Background: Using the Discriminative Dimensionality Reduction Tree (DDRTree) algorithm, a 2-dimensional tree model was constructed based on conventional clinical indicators to identify heart failure (HF) risk phenotypes with different pathophysiological characteristics from the general adult population and evaluate their association with HF prevalence and cardiovascular mortality risk.

Methods: A representative sample of 24 205 general US adults from the National Health and Nutrition Examination Survey 1999-2018 was analyzed. Eleven routine clinical indicators from multiple physiological dimensions were fed into the DDRTree algorithm to project each of them in a 2-dimensional space. The resulting tree structure revealed distinct HF risk phenotypes whose links to outcome HF prevalence and cardiovascular mortality risk were quantified with weighted logistic regression and Cox proportional-hazard models.

Results: Discriminant Dimensionality Reduction Tree identified 3 HF risk phenotypes: normal, metabolic (characterized by upregulated glucose-metabolism markers), and inflammatory (characterized by abnormalities in inflammatory and hepatic-renal markers). After fully adjusting for confounding factors, the metabolic phenotype was significantly associated with the risk of HF compared with the normal phenotype (odds ratio [OR] = 1.96, P < .001, 95% CI: 1.47-2.62), whereas it showed no significant correlation with the risk of cardiovascular death (hazard ratio [HR] = 0.92, P = .539, 95% CI: 0.70-1.20). In contrast, the inflammatory phenotype was significantly correlated with both HF risk (OR = 3.95, P < .001, 95% CI: 2.94-5.33) and cardiovascular mortality risk (HR = 1.47, P = .002, 95% CI: 1.16-1.87).

Conclusions: This study elucidates potential risk phenotypes influencing HF and cardiovascular mortality, providing new insights for precise HF risk stratification and targeted preventive interventions in the general population.