Latent Profiles of Allostatic Load and Their Association With Post-stroke Cognitive Impairment in Elderly Patients: A Nursing Risk Stratification Study

Post-stroke cognitive impairment (PSCI) is common, but the reproducibility and clinical meaning of allostatic load (AL)-based latent profiles remain uncertain. We examined empirical AL profiles and their associations with PSCI in older stroke survivors. This retrospective cross-sectional study included 356 adults aged 60 years or older with stroke. Latent profile analysis used 13 continuous biomarkers. One- through five-class diagonal Gaussian mixture models were fitted with 100 random starts; Bayesian information criterion guided selection among models meeting prespecified convergence, likelihood-replication, and minimum-class-size criteria. PSCI was defined as Mini-Mental State Examination (MMSE) < 27. Adjusted logistic regression, a continuous AL-score model, and biomarker-transformation and MMSE-cutoff sensitivity analyses were performed. The eligible four-profile solution had the lowest BIC (12,033.77; entropy = .908), with 25 (7.0%), 93 (26.1%), 160 (44.9%), and 78 (21.9%) participants. PSCI prevalence increased across profiles (20.0%, 23.7%, 45.6%, and 73.1%; p < . 001). After full adjustment, Profiles 3 (OR = 3.51, 95% CI [1.19, 10.34], p = .023) and 4 (OR = 11.54, 95% CI [3.24, 41.04], p < . 001), but not Profile 2, had higher PSCI odds than Profile 1. The continuous AL score was also associated with PSCI (OR = 1.21 per point, 95% CI [1.11, 1.32], p < . 001). Log transformation of high-sensitivity C-reactive protein and triglycerides selected three rather than four classes (adjusted Rand index = .788), meeting the prespecified instability criterion. Higher multisystem burden was associated with PSCI, but the number of latent profiles was transformation-sensitive. These profiles are exploratory empirical descriptions rather than validated clinical risk categories and require prospective external validation.