Social engagement, formal education, and cognitive outcomes among older adults in India: A longitudinal population‐based study Authors: Peiyao Zhu, Emma Nichols, Marco Angrisani, Alden L. Gross, Jinkook Lee, Lindsay C Kobayashi Published: 2026 ABSTRACT AbstractINTRODUCTIONWe estimated the associations between baseline social engagement and cognitive function and change among older adults in India, overall and by education.METHODSData were from 3869 adults aged ≥ 60 years from the Longitudinal Aging Study in India—Diagnostic Assessment of Dementia Wave 1 (2017–2019) and Wave 2 (2022–2024). Baseline social engagement was measured using a 12‐point index, dichotomized as higher (< 8) versus lower (≥ 8) engagement. General and domain‐specific cognitive scores were derived at both waves. Multivariable‐adjusted joint models accounted for mortality and evaluated effect modification by education.RESULTSHigher baseline social engagement was associated with higher general cognitive scores at baseline (0.145 standard deviation units; 95% confidence interval: 0.076, 0.214), but not with cognitive change over time. Education attainment did not show a consistent pattern of effect modification across cognitive domains.DISCUSSIONSocial engagement was associated with cognitive level at baseline but not the rate of cognitive change over time, with no consistent evidence of effect modification by education. BACKGROUND Alzheimer's disease and related dementias (ADRD) is a pressing public health issue for the global aging population. More than 55 million people are currently living with dementia, > 60% of whom lived in low‐ and middle‐income countries (LMICs) in 2020. 1 However, research on dementia in LMICs remains relatively scarce. India, a lower middle‐income country, has an estimated 8.8 million individuals aged ≥ 60 years living with dementia. 2 Because most approved treatments for dementia may be financially inaccessible to older adults living in India and other LMICs, 3 , 4 , 5 identifying modifiable factors that can reduce dementia risk at the population level is crucial for implementing early prevention, particularly in LMICs. Social engagement is a modifiable later life factor that may support cognitive health and reduce dementia risk. 6 It is thought to reduce dementia risk through contributing to later life cognitive reserve as it can be cognitively stimulating 7 by providing opportunities to activate and maintain cognitive processes through communication, social role participation, and cognitively stimulating interactions. 6 , 8 , 9 , 10 Cognitive reserve is a theoretical property of the brain thought to allow individuals to maintain cognitive function despite accumulating age‐related neuropathology by using pre‐existing cognitive processes or compensatory mechanisms. 11 , 12 It is dynamic and can accumulate across the lifespan. 13 , 14 In India and other LMICs, social networks, including family members and close friends, are an important source of social support and cognitively stimulating interactions for older adults, many of whom live with families in multigenerational households. Prior evidence from LMICs suggests that social engagement is positively associated with cognitive functioning, but much of this evidence is cross‐sectional and has not examined cognitive change over time, domain‐specific cognitive outcomes, or heterogeneity by educational attainment. 15 , 16 , 17 , 18 , 19 , 20 , 21 RESEARCH IN CONTEXT Systematic review: The authors reviewed the literature on social engagement and cognitive function and decline among older adults using PubMed and Google Scholar. Prior studies have generally shown that higher social engagement is associated with better cognitive function; however, many were cross‐sectional or conducted among populations in high‐income countries. Few studies examined effect modification by education attainment, considered domain‐specific cognitive outcomes, or included older populations in low‐ and middle‐income countries. Interpretation: In a large, population‐based longitudinal study of adults aged ≥ 60 years in India, higher baseline social engagement was associated with better cognitive performance across multiple domains but not subsequent cognitive change, with no consistent evidence of effect modification by education. Future directions: Future research should incorporate repeated measures of social engagement and longer follow‐up to better capture long‐term cognitive trajectories and explore the underlying mechanisms through which social engagement may influence cognitive aging. Systematic review: The authors reviewed the literature on social engagement and cognitive function and decline among older adults using PubMed and Google Scholar. Prior studies have generally shown that higher social engagement is associated with better cognitive function; however, many were cross‐sectional or conducted among populations in high‐income countries. Few studies examined effect modification by education attainment, considered domain‐specific cognitive outcomes, or included older populations in low‐ and middle‐income countries. Systematic review: The authors reviewed the literature on social engagement and cognitive function and decline among older adults using PubMed and Google Scholar. Prior studies have generally shown that higher social engagement is associated with better cognitive function; however, many were cross‐sectional or conducted among populations in high‐income countries. Few studies examined effect modification by education attainment, considered domain‐specific cognitive outcomes, or included older populations in low‐ and middle‐income countries. Interpretation: In a large, population‐based longitudinal study of adults aged ≥ 60 years in India, higher baseline social engagement was associated with better cognitive performance across