Pulmonary medicine · 2026
Optimizing Intubation Prediction in Pneumonia Patients: A Systematic Review and Meta-Analysis of Machine Learning Algorithms.
СтатьяСистематический обзорМетаанализНазвание и аннотация: Перевод в очереди
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Abdoli Elham, Eini Pooya, Farashi Sajjad, Farhadian Maryam
Аннотация
BACKGROUND: Pneumonia, including influenza, COVID-19, and community-acquired pneumonia, is a major global health burden associated with high morbidity, mortality, and frequent progression to respiratory failure requiring intubation. Early identification of patients at risk of endotracheal intubation is essential to improve outcomes and optimize ICU resource allocation, yet existing prognostic tools remain limited in predicting this need. This study evaluated the performance of machine learning (ML) algorithms in predicting endotracheal intubation among patients with pneumonia during hospital stay.
METHODS: We systematically searched five databases to evaluate the diagnostic accuracy of ML models. Pooled estimates of area under the receiver operating characteristic curve (AUROC), sensitivity, and specificity were calculated. Subgroup analysis and meta-regression were conducted. Risk of bias was assessed using PROBAST+AI and certainty of evidence with GRADE.
RESULTS: This systematic review of 34 studies (26 in meta-analysis) included 195,214 pneumonia patients. The pooled AUROC was 0.79 (95% CI: 0.75-0.82), with sensitivity of 0.74 (95% CI: 0.61-0.84), specificity of 0.71 (95% CI: 0.50-0.86), and a DOR of 7 (95% CI: 2-20), indicating moderate diagnostic accuracy. Heterogeneity was substantial across analyses (I2 = 90.45% for sensitivity and 94.58% for specificity). Risk of bias was lowest in development (59%) and highest in application domains (41% high risk). Despite a nonsignificant Deeks' test (p = 0.252), the funnel plot suggests selective publication of positive results, likely inflating the pooled AUROC. GRADE rated the evidence as moderate to low due to heterogeneity and imprecision.
CONCLUSION: ML algorithms demonstrate a modest and highly variable accuracy in predicting the need for endotracheal intubation among pneumonia patients. High heterogeneity and methodological variability highlight the need for standardized ML approaches before clinical adoption.
METHODS: We systematically searched five databases to evaluate the diagnostic accuracy of ML models. Pooled estimates of area under the receiver operating characteristic curve (AUROC), sensitivity, and specificity were calculated. Subgroup analysis and meta-regression were conducted. Risk of bias was assessed using PROBAST+AI and certainty of evidence with GRADE.
RESULTS: This systematic review of 34 studies (26 in meta-analysis) included 195,214 pneumonia patients. The pooled AUROC was 0.79 (95% CI: 0.75-0.82), with sensitivity of 0.74 (95% CI: 0.61-0.84), specificity of 0.71 (95% CI: 0.50-0.86), and a DOR of 7 (95% CI: 2-20), indicating moderate diagnostic accuracy. Heterogeneity was substantial across analyses (I2 = 90.45% for sensitivity and 94.58% for specificity). Risk of bias was lowest in development (59%) and highest in application domains (41% high risk). Despite a nonsignificant Deeks' test (p = 0.252), the funnel plot suggests selective publication of positive results, likely inflating the pooled AUROC. GRADE rated the evidence as moderate to low due to heterogeneity and imprecision.
CONCLUSION: ML algorithms demonstrate a modest and highly variable accuracy in predicting the need for endotracheal intubation among pneumonia patients. High heterogeneity and methodological variability highlight the need for standardized ML approaches before clinical adoption.