed across social groups. Although gender differences in UMN are well documented, less is known about how gender intersects with family and work roles to shape healthcare access.

We analyzed data from 230,925 adults aged ≥19 years in the 2024 Korea Community Health Survey. Outcomes were overall, time-related, and cost-related UMN. Multilevel Analysis of Individual Heterogeneity and Discriminatory Accuracy (MAIHDA) was applied using 144 intersectional strata defined by sex, household composition, employment status, age, and household income. We estimated variance partition coefficients, proportional change in variance, and area under the curve, along with main effects and intersectional interaction effects for each outcome.

Intersectional strata explained 18.6% of variation in time-related UMN and 27.3% in cost-related UMN, compared with 4.8% in overall UMN. Time-related UMN was concentrated among younger and middle-aged employed adults, but younger non-employed females living with children also had high risks and the largest intersectional interaction effects (ORs 1.59–2.10). Cost-related UMN was concentrated among middle-aged, low-income, non-employed adults living alone. However, non-employed middle-aged males living alone showed the strongest intersectional interaction effect even in the highest-income group (OR 2.60).

These patterns suggest that time- and cost-related UMN arise through different gendered work-family configurations that additive approaches may obscure, particularly among younger females living with children and non-employed middle-aged males living alone. Achieving equitable healthcare access may therefore require attention to the gendered organization of work, care, and household responsibilities.

Highlights

• •Gendered work-family roles shape healthcare access under universal health coverage.
• •Intersectional strata explained more variance in cause-specific vs overall unmet need.
• •Non-employed mothers experience time-related barriers comparable to employed adults.
• •Middle-aged men living alone show elevated cost-related barriers even at high income.
• •Intersectional analysis reveals hidden risk groups missed by additive approaches.

1. Introduction

Unmet medical need (UMN) refers to situations in which individuals perceive a need for healthcare but are unable to obtain appropriate or sufficient care (Allin et al., 2010). Despite ongoing efforts to improve healthcare access, UMN remains a persistent issue, even in high-income countries with established universal health coverage systems (OECD, 2025b). South Korea illustrates this tension. Although the National Health Insurance system provides near-universal population coverage and formal entitlement to care, coverage alone does not ensure equitable access, as it does not eliminate socially organized constraints such as employment conditions, caregiving responsibilities, and household resources (Hwang, 2018; Jung & Ha, 2021; Park et al., 2018). Accordingly, UMN remains substantial in South Korea, with approximately 8% of the population reporting such experiences (Korean Statistical Information Service (KOSIS), 2026), and is disproportionately concentrated among socioeconomically disadvantaged groups (Alemu et al., 2024; Kim et al., 2019; Maslyankov, 2024). Identifying which subgroups are most at risk and understanding the mechanisms underlying these disparities is therefore essential from a health equity perspective.

However, understanding the mechanisms underlying UMN—and translating them into actionable policy—is not straightforward, as UMN stems from a range of factors spanning multiple levels, including material resources, individual perceptions of health and care, and structural conditions (Andersen, 1995). Several previous studies have therefore suggested examining cause-specific UMN rather than overall UMN, on the grounds that distinguishing between barriers such as financial and time constraints can yield more targeted policy insights (Allin et al., 2010; Sanmartin et al., 2002). Yet even when specific causes are considered, universal policy measures or simple targeting criteria might be inadequate because the same barrier may reflect different experiences depending on individuals’ social positions.

Gender is one such social position shaping UMN. Numerous studies across diverse settings—including the United States, Canada, European countries, South Korea, and low- and middle-income countries—have consistently reported gender differences in the likelihood of experiencing UMN (Bryant et al., 2009; Cai & Bidulescu, 2023; Dawkins et al., 2021; Kim et al., 2019). Beyond simple gender disparities, several studies also suggest that gender may affect how other social factors shape UMN. For example, Jung and Ha's results stratified by sex show that household size is associated with UMN only among females (Jung & Ha, 2021). Similarly, according to Hwang's analysis, the employment in service and sales occupations is associated with acceptability-related UMN only among males (Hwang, 2018). A Canadian study reported that the adverse effects of being divorced or widowed were more pronounced among males (Tadiri et al., 2021). These findings suggest that the ways in which work and family contexts affect healthcare access may differ systematically by gender (note: Throughout this paper, we use female and male when referring to the measured sex variable and reporting empirical results, and women and men when discussing the broader gender system that this variable is theorized to reflect.)

From an intersectional perspective, gender and work/family roles may intersect to produce distinct experiences of UMN. Intersectionality, first theorized by Crenshaw, holds that social categories such as race, gender, and class are not independent axes but are mutually constitutive, producing forms of both disadvantage and advantage that cannot be reduced to the sum of their parts (Crenshaw, 1989). To illustrate, consider a woman with a parental role. Conventional additive regression approaches estimate the effect of gender while holding household composition at its reference category (without children), and the effect of parental status while holding gender at its reference category (male). These estimates are then summed to represent the expected UMN of a woman with a parental role compared with a man without a parental role. However, the experience of a woman with a parental role is unlikely to be adequately captured by adding the female disadvantage among those without children and the parental disadvantage among men. This is because the intersection of gender and parental role generates a qualitatively distinct experience that cannot be reduced to—or reconstructed from—its component axes alone.

This limitation of conventional additive approaches has important implications for policy. If the elevated UMN observed among women with children is driven primarily by the intersection of gender and parental role, then policies targeted at either women or parents risk misidentifying the populations most in need of intervention. When the specific contexts and lived experiences associated with each intersectional position are not considered, policymakers may fail to implement the most effective intervention strategies (Bowleg, 2012). By contrast, an intersectional approach has the potential to reveal the diversity of experiences across subpopulations, providing a stronger basis for targeted policy design. It enables more precise identification of at-risk populations and supports the development of policy responses that reflect the needs and lived experiences of targeted populations.