multiple domains but not subsequent cognitive change, with no consistent evidence of effect modification by education. Interpretation: In a large, population‐based longitudinal study of adults aged ≥ 60 years in India, higher baseline social engagement was associated with better cognitive performance across multiple domains but not subsequent cognitive change, with no consistent evidence of effect modification by education. Future directions: Future research should incorporate repeated measures of social engagement and longer follow‐up to better capture long‐term cognitive trajectories and explore the underlying mechanisms through which social engagement may influence cognitive aging. Future directions: Future research should incorporate repeated measures of social engagement and longer follow‐up to better capture long‐term cognitive trajectories and explore the underlying mechanisms through which social engagement may influence cognitive aging. Educational attainment may modify the association between social engagement and cognitive outcomes for multiple reasons. Education is the strongest known protective factor against dementia risk and, like social engagement, is thought to support cognitive reserve. 6 , 22 , 23 Educational attainment largely reflects structural conditions in early life, including government policies and prevailing socioeconomic conditions that determine access to, duration of, and quality of schooling, as well as sex inequities in access to education. 24 , 25 , 26 , 27 , 28 Education may also affect dementia risk through indirect pathways that operate across the life course, such as better adherence to health‐promoting behaviors. 29 , 30 , 31 , 32 In addition, formal education may shape the types of social activities and relationships available to older adults and the cognitive demands of those social activities. 33 , 34 , 35 As there is little prior evidence on this topic, the direction of any potentially modifying effect of education on the social engagement–cognition relationship is theoretically ambiguous. Social engagement may be especially beneficial for older adults without formal education by providing cognitively stimulating experiences that at least partially substitute for those typically gained through schooling, consistent with resource substitution. 36 Alternatively, formal education may amplify the cognitive demands and reserve‐building potential of social engagement, leading to stronger associations among those with more education, consistent with resource multiplication. 36 In India, ≈ 60% of adults > 60 years of age have no formal education and may be especially vulnerable to dementia risk. 2 Thus, the question of whether and in what direction formal education may modify the association of later‐life social engagement with cognitive outcomes is of practical importance for the targeted design of interventions to reduce population dementia risk in India and similar global settings. Using data from a national, population‐based longitudinal study of aging in India, we estimated the association between social engagement and both the level of cognitive function at baseline and the average rate of annual cognitive change over time. We also assessed whether formal educational attainment modified these associations. We hypothesized that older adults with greater social engagement have higher cognitive function scores at baseline and slower annual cognitive decline than those with less social engagement. We further hypothesized that these associations differ between individuals with no formal education and those with any formal education. METHODS Data sources The Longitudinal Aging Study in India (LASI) is a nationally representative, longitudinal study of adults aged ≥ 45 years in India, with a focus on health, economic, and social well‐being. 37 The Longitudinal Aging Study in India—Diagnostic Assessment of Dementia (LASI‐DAD) is a sub‐study of LASI focused on late‐life cognition and dementia. 38 LASI‐DAD includes 4096 LASI participants aged ≥ 60 at baseline, who completed the Harmonized Cognitive Assessment Protocol (HCAP), a comprehensive assessment battery designed to measure cognitive functioning and dementia. LASI‐DAD Wave 1 was fielded from 2017 to 2019, followed by Wave 2 from 2022 to 2024. Participants eligible for the current analysis were those enrolled in LASI‐DAD at Wave 1 who had complete data on social engagement, education, and all covariates at baseline. After excluding 227 (5.5%) participants who had missing data (24 missing sociodemographic factors, 21 missing childhood health status, 32 missing financial situation while growing up, 70 missing chronic conditions and mental health data, 80 missing social engagement data), the final analytic sample included 3869 individuals who contributed 6268 observations over follow‐up. Of these participants, 2399 completed the Wave 2 cognitive assessment, 899 died before Wave 2, and 571 did not participate in the Wave 2 follow‐up assessment. The mean follow‐up time in our analytic sample was 4.23 years (standard deviation [SD] = 1.25). See Figure S1 in supporting information for the study flow diagram. All participants provided informed consent. Measures Exposure: social engagement Social engagement was measured at baseline in LASI using a composite index based on four dimensions: (1) social contact, (2) participation in social events, (3) marital status and quality of relationship with spouse, and (4) household relationships (Appendix S1 in supporting information). This index was adapted from the composite measures developed by Steptoe et al. for the English Longitudinal Study of Ageing (ELSA). 