سيستم پيش بيني و ارزيابي طولاني مدت بيمارستاني براي ارزيابي مرگ و مير كوويد-19 و اثربخشي اعتبارسنجي

Longitudinal In-hospital Forecast an‎d eva‎luation System for Assessing COVID-19 Mortality an‎d Validation Effectiveness

رياضي كاربردي و علوم كامپيوتر

The acute respiratory disease caused by the COVID-19 virus, which originated in China at the end of 2019, not only placed a significant strain on healthcare resources, but also made the efficient management of time an‎d medical resources one of the major challenges during the pan‎demic era. The aim of this research is to design a predictive system for assessing the deterioration of hospitalized patients with COVID-19. To this end, data from Bellvitge University Hospital in Barcelona, Spain were analyzed to identify the most relevant features influencing disease severity. The dataset employed in this study comprises information from 2,285 patients with 288 variables. Subsequently, cause-specific Cox regression modeling was employed for prognostic modeling. Predictive variables were selec‎ted using three different approaches: a naive approach, a genetic algorithm optimization method, an‎d regularization techniques. In order to han‎dle missing data, multiple imputation was applied. Accordingly, seven distinct models were developed for mortality prediction. Models one an‎d two were based on the naive approach. Models three to five were developed using the genetic algorithm. Finally, model six (Ridge-penalized cause-specific Cox regression) an‎d model seven (Lasso-penalized cause-specific Cox regression) were implemented using regularization methods. The performance of the models was eva‎luated based on the area under the curve (AUC), employing DeLong et al. test for comparison. Results indicated that models six an‎d seven are statistically equivalent; however, both demonstrate superior performance compared to the other models. Model seven was selec‎ted as the final model due to its use of fewer predictive variables while maintaining high predictive accuracy. In seven model, age, smoking history, Charlson comorbidity index, hypertension, dyspnea, an‎d chronic heart failure were identified as significant predictors of mortality risk among patients with COVID-19. Additionally, reduced oxygen saturation was found to be a critical factor affecting both disease severity an‎d mortality. The AUC value of this model was calculated as 0.854. During model validation, the intercept an‎d slope of the calibration plot were obtained as −0.0069 an‎d 0.963, confirming good model calibration.

6