Unveiling digital phenotypes of acute on chronic heart failure through biomarker- comorbidity interactions: a proof of principle study

Background Acute on chronic heart failure (ACHF) is a complex syndrome with high morbidity, mortality, and healthcare costs. The optimisation of clinical outcomes in this patient population, characterised by comorbidities that exert a significant adverse influence on disease management, necessitates the implementation of personalised risk stratification guiding effective therapeutic strategies. Purpose We hypothesised that mathematically coupled biomarkers known to interact with Elixhauser Comorbidity Ensembles (ECEs) and associated mortality weights would produce synergies to both enhance risk stratification modeling, as well as allow for personalising treatment for ACHF patients. Methods Administrative, demographic and clinical course data within 24h of presentation were extracted under IRB exemption from electronic medical records between January 1, 2020 and September 30, 2024. Generalised regression identified putative features associated with hospital mortality (p<0.05). Youden’s index estimated optimal cut-point for Elixhauser [Mortality Index / Comorbidity Ensemble (n)] Ratio. Boosted Tree modeling estimated each ensemble feature’s proportion (%) of variance explained (VE) in overall model’s accuracy (R2) to estimate risk for hospital mortality. Continuous data were summarised with mean [SD]. Results Patients studied (n=2431) were similarly aged 73 [12] years including females (38%) and males (62%) distributed across Whites (91%), Blacks (4%) and other races (5%) experiencing 19% mortality. Boosted Tree model demonstrated high accuracy in predicting hospital mortality (R2 = 0.74, p<0.05; Area Under Receiver Operating Characteristic Curve = 0.98). Expressing an ECE ratio>4.1 was the most significant predictor contributing 48% VE. Several identified composite biomarkers also provided VE% including pro-BNP/hs-cTnl (19%), lactate/albumin (13%), BUN/albumin (12%), alkaline phosphatase/ albumin (4%), Hb / Albumin (2%), pro-BNP / [AST/SGOT / ALT/SGPT] (2%) and [albumin x platelet] (1%) also contributed significantly to the model's accuracy. Conclusion Integrating mathematically coupled biomarkers with Elixhauser ECEs and linked mortality risk may enhance risk stratification in ACHF patients. This data-driven approach could lead to more accurate risk prediction and personalised treatment strategies, ultimately improving clinical outcomes. The novelty of this approach lies in the integration of mathematically coupled biomarkers, which reflect different aspects of the host response to ACHF and treatment. By combining these ratios, we can capture a more comprehensive picture of the patient's physiological state and risk of mortality. These findings suggest that this approach has the potential to improve the care of ACHF patients and inform development of medical logic models for real-time tracking of clinical course.