However, previous research has not adequately adopted an intersectional approach to understanding how gender and role-based social positions shape UMN. Although several studies have applied intersectional analyses to examine healthcare access or UMN, most have centered on axes such as race/ethnicity, immigration status, or geographic residence rather than treating gender and its accompanying expectations as the primary focus (Anticona et al., 2023; He et al., 2024; Mahapatro et al., 2021; Rodriguez-Lopez et al., 2025; Vohra-Gupta et al., 2023). Missing from this literature is an examination of the two domains where gendered roles are most concretely lived out—the workplace and the family —the spheres that define much of how individuals spend their time and construct their identities in a gendered society. To our knowledge, UMN has rarely been examined through an intersectional framework that jointly considers gender, employment status, and household composition as intersecting axes.

One reason for the relative scarcity of intersectional studies focusing on gendered roles in understanding UMN is conceptual. Intersectionality was initially developed to examine interlocking systems of oppression and privilege, with traditional social categories including race, gender, and class (Crenshaw, 1989). More recently, intersectional health research has extended this framework to social positions such as household composition and employment status, particularly where these positions are embedded in gendered systems of work, care, and family responsibility. For example, Pomeroy and Fiori examined how gender and employment status combine with household-level demands to shape adult children's provision of informal care (Pomeroy & Fiori, 2025). Sesay et al. analyzed healthcare utilization inequalities in informal settlements by defining intersectional strata based on factors including head-of-household gender, marital composition, and engagement in income-generating activity (Sesay et al., 2026). These studies reflect a growing recognition that household and employment positions can function as analytically meaningful intersectional axes when they capture socially organized roles and constraints distributed along gendered lines.

Building on this perspective, we conceptualize household composition and employment status along with gender as jointly constituting social locations that assign distinct roles and expectations that may constrain healthcare seeking behavior. Care regime theory and gendered-labor literature suggest that welfare and labor-market institutions have historically assumed the care-free working man as the normative citizen. Accordingly, men's employment is constructed as a normative expectation of economic provision whereas women's employment is interpreted as secondary or conditional, and women are instead more closely tied to caregiving roles (Daly & Lewis, 2000; Hochschild & Machung, 2003; Lewis, 1992; Orloff, 1993). Such institutional arrangements influence individuals' behavior patterns and shape access to resources. The ‘motherhood penalty’ illustrates how the intersection of parental role and gender operates as a distinct vector of labor market discrimination (Budig & England, 2001; Correll et al., 2007; Ishizuka, 2021). Conversely, stay-at-home husbands may experience identity tensions due to their misalignment with breadwinner norms (Demantas & Myers, 2015; Kowalewska, 2024). In the context of UMN, these intersections may shape individuals' ability to allocate time, financial resources, and social support toward healthcare-seeking.

Beyond the conceptual challenges, empirical examination of intersectionality presents its own methodological difficulties. Approaches such as stratified analyses and interaction-term models, face inherent limitations: as the number of axes increases, stratified analyses suffer from reduced statistical power, while interaction models quickly become difficult to interpret. Multilevel Analysis of Individual Heterogeneity and Discriminatory Accuracy (MAIHDA) has emerged as a rigorous and scalable approach that addresses these limitations by modeling intersectional strata as random effects and stabilizing estimates for sparse groups through shrinkage (Evans, Borrell, et al., 2024; Evans, Leckie, et al., 2024). Therefore, we applied MAIHDA to accommodate multiple stratification axes and to obtain stable estimates even for underrepresented subgroups.

The present study examines how gender shapes individuals' experiences of UMN, with particular focus on their positions in the workplace and family. South Korea is particularly well-suited for this inquiry: it has one of the most pronounced gender gaps in employment and unpaid care work among OECD nations—an 18.9-percentage-point gap in labour force participation and a nearly threefold gap in daily household labor time (69 vs. 206 min)—making gendered role expectations an especially salient axis of healthcare access inequality (OECD, 2022, 2025a; Park et al., 2022; Stansbury et al., 2024; Statistics Korea (KOSTAT), 2024). To this end, we treat sex, household composition, and employment status as the primary intersectional axes to capture the gendered normative expectations individuals face across workplace and family domains. We further incorporate age and household income as additional axes: age to capture how these role expectations shift across the life course, and household income to reflect how material resources alter intersecting experiences of UMN. We pursue three aims: (1) to examine how each intersectional axis contributes to cause-specific (time-related and cost-related) UMN; (2) to assess the extent to which intersecting social positions generate inequalities in UMN; and (3) to identify which intersectional strata bear a disproportionate burden of UMN and reveal the specific social locations where intersectional interaction effects are particularly pronounced.

2. Methods

2.1. Data source

This study used data from the 2024 wave of the Korea Community Health Survey (KCHS), a nationwide annual cross-sectional survey conducted by the Korea Disease Control and Prevention Agency. The KCHS targets all adults aged 19 years or older residing in Korea and selects participants using a two-stage stratified probability sampling design: survey areas are first stratified by the catchment areas of community health centers, sample points are then selected via probability-proportional-to-size systematic sampling, and sample households within each point are selected via systematic sampling (Kang et al., 2015). Data collection for the 2024 wave was conducted between May 16 and July 31, 2024. The survey collects information on healthcare utilization, health behaviors, and sociodemographic characteristics of adults aged 19 years or older across all districts in Korea. Of the 231,728 individuals who participated in the 2024 survey, those with missing information on the outcome or stratification variables were excluded, resulting in a final study sample of 230,925 participants. This study was reviewed and approved as exempt by the Institutional Review Board of the first author's institution (IRB details withheld for peer review).

