Perceived social support and life engagement among older adults using thermal health tourism services: the moderating role of aging anxiety

Introduction

Population aging is a central public health and social policy issue. The World Health Organization (1) estimates that the number of people aged 60 years and older will rise from 1.1 billion in 2023 to 1.4 billion by 2030, and the United Nations projects continued growth in the oldest-old population (2). Later-life health encompasses mental health, social connectedness, everyday functioning, and longevity (3, 4). Life engagement and social connection therefore represent central dimensions of healthy aging.

Perceived social support is consistently associated with later-life well-being. Stronger social support is associated with better quality of life (5), and loneliness partly accounts for the association between support and physical and mental outcomes (6). Institutional research also links social support with psychological pathways involved in successful aging (7). Support from family, friends, and close others occupies distinct positions within an older adult's social network (8). Living arrangements, widowhood, retirement, and health limitations shape the availability and salience of these ties.

Life engagement reflects meaningful goals, sustained involvement, and the value assigned to daily activities (9, 10). Purpose-related functioning is associated with psychological well-being and health-promoting behavior in older age (11, 12). Longitudinal evidence links favorable social-network types with stronger purpose in life (8), and social participation is associated with mental well-being through purpose in life (13). These findings place social context within the empirical study of later-life engagement.

Aging-related perceptions form the psychological context in which social resources are experienced. Positive self-perceptions of aging are associated with quality of life (4) and psychological well-being (14), while anxiety symptoms are common and underdetected in older populations (3, 15). Interventions directed at self-perceptions of aging have produced improvements in psychosocial outcomes (16), and socioeconomic position is associated with aging anxiety through stressors and aging perceptions (17). These findings support a resource-threat framework in which perceived support represents a relational resource, aging anxiety represents an aging-related threat appraisal, and life engagement represents meaningful involvement in life.

Thermal and wellness travel provides a health-seeking service context for studying these psychosocial associations. Tourism research links later-life travel with healthy aging, social participation, and age-friendly service design (18). Spa motivations are associated with post-trip life satisfaction among older visitors (19), and wellness tourism experiences are associated with restorative appraisals and well-being (20). Existing studies emphasize motivations, segmentation, life satisfaction, and broad well-being outcomes. Source-specific support, aging anxiety, and life engagement remain less developed within this recruitment context.

The study addressed three empirical gaps: the association between perceived social support and life engagement among older thermal-service users; the interaction between perceived support and aging anxiety; and the overlapping contributions of significant-other, family, and friend support, which have distinct associations with later-life adaptation and purpose (6, 8).

The resource-threat framework generated three hypotheses. Higher perceived social support was expected to be associated with higher life engagement (H1), higher aging anxiety was expected to be associated with lower life engagement (H2), and the positive association between perceived social support and life engagement was expected to be smaller at higher levels of aging anxiety (H3). Source-specific support analyses were designated secondary because the MSPSS subscales represent correlated components of the same support system. The thermal facility served as the recruitment setting; facility characteristics were not modeled as explanatory exposures.

Methods

Study design and setting

This cross-sectional, questionnaire-based observational study followed the Strengthening the Reporting of Observational Studies in Epidemiology recommendations (21). Participants were recruited at a single thermal health tourism facility in Türkiye during the approved field period. The facility provided thermal health and wellness services to older adults and served solely as the recruitment setting for the study. The facility name and more specific geographical information are not reported to preserve institutional confidentiality. No facility-level exposure was included in the analysis. All observations came from one site; therefore, facility codes, facility fixed effects, and cluster-robust standard errors were not applicable.

Participants and procedure

A facility-based convenience sample included service users aged 65 years or older who understood the study information, provided digital informed consent, and completed the electronic questionnaire in Turkish. Age eligibility was self-reported at enrollment. No formal cognitive screening instrument was administered. The exact age category was unavailable for 17 enrolled participants and was retained as missing in the analytic file. The source dataset contained submitted questionnaires; counts of people approached, declining, or leaving the form before submission were not recorded, so an approach-based participation rate was unavailable.

