Phenotypic Consensus Clustering and Treatment Heterogeneity Analysis in Critically Ill Patients with Comorbid Type 2 Diabetes Mellitus

Abstract

Aims

This study aimed to establish a phenotypic clustering model for critically ill patients with comorbid type 2 diabetes mellitus (CIP T2DM), define distinct subtypes, and analyze differences in clinical characteristics and treatment response.

Methods

Patients from the Medical Information Mart for Intensive Care IV (MIMIC-IV) database were phenotyped based on demographic, physiological, and biochemical parameters, along with critically ill scores, using a consensus clustering algorithm. Subtype validity was assessed using the eICU Collaborative Research Database (eICU). High-frequency combination drug regimens were then extracted to reveal therapeutic heterogeneity among subtypes.

Results

Three subtypes were identified in the MIMIC-IV cohort (n = 6349). The in-group proportions of 0.957, 0.898, and 0.836 in the eICU cohort (n = 1425) show high consistency. These three subtypes are: a) CIP T2DM with severe infection (CIP T2DM-SI) with a mortality of 16.7 % and a post-medication blood glucose (PMBG, first measured 24 h after medication administration) of 8.25 mmol/L; b) CIP T2DM with organ failure (CIP T2DM-OF) with 18.6 % mortality and 8.03 mmol/L PMBG; and c) CIP T2DM under monitoring for continuous observation and evaluation (CIP T2DM-UM) with 8.87 % mortality and 6.98 mmol/L PMBG. Moreover, the three subtypes showed different in-hospital mortality risks under the same medication regimen.

Conclusions

Three phenotypes were identified in CIP T2DM, showing significant heterogeneity in clinical characteristics and prognosis. Personalized interventions for these subtypes may help reduce adverse events and guide precise treatment in practice.