2.2. Outcome

The outcome of interest was UMN. Overall UMN was assessed using a self-reported item asking, “During the past 12 months, was there any time when you needed medical care (excluding dental care), including examinations or treatment, but did not receive it?” Responses were dichotomized into having experienced UMN and not having experienced UMN, with the latter category including respondents who reported no perceived need for medical care during the reference period.

Two cause-specific UMN outcomes were additionally examined: cost-related UMN and time-related UMN. Cost-related UMN was defined as respondents who reported that their UMN was due to financial reasons (i.e., the cost of medical care was burdensome). Time-related UMN was defined as respondents who reported time constraints, such as inconvenient clinic hours, inability to leave work, or lack of childcare. For each outcome, responses were dichotomized into the relevant reason for UMN versus all other responses (including no UMN, UMN due to other reasons, or no perceived need for medical care).

2.3. Stratification variables

Following Collins's conceptualization of intersectionality, we selected stratification variables that are theoretically relevant to the research question, with the aim of keeping the resulting intersectional strata conceptually grounded and closely aligned with the research focus (Collins & Bilge, 2020; Collins et al., 2021). The selected variables have been consistently reported to be associated with UMN in previous studies (Huh & Lee, 2016; Hwang, 2018; Jung & Ha, 2021; Kim et al., 2019). Age was categorized into 19–39, 40–64, and ≥65 years, representing young adulthood, middle age, and older adulthood. Self-reported sex was used as a proxy for gender, representing participants' positions within the socially constructed gender system. In the Korean context, gendered expectations—particularly those related to caregiving and labor—are strongly structured along binary sex categories, making sex a reasonable proxy for gender in this study. However, we acknowledge that this operationalization has limitations: sex and gender do not always align, and a binary classification does not fully capture gender diversity, including transgender and non-binary individuals, who may be misclassified in survey data. Household composition was categorized into living alone, living with spouse only, living with children, and living with others. The “living with children” category included individuals living with their children in a two-generation household. The “living with others” category included individuals living with parents or extended family members. Employment status was dichotomized as “employed” or “not employed” with the latter category including both unemployed and economically inactive individuals. Household income was equivalized by dividing total household income by the square root of household size, consistent with the approach used by the OECD (OECD, 2024). The equivalized household income was then categorized into tertiles to represent the level of available economic resources. The selected five variables were cross-classified to create 144 unique intersectional strata. Each respondent was assigned to exactly one stratum.

2.4. Statistical analysis

We first described the distribution of the stratification variables in the overall sample. Variables were summarized as counts and percentages. To assess the empirical support for the main analysis, we also summarized the sample size distribution of the intersectional strata, grouping strata according to the number of individuals they contained (<10, 10–19, 20–49, 50–99, and ≥100).

For the main analysis, we applied the MAIHDA framework using two-level logistic regression models, with individuals (level 1) nested within intersectional strata (level 2). We fitted two models for each outcome: (1) a null model and (2) a main-effect model.

The following model equation was used for the null model:

$$\log \left(\frac{{\mathrm{\pi }}_{ij}}{1-{\mathrm{\pi }}_{ij}}\right)={\beta }_0+u_j$$

where, ${\mathrm{\pi }}_{ij}$ is the probability of UMN for individual $i$ in stratum $j$, ${\beta }_0$ is the overall intercept, and $u_j$ is the random intercept for stratum $j$. This model captures the overall between-stratum variance without adjusting for any main effect.

The main-effect model is specified as:

$$\log \left(\frac{{\mathrm{\pi }}_{ij}}{1-{\mathrm{\pi }}_{ij}}\right)={\beta }_0+\sum _k{\beta }_k{X}_{kj}+u_j$$

where, $X_{kj}$ denotes a set of indicator variables for the categories of the five stratification variables for stratum $j$ (Evans, Leckie, et al., 2024). This model estimates the main additive associations of each stratification variable with UMN while retaining the random intercepts $u_j$ to capture residual intersectional interaction effects beyond additivity.

Stratum-level variance was estimated for each model, and variance decomposition was conducted on the latent response scale for logistic models. In this framework, the binary outcome is conceptualized as a dichotomization of an underlying continuous latent variable, whose level-1 error follows a standard logistic distribution with variance fixed at π²/3. The variance partition coefficient (VPC) was calculated to quantify the proportion of total variance in UMN attributable to between-stratum differences. In the null model, the VPC reflects the extent to which UMN varies across intersectional strata, whereas in the main-effect model, the VPC represents residual between-stratum differences after accounting for the additive main effects of the stratification variables. The proportional change in variance (PCV) was calculated to quantify the extent to which the additive main effects explain between-stratum variance by comparing the VPC of the main-effect model to that of the null model. The area under the receiver operating characteristic curve (AUC) was used as a measure of discriminatory accuracy, reflecting the ability of the model to distinguish between individuals who did and did not experience UMN. AUC values range from 0.5, indicating discriminatory accuracy no better than chance, to 1, indicating perfect discrimination. From the main-effect model, odds ratios (ORs) and 95% confidence intervals were obtained for the fixed effects of the stratification variables.

$$VPC=\frac{{\sigma }_u^2}{{\sigma }_u^2+{\sigma }_e^2}$$

$$PCV=\frac{{\sigma }_{u,null}^2-{\sigma }_{u,main}^2}{{\sigma }_{u,null}^2}\times 100$$

In the above equation, ${\sigma }_u^2$​ denotes the between-stratum variance of the random intercepts, ${\sigma }_e^2={\pi }^2/3$ denotes the level-1 residual variance on the latent logistic scale, and ${\sigma }_{u,null}^2$ and ${\sigma }_{u,main}^2$​ denote the between-stratum variances estimated from the null and main-effect models, respectively.

