Evaluating AI’s Role in Mortality Prediction for Patients With Severe COVID-19
A large subset of patients with COVID-19 develop severe complications, necessitating advanced interventions and contributes to high mortality rates. Artificial intelligence (AI) has shown promise in predicting severe disease, but its efficacy in determining the outcomes of severe COVID-19 cases remains unclear. A review published in Frontiers in Public Health assessed AI’s potential in predicting mortality in patients with severe COVID-19 and identified factors affecting model performance.
Researchers analyzed data from 19 studies comprising 26 predictive models. Eligible studies utilized AI to predict mortality in COVID-19 patients diagnosed with severe disease and treated in intensive care settings. Bias was assessed using the PROBAST 2019 tool, and statistical analyses, including meta-analysis, publication bias assessment, and sensitivity analysis, were performed using Stata 16 software.
The analysis revealed that AI models combining clinical and imaging data exhibited the highest predictive performance, achieving an overall sensitivity of 0.81, specificity of 0.77, and an area under curve (AUC) of 0.88. Models relying solely on clinical data demonstrated a similar AUC of 0.88, whereas imaging-only models showed lower performance, with an AUC of 0.75. Subgroup analysis indicated that studies conducted in high-income countries achieved significantly higher specificity in both imaging and combined-data models, Models focusing on patients outside of the intensive care unit (ICU) demonstrated greater predictive specificity. Deep learning models exhibited higher specificity compared with traditional machine learning approaches, likely due to their ability to identify complex, nonlinear patterns in high-dimensional datasets.
The researchers concluded that AI has shown substantial potential in predicting mortality among patients with severe COVID-19, which can aid providers in COVID-19 management and treatment. Accuracy and quick turnaround times position AI as a cost-effective strategy for managing COVID-19. However, the applicability of these models in clinical practice remains uncertain due to variations in performance across populations sand settings. The study authors urged future research to refine AI methodologies, improve predictive accuracy, and ensure equitable implementation across diverse health care systems.
“The application of AI extends beyond COVID-19; it is being utilized in diagnosing, screening, and managing other diseases, aligning with the objectives of tertiary prevention strategies,” the researchers concluded. “Therefore, conducting high-quality, large-scale, multicenter studies is imperative for advancing the field of AI in healthcare and ensuring its effective implementation.”
Reference
Qin C, Ma H, Hu M, Xu X, Ji C. Performance of artificial intelligence in predicting the prognosis of severe COVID-19: a systematic review and meta-analysis. Front Public Health. 2024;12:1371852. doi:10.3389/fpubh.2024.1371852
