1. Introduction

Population ageing is no longer a concern limited to high-income countries. Several low- and middle-income countries (LMICs) are undergoing significant demographic changes. Improvements in survival rates and better management of acute and chronic conditions have increased life expectancy in many LMICs (Cadmus et al., 2021). Consequently, the share of the population aged 50 years and above is steadily rising in these countries. With constrained health system capacity, LMICs now face twin challenges: primordial health issues such as infectious diseases and child mortality, along with the growing burden of age-related health conditions such as non-communicable diseases (NCDs) (Piot et al., 2016). These shifts may necessitate augmenting service capacity and health care financing mechanisms to meet the needs of the elderly population.

India is a classic example of this demographic transition. According to estimates from the Ministry of Statistics and Programme Implementation, the share of the population aged 60 years and above is projected to increase from 7.4% in 2001 to 13.2% in 2031 (National Statistical Office, 2021). On the other hand, the proportion of the population aged 0–14 years is expected to decline from 35.3% in 2001 to 21.7% in 2031 (National Statistical Office, 2021). Moreover, estimates suggest that India’s youth population (aged 0–29 years) will decline from 27.9% in 2016 to 22.7% by 2036 (National Statistical Office, 2022). This shift in demographic patterns toward older age groups is likely to exacerbate the burden of NCDs in the country. In addition to this demographic shift, India has been undergoing an epidemiological transition, leading to an increase in the prevalence of NCDs. Estimates indicate that mortality attributable to NCDs increased from 37% in 1990 to 61% in 2016 (Indian Council of Medical Research et al., 2017). More recent estimates further highlight the scale of this burden. In 2023, NCDs accounted for approximately 70.81% of total deaths, 63.37% of disability-adjusted life years (DALYs), and 73.1% of years lived with disability (YLDs) in India. These figures highlight the substantial mortality and morbidity burden associated with chronic diseases (Institute for Health Metrics and Evaluation [IHME], 2024). It is also crucial to highlight that the economic strain associated with NCDs is much more than that of acute conditions.

Health systems that have traditionally focused on episodic and inpatient care have limited capacity to address chronic multimorbidity. In countries like India, where public health spending is low and primary care is fragmented (Ghosh, 2023), managing chronic diseases often falls financially on households. OOP payments are the main mechanism through which healthcare is funded. While there has been a recent impetus toward publicly funded health insurance schemes such as the Prime Minister Jan Arogya Yojana (PM-JAY), they mainly focus on covering inpatient care (Goswami & Bagdi, 2024). Outpatient services, which are the primary mode of managing most chronic conditions, are largely paid for through OOP expenses. As a result, families have the least financial protection for the care they use the most (Murphy et al., 2019). Given the changing demographic and epidemiological trends, India’s domestic policies will need to undergo significant adjustments in how NCDs are diagnosed, treated, and managed among the elderly population.

2. Literature review

Unlike acute conditions which are largely episodic and one-time occurrences, NCDs require continuous monitoring and treatment. Another issue that merits attention in the context of NCDs and ageing is multimorbidity. Multimorbidity refers to the presence of two or more NCDs in the same individual. A substantial body of clinical evidence indicates that multimorbidity is a common characteristic of NCDs and contributes significantly to the physiological and economic burden associated with them. Multimorbidity may further aggravate the existing challenges posed by NCDs, population ageing, and long-term disease management. For instance, using nationally representative data from the Study on Global Ageing and Adult Health across six middle-income countries, Lee et al. (2015) show that NCD multimorbidity rises sharply with age and is associated with significantly higher healthcare utilisation and out-of-pocket (OOP) expenditures. For India and China in particular, multimorbidity substantially increases outpatient visits and OOP spending, with medicines constituting the largest share of expenses. Using primary data from multimorbid patients in a facility-based cross-sectional study in Addis Ababa, Ethiopia, Habtemichael et al. (2024) found high annual OOP spending (around $500 per patient) and a substantial incidence of catastrophic health expenditure (CHE), particularly among uninsured patients. Moreover, using household survey data from 1,276 households in Pokhara Metropolitan City, Nepal, Kafle et al. (2025) observed that nearly 10% of households incurred CHE due to OOP spending on NCD care. They further demonstrate that OOP spending worsened poverty rates, and the financial burden fell more heavily on poorer households and those with more than one member living with NCDs.

