High need, low use: separating multidimensional need from formal healthcare use among older adults in China using CHARLS 2020

BACKGROUND: Evidence on healthcare utilization among older adults in China has grown rapidly, but less is known about whether recent formal healthcare contact is aligned with multidimensional need. Because high need and low service use represent distinct processes, this study separated need status from service-use status and examined low formal healthcare use among older adults classified as high need. METHODS: We conducted a cross-sectional secondary analysis of CHARLS Wave 5 (2020), restricted to respondents aged 60 years or older. High need was defined strictly as at least two of four observed need indicators: poor self-rated health, any activities-of-daily-living (ADL) limitation, multimorbidity, and high depressive symptoms (CES-D-10 > = 10). Low formal healthcare use was defined as no outpatient service in the past month and no hospitalization in the past year. We first described a four-cell classification of need status by use status and then fitted the primary model for low use among high-need respondents. Models used logistic regression with community-clustered robust standard errors. We report odds ratios, average marginal effects, predicted probabilities, individual-weighted descriptive estimates, and sensitivity analyses including modified Poisson regression, individual-weighted regression, categorical education, inverse probability weighting, multiple imputation, and an IADL-based threshold validation. RESULTS: Among 11,473 age-eligible respondents, 9,103 had complete information for the strict need-use classification. The high-need/low-use screening-proxy group comprised 2,516 respondents, representing 27.6% of the strict outcome-observed sample and 26.5% after applying the individual response-adjusted CHARLS weight. Among 4,434 high-need respondents, 2,693 had complete covariate data for the primary model; 1,549 (57.5%) reported low formal healthcare use. In the primary model, rural residence was not associated with low use among high-need respondents (adjusted OR 1.02, 95% CI 0.84-1.23; average marginal effect + 0.4% points). Male sex (OR 1.00, 95% CI 0.85-1.17), cognition (OR 1.02 per point, 95% CI 0.99-1.05), education, income, marital status, and surviving children also showed no strong association with low use in the high-need subgroup. By contrast, a sequential model showed that rural residence, lower income, lower cognition, older age, female sex, and not being married/cohabiting were associated with being classified as high need. The need-score threshold of > = 2 provided a pragmatic balance of sensitivity (0.78) and specificity (0.61) against IADL limitation as an external vulnerability marker; the AUC of the strict need score for IADL limitation was 0.745. CONCLUSIONS: A substantial subgroup of older adults in CHARLS 2020 had multidimensional need and no recent outpatient or inpatient use. However, once the analysis was restricted to high-need respondents, rural residence, sex, and cognition were not strong predictors of low formal use. The results support using the high-need/low-use category as a transparent screening proxy for possible need-use misalignment, while cautioning against interpreting full-sample composite-outcome models as determinants of low service use among high-need older adults.