Comparison of artificial intelligence and logistic regression models for mortality prediction in acute respiratory...

The application of artificial intelligence (AI) in predicting the mortality of acute respiratory distress syndrome (ARDS) has garnered significant attention. However, there is still a lack of evidence-based support for its specific diagnostic performance. Thus, this systematic review and meta-analysis was conducted to evaluate the effectiveness of AI algorithms in predicting ARDS mortality.

We conducted a comprehensive electronic search across Web of Science, Embase, PubMed, Scopus, and EBSCO databases up to April 28, 2024. The QUADAS-2 tool was used to assess the risk of bias in the included articles. A bivariate mixed-effects model was applied for the meta-analysis. Sensitivity analysis, meta-regression analysis, and tests for heterogeneity were also performed.

Eight studies were included in the analysis. The sensitivity, specificity, and summarized receiver operating characteristic (SROC) of the AI-based model in the validation set were 0.89 (95% CI 0.79-0.95), 0.72 (95% CI 0.65-0.78), and 0.84 (95% CI 0.80-0.87), respectively. For the logistic regression (LR) model, the sensitivity, specificity, and SROC were 0.78 (95% CI 0.74-0.82), 0.68 (95% CI 0.60-0.76), and 0.81 (95% CI 0.77-0.84). The AI model demonstrated superior predictive accuracy compared to the LR model. Notably, the predictive model performed better in patients with moderate to severe ARDS (SAUC: 0.84 [95% CI 0.80-0.87] vs. 0.81 [95% CI 0.77-0.84]).

The AI algorithms showed superior performance in predicting the mortality of ARDS patients and demonstrated strong potential for clinical application. Additionally, we found that for ARDS, a highly heterogeneous condition, the accuracy of the model is influenced by the severity of the disease.