Prediction Models for Acute Kidney Injury in Stroke Patients: A Systematic Review

TL;DR

This paper reviews prediction models for acute kidney injury in stroke patients, finding that while some models show promise, they need improvement to reduce bias and increase reliability.

Contribution

A systematic review of AKI prediction models in stroke patients, highlighting variability in performance and the need for standardized development.

Findings

35 prediction models were identified, with AUC values ranging from 0.428 to 1.000.

Common predictors include hypertension, serum creatinine, age, and NIHSS score.

Models showed significant performance variability and high risk of bias.

Abstract

Introduction

To systematically identify and synthesize the research on prediction models for acute kidney injury (AKI) in stroke patients. Methods: CNKI, Wanfang, VIP, CBM, PubMed, Cochrane Library, Embase, and Web of Science were searched from inception to April 26, 2025. The fundamental characteristics of the included studies were extracted, including model construction, predictors, model performance, and presentation methods. Results: A total of 35 prediction models were identified in this systematic review, with area under the curve (AUC) values ranging from 0.428 to 1.000. Seven studies performed external validation. Common predictors included hypertension, serum creatinine levels, age, diuretic use, mechanical ventilation, and the National Institutes of Health Stroke Scale score (NIHSS). Conclusions: The risk prediction model for AKI in stroke patients still needs to be developed. Despite demonstrating promising predictive capability, the models exhibited significant performance variability and an overall high risk of bias. Future research requires standardized development and validation of models to develop reliable prediction tools with minimal bias and enhanced applicability.

Predictive models can assess the risk of acute kidney injury following stroke. However, there remains a need to develop tools with lower bias and higher applicability in the future.

Figures