Lung involvement percentage in patients with COVID-19 during the Omicron wave in China: a SHAP-explained machine learning study from a single center.
Following the lifting of China's stringent lockdown policy on December 7, 2022, COVID-19 cases surged in a pattern, creating unprecedented strain on healthcare systems. The Omicron variant, characterized by high transmissibility and rapid spread, led to a sharp rise in infections. Understanding its clinical impact-particularly on lung involvement percentage-is crucial for optimizing patient care under such outbreak conditions. This study aimed to assess the extent of lung involvement percentage during the outbreak and its major associations.
The hospital's daily computed tomography examination volume was quantified using artificial intelligence-based pulmonary inflammation analysis software and used as an indicator of epidemic intensity. Associations between lung involvement percentage and age, sex, and daily case counts were evaluated using GEE Logistic Regression, complemented by machine learning models. Model interpretation was performed using SHapley Additive exPlanations.
GEE Logistic regression demonstrated that age was strongly associated with lung involvement (OR 1.0813, 95% CI 1.0703-1.0925, p < 0.0001), while daily case counts also showed a small but significant independent association (OR 1.0033, 95% CI 1.0018-1.0047, p < 0.0001). Sex exhibited only minimal association (OR 0.8098, 95% CI 0.6983-0.9391, p = 0.0053). Complementary machine learning analyses, including gradient boosting, identified age as the dominant contributor, followed by daily case counts with a small effect and sex with minimal contribution. SHAP analysis provided interpretable insights into how each feature influenced model predictions at both global and individual levels.
During the Omicron surge, greater age and higher daily case counts were associated with higher lung involvement percentage. These associations highlight the relevance of demographic and epidemic factors in characterizing pulmonary findings during large-scale outbreaks.
The hospital's daily computed tomography examination volume was quantified using artificial intelligence-based pulmonary inflammation analysis software and used as an indicator of epidemic intensity. Associations between lung involvement percentage and age, sex, and daily case counts were evaluated using GEE Logistic Regression, complemented by machine learning models. Model interpretation was performed using SHapley Additive exPlanations.
GEE Logistic regression demonstrated that age was strongly associated with lung involvement (OR 1.0813, 95% CI 1.0703-1.0925, p < 0.0001), while daily case counts also showed a small but significant independent association (OR 1.0033, 95% CI 1.0018-1.0047, p < 0.0001). Sex exhibited only minimal association (OR 0.8098, 95% CI 0.6983-0.9391, p = 0.0053). Complementary machine learning analyses, including gradient boosting, identified age as the dominant contributor, followed by daily case counts with a small effect and sex with minimal contribution. SHAP analysis provided interpretable insights into how each feature influenced model predictions at both global and individual levels.
During the Omicron surge, greater age and higher daily case counts were associated with higher lung involvement percentage. These associations highlight the relevance of demographic and epidemic factors in characterizing pulmonary findings during large-scale outbreaks.
Authors
Ma Ma, Ye Ye, Pan Pan, Shen Shen, Wang Wang, Song Song, Shen Shen, Zhu Zhu, Chen Chen, Shi Shi, Ye Ye, Qin Qin, Ren Ren, Luo Luo, Xu Xu, Zhao Zhao, Zhu Zhu, Zhou Zhou, Zhu Zhu, Zhang Zhang
View on Pubmed