QUANTIFYING THE CONTRIBUTION OF CONFOUNDERS TO THE ASSOCIATION BETWEEN LIVING ALONE AND SUICIDAL IDEATION/PLANNING AMONG KOREAN OLDER ADULTS: A COHORT STUDY
Author(s)
Jimin Do, MS1, Hae Sun Suh, MA, MS, PhD2.
1Department of Regulatory Science, Graduate School, Kyung Hee University, Seoul, Korea, Republic of, 2College of Pharmacy, Kyung Hee University, Seoul, Korea, Republic of.
1Department of Regulatory Science, Graduate School, Kyung Hee University, Seoul, Korea, Republic of, 2College of Pharmacy, Kyung Hee University, Seoul, Korea, Republic of.
OBJECTIVES: Prior Korean studies have identified living alone as a risk factor for suicidal ideation among older adults, yet this association varies depending on which covariates are adjusted for. This study aimed to quantify each covariate's contribution to confounding using Shapley value decomposition.
METHODS: We conducted a cohort study using Waves 19(2024) and 20(2025) of the Korea Welfare Panel Study(KoWePS). The study population comprised adults aged ≥65 years participating in both waves. Single-person household status and covariates were assessed at Wave 19, and past-year suicidal ideation/planning was assessed at Wave 20. Participants with suicidal ideation/planning at Wave 19 were excluded. Covariates included sociodemographic factors(sex, age, education, marital status, region), economic factors(household income, welfare benefit receipt), health factors(self-rated health, chronic disease, depression[CES-D]), and social factors(family and social relationship satisfaction, volunteering). Two analytic approaches were applied: (1) a TabPFN v3.0 S-learner to estimate the average associational effect(ATE) of living alone on suicidal ideation/planning, with 95% CIs via bootstrapping; and (2) ConfoundingSHAP to quantify each covariate's contribution to confounding via Shapley values(φ), defined as the weighted-average change in the effect across all covariate subsets. Analyses were conducted without and with adjustment for CES-D to address its dual role as a confounder and mediator.
RESULTS: A total of 5,171 participants were included, of whom 1,857 lived alone. Suicidal ideation/planning was reported in 1.7% of those living alone and 0.5% of those living with others. The ATE was +0.386%p(95% CI: 0.365-0.407). Without adjustment for CES-D, ConfoundingSHAP identified welfare benefit receipt as the largest contributor to confounding(φ=+0.139%p). After adjustment for CES-D(φ=+0.017%p), its contribution decreased to +0.125%p, suggesting partial confounding by depressive symptoms.
CONCLUSIONS: This study demonstrated that the observed association between living alone and suicidal ideation/planning was influenced by confounding. Quantifying individual confounder contributions provides a clearer basis for interpreting associations and evaluating confounding.
METHODS: We conducted a cohort study using Waves 19(2024) and 20(2025) of the Korea Welfare Panel Study(KoWePS). The study population comprised adults aged ≥65 years participating in both waves. Single-person household status and covariates were assessed at Wave 19, and past-year suicidal ideation/planning was assessed at Wave 20. Participants with suicidal ideation/planning at Wave 19 were excluded. Covariates included sociodemographic factors(sex, age, education, marital status, region), economic factors(household income, welfare benefit receipt), health factors(self-rated health, chronic disease, depression[CES-D]), and social factors(family and social relationship satisfaction, volunteering). Two analytic approaches were applied: (1) a TabPFN v3.0 S-learner to estimate the average associational effect(ATE) of living alone on suicidal ideation/planning, with 95% CIs via bootstrapping; and (2) ConfoundingSHAP to quantify each covariate's contribution to confounding via Shapley values(φ), defined as the weighted-average change in the effect across all covariate subsets. Analyses were conducted without and with adjustment for CES-D to address its dual role as a confounder and mediator.
RESULTS: A total of 5,171 participants were included, of whom 1,857 lived alone. Suicidal ideation/planning was reported in 1.7% of those living alone and 0.5% of those living with others. The ATE was +0.386%p(95% CI: 0.365-0.407). Without adjustment for CES-D, ConfoundingSHAP identified welfare benefit receipt as the largest contributor to confounding(φ=+0.139%p). After adjustment for CES-D(φ=+0.017%p), its contribution decreased to +0.125%p, suggesting partial confounding by depressive symptoms.
CONCLUSIONS: This study demonstrated that the observed association between living alone and suicidal ideation/planning was influenced by confounding. Quantifying individual confounder contributions provides a clearer basis for interpreting associations and evaluating confounding.
Conference/Value in Health Info
2026-11, ISPOR Europe 2026, Vienna, Austria
Value in Health, Volume 29, Issue 12S
Code
MSR133
Topic
Epidemiology & Public Health, Methodological & Statistical Research
Topic Subcategory
Artificial Intelligence, Machine Learning, Predictive Analytics, Confounding, Selection Bias Correction, Causal Inference
Disease
Geriatrics, Mental Health (including addiction)