Following established MAIHDA practice, random intercepts (uⱼ) from the main-effect model were interpreted as capturing intersectional interaction effects, defined as departures from what would be expected based solely on the additive main effects of stratification variables (Evans, Leckie, et al., 2024). Caterpillar plots on the OR scale of intersectional interaction effects were used to identify strata exhibiting large positive or negative deviations from the main effects. Strata were ranked based on the magnitude of the estimated effects. It should be noted that “intersectional interaction effects” do not represent causal effect estimates. While we use the term “effect” in line with its usage in “fixed effects” or “random effects,” it is employed here in a descriptive sense rather than to make claims about causal inference. We further note that the absence of a statistically significant intersectional interaction effect does not imply the absence of intersectionality. What we refer to as a main effect is an average additive gradient across strata, which may itself harbor heterogeneity in how a given axis operates across intersectional positions. Thus, the intersectional interaction effects should not be understood as separate effects added onto main effects assumed to operate uniformly. Instead, the interaction effects indicate, in quantitative terms, how much of a stratum's pattern departs from the sum of average additive gradients alone.

We also conducted sensitivity analyses. First, hypertension and diabetes were included as level-1 fixed effects in the main-effect model to assess whether adjustment for underlying health conditions altered the observed patterns. Second, analyses were restricted to individuals reporting a perceived need for medical care to evaluate whether the observed patterns were robust to excluding individuals with no perceived need for medical care. Analyses were conducted using the lme4 package in R (version 4.4.2) (Bates et al., 2015).

3. Results

3.1. Sample characteristics

A total of 230,925 participants were included in the analysis (Table 1). The largest age group was 40–64 years (44.4%), followed by ≥ 65 years (36.1%), while 19–39 years accounted for 19.5% of the sample. Females comprised a slightly larger proportion of the sample (54.2%) than males (45.8%). Living with a spouse only was the most common household type (33.0%), followed by living with children (27.5%), living with others (20.2%), and living alone (19.3%). Overall, 64.5% of participants were employed. Cross-classification of five variables yielded 144 intersectional strata. Only one stratum had fewer than 20 individuals, and three had fewer than 50. The majority of strata (91.0%) contained at least 100 individuals (Table S1).

Table 1.

Characteristics of the study population.

Characteristics	N = 230,925
Age
≥ 65	83,408 (36.1%)
40-64	102,463 (44.4%)
19-39	45,054 (19.5%)
Sex
Male	105,838 (45.8%)
Female	125,087 (54.2%)
Household Composition
Living with spouse only	76,174 (33.0%)
Living alone	44,681 (19.3%)
Living with children	63,425 (27.5%)
Living with others	46,645 (20.2%)
Employment Status
Not employed	81,952 (35.5%)
Employed	148,973 (64.5%)
Household Income
Highest	77,008 (33.3%)
Middle	77,094 (33.4%)
Lowest	76,823 (33.3%)

3.2. Main effects of stratification variables

In MAIHDA analysis, the main effects of the stratification variables differed by the cause of UMN (Table 2). Overall, the odds of UMN were higher among younger adults (OR 1.52; 95% CI 1.37–1.69) and middle-aged adults (OR 1.39; 95% CI 1.27–1.53) compared with older adults. Females had higher odds of UMN than males (OR 1.34; 95% CI 1.23–1.45). Compared with those living with a spouse only, individuals living alone (OR 1.56; 95% CI 1.39–1.76), living with children (OR 1.32; 95% CI 1.18–1.48), and living with others (OR 1.13; 95% CI 1.01–1.27) had higher odds of UMN. Participants who were employed were also more likely to experience UMN than those not employed (OR 1.34; 95% CI 1.23–1.46). The lowest income group had higher odds of UMN (OR 1.43; 95% CI 1.29–1.58), whereas the middle-income group was not statistically significant compared with the highest tertile.

Table 2.

Estimates from MAIHDA models for overall and cause-specific UMN.

	Overall UMN		Time-related UMN		Cost-related UMN
	Null Model	Main Effect Model	Null Model	Main Effect Model	Null Model	Main Effect Model
Age (ref: ≥ 65)
40-64		1.39 (1.27, 1.53)		2.27 (1.95, 2.64)		1.21 (0.92, 1.60)
19-39		1.52 (1.37, 1.69)		3.05 (2.59, 3.60)		0.95 (0.68, 1.32)
Sex (ref: Male)
Female		1.34 (1.23, 1.45)		1.58 (1.34, 1.79)		0.96 (0.75, 1.21)
Household Composition (ref: Living with spouse only)
Living alone		1.56 (1.39, 1.76)		1.15 (0.97, 1.38)		3.60 (2.56, 5.05)
Living with children		1.32 (1.18, 1.48)		1.46 (1.23, 1.72)		1.62 (1.14, 2.31)
Living with others		1.13 (1.01, 1.27)		1.02 (0.86, 1.22)		1.51 (1.05, 2.17)
Employment Status (ref: Not employed)
Employed		1.34 (1.23, 1.46)		3.06 (2.67, 3.50)		0.55 (0.43, 0.71)
Household Income (ref: Highest)
Middle		0.97 (0.88, 1.07)		0.95 (0.82, 1.10)		0.95 (0.68, 1.32)
Lowest		1.43 (1.29, 1.58)		0.97 (0.84, 1.13)		4.55 (3.39, 6.10)
Variance of Random Effects
Stratum-Level	0.17	0.03	0.75	0.06	1.24	0.22
Individual-Level	3.29	3.29	3.29	3.29	3.29	3.29
Summary Statistics
VPC	4.8%	0.9%	18.6%	1.8%	27.3%	6.2%
PCV		81.4%		91.9%		82.5%
AUC	0.60	0.60	0.69	0.69	0.81	0.80

Notes. Fixed effects are presented as odds ratios with 95% confidence intervals.