The standardized electronic questionnaire contained a sociodemographic form and three psychometric instruments. Study information and digital consent preceded the survey items. Participation was voluntary, responses were confidential, and discontinuation carried no penalty. The electronic form applied the same item wording, response options, and order to all respondents. The analytic dataset contained no interviewer identifier or directly identifying participant field.

Measures

Perceived social support

Perceived social support was assessed with the 12-item Multidimensional Scale of Perceived Social Support (MSPSS) (22) and its Turkish version (23). Items use a 7-point response scale and form significant-other, family, and friend subscales. The total score was the mean of available items when at least 10 of 12 items were answered. Each four-item subscale required all four responses. Higher scores indicated greater perceived social support.

Life engagement

Life engagement was measured with the six-item Life Engagement Test (24) and its Turkish adaptation (25). Responses ranged from 1 (strongly disagree) to 5 (strongly agree). Items 1, 3, and 5 were reverse-scored. The life engagement score was the mean of available items when at least five of six items were answered. Higher scores indicated greater life engagement.

Relational aging anxiety

Aging-related concerns were measured with the 16-item Turkish Relational Aging Anxiety Scale (26), developed from the relational ageism framework (27). Responses ranged from 1 to 5. Items 3, 8, 11, and 13 were retained in the field direction; Items 1, 2, 4, 5, 6, 7, 9, 10, 12, 14, 15, and 16 were reverse-scored so that higher values indicated greater aging anxiety. The total score was the mean of available items when at least 13 of 16 items were answered. The total score was treated as an operational composite; subdimension scores were excluded from hypothesis testing.

Sociodemographic and study-specific variables

Covariates were age group (65–74, 75–84, ≥85 years), sex, educational level, thermal-facility visit frequency, primary visit purpose, chronic disease status, living arrangement, and social insurance status. All covariates were categorical. Reference categories were 65–74 years, men, university/postgraduate education, one visit every several years, physical treatment, chronic disease present, living with a spouse, and social insurance present. The model contained 17 covariate indicator variables.

Data management and statistical analysis

The source file contained 425 submitted questionnaires. Data screening covered response ranges, missing values, response structure, and exact duplicate patterns. Duplicate detection used all 42 non-timestamp fields because participant identifiers were absent. Ten later records matched an earlier record on every non-timestamp field; the first record in each exact-match sequence was retained, yielding 415 unique analytic records. Scale means were computed with the prespecified 80% item-completion rules. Primary regression used complete cases across the outcome, focal predictors, and covariates. A strict complete-item sensitivity analysis required all MSPSS, LET, and aging-anxiety items. A second sensitivity analysis retained the 10 exact-match records.

Cronbach's alpha was calculated among respondents with complete item data for each scale. Means and standard deviations used all scores meeting the 80% rule. Categorical variables were summarized with frequencies, percentages, and explicit missing counts. Pearson correlations used pairwise available scores, with the exact pairwise denominator reported in each table cell. Primary-model complete cases were compared with excluded records on available scale scores using Welch tests and on categorical variables using chi-square tests.

The primary model regressed life engagement on mean-centered perceived social support, mean-centered aging anxiety, their product term, and the prespecified covariates. Ordinary least squares coefficients were reported as unstandardized b values with HC3 heteroskedasticity-consistent standard errors (28). Mean-centering preceded construction of the interaction term (29). The model included 20 predictors plus the intercept. Conditional support slopes were estimated at the aging-anxiety mean and at one standard deviation below and above the mean. The Johnson-Neyman procedure identified the aging-anxiety values for which the adjusted support slope differed from zero. A sensitivity calculation based on G*Power conventions (30) showed that n = 384, α = .05, one tested interaction term, and 20 total predictors corresponded to 80% power for f² = .021.

Sensitivity analyses comprised Huber M-estimation, median quantile regression, strict complete-item scoring, and duplicate-retained estimation. Participant-level nonparametric bootstrap resampling used 2,000 samples, a percentile confidence interval, and random seed 20260731 for the interaction term (31). A formal three-degree-of-freedom support-by-living-arrangement interaction tested slope heterogeneity. Secondary source-specific models entered each MSPSS subscale separately and then simultaneously, with aging anxiety and the covariates retained. General dominance weights averaged incremental R² across every subset of the three support sources. Commonality coefficients decomposed the source-attributable incremental R² into unique, pairwise shared, and three-way shared components. Dominance and commonality intervals used 500 participant-level bootstrap samples with seed 20260731. All tests were two-sided with α = .05.