39 , 40 Social contact was assessed by evaluating the respondent's relationship and the frequency of contact with friends (range: 0–2). Participation in social events was measured based on the frequency of involvement in social activities such as playing indoor games, visiting friends or relatives, attending religious events, and participating in community or organization group meetings (range: 0–4). Marital status and quality of relationship with a spouse or partner were also incorporated into the composite index (range: 0–2). Individuals who were widowed or divorced were classified as not married; however, because they may maintain broader social networks and receive social support from adult children 41 we thus incorporated household relationships based on the respondent's household living arrangements and the quality of their relationships with co‐residing household members, in recognition that older adults often live with their adult children in multi‐generational households, especially if they are not married (range: 0–4). The social isolation index ranged from 0 to 12 points, with higher scores indicating greater social isolation and lower social engagement. We classified participants as having lower social engagement if their social isolation index score was ≥ 8, which identified individuals in approximately the top tertile of social isolation and was consistent with prior work in LASI‐DAD that adapted the ELSA social isolation index to the Indian context. 39 Participants with scores < 8 were classified as having higher social engagement. Effect modifier: educational attainment All respondents were asked to report the highest level of education they had attained during the baseline LASI‐DAD interview. Response options included never attended school, less than primary school, primary school completed, middle school completed, secondary school/matriculation completed, higher secondary/intermediate/senior secondary school completed, diploma or certificate, graduate degree, postgraduate degree or above, and professional course or degree. To reflect the limited level of formal education among older adults in India, 2 we categorized education status as “no formal education” or “any formal education” given the observed distribution of educational attainment (49% of the sample had no formal education). Outcome: general and domain‐specific cognitive function scores The primary outcome of our longitudinal model was a factor score for general cognitive function (GCF), based on all available measures in the LASI‐DAD HCAP battery at each of Waves 1 and 2. 42 We also separately examined four cognitive domains: memory, visuospatial ability, executive function, and verbal fluency. These outcomes were represented as factor scores scaled to a mean of 0 and a SD of 1 at Wave 1, with the factor scores at Wave 2 scaled to the Wave 1 distribution to facilitate comparison. 42 , 43 Covariates We considered sociodemographic factors, early‐life factors, and health conditions as potential confounders of the relationship between social engagement and later life cognitive function and change. All covariates were assessed in the baseline LASI interview. Sociodemographic factors included age (mean centered, continuous in years), age2 (continuous, to account for non‐linear effects), sex (male, female), household per capita consumption quintile, rurality of residence (urban community, rural village), religion (Hindu, Muslim, Christian/other/none), and caste (scheduled caste, scheduled tribe, other backward class, none/other). Early‐life factors were included based on a life course framework, as childhood health and socioeconomic conditions may influence both later life cognitive outcomes and opportunities for social engagement through providing education and socioeconomic resources across the life course, in addition to the development and maintenance of social relationships. 44 , 45 , 46 Early‐life factors included self‐reported childhood health (very good, good, fair, poor, very poor), family financial status during childhood (well off, average, poor), and each of father's and mother's education status (no education, any education, missing). We retained a missing indicator for father's and mother's education as prior research indicates that informative missingness is common in measures of parental education, which may result in bias if individuals with missing observations on this factor are excluded. 47 , 48 Health conditions included depression (yes, no) assessed using the Short Form Composite International Diagnostic Interview (CIDI‐SF), 49 and self‐reported difficulties with activities of daily living (ADLs; yes, no). In addition, we categorized individual chronic conditions (self‐reported physician diagnosis of hypertension, heart problems, stroke, lung disease, cancer, diabetes, psychiatric problems, osteoporosis, high cholesterol, and arthritis) into three groups (none, 1–2 conditions, ≥ 3 conditions). Statistical analysis We described characteristics of the sample, overall and according to level of social engagement. Among the 3869 participants included in the analytical sample, 899 (23%) died between LASI‐DAD Wave 1 and Wave 2, a period which included the COVID‐19 pandemic. 50 To account for potential selection bias due to differential mortality, we used a multivariable‐adjusted joint model to account for mortality while assessing the relationships of social engagement with cognitive function level and change. 