UMN, unmet medical need; VPC, variance partition coefficient; PCV, proportional change in variance; AUC, area under the receiver operating characteristic curve; MAIHDA, multilevel analysis of individual heterogeneity and discriminatory accuracy.

For time-related UMN, the associations with age, sex, and employment status were more pronounced than in the overall UMN model. Younger adults (OR 3.05; 95% CI 2.59–3.60) and middle-aged adults (OR 2.27; 95% CI 1.95–2.64) had higher odds than older adults. Females and employed individuals also showed elevated odds compared with males (OR 1.58; 95% CI 1.34–1.79) and non-employed individuals (OR 3.06; 95% CI 2.67–3.50), respectively. Household income was no longer statistically significant. Among household composition categories, living with children was the only category that remained significant and had a larger OR than in the overall UMN model (OR 1.46; 95% CI 1.23–1.72).

For cost-related UMN, the associations with household composition were larger in magnitude across all categories compared with the overall UMN model, with living alone showing the largest OR (3.60; 95% CI 2.56–5.05). The effect of the lowest income group was also more pronounced compared with the highest income group (OR 4.55; 95% CI 3.39–6.10), whereas the middle-income group remained not statistically significant. In contrast, the direction of the association for employment status was reversed, with employed associated with lower odds of cost-related UMN (OR 0.55; 95% CI 0.43–0.71). Age and sex were not significantly associated with cost-related UMN.

3.3. Variance partitioning and discriminatory accuracy

The proportion of variance explained by the strata was higher for cause-specific UMN than for overall UMN. When examining overall UMN, the VPC was 4.8% in the null model and decreased to 0.9% after adjusting for the main effects of the stratification variables, corresponding to a PCV of 81.4%. For time-related UMN, the VPC was 18.6% in the null model, which decreased to 1.8% after including the main effects, corresponding to a PCV of 91.9%. For cost-related UMN, the VPC was 27.3% in the null model and decreased to 6.2% after adjusting for the main effects, corresponding to a PCV of 82.5%. In the main-effects models, AUCs were 0.60 for overall UMN, 0.69 for time-related UMN, and 0.80 for cost-related UMN.

3.4. Intersectional interaction effects

Fig. 1 shows the predicted probabilities of time-related UMN from the main-effect model. Overall, employed strata exhibited higher predicted probabilities than their non-employed counterparts within the same age groups, with the highest probabilities observed among younger and middle-aged employed adults across both sexes. Among younger employed adults, strata living with children showed notably higher predicted probabilities than those with other household compositions. Notably, among non-employed younger females living with children, predicted probabilities were markedly elevated and comparable to those of employed younger adults—including males living with children and females living with a spouse—a pattern not observed among non-employed younger males. These three strata of non-employed younger females living with children across different income levels ranked first, second, and third in the intersectional-effect caterpillar plot (Fig. 2). The two most extreme deviations from the additive expectation corresponded to the middle- and lowest-income strata, followed by the highest-income stratum. The corresponding intersectional interaction effects were 2.10 (95% CI 1.63–2.72), 2.06 (95% CI 1.50–2.84), and 1.59 (95% CI 1.18–2.15), respectively (Table S2).

Fig. 1.

Predicted probabilities of time-related UMN across intersectional strata defined by sex, age, employment status, household composition, and household income.

Fig. 2.

Caterpillar plot of intersectional interaction effects for time-related UMN. The red points indicate the strata with the four highest intersectional interaction effects. The highest-ranking strata were non-employed younger females living with children, including those in the middle-income tertile (OR 2.10, 95% CI 1.63–2.72), lowest-income tertile (OR 2.06, 95% CI 1.50–2.84), and highest-income tertile (OR 1.59, 95% CI 1.18–2.15). The fourth highest intersectional interaction effect was observed among employed older females living alone in the middle-income tertile (OR 1.48, 95% CI 1.06–2.07).

Fig. 3 presents the predicted probabilities of cost-related UMN estimated from the main-effect model. Strata with the lowest income consistently exhibited the highest predicted probabilities. Within the lowest income group, individuals who were not employed generally showed higher predicted probabilities than those who were employed. Among both females and males, middle-aged individuals with the lowest income who were not employed exhibited particularly high predicted probabilities when living alone, a pattern that was especially pronounced among males. Among middle-aged males, a similar spike associated with living alone was observed among those with the lowest income who were employed and those with the highest income who were not employed. Notably, the latter stratum ranked sixth in predicted probability of cost-related UMN and first in the intersectional interaction effect across all strata (OR 2.60, 95% CI 1.31–5.16, Table S3). In addition, these three middle-aged male strata living alone corresponded to three of the four strata with the highest odds ratios in the intersectional-effect caterpillar plot (Fig. 4).

Fig. 3.

Predicted probabilities of cost-related UMN across intersectional strata defined by sex, age, employment status, household composition, and household income.

Fig. 4.

Caterpillar plot of intersectional interaction effects for cost-related UMN. The red points indicate the strata with the four highest intersectional interaction effects. The first-, second-, and fourth-ranked strata were middle-aged males living alone, including non-employed individuals in the highest-income tertile (OR 2.60, 95% CI 1.31–5.16), non-employed individuals in the lowest-income tertile (OR 2.21, 95% CI 1.77–2.76), and employed individuals in the lowest-income tertile (OR 2.07, 95% CI 1.41–3.06). The third-ranked stratum was employed older females living with children in the lowest-income tertile (OR 2.21, 95% CI 1.25–3.90).