The issue of NCD prevalence and its association with wealth or affluence has been widely discussed. In the early phases of the epidemiological transition in developing countries, NCDs have often been found to be more prevalent among the wealthier sections. Using data from the Longitudinal Ageing Study in India (LASI 2017–18), Khura et al. (2022) found that hypertension and diabetes are more prevalent among wealthier groups in India, and that the rich have improved access to preventive care and timely treatment. Using nationally representative biomarker and survey data from the Study on Global Ageing and Adult Health (SAGE), Lloyd-Sherlock et al. (2017) showed that hypertension does not consistently conform to the “diseases of affluence” narrative. While hypertension is primarily concentrated among the rich in India, significant gaps persist in diagnosis and clinical management among socioeconomically disadvantaged populations. Furthermore, using data from the World Health Survey, Hosseinpoor et al. (2012) found an opposite socioeconomic gradient, with most NCDs, such as angina, arthritis, asthma, and depression, disproportionately concentrated among poorer and less-educated groups in LMICs, whereas diabetes was more prevalent among wealthier and more educated populations.

This NCD–wealth gradient raises an important policy concern. Although NCDs are increasingly prevalent across different income groups, their financial consequences are not evenly distributed. While the disease burden may appear similar across wealth levels, poorer households are more vulnerable to the associated economic shocks due to limited financial protection (Alam & Mahal, 2014). In contrast, wealthier households are better able to absorb healthcare costs. As a result, even if health outcomes become more evenly distributed, financial hardship may become increasingly concentrated among poorer households (Niessen et al., 2018). In this case, a similar disease burden does not imply similar financial protection; rather, inequality may shift from health outcomes to financial vulnerability. To assess whether such a shift is occurring, a long-term analysis of both health and financial outcomes is required.

From a health economics viewpoint, it is crucial to study health shocks and their financial impacts together rather than separately. Cross-sectional studies have shown links between multimorbidity and higher OOP costs. However, these studies do not reveal how the same households adjust when new chronic conditions arise. They also do not distinguish between newly developed multimorbidity and pre-existing health conditions. Longitudinal panel data allow us to observe changes within households over time and examine how new chronic conditions affect financial risks relative to their initial levels.

This study uses longitudinal data from Waves 2 and 3 of the WHO’s SAGE India surveys to explore the relationship between wealth, multimorbidity, outpatient OOP spending, and CHE among adults aged 50 years and above (WHO, 2015; WHO, 2020). By linking the two waves at the household level, we construct a panel dataset and create a cumulative count of major NCDs. This enables us to track how disease burden and financial risks evolve over time within the same households, something cross-sectional data cannot capture. The panel structure further allows us to distinguish differences in initial wealth from the financial changes that occur when households develop new NCDs.

This study addresses three related questions. First, how is the burden of NCDs distributed across wealth groups over time, and does this pattern suggest convergence or divergence? Second, how does wealth shape trends in outpatient OOP spending, particularly after accounting for multimorbidity and its onset? Third, does the risk of CHE change differently across socioeconomic groups once we consider the onset of new diseases and the passage of time?

Based on the existing literature and the health financing structure in India, we examine three broad hypotheses. First, the burden of NCDs may increasingly shift toward lower wealth groups over time, reflecting a possible convergence in epidemiological risk. Second, poorer households experiencing multimorbidity may not increase outpatient healthcare spending proportionately in response to worsening health, reflecting financial constraints in healthcare utilisation. Third, the risk of CHE may become increasingly concentrated among poorer households over time, even after accounting for multimorbidity.

This study differs from prior cross-sectional studies such as Kafle et al. (2025), which examined CHE associated with NCD-related OOP expenditure in Nepal using household survey data, and Puri and Singh (2022), which analysed the prevalence and patterns of multimorbidity among older adults in India using Wave 1 of LASI, in several important ways. First, by using longitudinal panel data, we examine changes within the same households over time rather than cross-sectional differences alone. Second, we analyse both multimorbidity burden and financial outcomes jointly, including outpatient OOP expenditure and CHE. Third, we distinguish between existing and incident multimorbidity, which allows us to examine how newly developed chronic disease burden is associated with financial risk.