51 Our analyses included two sub‐models: a longitudinal sub‐model for rate of cognitive change and a survival sub‐model for time to death. Longitudinal sub‐model The longitudinal sub‐model of the joint model was a linear mixed‐effects model to estimate the associations between social engagement level and domain‐specific cognitive function over time. Given the availability of two cognitive assessments, the time coefficient and time interaction terms were interpreted as average annual differences in cognitive change between Wave 1 and Wave 2 interviews, using the exact number of years between interviews for each study participant. The general formula of the model is: Cognitionit=mit+eit where: Cognitioni(t) represents the general or domain‐specific cognitive factor score for individual 𝑖 at time t. mi(t)=β0+xiT(t)β+b0i+ziT(t)bi is the trajectory function for the true unobserved value of cognitive function for the individual 𝑖 at time t. ei(t) is the residual error term. Cognitioni(t) represents the general or domain‐specific cognitive factor score for individual 𝑖 at time t. Cognitioni(t) represents the general or domain‐specific cognitive factor score for individual 𝑖 at time t. mi(t)=β0+xiT(t)β+b0i+ziT(t)bi is the trajectory function for the true unobserved value of cognitive function for the individual 𝑖 at time t. mi(t)=β0+xiT(t)β+b0i+ziT(t)bi is the trajectory function for the true unobserved value of cognitive function for the individual 𝑖 at time t. ei(t) is the residual error term. ei(t) is the residual error term. The model included social engagement, education attainment, the covariates listed in section 2.2.4, a time variable representing the time in years between study interviews, and an interaction between social engagement and time. Statistically significant (P < 0.05) covariate‐by‐time interactions were included to adjust for potential confounding of slope effects. To evaluate the modifying effect of education on the relationships between social engagement and cognitive outcomes, we included an interaction between social engagement and education attainment. To further assess whether this effect changes over time, we incorporated a three‐way interaction among social engagement, education attainment, and time. The final model was specified as: mit=β0+β1SocialEngagementi+β2Timeit+β3SocialEngagementi∗Timeit+βnCovariatei+βnCovariatei∗Timeit+βnEducationi∗Timeit+βnSocialEngagementi∗Educationi+βnSocialEngagementi∗Educationi∗Timeit+b0i+b1iTimeit Survival sub‐model The outcome for the survival model was time to death, measured in years since the LASI‐DAD Wave 1 interview. The dates of the Wave 1 and Wave 2 interviews, as well as the date of death, were recorded in LASI‐DAD. For participants who completed both waves of interviews and were still alive, their survival time was censored at the date of the Wave 2 interview. For those who died between waves, time‐to‐event was calculated as the time between the date of the Wave 1 interview to the recorded date of death. For individuals who were alive but did not complete the Wave 2 interview, or for those who died without a recorded date of death, we censored their follow‐up time at the midpoint of the first and last Wave 2 interviews. To estimate mortality risk, we fitted a Cox proportional hazards model with social engagement, education attainment, their interaction, and the covariates listed in section 2.2.4. The model was specified as: hit=h0texpγTwi=h0texpγ1SocialEngagementi+∑nγnCovariateni where ℎ0 (𝑡) is the baseline hazard function, ℎ𝑖 (𝑡) is the hazard of mortality for individual 𝑖 at time t, and 𝑤𝑖 represents a vector of covariates with corresponding regression coefficients 𝛾. Joint models We used a joint modeling framework to link the longitudinal and survival processes through a shared random effect. 52 , 53 The shared random effect links individual differences in cognitive function over time with mortality risk. 54 , 55 The general formula of the model is: ht=h0texpγTwi+αTbi where α quantifies how individual‐level deviations in cognitive function (random effects bi) are associated with mortality risk. We implemented the joint model using the INLAjoint package (version 24.3.25) developed by Rustand et al. in R, 54 which approximates Bayesian inference using the integrated nested Laplace approximation (INLA). INLA provides fast, accurate approximations to the posterior distributions of model parameters and avoids time‐consuming sampling. 56 We modeled mortality as the time‐to‐event outcome, rather than incident dementia, because the survival sub‐model was specified to address potentially informative censoring of cognitive observations due to mortality rather than to estimate dementia incidence. Sensitivity analysis We conducted six sensitivity analyses. First, we repeated the analyses using a continuous social engagement score, constructed as the reverse‐coded social isolation index, with a higher score indicating greater social engagement. Second, to assess potential heterogeneity within the “any formal education” category, we repeated the joint models using a three‐level educational attainment measure (no formal education, less than primary school or primary school completed, and middle school or higher). Third, we excluded self‐reported ADLs from the models to reduce potential reverse causation, as ADL limitations may reflect functional consequences of cognitive decline rather than a pre‐existing confounder. Fourth, we expanded the analytic sample to include participants from the Wave 2 refresher cohort, increasing the sample size from 3869 to 5833 individuals. Because these participants had only baseline cognitive assessments from their Wave 2 interviews but no follow‐up cognitive or mortality information, we applied a multivariable‐adjusted mixed‐effects model rather than a joint model for this sensitivity analysis. To account for differences in the interval between the social engagement assessment and the baseline cognitive evaluation across cohorts, we incorporated a baseline‐wave indicator in this model. Fifth, we repeated the longitudinal analysis using a generalized estimating equation (GEE) model weighted for the inverse probabilities of mortality and attrition as an alternative approach to addressing mortality‐related attrition. We applied the LASI‐DAD Wave 2 attrition weight, which jointly accounts for study attrition and mortality, and used the same