Reserve-informed cardiovascular-kidney-metabolic risk stratification for cardiometabolic multimorbidity progression in middle-aged and older adults: a multi-cohort study

Background Cardiovascular-kidney-metabolic (CKM) assessment may not capture heterogeneity in vulnerability to cardiometabolic multimorbidity (CMM). We evaluated whether objective functional reserve could provide a complementary phenotyping layer. Methods We analysed CHARLS as the development cohort, HRS VBS as the primary supportive replication cohort, ELSA as a simplified transportability cohort, and SHARE W6 DBS as a grip-only robustness cohort. Equal-weight cohort-specific CKM domain counts were crossed with age- and sex-standardized objective functional reserve. Incident CMM was defined as the first follow-up observation with at least two reported cardiometabolic conditions. Cox models were supplemented by discrete-time, common-domain, common-threshold three-domain, grip-only, interaction, CMM-or-death, selection-weighted, landmark, subgroup, and five-year risk-stratification analyses; Fine–Gray and cause-specific models were used where mortality timing was adequate. In the five-year CMM analyses, deaths before CMM were counted separately rather than classified as controls; 1,000 bootstrap resamples were used. Results Descriptive samples comprised 3,238 CHARLS, 1,627 HRS VBS, 1,927 ELSA, and 12,641 SHARE participants, with 284, 117, 73, and 701 incident CMM events. High CKM burden plus low reserve was associated with incident CMM in CHARLS (hazard ratio [HR] 3.52, 95% CI 2.47–5.00), HRS VBS (HR 1.71, 1.03–2.84), ELSA (HR 3.02, 1.62–5.63), and SHARE (HR 1.67, 1.35–2.06). Fine–Gray subdistribution HRs were 1.68 (1.02–2.74) in HRS VBS and 1.68 (1.36–2.09) in SHARE. Statistical interaction was not demonstrated on the multiplicative or approximate additive scale. Five-year discrimination gains after adding reserve were small (apparent ΔAUC 0.0011–0.0150). Conclusions A joint CKM–functional-reserve phenotype identified participants with higher reported CMM incidence. Statistical interaction was not demonstrated on either the multiplicative or approximate additive scale, and the analyses did not establish a clinically deployable prediction tool. Disease outcomes were primarily self-reported. Mortality ascertainment did not support a uniform four-cohort competing-risk analysis. Findings support further evaluation of the phenotyping concept.