Ethical considerations

The institutional ethics committee approved the study before participant recruitment, and the participating thermal facility authorized data collection. Digital informed consent was obtained before questionnaire access. Participation was voluntary, withdrawal carried no penalty, and the analytic dataset contained no direct identifiers. The ethics approval identifiers and approved field dates were entered consistently in the submission record and study documentation.

Results

The source file contained 425 questionnaires. Exact-pattern screening removed 10 later duplicate records and retained 415 unique records. Table 1 presents the sample characteristics and missing counts. The primary scale-score denominators were 408 for MSPSS total, 410 for life engagement, and 409 for aging anxiety. Thirty-one records lacked at least one variable required by the adjusted model, producing a primary-model sample of 384.

Table 1. Sample characteristics after duplicate removal.

Note. Percentages use the deduplicated sample (N = 415); missing responses are displayed for every variable with missing data. Sex had no missing values.

Table 2 presents score distributions and internal consistency. Cronbach's alpha was.978 for MSPSS total,.819 for life engagement, and.929 for aging anxiety. Mean scores were 4.62 (SD = 1.98) for MSPSS total, 3.67 (SD = 0.89) for life engagement, and 2.49 (SD = 0.87) for aging anxiety. The MSPSS source subscales also showed high internal consistency (α = .967-.971).

Table 2. Internal consistency and descriptive statistics for study measures.

α was calculated from complete item sets; α denominators are shown in parentheses. Score n reflects the prespecified item-completion rule. MSPSS, Multidimensional Scale of Perceived Social Support.

Table 3 and Figure 1 present the pairwise correlations. MSPSS total was positively correlated with life engagement (r = .440) and negatively correlated with aging anxiety (r = -.477). Aging anxiety was negatively correlated with life engagement (r = -.411). Family support had the largest bivariate correlation with life engagement (r = .452), followed by friend support (r = .407) and significant-other support (r = .380). Correlations among the three support sources ranged from.808 to.861.

Table 3. Pearson correlations among perceived social support, life engagement, and total aging anxiety.

Lower-triangle cells report Pearson r with the pairwise available-case n in parentheses. Variable labels: 1 = MSPSS total; 2 = significant-other support; 3 = family support; 4 = friend support; 5 = life engagement; 6 = aging anxiety.

The adjusted model was statistically significant, robust F(20, 363) = 12.72, p < .001, with R² = .349 and adjusted R² = .313 (Table 4). At the mean aging-anxiety score, perceived social support was positively associated with life engagement [b = 0.106, 95% CI (0.048, 0.164), p < .001]. At the mean support score, aging anxiety was negatively associated with life engagement [b = −0.326, 95% CI (−0.474, −0.179), p < .001]. The interaction was negative [b = −0.061, 95% CI (−0.118, −0.005), p = .034] and added ΔR² = .014.

Table 4. Fully adjusted regression models for life engagement.

Coefficients are unstandardized b values from the primary ordinary least squares model with HC3 standard errors (n = 384). Perceived social support and aging anxiety were mean-centered. Reference categories are shown in the predictor labels. Model R² = .349; adjusted R² = .313; interaction ΔR² = .014; robust F(20, 363) = 12.72, p < .001.

The conditional support slope was 0.160 [95% CI (0.079, 0.241), p < .001] at one standard deviation below the aging-anxiety mean, 0.106 [95% CI (0.048, 0.164), p < .001] at the mean, and 0.052 [95% CI (−0.020, 0.124), p = .154] at one standard deviation above the mean. The Johnson-Neyman boundary was 3.15 on the observed 1–5 aging-anxiety scale; the positive adjusted support slope was statistically significant below this value (Figure 2).