fixed‐effect structure as in the primary longitudinal sub‐model. Finally, we compared baseline characteristics across participants who completed the Wave 2 cognitive assessment, died before Wave 2, or were alive but did not complete the Wave 2 cognitive assessment. RESULTS Our analytical sample included 3869 participants (mean age [SD]: 69.61 [7.44] years; n = 1897 [49%] with no formal education; Table 1). Of them, 67.2% (2598/3869) were classified as having higher social engagement. Most participants identified as Hindu and belonged to the other backward class or none/other caste categories (Table 1). Compared to those with lower social engagement, those with higher social engagement were younger (mean age 68.55 vs. 71.80 years; P < 0.001); more likely to be male (54% vs. 29%; P < 0.001); and more likely to report higher household per capita consumption (P < 0.001), more favorable childhood financial circumstances (P = 0.010), and better childhood health (P = 0.026). They were also more likely to have attained any formal education (58% vs. 37%; P < 0.001) and to have parents with any education (P < 0.001). Social engagement status did not differ significantly by rural residence, depressive symptoms, or the presence of chronic conditions. Those with higher social engagement were less likely to report difficulties with activities of daily living (20% vs. 27%; P < 0.001; Table 1). TABLE 1Participant characteristics, overall and by baseline social engagement level, harmonized Longitudinal Aging Study in India—Diagnostic Assessment of Dementia (LASI‐DAD), 2017–2019. TotalLess social engagementMore social engagement p value 2 N = 3869 1 N = 1271/3869 (32.85%) 1 N = 2598/3869 (67.15%) 1 Survival status <0.001Alive2970 (77%)908 (71%)2062 (79%) Died899 (23%)363 (29%)536 (21%) Age 69.61 (7.44)71.80 (8.49)68.55 (6.60)<0.001 Sex <0.001Female2104 (54%)901 (71%)1203 (46%) Male1765 (46%)370 (29%)1395 (54%) Religion 0.003Hindu3074 (79%)1050 (83%)2024 (78%) Muslim485 (13%)133 (10%)352 (14%) None/others310 (8.0%)88 (6.9%)222 (8.5%) Caste <0.001None/others1315 (34%)365 (29%)950 (37%) Scheduled caste714 (18%)268 (21%)446 (17%) Scheduled tribe201 (5.2%)73 (5.7%)128 (4.9%) Other backward class1639 (42%)565 (44%)1074 (41%) Household per capita consumption <0.001Q1724 (19%)309 (24%)415 (16%) Q2808 (21%)270 (21%)538 (21%) Q3797 (21%)253 (20%)544 (21%) Q4769 (20%)229 (18%)540 (21%) Q5771 (20%)210 (17%)561 (22%) Childhood health 0.026Very good2086 (54%)649 (51%)1437 (55%) Good1392 (36%)477 (38%)915 (35%) Fair/poor/very poor391 (10%)145 (11%)246 (9.5%) Family financial status during childhood 0.010Pretty well off413 (11%)112 (8.8%)301 (12%) Average2047 (53%)666 (52%)1381 (53%) Poor1409 (36%)493 (39%)916 (35%) Rural residence 0.139Urban community1461 (38%)459 (36%)1002 (39%) Rural village2408 (62%)812 (64%)1596 (61%) Highest education level <0.001Never attended school1897 (49%)796 (63%)1101 (42%) Less than primary school (standard 1–4)519 (13%)171 (13%)348 (13%) Primary school (standard 5–7)499 (13%)128 (10%)371 (14%) Middle school (standard 8–9)301 (7.8%)55 (4.3%)246 (9.5%) Secondary school (standard 10–11)363 (9.4%)79 (6.2%)284 (11%) Higher secondary (standard 12)113 (2.9%)19 (1.5%)94 (3.6%) Above standard 12177 (4.6%)23 (1.8%)154 (5.9%) Mother's education <0.001None3341 (86%)1119 (88%)2222 (86%) Any formal education419 (11%)103 (8.1%)316 (12%) Missing109 (2.8%)49 (3.9%)60 (2.3%) Father's education <0.001None2677 (69%)944 (74%)1733 (67%) Any formal education1007 (26%)255 (20%)752 (29%) Missing185 (4.8%)72 (5.7%)113 (4.3%) Depression 0.071No3548 (92%)1151 (91%)2397 (92%) Yes321 (8.3%)120 (9.4%)201 (7.7%) ADL <0.001No2994 (77%)923 (73%)2071 (80%) Yes875 (23%)348 (27%)527 (20%) Chronic conditions 0.105None1587 (41%)494 (39%)1093 (42%) 1–2 conditions1920 (50%)646 (51%)1274 (49%) ≥ 3 conditions362 (9.4%)131 (10%)231 (8.9%) Abbreviations: ADL, activities of daily living; SD, standard deviation. 1 Mean (SD); n (%). 2 Welch two‐sample t test; Pearson chi‐squared test.John Wiley & Sons, Ltd. Participant characteristics, overall and by baseline social engagement level, harmonized Longitudinal Aging Study in India—Diagnostic Assessment of Dementia (LASI‐DAD), 2017–2019. Abbreviations: ADL, activities of daily living; SD, standard deviation. Mean (SD); n (%). Welch two‐sample t test; Pearson chi‐squared test. Table 2 presents the results from multivariable‐adjusted joint models for the association between social engagement and general and domain‐specific cognitive function. Higher social engagement was associated with better general cognitive function (0.145 SD units, 95% confidence interval [CI]: 0.076, 0.214), executive function (0.104 SD units, 95% CI: 0.036, 0.171), memory (0.110 SD units, 95% CI: 0.031, 0.189), and visuospatial ability (0.187 SD units, 95% CI: 0.120, 0.254), compared to lower social engagement (Table 2, Figures 1 and 2). A similar but non‐statistically significant pattern was observed for language (0.064 SD units, 95% CI: −0.009, 0.136; Table 2, Figures 1 and 2). To contextualize the magnitude of these estimates, for example, the 0.145‐SD higher baseline general cognitive function score associated with higher social engagement was approximately equivalent to 1.8 years of annual cognitive decline as estimated in the model with general cognitive function as the outcome (−0.082 SD units per year; 95% CI: 0.145, 0.082; Table 2, Figures 1 and 2). Participants with higher social engagement had slower annual change in general cognitive function (0.006 SD units; 95% CI: −0.010, 0.021), executive function (0.004 SD units, 95% CI: −0.009, 0.016), memory (0.004 SD units; 95% CI: −0.011, 0.020), and language (0.014 SD units; 95% CI: −0.004, 0.033) than those with lower social engagement, whereas for visuospatial ability the estimated difference was slightly negative (−0.003 SD units; 95% CI: −0.012, 0.007). However, these longitudinal associations were weak in magnitude and imprecise, with all credible intervals crossing the null (Table 2; Figures 1, 2). Interaction terms between social engagement and education attainment were not statistically significant across cognitive domains, except for visuospatial ability. The negative interaction (visuospatial ability: β = −0.116; 95% CI: −0.204, −0.029) indicated that the positive association between higher social engagement and baseline performance was weaker among adults with no formal education than among those with any formal education (Table 2). No statistically significant associations were observed for the three‐way interactions involving time in any cognitive domain (Table 2). TABLE 2Multivariable joint model results for the association between baseline social engagement level and domain‐specific cognitive function: Harmonized Longitudinal Aging Study in India—Diagnostic Assessment of Dementia (LASI‐DAD), 2017–2024.CharacteristicsGeneral cognitive functionExecutive functionMemory functionLanguage functionVisuospatial abilityLongitudinal modelSurvival modelLongitudinal modelSurvival modelLongitudinal modelSurvival modelLongitudinal modelSurvival modelLongitudinal modelSurvival modelBeta95% CIHR95% CIBeta95% CIHR95% CIBeta95% CIHR95% CIBeta95% CIHR95% CIBeta95% CIHR95% CI Social engagement level Less social engagementrefrefrefrefrefrefrefrefrefrefrefrefrefrefrefrefrefrefrefrefMore social engagement0.1450.076, 0.2140.8470.677, 1.0460.1040.036, 0.1710.8500.681, 1.0470.1100.031, 0.1890.8470.679, 1.0440.064−0.009, 0.1360.8540.684, 1.0520.1870.120, 0.2540.8470.681, 1.043 Time −0.082−0.097, −0.067——−0.082−0.095, −0.069——−0.063−0.080, −0.047——−0.041−0.068, −0.015——−0.030−0.039, −0.021—— Time x social engagement level Time x less social engagementrefref——refref——refref——refref——refref——Time x more social engagement0.006−0.010, 0.021——0.004−0.009, 0.016——0.004−0.011, 0.020——0.014−0.004, 0.033——−0.003−0.012, 0.007—— Educational attainment Any formal educationrefrefrefrefrefrefrefrefrefrefrefrefrefrefrefrefrefrefrefrefNo formal education−0.744−0.822, −0.6651.3121.030, 1.648−0.685−0.762, −0.6071.3011.022, 1.632−0.587−0.677, −0.4971.2831.008, 1.609−0.400−0.483, −0.3171.3041.025, 1.636−0.707−0.783, −0.6301.3131.032, 1.645 Time x educational attainment Any formal education x timerefref——refref——refref——refref——refref——No formal education x time0.0450.028, 0.063——0.0230.009, 0.036——0.0350.018, 0.052——0.0480.028, 0.069——0.001−0.010, 0.011—— Social engagement level x educational attainment Less social engagement x any formal educationrefrefrefrefrefrefrefrefrefrefrefrefrefrefrefrefrefrefrefrefMore social engagement x no formal education−0.048−0.138, 0.0421.1270.843, 1.475−0.069−0.158, 0.0201.1190.839, 1.4630.006−0.097, 0.1101.1170.838, 1.4600.019−0.076, 0.1141.1030.828, 1.441−0.116−0.204, −0.0291.1190.841, 1.459 Social engagement level x educational attainment x time Less social engagement x any formal education x timerefref——refref——refref——refref——refref——More social engagement x no formal education x time−0.014−0.034, 0.006——−0.001−0.017, 0.015——−0.009−0.029, 0.012——−0.017−0.041, 0.007——−0.005−0.017, 0.007—— α ——0.5130.452, 0.576——0.5730.511, 0.647——0.6250.562, 0.691——0.6230.547, 0.704——0.6980.625, 0.780 Note: All models were adjusted for baseline covariates: age (and age2), sex, religion, caste, household per‐capita consumption, childhood health, childhood financial status, rural residence, parental education, depression, ADL difficulty, and chronic conditions. Domain‐specific model time interaction terms: General cognition model: time × age; time × religion; time × childhood health; time × rural residence. Executive function model: time × caste; time × religion; time × childhood health; time × rural residence. Memory model: time × age; time × religion; time × childhood health; time × ADL; time × rural residence; time × chronic conditions. Language model: time × age; time × religion; time × childhood health; time × childhood financial status; time × rural residence; time × household consumption. Visuospatial model: time × age; time × sex; time × religion; time × childhood health.Abbreviations: ADL, activities of daily living; CI, credible interval; HR, hazard ratio.John Wiley & Sons, Ltd. Multivariable joint model results for the association between baseline social engagement level and domain‐specific cognitive function: Harmonized Longitudinal Aging Study in India—Diagnostic Assessment of Dementia (LASI‐DAD), 2017–2024. Note: All models were adjusted for baseline covariates: age (and age2), sex, religion, caste, household per‐capita consumption, childhood health, childhood financial status, rural residence, parental education, depression, ADL difficulty, and chronic conditions. Domain‐specific model time interaction terms: General cognition model: time × age; time × religion; time × childhood health; time × rural residence. Executive function model: time × caste; time × religion; time × childhood health; time × rural residence. Memory model: time × age; time × religion; time × childhood health; time × ADL; time × rural residence; time × chronic conditions. Language model: time × age; time × religion; time × childhood health; time × childhood financial status; time × rural residence; time × household consumption. Visuospatial model: time × age; time × sex; time × religion; time × childhood health. Abbreviations: ADL, activities of daily living; CI, credible interval; HR, hazard ratio. Predicted general cognitive function scores (2017–2024) by social engagement level and education status—the harmonized Longitudinal Aging Study in India—Diagnostic Assessment of Dementia (LASI‐DAD). Predicted domain‐specific cognitive function scores (2017–2024) by social engagement level and education status—the harmonized Longitudinal Aging