By answering these questions, the study examines whether population ageing in India is accompanied by equitable financial protection or by widening inequality in financial risk. Beyond documenting spending patterns, it seeks to identify whether emerging financial vulnerabilities are driven primarily by epidemiological changes or by structural features of the health financing system. The findings have important policy relevance for strengthening financial protection in outpatient chronic disease management and for designing insurance mechanisms that address the growing burden of multimorbidity.

3. Methods

The present study uses secondary data from Waves 2 and 3 of the Indian arm of SAGE surveys conducted by the World Health Organization (WHO). SAGE is a nationwide survey that examines indicators related to health, well-being, and ageing among adults aged 50 and older, with a comparison group of adults aged 18 to 49. Wave 2 and Wave 3 followed similar survey design and sampling methods. Wave 2 was implemented in India in 2015 and included 8,152 households and 9,116 individuals from six states: Assam, Karnataka, Maharashtra, Rajasthan, Uttar Pradesh, and West Bengal. Wave 3 was conducted between 2019 and 2020, involving 6,073 households and 7,885 individuals from the same six states. In Wave 3, eligible respondents from earlier waves were re-interviewed and additional respondents were included to maintain the longitudinal design.

We combined the two waves at the individual level using unique personal identifiers and created a two-period panel dataset. Individuals who participated in both waves were included to form a balanced panel. After setting a baseline age limit (age 50 and older in Wave 2) and retaining respondents present in both rounds, the final sample included 6,172 individuals observed across both waves. This implies an attrition rate of approximately 32% from the initial Wave 2 sample.

The NCD burden was estimated using self-reported diagnoses of diabetes, hypertension, heart disease, chronic lung disease, asthma, arthritis, and depression. Binary indicators were coded as 1 if the condition was reported and 0 otherwise. For each wave, an NCD count variable was created, which summed these conditions for each individual. Multimorbidity was defined as having two or more conditions, and incident multimorbidity was when an individual transitioned from having fewer than two conditions in Wave 2 to having two or more conditions in Wave 3.

The NCD variables were based on self-reported lifetime medical diagnosis. Respondents were asked whether they had ever been diagnosed or told by a health professional that they had specific conditions such as diabetes, hypertension, asthma, chronic lung disease, heart disease, arthritis, and depression. Outpatient OOP expenditure was measured using self-reported spending incurred during the respondent’s most recent outpatient visit, including provider fees, medicines, tests, transport, and other related expenses.

We assume OOP expenditure to represent the total direct financial burden borne by the individual during the reference period. Expenditure values were adjusted for consistency across survey rounds. Because expenditure is right-skewed, the dependent variable in linear models was defined as the winsorized natural logarithm of outpatient spending. Observations with zero or negative expenditure accounted for approximately 2% of the sample and were excluded before log transformation. CHE was defined as outpatient OOP costs that exceeded 10% of total household spending, which aligns with the standard practice in health economics literature (Wagstaff et al., 2018; Akazili et al., 2025). Wealth quintiles are based on the household wealth index provided in the SAGE dataset. The index is constructed using information on household assets and dwelling characteristics following the standard SAGE methodology, and households are classified into quintiles ranging from poorest to richest (Sekher et al., 2024). The poorest quintile is used as the reference category in all regression analyses. Age was included as a continuous control variable in our models to adjust for potential age-related heterogeneity in disease burden and health expenditure.

All analyses use the two-period panel dataset constructed from Waves 2 and 3. Linear regression models with interaction terms between wealth quintile and survey wave are used to examine inequality in NCD burden and financial outcomes. Although the dataset has a panel structure, our primary objective is to examine wealth gradients and how they evolve over time. Since wealth quintiles are largely time-invariant across the two waves, using fixed effects models would eliminate this key variable and prevent us from studying differences across wealth groups. Random effects models are also not preferred because they rely on the assumption that unobserved individual effects are uncorrelated with the explanatory variables. This assumption may not hold in this context, especially because wealth and unobserved factors related to health are likely correlated. Therefore, we use pooled regressions with wave indicators and interaction terms to capture both cross-sectional differences across wealth quintiles and changes over time.

