Clinical data and quantitative CT parameters combined with machine learning to predict short-term prognosis of severe...

This study aims to evaluate the effectiveness of integrating clinical data and quantitative CT parameters with machine learning techniques in forecasting the short-term outcomes of severe COVID-19 in elderly patients.

In this retrospective study, we analyzed the clinical profiles and chest quantitative CT parameters of 239 elderly patients with severe COVID-19 admitted for treatment. The cohort included 61 deceased patients (death group) and 178 who recovered and were discharged (survival group). The participants were randomly assigned into a training group (n = 167) and a validation group (n = 72). Quantitative CT parameters were measured using the 3D-Slicer software. Univariate and multivariate logistic regression analyses identified independent risk factors for mortality. Predictive models were developed employing four machine learning algorithms: Logistic Regression (LR), Random Forest (RF), Decision Tree (DT), and Support Vector Machine (SVM).

Both univariate and multivariate logistic regression analyses revealed age, hypersensitive C-reactive protein (hs-CRP), and solid organ volume percentage (SOV%) as independent predictors of mortality. The Area Under the Curve (AUC) values for the LR, RF, DT, and SVM models in the training group were 0.795, 0.726, 0.854, and 0.589, respectively; for the validation group, they were 0.817, 0.634, 0.869, and 0.754, respectively. The DT algorithm outperformed other models in both the training and validation groups, emerging as the most effective predictive model in this study.

The combination of clinical data and quantitative CT parameters with machine learning approaches is highly valuable in predicting the short-term prognosis of severe COVID-19 in the elderly. Among the various models tested, the Decision Tree algorithm-based model proved to be the most accurate and reliable in this context.