Sensitivity analyses revealed broadly similar results. When hypertension and diabetes were additionally adjusted for, the overall findings remained largely unchanged, including the pattern of main effects, the magnitude of VPC, PCV, and AUC. For time-related UMN, the ranking of the four highest intersectional interaction effects across strata remained the same. For cost-related UMN, the ranking of the top intersectional interaction effects shifted slightly: non-employed middle-aged men living alone in the lowest-income tertile moved from second to third, whereas employed older women living with children in the lowest-income tertile moved from third to second (Table S4; Fig. S1–S4). When individuals reporting no perceived medical need were excluded, the findings were similarly robust, with the pattern of main effects, the magnitude of VPC, PCV, and AUC, and the ranking of four highest intersectional interaction effects across strata remaining unchanged (Table S5; Fig. S5–S8).

4. Discussion

This study examined cause-specific UMN through a gender lens, conceptualizing sex, household composition, employment status, age and household income as intersectional axes structuring UMN experiences. Three findings stand out. First, the main effects of intersectional axes differed across causes of UMN: while sex and age showed larger associations with time-related UMN, household composition and income were more strongly associated with cost-related UMN. The direction of the association for employment status was reversed between the two causes. Second, intersectional strata explained substantially more variation in cause-specific than in overall UMN. Third, mapping predicted probabilities across strata revealed distinct risk profiles for each cause. Time-related UMN was most concentrated among younger and middle-aged employed adults, with an additional cluster of elevated risk among younger non-employed females living with children. Cost-related UMN, by contrast, was disproportionately concentrated among the lowest-income strata—particularly middle-aged individuals who were not employed and living alone—with the effect of living alone being especially pronounced among middle-aged males.

The associations observed for stratification variables and UMN were broadly consistent with previous studies. A systematic review of Korean studies reported higher odds of overall UMN among females, younger individuals, those without a spouse, individuals with lower household income, and those participating in the labor market (Kim et al., 2019). Similar patterns have been documented in other high-income countries. For example, a systematic review of Canadian studies and a study comprising 29 European countries have likewise reported higher odds of UMN among females, younger individuals, and those with lower income, with additional evidence linking UMN to employment and certain household contexts such as living with children (Alemu et al., 2024; Chaupain-Guillot & Guillot, 2015). A South Korean study concluded that time-related UMN was more common among females, younger individuals, and those participating in the labor market whereas cost-related UMN was more common among older adults, those without a spouse, and individuals with lower household income (Huh & Lee, 2016), a pattern broadly consistent with our findings.

The VPC estimates showed that intersectional strata explained substantially more variation in cause-specific UMN than in overall UMN. This divergence likely reflects that intersectional strata carry distinct implications across different causes of UMN. When cause-specific UMNs are collapsed into a single aggregate outcome, the effects associated with different causes may partially offset one another, thereby attenuating the apparent contribution of intersectional positions. However, even the 4.8% observed for overall UMN is not trivial compared to other MAIHDA studies, where VPCs below 5% are commonly reported and rarely exceed 10% (Evans, Leckie, et al., 2024). Notably, given that no need-related or health condition variables were included in the model, the observed level of between-stratum variation is meaningful. Yet, examining only the overall UMN would have obscured the finding that intersecting positions defined by five social variables account for roughly one-fifth to one-quarter of the variation in cause-specific UMN. Furthermore, the higher AUC values observed for cause-specific UMNs suggest that disaggregating UMN by underlying barriers improves the identification of high-risk populations and supports more targeted policy interventions.

When the main effects of the intersectional axes were accounted for, residual VPCs attributable to intersectional interaction effects were substantially reduced, with PCVs exceeding 80% for all outcomes. This suggests that most between-stratum variation is captured by the average additive gradients associated with the individual axes, with relatively small residual variation attributable to departures from those additive patterns. However, small residual VPCs should not be interpreted as evidence that intersectional interaction effects are uninformative. In MAIHDA, an important analytical value lies in identifying strata that deviate from the general pattern characterized by these average additive gradients. When such deviations are concentrated in a handful of strata, these strata can then be prioritized as localized pockets of disadvantage or advantage for further investigation and intervention. Under these conditions where the general patterns shaped by average additive gradients explain the majority of strata, small residual VPCs are an expected feature rather than a limitation (Evans, Leckie, et al., 2024). In this study, focusing on such strata enabled the identification of specific groups that warrant particular attention—namely, non-employed younger females living with children for time-related UMN, and non-employed middle-aged males living alone, including those in the highest-income tertile, for cost-related UMN.

The patterns of time-related UMN were broadly consistent with established literature on time poverty and time use. Across both sexes, employed adults in young-to-middle adulthood exhibited the highest predicted probabilities of time-related UMN regardless of income level. These years correspond to the prime working-age population, and market participation in this life stage may be particularly important in predicting UMN in the South Korean context, given long working hours and limited access to sick leave (Carcillo et al., 2024; Lee, 2025a, Lee, 2025b). Within this group, additional variation emerged by household composition. Among employed young adults, predicted probabilities were markedly higher among those with children compared to those living with a spouse only. This pattern likely reflects the dual burden of paid work and caregiving responsibilities, which may be especially pronounced in young adulthood when childcare demands are at their peak (Rubiano-Matulevich & Viollaz, 2019). This burden appears to be further stratified by sex. The three strata with the highest predicted probabilities were young females living with children who were employed. Prior studies suggest that even after accounting for living arrangement and employment status, women consistently spend more time on unpaid household and care work than men, including care extended to relatives and social networks outside the household (Ervin et al., 2022; Pomeroy & Fiori, 2025; Ragnarsdóttir et al., 2023; Román, 2025). While this may partly explain the overall sex difference, whether the mechanisms underlying this gradient operate uniformly across all strata or vary depending on intersecting factors warrants further investigation.