Extended models further included multimorbidity and incident multimorbidity, along with their interaction terms, to examine heterogeneity across changes in multimorbidity status. Logistic regression models estimated the probability of incurring CHE. Odds ratios and 95% confidence intervals are reported for these models. Robust standard errors were applied to all specifications. For robustness, models were re-estimated while controlling for inpatient use to check if the observed patterns were affected by hospital stays. We observed no significant changes in our results. All analyses were conducted using Stata / MP 19.5.

4. Results

Descriptive statistics from our analysis revealed that the average number of individual NCD count was 0.626 in Wave 2 and 0.644 in Wave 3. The median for outpatient OOP spending was INR 490 in Wave 2 and INR 500 in Wave 3. The average age of the sample was 62.62 years in Wave 2 and 60.13 years in Wave 3. Table 1 presents the mean number of self-reported NCDs per individual aged 50 years and above by wealth quintile across Wave 2 and Wave 3. The poorest, poorer, and middle quintiles experienced an increase in mean NCD count over time, whereas the richest quintile experienced a decline, from 0.79 in Wave 2 to 0.63 in Wave 3. At the aggregate level, the overall mean NCD count increased slightly from 0.63 to 0.64, indicating a modest rise in chronic disease burden in the panel sample. It is important to highlight that the difference between the poorest and richest quintiles narrowed across waves.

Table 2 shows the results of the regression analysis, which uses the winsorized log of outpatient OOP spending as the dependent variable. The regression includes wealth quintile, survey wave, their interactions, number of NCDs, and age, along with robust standard errors. Winsorization was applied at the top 0.1 percentile to lessen the impact of extreme values. In the raw data, these extreme values reached several million rupees, while most households reported much lower expenditures. This method helps ensure the results show general trends instead of being skewed by a few outliers.

At the start (Wave 2), there is a notable wealth gradient in OOP spending. Households in the poorer, middle, richer, and richest quintiles spent 0.186 (p < 0.05), 0.142, 0.257 (p < 0.01), and 0.421 (p < 0.01) log points more compared to the Poorest quintile. These coefficients suggest that the richest households spent about 52% (because exp(0.421) − 1 ≈ 52%) more on outpatient care than the poorest households. This suggests a significant wealth gradient in outpatient financial burden at baseline. The count of NCDs was a strong predictor of spending: each additional NCD linked to a 0.265 log-point rise in OOP spending, which corresponds to about a 30% increase. Age did not have a significant effect after considering wealth and NCDs.

The interaction terms between wealth and Wave 3 suggest some narrowing in differences in OOP spending over time. The negative and significant interaction terms for the richer and richest quintiles indicate a relative drop in outpatient spending in Wave 3 compared to the poorest quintile. This suggests some narrowing in differences in outpatient OOP spending across wealth groups over time, driven by a relative reduction among the richer and richest quintiles. The estimates indicate higher outpatient spending among wealthier households at baseline and a positive association between NCD count and outpatient spending.

As shown in Figure 1, the predictive margins indicate a clear wealth gradient in Wave 2, with higher predicted spending among richer quintiles. In Wave 3, the gradient becomes substantially flatter and spending levels are more compressed across quintiles. This pattern suggests reduced inequality in outpatient spending and partial convergence over time.

Figure 1Predictive margins of log outpatient out-of-pocket expenditure by wealth quintile and survey wave with 95% confidence intervals 

Source: Authors’ calculations using WHO SAGE India Waves 2 and 3 data.

OOP expenditure for outpatient care is a key indicator of the financial burden households face in managing non-communicable diseases (NCDs). To examine how wealth and multimorbidity shape outpatient OOP spending over time, we estimate a linear regression of log-transformed OOP (ln(OOP)) on wealth quintiles, survey wave, and multimorbidity status (2+ NCDs), including all two-way and three-way interactions. Age is included as a control variable, and robust standard errors are used. This specification allows us to distinguish baseline differences in Wave 2 from changes between Wave 2 and Wave 3 across socioeconomic and disease groups. Table 3 reports the regression results.

As shown in Table 3, a clear socioeconomic gradient in outpatient spending is observed at baseline (Wave 2) among households without multimorbidity (0–1 NCDs). Relative to the poorest quintile, richer households spend approximately 0.29 log points more and richest households 0.43 log points more. Moreover, both these differences are statistically significant. This indicates higher outpatient expenditures among wealthier households at baseline. Multimorbidity is associated with higher spending among poorest households (0.25 log points), even though this difference is not statistically significant.

