Prediction of Delirium Risk in Mild Cognitive Impairment Using Time-Series data, Machine Learning and Comorbidity Patterns -- A Retrospective Study

TL;DR

This retrospective study uses MIMIC-IV v2.2 data to examine delirium risk in patients with mild cognitive impairment (MCI) through comorbidity profiling, survival analysis, and time-series prediction. It identifies distinct comorbidity patterns that elevate delirium risk in MCI and demonstrates that a lightweight LSTM model, trained on longitudinal EHR data with SMOTENC oversampling, achieves strong predictive performance (AUROC $0.93$, AUPRC $0.92$) with calibrated probabilities. Kaplan-Meier analysis shows significantly higher delirium risk and faster delirium onset in the MCI group compared with non-MCI. The work supports integrating time-series ML approaches into risk stratification workflows to mitigate delirium burden in cognitively vulnerable populations.

Abstract

Delirium represents a significant clinical concern characterized by high morbidity and mortality rates, particularly in patients with mild cognitive impairment (MCI). This study investigates the associated risk factors for delirium by analyzing the comorbidity patterns relevant to MCI and developing a longitudinal predictive model leveraging machine learning methodologies. A retrospective analysis utilizing the MIMIC-IV v2.2 database was performed to evaluate comorbid conditions, survival probabilities, and predictive modeling outcomes. The examination of comorbidity patterns identified distinct risk profiles for the MCI population. Kaplan-Meier survival analysis demonstrated that individuals with MCI exhibit markedly reduced survival probabilities when developing delirium compared to their non-MCI counterparts, underscoring the heightened vulnerability within this cohort. For predictive modeling, a Long Short-Term Memory (LSTM) ML network was implemented utilizing time-series data, demographic variables, Charlson Comorbidity Index (CCI) scores, and an array of comorbid conditions. The model demonstrated robust predictive capabilities with an AUROC of 0.93 and an AUPRC of 0.92. This study underscores the critical role of comorbidities in evaluating delirium risk and highlights the efficacy of time-series predictive modeling in pinpointing patients at elevated risk for delirium development.

Figures (4)

• Figure 1: Study cohort selection flowchart from the MIMIC-IV v2.2 database. The flowchart outlines the inclusion and exclusion criteria used to identify the study population, the population was stratified into MCI and non-MCI cohorts, and each group was further divided based on the presence or absence of delirium
• Figure 2: A Lightweight LSTM-Based Architecture for Predicting Delirium Onset Using Sequential Data
• Figure 3: The Kaplan-Meier survival curves comparing the time to delirium onset between the Non-MCI and MCI cohorts. The plot shows a significantly lower survival probability (delirium-free duration) over time in the MCI cohort, indicating a higher risk of developing delirium compared to the Non-MCI group.
• Figure 4: The Receiver Operating Characteristic (ROC) curve for the validation cohort, illustrating the model’s ability to discriminate between patients who developed delirium and those who did not. A high area under the curve (AUC) indicates strong predictive performance..