A particularly notable pattern emerged among younger females who were not employed and living with children. This group was the only young or middle-aged group whose predicted probabilities of time-related UMN were comparable to those of their employed counterparts. This deviation from the general pattern, alongside substantial intersectional interaction effects, suggests a distinct experience arising at the intersection of female sex, parental role, and non-employment in this life stage. One possible explanation lies in the ideology of intensive mothering—a cultural script that prescribes continuous maternal availability and prioritization of children's needs (Batram-Zantvoort et al., 2022; Hays, 1996). Within this framework, normative expectations may render non-employment as availability for caregiving rather than relief from time pressure, thereby reinforcing expectations of full-time caregiving (Batram-Zantvoort et al., 2022; Hays, 1996; Johnston & Swanson, 2006; Liss et al., 2013). Such norms have been shown to generate guilt when mothers prioritize their own needs over caregiving responsibilities, leading to the deferral of healthcare among mothers (Damaske, 2022; McArthur et al., 2024). This potential interpretation is consistent with the well-documented gender asymmetry in unpaid care work. In Korea, among single-earner households with children under 18 years, non-employed women spend approximately 7 h per day on unpaid household labor, roughly double the time spent by non-employed men when women are the single-earner (Statistics Korea (KOSTAT), 2025). Similar patterns have been reported in the United States, where stay-at-home mothers perform 76.2% of total housework time, whereas stay-at-home fathers account for 55.9% (Pew Research Center, 2013). Furthermore, international evidence indicates that unemployment among men does not translate into equivalent increases in unpaid care work (Damaske, 2022; Gough & Killewald, 2011), suggesting that gendered role expectations operate in both directions: directing women toward child-centered domestic routines while simultaneously discouraging men from assuming caregiving responsibilities. While prior qualitative research has extensively documented the lived experiences of mothers navigating caregiving responsibilities, quantitative evidence on the intersectional vulnerability of non-employed young women living with children remains limited (Gaur et al., 2024; Jain et al., 2024; McArthur et al., 2024). Our findings extend this qualitative evidence by quantifying the elevated risks of time-related UMN at this intersection using large, nationally representative data.

Cost-related UMN appears to be primarily shaped by economic vulnerability, with middle age and living alone carrying particular implications. Income reflects the availability of financial resources, whereas non-employment may indicate financial insecurity, making this combination particularly disadvantageous for cost-related healthcare access. National trend analyses reporting persistent income-related disparities in UMN in Korea support this interpretation (Kim et al., 2024, 2026). Across both sexes, middle-aged individuals in this group exhibited the highest levels of cost-related UMN within comparable household compositions. From a life-course perspective, midlife represents a critical period for accumulating household wealth while simultaneously bearing substantial financial responsibilities, including supporting both children and aging parents (Cheung & Lucas, 2015; Nam et al., 2025). In this context, individuals who are both low-income and non-employed may experience intensified financial pressure, potentially reducing their willingness or ability to pay for healthcare services (Infurna et al., 2021; Lachman et al., 2015). Notably, although age alone may not show a strong independent association with cost-related UMN, it appears to play a meaningful role when interacting with household income and employment status. This vulnerability is further exacerbated among those living alone. As health expenditures are typically incurred and assessed at the household level, the absence of co-residing household members may limit opportunities for intra-household financial risk pooling and reduce buffering against healthcare costs (OECD, 2025b). In addition, the intersection of non-employment and living alone during midlife—a life stage typically characterized by high labor force participation and strong family involvement—may also reflect diminished social support, which has been consistently associated with unmet healthcare needs (de Beeck et al., 2025; Hughes & Waite, 2002; Yang et al., 2024).

The effect of living alone was more pronounced among middle-aged males. Notably, elevated levels of cost-related UMN were observed among middle-aged males who were not employed and living alone, even in the highest-income group. This subgroup also demonstrated the largest intersectional interaction effect. This pattern suggests that cost-related UMN in this group may reflect perceived affordability or the prioritization of healthcare spending, rather than objective financial constraints alone. Gendered divisions of household labor have historically positioned women as primary managers of health-related practices, including monitoring health status and facilitating healthcare use (Damaske, 2022; Reczek et al., 2018). Accordingly, men may rely more on spousal co-residence for healthcare engagement, and without this support, men living alone may be less likely to perceive healthcare spending as worthwhile or to act on health needs. In addition, within the Korean context, men are more strongly associated with traditional breadwinner roles compared to countries with similar GDP levels (Choi & Qian, 2023; Lee, 2025a, Lee, 2025b). Non-employment or low income during midlife may therefore represent a disruption of expected social roles, potentially further contributing to lower prioritization of healthcare spending (Amiri, 2022; Heyne & Vossemer, 2023). Together, these findings suggest that cost-related UMN may be shaped by the intersection of economic vulnerability, gendered role expectations, and household context. However, the underlying mechanisms warrant further investigation, as middle-aged men are not typically considered a high-risk group for cost-related healthcare access.

These findings have implications for how healthcare access inequities are identified and addressed under universal health coverage. They suggest that approaches based only on average associations or single-axis categories may obscure localized patterns of vulnerability, leading to both under-identification of certain at-risk groups and overgeneralization of risk within broad social categories. In our analysis, for example, elevated time-related UMN among non-employed younger females living with children and elevated cost-related UMN among some middle-aged males living alone would have been difficult to detect without an intersectional perspective. These patterns support the value of stratum-sensitive approaches that attend to the specific social contexts in which barriers to care arise. For time-related UMN, interventions focused only on formal workplaces may be insufficient, as they are unlikely to reach non-employed young mothers. Policies that reduce caregiving constraints, expand flexible service access, or improve childcare-related support may therefore be relevant for these groups. For cost-related UMN, the elevated risk observed among middle-aged males living alone, including those not conventionally defined as economically disadvantaged, warrants further qualitative investigation. Such inquiry could help clarify whether these patterns reflect perceived affordability, social isolation, household-level health management, or other mechanisms, thereby informing more contextually grounded policy responses and public health campaigns.