The Wave 3 main effect indicates that poorest households without multimorbidity experience an increase of 0.285 log points in outpatient OOP between waves, corresponding to approximately a 33% rise (exp(0.285) ≈ 1.33). Two-way interactions between wealth and wave show that this increase is smaller for wealthier households without multimorbidity. In particular, the negative interaction with the richest quintile (−0.439, p < 0.01) indicates a significantly slower growth rate relative to the poorest group. This pattern may indicate partial convergence in outpatient spending across wealth groups over time.

The interaction between Wave 3 and multimorbidity is positive but not statistically significant. This means that the poorest households with two or more non-communicable diseases (NCDs) do not have a different spending pattern compared to those without multimorbidity. However, the three-way interaction for the poorest group is negative and statistically significant (−0.596, p < 0.05). This suggests that multimorbid households in this group see a smaller increase in outpatient spending over time compared to the poorest households without multimorbidity. This may reflect limited spending growth among lower-income households with multiple chronic conditions.

Age does not significantly predict outpatient OOP expenditure. The results in Table 3 show that while wealthier households spend more on outpatient care, the growth in spending over time mainly occurs among poorer households without multimorbidity. Moreover, multimorbidity affects this growth in lower-income groups. These findings indicate that differences in outpatient spending across wealth groups may have narrowed over time.

OOP costs for outpatient care highlight the financial strain families face when managing chronic diseases. To see how newly diagnosed multiple conditions affect these costs across different income levels, we estimated a linear regression model. By tracking households that developed two or more conditions over time, we examined the association between incident multimorbidity and outpatient OOP spending.

Our results in Table 4 show a clear wealth gap in initial spending. Even among families without new conditions, those with more financial resources spent significantly more than the poorest households. This may reflect differences in healthcare access and ability to pay. However, by the third wave, even the poorest households without new diagnoses saw a cost increase of about 31%, suggesting an overall increase in outpatient healthcare expenditure over time.

We also found that wealthier households had smaller percentage increases in spending over time compared to poorer families. Their higher baseline spending may have helped them manage rising costs without drastic relative increases. In contrast, poorer households experienced larger percentage increases in spending over time. An important finding relates to the effect of new diagnoses. Households with incident multimorbidity showed higher estimated spending, although the coefficient was not statistically significant. This may indicate additional financial burden associated with the onset of multiple chronic conditions. Medical expenses may increase with the onset of multiple chronic conditions, including among households with limited financial resources.

Interestingly, there was limited evidence of significant differences across wealth groups in the association between incident multimorbidity and spending. Regardless of their financial situation, the onset of multiple conditions placed a comparable extra burden on households. Age played only a minor role once we considered wealth and disease status, suggesting that disease burden may be more strongly associated with expenditure than age alone. In summary, these findings suggest potential financial vulnerability among households experiencing new chronic disease burdens. Since this health crisis affects households across all income levels, targeted financial support, such as subsidized services or expanded insurance coverage, may help reduce financial burden.

As a robustness check, we additionally control for inpatient utilization. Appendix Table A1 shows that the wealth gradient and convergence in outpatient OOP remain substantively unchanged after including an indicator for any inpatient care, indicating that the results are not driven by differences in inpatient utilization.

Table 5 shows the percentage of households facing CHE. This happens when outpatient costs exceed 10% of a family's income. The descriptive patterns suggest widening differences across wealth groups. For the poorest families, the proportion increased noticeably. The rate of CHE rose from 12.1% in Wave 2 to 21% in Wave 3. By the second wave, nearly one in five of the poorest households faced costs classified as CHE. The “poorer” group experienced a similar, though smaller, increase, going from 12.3% to 17.9%. On the contrary, the middle and wealthier groups remained relatively stable. Meanwhile, the wealthiest households saw their risk of CHE drop sharply, from 13.3% to only 5.7%.

These patterns suggest weaker financial protection among poorer households. This pattern is noteworthy because wealthier households typically report higher healthcare spending in absolute terms. However, when you compare that spending to their income, the burden increasingly falls on the poor. These descriptive results suggest increasing differences in financial vulnerability across wealth groups over time.

