Risk prediction models for sepsis-associated encephalopathy: a systematic evaluation and meta-analysis

TL;DR

This study evaluates the quality and effectiveness of risk prediction models for sepsis-associated encephalopathy, finding that while some models show promise, they all have significant biases.

Contribution

The study provides a systematic evaluation and meta-analysis of existing SAE risk prediction models, highlighting their limitations.

Findings

Age and SOFA score are the most commonly used and significant predictors in SAE models.

The pooled AUC for validated models was 0.83, indicating fair discrimination.

All studies had a high risk of bias, mainly due to data source and analysis issues.

Abstract

The number of risk prediction models for sepsis-associated encephalopathy (SAE) is increasing, while the quality and applicability of these models in clinical practice and future research remain uncertain.

To systematically review published studies on SAE risk prediction models.

Systematic review and meta-analysis of observational studies.

A systematic search of PubMed, Web of Science, Embase, Wanfang, VIP, and CNKI databases was conducted from inception to April 2, 2025, to identify studies on SAE risk prediction models. Two independent reviewers screened the studies and extracted data. The Prediction model Risk Of Bias Assessment Tool (PROBAST) was applied to evaluate the risk of bias and applicability of the included studies.

A total of 1,994 studies were identified, and 10 were included after screening. The reported incidence of SAE ranged from 15.16% to 63.3%. Age and Sequential Organ Failure Assessment (SOFA) score are the most frequently adopted factors with significant predictive value, both of which were incorporated into five models. Both the SOFA score and age were significantly associated with SAE. In studies with available data, the odds ratio (OR) for age ranged from 1.084 to 1.018, while that for SOFA score ranged from 1.246 to 2.416. The area under the receiver operating characteristic curve (AUC) for the 10 studies ranged from 0.743 to 0.975. All studies were found to have a high risk of bias, primarily due to inappropriate data sources and deficiencies in the analysis domain. The pooled AUC for the six validated models was 0.83 (95% confidence interval [0.77–0.89]), indicating fair discrimination.

Although the included studies reported some discrimination in the SAE prediction models, all were found to have a high risk of bias according to the PROBAST checklist.

This study protocol was registered on PROSPERO (registration number: CRD420251012485).

Figures