Study in India—Diagnostic Assessment of Dementia (LASI‐DAD). In the survival sub‐models, higher social engagement was associated with a lower hazard of mortality, although the credible interval crossed the null (e.g., general cognitive function hazard ratio = 0.847; 95% CI: 0.677, 1.046; Table 2 ). Higher cognitive function scores were associated with a lower likelihood of mortality during the follow‐up (e.g., general cognitive function α = 0.513, 95% CI: 0.452, 0.576; Table 2). Results from the sensitivity analysis using a continuous social engagement score were consistent with the main findings (Table S1 in supporting information). Higher social engagement was associated with better baseline cognitive performance across domains. The education by social engagement interaction was significant for visuospatial ability and executive function, while no three‐way interactions involving time were statistically significant. In the three‐level education sensitivity analysis, all credible intervals for three‐way interaction included the null, consistent with the primary findings for cognitive change. For baseline memory, positive social engagement‐by‐education interactions were observed in both lower education groups versus the middle school or higher group. In contrast, the visuospatial interaction observed in the primary analysis was attenuated and included the null. Thus, the domain‐specific interaction patterns differed across education parameterizations, but neither parameterization showed consistent effect modification across cognitive domains (Table S2 in supporting information). Findings from the additional sensitivity analyses (i.e., excluding self‐reported ADLs, expanding the analytic sample to include the Wave 2 refresher cohort, and GEE models weighted for the inverse probabilities of mortality and attrition) were also consistent with the main results (Tables S3–S5 in supporting information). Finally, participants who died before Wave 2 had lower baseline general and domain‐specific cognitive function scores, lower social engagement scores, older age, and a higher prevalence of ADL difficulty and chronic conditions (P values < 0.05; Table S6 in supporting information), supporting the use of joint models to handle informative censoring due to mortality when estimating cognitive change. DISCUSSION Using data from a large, population‐based longitudinal study of older adults in India, we found individuals with higher baseline social engagement exhibited better baseline performance in general cognitive function and cognitive domains of executive function, memory, and visuospatial ability. However, baseline social engagement was not associated with the rate of cognitive change over time. These findings suggest social engagement may be related to the accumulation or maintenance of cognitive reserve and a higher level of cognitive functioning in later life but has limited influence on the aging‐related cognitive change. Overall, there was no consistent evidence that educational attainment modified associations between social engagement and cognitive outcomes, in contrast to both resource substitution and resource multiplication theories. 36 Our results underscore the value of both early‐life education and later life social engagement for later life cognitive function in India. , Our finding that social engagement is associated with better cognitive function is consistent with previous evidence from both high‐income country 57 , 58 and LMIC 19 , 59 , 60 settings. For example, analyses from the nationally representative China Health and Retirement Longitudinal Study (CHARLS) demonstrated that participation in a variety of social activities, including card games, television watching, and outdoor activities, are positively associated with later life cognitive function, with loneliness proposed as a potential mediating pathway. 59 Similarly, a cross‐national study spanning six LMICs reported that higher levels of social participation were linked to lower odds of mild cognitive impairment. 19 However, these studies were cross‐sectional and limited in their ability to assess long‐term trajectories of cognitive change. Our study builds on this evidence by using longitudinal data to estimate cognitive change in a large cohort of older adults in India. While our results warrant replication in other populations and settings, they suggest that social engagement is associated with higher levels of cognitive function in later life, potentially reflecting cognitive reserve, rather than with the neurobiological aging processes that drive later life cognitive decline. 61 Consistent with our findings but in a different population and setting, the Whitehall II cohort study of British civil servants reported that cognitive reserve‐related factors such as occupation and education were strongly associated with baseline cognitive function but not the slope of decline over a 10‐year period. 62 Similar patterns have been observed in other longitudinal cohorts, in which cognitive reserve proxies show strong associations with baseline cognitive level but little or inconsistent influence on subsequent rates of decline. 63 , 64 , 65 Together, these results support the view that cognitive reserve‐related factors, including social engagement, may assist in prolonging the maintenance of functioning in the presence of neurodegeneration but may not alter the pace of decline once the compensatory mechanisms are overwhelmed. 62 , 66 Another finding of this study was the domain‐specific interaction between social engagement and educational attainment observed for visuospatial ability. The positive association between higher social engagement and baseline visuospatial performance was weaker among adults with no formal education than among those with any formal education. One plausible explanation for this domain‐specific pattern is that visuospatial performance may be particularly sensitive to formal schooling and literacy‐related experiences. In the harmonized LASI‐DAD, visuospatial function was assessed using constructional praxis tasks, including drawing a circle, rectangle, cube, and diamond, as well as copying interlocking pentagons. 