Clinical clusters during acute illness predict long-term mortality in older patients

Area of Science:

• Gerontology
• Computational Biology
• Clinical Medicine

Background:

• Acute illness decompensation lacks individualized treatment approaches for older patients.
• Multi-modal and high-dimensional data offer potential for deriving clinically meaningful patient clusters.
• This study explores cluster-driven and high-dimensional predictors for acute illness decompensation, including brain involvement.

Purpose of the Study:

• To test the hypothesis that cluster-driven and high-dimensional predictors can be constructed for clinical deployment.
• To gain mechanistic insights into the pathophysiology of acute illness decompensation, particularly affecting the brain.
• To improve the prediction of adverse outcomes in older adults experiencing acute illness.

Main Methods:

• Harmonization of two independent prospective cohort studies (DELPHIC and DECIDE) for training and testing.
• Application of T-stochastic neighbor embedding and agglomerative hierarchical clustering to identify patient subtypes.
• Comparison of predictive performance for brain decompensation and 2-year mortality using cluster-driven and high-dimensional models, with SHAP values for input contribution analysis.

Main Results:

• Identification of three broad clinical subtypes in older adults during acute illness, with contributions from both baseline and acute variables.
• Significant improvement in the area under the receiver operating characteristic curve (AUROC) for predicting brain decompensation, from 0.563 (baseline) to 0.641 (cluster-driven) and 0.797 (high-dimensional).
• Sleep-wake cycle disturbance identified as the primary predictor of delirium; physiological fluctuations, especially cognitive, predicted long-term mortality.

Conclusions:

• Robust, generalizable, and clinically useful clusters for older patients during acute illness are identifiable.
• Demonstration of a proof-of-concept for a longitudinal approach to defining and modeling acute illness.
• Highlighting the potential of high-dimensionality and multi-modality for predicting adverse outcomes and identifying sleep-wake cycle disturbances as a key target for delirium research.