In this study, we conceptualized household composition and employment status as intersectional dimensions through which gendered work-family roles are socially organized. In the Korean gender regime, they mark social locations through which expectations around caregiving, breadwinning, dependency, and household responsibility are distributed and experienced. This approach is consistent with Collins’ conceptualization of intersectionality as an analytic framework that emphasizes the flexible selection of categories that best explain the issue at hand (Collins & Bilge, 2020). At the same time, Collins cautions that this flexibility should not lead to the uncritical inclusion of categories, and that each category and its relationships should be empirically studied and theorized within the specific social context under study (Collins et al., 2021). Accordingly, our analysis treats household composition and employment status as meaningful axes because they are embedded in institutional and normative arrangements that shape healthcare-seeking behavior in South Korea. By situating these axes within the intersectional structure of UMN, we identified localized vulnerabilities that single-axis or purely additive approaches would have obscured.

This study has several limitations. First, the cross-sectional nature of the data and the descriptive application of MAIHDA limit causal interpretation. Second, UMN was measured based on a single self-reported primary reason, requiring respondents to select only one cause among multiple possible barriers. As a result, the prevalence of cause-specific UMN may be underestimated or distorted, particularly for subgroups facing compounded constraints. Third, our interpretations primarily focused on the predisposing factors characterizing each intersectional stratum, which may have led to insufficient consideration of need-related factors. To address this limitation, we showed similar results from sensitivity analyses in two ways: by excluding individuals who reported no medical need during the reference period and adjusting for chronic conditions. However, the latter analysis adjusted only for hypertension and diabetes, and residual confounding by other unmeasured health conditions cannot be ruled out. Fourth, household composition was indirectly derived from multiple variables, including household head reports of household composition and marital status, which limited the ability to capture more nuanced family structures. In particular, individuals with complex caregiving roles—such as those in “sandwich generation” positions—were grouped into “living with others” category alongside those under parental support, potentially obscuring meaningful heterogeneity. Similarly, the dichotomization of employment status combined heterogeneous subgroups—including the unemployed, retired individuals, students, and those engaged solely in homemaking—into a single “not employed” category, which may obscure differences in the mechanisms underlying UMN across these subgroups. Fifth, as this study is based on nationally representative data from Korea, the generalizability of the findings to other sociocultural contexts may be limited. Future research in countries with similar gender norms and high levels of healthcare coverage may provide further insight into the extent to which these patterns are observed across comparable contexts. Sixth, the possible role of unadjusted socioeconomic and contextual variables—such as assets, reason for non-employment (e.g., retirement or job loss), and regional healthcare access—cannot be ruled out. Accordingly, the gendered work-family interpretation should be understood as one plausible explanatory framework among several.

Despite these limitations, our study is among the first quantitative analyses to examine gendered experiences of cause-specific UMN through an intersectional lens, incorporating role-based social positions in work and family as structuring dimensions. By revealing distinct patterns of vulnerability, our findings suggest that time-related and cost-related barriers to care may be experienced differently across social locations within Korea's gendered work-family role regime. In particular, the intersecting positions that deviated from the overall pattern—such as non-employed younger females living with children and some middle-aged males living alone—suggest that effective policy responses may benefit from stratum-sensitive approaches grounded in the contextual barriers faced by these subpopulations, rather than relying on average effects or single-axis targeting strategies. Future research should further investigate possible explanations for these patterns through qualitative inquiry into how intersecting social positions shape healthcare-seeking experiences and longitudinal designs that can examine how these barriers to care evolve over time. Such work would help clarify the broader implication of our findings: achieving equitable healthcare access under universal health coverage may require attention not only to formal entitlement or affordability, but also to the gendered organization of work, care, household responsibility, and social support. Finally, our findings underscore the value of intersectional approaches, including MAIHDA, in uncovering otherwise obscured patterns of vulnerability in complex healthcare behaviors such as UMN.

CRediT authorship contribution statement

Hanyul Lee: Writing – original draft, Visualization, Methodology, Formal analysis, Data curation, Conceptualization. Wonbin Choi: Writing – review & editing, Investigation, Conceptualization. Woojoo Lee: Writing – review & editing, Validation, Methodology. Hongsoo Kim: Writing – review & editing, Supervision, Project administration, Methodology, Conceptualization.

Ethics declaration

The authors have NOT obtained informed consent from participants or their legal representatives. Informed consent was not obtained for the present study because the analysis used secondary, de-identified data from the Korea Community Health Survey and involved no direct contact with participants.

This study was performed in compliance with relevant laws, regulatory frameworks and guidelines where the research took place. The Institutional Review Board of Seoul National University granted an exemption. The authors provided the following explanation: The requirement for informed consent for the present study was waived because the analysis used de-identified secondary data and involved no direct contact with participants.

Ethical statement

This study used secondary, de-identified data from the Korea Community Health Survey (KCHS), conducted by the Korea Disease Control and Prevention Agency (KDCA). This study was reviewed and approved as exempt by the Institutional Review Board of Seoul National University (IRB No. E2604/001-001). The requirement for informed consent for the present study was waived because the analysis used de-identified secondary data and involved no direct contact with participants.

Declaration of generative AI and AI-assisted technologies in the manuscript preparation process

During the preparation of this work, the authors used ChatGPT (OpenAI, GPT-5.3) in order to polish English language. After using this tool, the authors reviewed and edited the content as needed and take full responsibility for the content of the published article.

Funding statement

This work was supported by the National Research Foundation of Korea (BK21 Center for Integrative Response to Health Disasters, Graduate School of Public Health, Seoul National University)(NO.419 999 0514025).

Conflict of interest

The authors have no conflict of interest to declare.

Appendix A. Supplementary data

The following is the Supplementary data to this article:

Data availability

Data will be made available on request.