42 These tasks require graphomotor coordination, spatial organization, familiarity with geometric forms, and literacy‐related visual processing, all of which may be shaped by formal schooling. 67 , 68 , 69 A study among Korean older adults found that visuo‐constructional performance differed significantly by literacy and education level, with the uneducated–illiterate group scoring substantially lower than both educated–literate and uneducated–literate groups. 67 Neuroimaging research further demonstrates that learning to read and acquiring basic schooling boost reorganization of the visual cortices and enhance neural responses across visual stimulus categories, 68 offering biological plausibility for an education modification specific to visuospatial processing. However, this interaction should be interpreted with caution. The visuospatial interaction was attenuated and included the null when education was categorized into three levels, whereas positive interactions were observed for baseline memory in both lower education groups. Moreover, interaction estimates for general cognition, executive function, and language were small, inconsistent in direction, and statistically non‐significant. Therefore, the visuospatial interaction may reflect a domain‐specific education mechanism rather than evidence of a broader modifying effect of education on the relationship between social engagement and cognitive function. Limitations and strengths This study has limitations. First, cognitive function was measured at only two time points, which limits the precision of slope estimates and precludes the ability to capture non‐linear patterns of cognitive change. Second, reverse causation is possible because individuals with poorer cognitive function or cognitive impairment at baseline may have withdrawn from social activities or been less likely to maintain social relationships. We were unable to explore this possibility, as social engagement was measured once at baseline in core LASI, which currently only has a single wave available. Future studies should consider incorporating time‐varying measures of social engagement and additional follow‐up waves to better characterize long‐term trends and more accurately examine the dynamic relationship between social engagement and cognitive aging. The generalizability of these findings should be considered in light of the cohort's educational and sociocultural context. Nearly half of participants had no formal education, reflecting the historical educational opportunities of the current generation of older adults in India. By including a large proportion of older adults with no formal education, this study extends evidence to a population that remains underrepresented in cognitive aging research. However, additional research is needed to understand the extent to which the current findings may generalize to younger birth cohorts in India or to more highly educated populations. This study also has several strengths. This is the first investigation of social engagement and cognitive function and change in a large, population‐based study of older adults in India, the world's most populous country with a rapidly aging population. We used a joint modeling approach that simultaneously accounted for mortality over the follow‐up while estimating cognitive change over time, reducing the likelihood of selection bias due to differential mortality. Our use of general and domain‐specific cognitive outcomes enabled us to capture heterogeneity across cognitive domains and provided more nuanced insights into underlying mechanisms. The general and domain‐specific cognitive scores that we used have been validated against clinical diagnoses and are associated with polygenic risk scores for Alzheimer's disease in the LASI‐DAD sample, underscoring their clinical relevance. 70 , 71 Further, these cognitive scores are harmonized with those used in the US Health and Retirement Study HCAP and its network of International Partner Studies, facilitating comparison to other studies in future research. Finally, our measurement of social engagement incorporated specific factors that were culturally relevant to the Indian context, enhancing the validity of our findings and supporting their applicability to other LMICs with similar social structures and education distributions. Conclusion Understanding modifiable risk and protective factors for dementia among older adults in LMICs is a critical need given rapidly aging populations. In a large, population‐based cohort study of older adults in India, we found that higher baseline social engagement was consistently associated with better baseline cognitive function across multiple domains but not cognitive change over time, with no consistent evidence that these associations differed by formal education attainment. These findings suggest social engagement may be a potentially modifiable population‐level factor relevant to cognitive health in India and may offer insights for other LMICs with similar social structures and education systems. Future research with repeated assessments of social engagement and extended follow‐up is needed to clarify its associations with long‐term cognitive aging. CONFLICT OF INTEREST STATEMENT The authors declare no conflicts of interest. Author disclosures are available in the Supporting Information. CONSENT STATEMENT All human subjects provided informed consent to participate in this research. Supporting information Supporting Information Supporting Information