Table 6 provides odds ratios from a logistic regression for CHE across wealth quintiles and survey waves. In Wave 2 there is no significant difference in the odds ratios of CHE among wealth groups compared to the poorest households. This indicates that there is no clear wealth pattern initially. However, the odds of facing CHE nearly double in Wave 3. Households in Wave 3 are almost twice as likely to experience CHE compared to those in Wave 2 (OR = 1.94; 95% CI: 1.16, 3.25). Moreover, the interaction terms between wealth and Wave 3 show significant variation in this change over time. Middle-income households see a much smaller rise in CHE risk compared to the poorest (OR = 0.41; 95% CI: 0.20, 0.86). On the other hand, wealthiest households experienced a smaller relative increase in the odds of CHE over time (OR = 0.20; 95% CI: 0.09, 0.46). These results suggest that poorer households faced a bigger increase in CHE, while wealthier groups had a more limited increase. Age is positively linked to CHE, even though the magnitude of that effect is small. These findings suggest increasing disparities in CHE risk across wealth groups over time.

Table 7 presents data on newly acquired multimorbidity and its impact on CHE. The results reveal no significant difference in the odds of facing catastrophic expenditures among different wealth groups at the beginning. This suggests limited evidence of systematic wealth differences in CHE risk at baseline. By Wave 3, there is a general trend toward a higher likelihood of facing these extreme costs, although the evidence is only marginally significant.

An important pattern emerges from the interaction terms between wealth and survey wave. While the risk of catastrophic spending increased for many, middle-income and wealthiest households experienced much smaller rises compared to the poorest families. This may indicate widening differences in financial protection across wealth groups, with poorer households showing relatively larger increases in CHE risk over time.

Interestingly, the onset of new chronic diseases, or incident multimorbidity, did not have a statistically significant effect on its own, and the wide confidence intervals suggest we should approach this finding with caution. Likewise, the interactions between new diseases and wealth were not significant. These results provide limited evidence that incident multimorbidity alone explains differences in CHE risk across wealth groups. Although age seems to be slightly associated with a higher risk of these costs, the effect in this model is modest. To conclude, the findings suggest that wealth-related differences remain important in understanding CHE risk over time.

The results remain robust in models that include inpatient utilization. Appendix Table A2 shows that differences in CHE across wealth groups remain even after controlling for inpatient utilization. The wide confidence intervals suggest limited statistical power for detecting heterogeneity in incident multimorbidity effects.

5. Discussion

Using longitudinal panel data from waves 2 and 3 of WHO’s SAGE India surveys, this study makes one of the first attempts to understand how multimorbidity and healthcare expenditures interact over time among adults aged 50 years and above. We construct a cumulative NCD count (depression, arthritis, diabetes, asthma, hypertension, chronic lung disease, and heart disease) and analyse the evolving burden of chronic disease within the same individuals across survey waves. By tracking the same individuals across time, the SAGE surveys are able to capture changes in both disease burden and financial outcomes. Our results suggest a pattern in which the epidemiological burden of NCDs appears to converge across socioeconomic groups, while financial risk becomes increasingly differentiated across wealth quintiles.

The analysis shows that mean individual NCD counts are rising among the poorer wealth quintiles while declining among the richest. In an ageing population, where disease typically increases across all groups, the drop among individuals in the wealthiest quintile is particularly noticeable. This pattern may reflect the influence of factors beyond age alone on disease burden across wealth groups. Wealthier households may have earlier diagnoses, ongoing access to and adherence to medications, preventive care, and necessary lifestyle changes (Di Cesare et al., 2013). These factors may be associated with better disease control and slower progression to multiple conditions among wealthier groups (Nugent, 2008). The relative shift in NCD burden toward lower socioeconomic groups is consistent with evidence on the epidemiological transition in LMICs, which suggests that NCDs increasingly affect not only affluent populations but also economically vulnerable groups. However, this convergence does not mean there is equal access to the financial means needed to diagnose, treat, and manage NCDs. In other words, while disease risk may be more evenly spread, the resources to manage that risk are still distributed disproportionately.