The interaction estimate remained negative in Huber M-estimation [b = −0.051, 95% CI (−0.089, −0.013), p = .008] and the 2,000-sample participant bootstrap [mean b = −0.062, percentile 95% CI (−0.115, −0.012)]. Its confidence interval included zero in median quantile regression [b = −0.042, 95% CI (−0.086, 0.003), p = .069] and in the strict complete-item analysis [n = 334; b = −0.057, 95% CI (−0.120, 0.007), p = .081]. Retaining exact-match submissions produced a similar estimate [n = 392; b = −0.058, 95% CI (−0.114, −0.003), p = .039].

The Breusch-Pagan test detected heteroskedasticity, LM(20) = 61.42, p < .001, supporting HC3 inference. Ramsey RESET showed no evidence of functional-form misspecification, F(1, 362) = 0.13, p = .722. The maximum variance inflation factor was 1.70, maximum Cook's distance was.034, and maximum leverage was.113. A formal support-by-living-arrangement interaction was not significant, robust F(3, 360) = 1.44, p = .230. Available focal scale means were similar between included and excluded records (all p ≥ .136); categorical distributions differed for sex, education, chronic disease, and living arrangement (all p ≤ .016).

The source-specific analysis used 358 records with complete four-item MSPSS subscales and complete model variables (Table 5). Source correlations ranged from.808 to.861, and source-model variance inflation factors ranged from 3.69 to 5.24. Each source was positively associated with life engagement in separate adjusted models: significant-other support b = 0.070, family support b = 0.093, and friend support b = 0.090. In the simultaneous model, the three source coefficients had confidence intervals containing zero. The three sources increased R² from.318 to.346 (incremental R² = .028). General dominance weights allocated 41.2% of this increment to family support, 35.0% to friend support, and 23.8% to significant-other support; the bootstrap intervals overlapped.

Table 5. Support-source overlap, separate adjusted models, and dominance analysis.

Source models included aging anxiety and all primary-model covariates (n = 358). VIF, variance inflation factor. General dominance weights sum to the source-attributable incremental R² (.028); 95% intervals use 500 participant-level bootstrap samples.

Commonality analysis allocated 61.6% of the source-attributable incremental R² to variance shared across all three support sources (Table 6). Unique components were 0.0001 for significant-other support, 0.0027 for family support, and 0.0014 for friend support. Pairwise shared components were positive and small. The decomposition shows that most source-related explanatory variance was common to the three MSPSS subscales.

Table 6. Commonality analysis of MSPSS support-source subscales in relation to life engagement.

The baseline model contained aging anxiety and all primary-model covariates. Baseline R² = .318, full-model R² = .346, source-attributable incremental R² = .028, n = 358. Confidence intervals use 500 participant-level bootstrap samples. Components sum to the incremental R²; negative values, when present in a bootstrap sample, represent suppressor-related overlap.

Discussion

Three findings summarize the analysis. Perceived social support was positively associated with life engagement, aging anxiety was negatively associated with life engagement, and the adjusted support-life engagement association was smaller at higher aging-anxiety values. The interaction explained 1.4% additional variance and was statistically significant in the HC3 model, Huber model, and participant bootstrap. The median quantile and strict complete-item confidence intervals included zero. Source-specific results were dominated by variance shared across significant-other, family, and friend support.

The positive support-life engagement association is consistent with longitudinal evidence linking meaning in life with health, emotional well-being, and daily functioning (32). Active social engagement is associated with health-related quality of life (33), and regular social interaction is associated with affective well-being (34). Nursing-home research also links social support with successful aging through meaning in life (7). The present cross-sectional association extends this pattern to older adults voluntarily using a thermal health tourism service.

The negative interaction is consistent with a resource-threat framework. Research in older adults with chronic illness has documented joint associations among social support, psychological well-being, and burdensomeness (35). Intergenerational support also shows distinct emotional and practical associations with mental health (36). In the present sample, the adjusted support slope declined from 0.160 at low aging anxiety to 0.052 at high aging anxiety, and statistical significance ended at an aging-anxiety score of 3.15. This pattern describes conditional association within the observed sample.

