# Multimorbidity patterns are associated with postoperative delirium in older patients undergoing non-cardiac surgery: an observational study

## TL;DR

This study found that specific combinations of chronic diseases in older patients are linked to a higher risk of post-surgery delirium, with frailty playing a mediating role.

## Contribution

The study identifies distinct multimorbidity patterns and their association with postoperative delirium through latent class analysis and mediation modeling.

## Key findings

- Three distinct multimorbidity subgroups were identified with significant differences in frailty, cognition, and delirium risk.
- Subgroup 2 showed a mediated effect of delirium through frailty after adjusting for age and cognitive performance.
- A multinomial logistic regression model accurately predicted subgroup membership with high AUROC and AUPRC scores.

## Abstract

Multimorbidity is associated with adverse outcomes among older adult surgical patients, yet its role in postoperative delirium (POD) remains unclear. In the present study, we hypothesized that distinct pattern of multimorbidity is associated with increased incidence of POD.

From January 2024 to December 2024, 819 older adult patients were recruited at the Second Affiliated Hospital of Nanjing Medical University. Latent class analysis was used to identify patient subgroups based on disease composition. Mediation effect analysis explored the relationship between subgroups, Edmonton frail scale (EFS), and cognitive performance. Multinomial logistic regression model was employed to predict the subgroup to which patients with different diseases belong.

Three clinically distinct multimorbidity subgroups were identified. Significant differences in EFS, mini-mental state examination (MMSE), and POD were observed among subgroups (p < 0.05). After adjustment for age and MMSE, we found that subgroup 2 mediated the occurrence of POD through frailty [Indirect effect = 0.043; (95%CI = 0.019 ~ 0.070)]. Multinomial logistic regression model demonstrated good predictive power for subgroups, with AUROC scores as follows: subgroup 1 = 0.993, subgroup 2 = 0.977, and subgroup 3 = 0.990. The AUPRC scores were also strong, with subgroup 1 = 0.995, subgroup 2 = 0.886, and subgroup 3 = 0.974.

We identified a specific pattern of multimorbidities significantly associated with frailty, cognitive impairment, and POD risk. The high-risk subgroup’s effect on POD was partially mediated by frailty. Multinomial logistic regression model accurately predicted subgroup membership, offering a potential tool for preoperative risk stratification.