The findings on healthcare spending reveal an important pattern. Wealthier households spent more on outpatient care at the start, but this difference narrowed by Wave 3. Our results show that households in the lower wealth quintiles had higher outpatient costs in Wave 3 compared to Wave 2. This suggests that the gap in spending between wealth levels narrowed as the financial burden of outpatient care increased for the lower quintiles. At first glance, this trend might seem like improved equity in healthcare use. However, when we include multimorbidity in the analysis, a different picture appears. We found that lower wealth households dealing with multiple chronic diseases did not increase their outpatient spending proportionately in response to worsening health. The significant negative interaction effects show that poorer households responded less to multimorbidity in their healthcare spending. In simpler terms, poorer households facing multiple illnesses do not increase their healthcare use as much. This pattern may reflect constrained healthcare utilization among poorer households. In situations where healthcare is largely funded through OOP payments, spending may reflect financial ability rather than actual medical need (Wagstaff et al., 1999). This suggests that the apparent convergence in healthcare spending could obscure ongoing inequities in access and use. Therefore, the narrowing gap in wealth and healthcare spending does not necessarily imply better financial protection; it may instead reflect unmet healthcare needs among poorer households.

An important strength of this study is its long-term design, which allows us to analyse incident multimorbidity. Our results show that incident multimorbidity leads to a significant rise in OOP spending on healthcare across households in all wealth groups. It is noteworthy that this proportional increase is quite uniform. This indicates that the onset of chronic disease acts as a major financial shock across different wealth levels in a system largely reliant on OOP financing. However, a similar proportional increase does not mean that the welfare impacts are the same. For example, a 50% rise in healthcare spending has vastly different effects based on household resources. Wealthier households may adjust their discretionary spending, while poorer households might have to cut back on essential items, deplete assets, or increase debt and informal borrowing (Pham & Pham, 2022). The panel structure allows changes in expenditure associated with chronic disease onset to be observed over time.

One important finding relates to CHE. For wave 2, there was no clear wealth gradient in CHE. Economic vulnerability to healthcare costs seemed fairly evenly spread across quintiles. By Wave 3, CHE almost doubled among the poorest households but sharply declined in the richest households. Regression models confirm that wealthier households faced much smaller increases in the likelihood of CHE over time. This pattern may reflect uneven financial protection across socioeconomic groups (Bose & Banerjee, 2019). The lack of a strong initial gradient is crucial because it suggests that the differences seen later are not just a continuation of established inequality. Instead, they show a growing gap in financial protection over time. Several factors could explain this widening divergence in financial protection among wealth quintiles. Wealthier households may be increasingly using private insurance and accessing personal savings to cover outpatient care, while poorer households continue to face ongoing outpatient costs without adequate financial support (Subramanian et al., 2018). The widening gap indicates that protection from financial risk is increasing for wealthier households while vulnerability is worsening for those at the bottom.

Furthermore, even after considering incident multimorbidity in the CHE models, wealth-based protection remains strong. Therefore, differences in CHE continue regardless of any newly identified disease burden. This suggests that broader socioeconomic and health system factors may contribute to inequalities in CHE (McKee et al., 2014). The rise in CHE among poorer households cannot be explained solely by higher disease rates. This points to financial hardship being influenced by the unequal ability to manage healthcare costs.

The financing setup of the health system is significant here. In India, publicly funded insurance programs have traditionally focused on inpatient hospitalization. However, the management of chronic NCDs mostly occurs on an outpatient basis. Conditions like diabetes, hypertension, heart disease, and chronic lung disease need ongoing clinical monitoring and long-term medication. These costs are recurring and only partially covered by existing programs. As a result, households in lower wealth groups still depend heavily on OOP payments to handle chronic conditions. As people age, their health needs may increase, but their income security may decrease. Continuous outpatient costs may worsen economic vulnerability. Thus, managing outpatient NCDs makes lower-income households particularly susceptible to CHE. Therefore, our findings highlight a clear contrast: while the burden of NCDs is becoming more evenly spread across wealth groups, financial protection against that burden is increasingly unequal.

