Table 1_Risk prediction models for mortality in patients with severe pneumonia: a systematic review and meta-analysis.docx

BackgroundThe number of risk prediction models for mortality in patients with severe pneumonia (SP) is increasing, while the quality and clinical applicability of these models remain unclear. This study aimed to systematically review published research on risk prediction models for mortality in patients with SP.
MethodsPubMed, Embase, Cochrane Library, and Web of Science were searched from inception to August 31, 2024. Data from selected studies were extracted, including study design, participants, diagnostic criteria, sample size, predictors, model development, and performance. The prediction model risk of bias assessment tool was used to assess the risk of bias and applicability. A meta-analysis of the area under the curve (AUC) values from validated models was conducted using Stata 17.0 software.
ResultsA total of 22 prediction models from 18 studies were included in this review, including 15 logistic regression models, two cox proportional regression hazards models, two classification and regression trees, one light gradient boosting machine, and one multilayer perceptron. The reported AUC values ranged from 0.713 to 0.952. Seventeen studies were found to have a high risk of bias, primarily due to inappropriate data sources and poor reporting of the analysis domain. The pooled AUC value of five validated models was 0.85 (95% confidence interval: 0.81–0.88), indicating a fair level of discrimination.
ConclusionAlthough the included studies reported that the risk prediction models for mortality in patients with SP exhibited a certain level of discriminative ability, most of these models were found to have a high risk of bias. Future studies should focus on developing new models with larger sample sizes, rigorous study designs, and multicenter external validation.
Systematic review registrationhttps://www.crd.york.ac.uk/PROSPERO/view/CRD42024589877, identifier: CRD42024589877.

背景
重症肺炎（severe pneumonia, SP）患者死亡风险预测模型的数量日益增多，但此类模型的质量与临床应用价值仍不明确。本研究旨在对已发表的重症肺炎患者死亡风险预测模型相关研究进行系统评价。

方法
本研究检索了PubMed、Embase、Cochrane图书馆及Web of Science数据库，检索时限为建库至2024年8月31日。提取纳入研究的相关数据，包括研究设计、研究对象、诊断标准、样本量、预测因子、模型构建及模型性能。采用预测模型偏倚风险评估工具（Prediction model risk of bias assessment tool）对偏倚风险与适用性进行评价。使用Stata 17.0软件对验证模型的受试者工作特征曲线下面积（area under the curve, AUC）值进行荟萃分析。

结果
本系统评价共纳入18项研究中的22个预测模型，其中15个为逻辑回归模型、2个为Cox比例风险回归模型、2个为分类与回归树模型、1个为轻量梯度提升机模型，另有1个为多层感知机模型。报告的AUC值范围为0.713~0.952。其中17项研究存在较高偏倚风险，主要原因是数据来源不当及分析领域报告不完整。5个验证模型的合并AUC值为0.85（95%置信区间：0.81~0.88），提示模型具备中等程度的区分度。

结论
尽管纳入研究显示重症肺炎患者死亡风险预测模型具备一定的区分能力，但多数模型存在较高偏倚风险。未来研究应聚焦于开发更大样本量、研究设计严谨且经过多中心外部验证的新型预测模型。

系统评价注册信息
https://www.crd.york.ac.uk/PROSPERO/view/CRD42024589877，注册标识：CRD42024589877。