The interaction magnitude was modest. Aging-related negative cognitions are associated with poorer psychological well-being (37), and psychological interventions in older populations are associated with improvements in emotional symptoms and quality of life (38). The current data identify aging anxiety as a variable that accompanies heterogeneity in the support-life engagement association. Longitudinal and intervention studies are required to estimate direction and clinical relevance.

The source-specific coefficients reflect substantial shared variance. Collinearity increases coefficient uncertainty when correlated predictors enter the same model (39). The MSPSS source correlations ranged from.808 to.861, and simultaneous source coefficients had wide intervals containing zero. Separate adjusted models produced positive coefficients for every source, consistent with prior source-specific research (40). These estimates describe correlated dimensions of perceived support and provide no basis for an isolated causal effect of one source.

Dominance analysis compares relative predictor contributions across subset models (41), and variance-decomposition methods quantify shared explanatory variance (42, 43). Family support had the largest point estimate for the general dominance weight, followed by friend and significant-other support. The bootstrap intervals overlapped, and the three-way shared commonality component represented 61.6% of the incremental R². The ranking is a sample-specific description of overlapping support dimensions.

Family and intergenerational ties remain prominent sources of support in later life (36, 44). Friend support also has distinct associations with social participation and loneliness (45). In this sample, family and friend support accounted for the two largest dominance weights, while significant-other support retained a positive separate-model coefficient. The overlapping intervals and commonality results place the three estimates within a shared support structure.

Later-life tourism research connects travel with healthy aging, social participation, and meaningful experience (46, 47). Wellness-tourism research also reports associations among service experiences, restorative appraisal, life satisfaction, and well-being (20). The present facility served as a health-seeking recruitment context. The analysis measured participant psychosocial characteristics and contained no facility-environment exposure. The findings therefore characterize voluntary older users of one thermal service setting.

The results identify two concurrent psychosocial correlates of life engagement: perceived support and aging anxiety. Related work documents social support within psychological-strain models (35) and emotional-symptom interventions among community-dwelling older adults (38). Family-support research also demonstrates strong cultural and contextual variation (48). Assessment studies in thermal-service populations can incorporate both support structure and aging-related concern, with longitudinal follow-up used to establish temporal order.

Limitations

The study used established instruments, an adjusted interaction model, complete covariate reporting, HC3 inference, conditional slopes, and prespecified sensitivity analyses. Its cross-sectional design establishes no temporal or causal direction. All primary variables were self-reported, and common-method variance remains possible. Recruitment at one thermal facility produced a self-selected group of health-seeking older adults; generalizability to the wider older population and other facilities is limited. The invitation denominator was unavailable. Thirty-one records were excluded from the primary model, and exclusion was associated with several categorical characteristics. The strict complete-item interaction confidence interval included zero. The aging-anxiety total was treated as an operational composite without a new factor-analytic validation. The high MSPSS reliability and source correlations indicate substantial item and subscale overlap. Dominance and commonality results are sample-specific variance decompositions without causal or clinical ranking.

Conclusion

Among older adults voluntarily using one thermal health tourism facility, perceived social support was positively associated with life engagement and aging anxiety was negatively associated with life engagement. The positive support-life engagement association was smaller at higher aging-anxiety levels, with a modest increment in explained variance. Sensitivity analyses produced mixed evidence for the interaction. Significant-other, family, and friend support contributed mainly through shared variance. These findings define a cross-sectional psychosocial pattern for this recruitment context.

Generative AI statement

The author(s) declared that generative AI was used in the creation of this manuscript. ChatGPT using GPT-5.6 Sol was used for language editing, statistical-code review, and figure preparation. The author verified the scoring rules, analytic decisions, numerical results, citations, and final text and accepts full responsibility for the integrity and accuracy of the work.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher's note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Supplementary material

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/frhs.2026.1901204/full#supplementary-material

Supplementary material Supplementary material at the source

Footnotes

Edited by: Saif Ul Islam, University of Warwick, United Kingdom

Reviewed by: Sorina Corman, Lucian Blaga University of Sibiu, Romania

Donatila Souza, State University of Campinas, Brazil