Our findings highlight the need for stronger outpatient financial protection within existing public health financing frameworks in India. For example, expanding insurance support to cover outpatient consultations, diagnostics, and medicines for chronic diseases under schemes such as PM-JAY could help reduce recurrent OOP spending among older adults. Since management of NCDs depends substantially on recurrent outpatient visits and long-term medication adherence, policies aimed at improving the availability of affordable essential medicines through public facilities may be particularly important in reducing dependence on private OOP purchases. Strengthening primary healthcare delivery through Health and Wellness Centres may further improve continuity of care for chronic disease management, particularly among poorer households with multimorbidity. The findings may also have implications for existing national programmes focused on ageing and chronic disease management. For instance, the National Programme for Health Care of the Elderly (NPHCE) and the National Programme for Prevention and Control of Non-Communicable Diseases (NP-NCD) already aim to strengthen screening, prevention, and long-term management of chronic conditions among older adults. Integrating outpatient financial protection into these programs, including access to affordable diagnostics, medicines, and follow-up care for multimorbidity, could help reduce the growing burden of CHE among poorer households.

Our study has some limitations that merit attention. First, self-reported NCD data might understate actual disease prevalence, especially among lower wealth groups where access to diagnosis is limited. This underreporting could weaken the observed socioeconomic differences in multimorbidity. Second, the analysis looks only at outpatient OOP expenditure and does not include indirect costs like lost income or unpaid caregiving. Third, given that inpatient spending may affect CHE expenditure, we conducted robustness checks controlling for inpatient utilisation. The results remained largely unchanged, which suggested that hospitalisation does not fully explain the disparities in financial risk.

Fourth, the empirical approach used in the paper does not fully control for unobserved time-invariant individual heterogeneity. As a result, estimated associations may be biased if such factors are correlated with both wealth and health outcomes. Furthermore, we acknowledge the potential simultaneity between diagnosis and healthcare expenditure. The fact that NCDs are determined by self-reported diagnosis may indicate access to healthcare. Although instrumental variable approaches could address this concern, suitable instruments are not available in the dataset. Since most observable factors that influence diagnosis are also directly related to healthcare expenditure, suitable instruments that satisfy both relevance and exclusion criteria are not available in the dataset. Our results should be interpreted as associations, not causal effects, for both of these reasons. Fifth, observations with zero or negative outpatient expenditure constituted a very small share of the sample (approximately 2%) and were excluded before log transformation. However, zero expenditure may reflect lack of access to healthcare or unmet need, especially among poorer households. Excluding these observations may bias the results to reflect individuals who utilize healthcare services.

Sixth, it is important to note that an attrition rate of approximately 32% from the initial Wave 2 sample may bias the results if those lost to follow-up differ from those retained. This may include non-response as well as mortality between waves. If poorer or less healthy individuals are more likely to be lost, the final sample may over-represent relatively healthier or economically better-off individuals.

To sum up, these findings speak directly to the three central questions of this study. First, the burden of NCDs appears to be gradually shifting toward lower wealth groups, pointing to a convergence in epidemiological risk. Second, patterns of outpatient spending suggest that poorer households do not increase healthcare use proportionately when faced with worsening multimorbidity. Third, the risk of CHE has become increasingly concentrated among poorer households over time, highlighting a growing divergence in financial protection even as disease burden becomes more evenly distributed.

6. Concluding observations

This study highlights an important shift in the wealth gradient of NCDs in the context of India’s ongoing demographic and epidemiological transition. By merging Wave 2 and Wave 3 of the WHO’s SAGE surveys and constructing a panel dataset, we examine how changes in multimorbidity are linked to changes in OOP spending and CHE over time. The analysis focuses on individuals aged 50 years and above at baseline, allowing us to capture the dynamics of ageing and chronic disease progression within the same cohort over time.

Our findings show that while NCD burden and multimorbidity are increasingly concentrated in poorer households in Wave 3 of the survey, marking a departure from Wave 2, financial protection has not improved for the poorer sections. Instead, increases in outpatient OOP expenditure and the risk of facing CHE are disproportionately concentrated among poorer groups. This suggests that even as the epidemiological transition leads to a wider prevalence of NCDs across socioeconomic groups, the economic consequences remain largely uneven.

In the context of population ageing and demographic transition, these patterns have important policy implications. As multimorbidity in NCDs becomes more common, health systems must move beyond inpatient models of care and strengthen financial risk protection for chronic care. Expanding outpatient coverage, improving continuity of care, and designing targeted protection against NCD-related CHE for the elderly and vulnerable households are essential. Without such adjustments, the combined pressures of ageing, NCD burden, and rising OOP payments may further worsen socioeconomic inequality in financial risk.