Factors affecting prolonged SARS‐CoV‐2 infection and development and validation of predictive nomograms

Abstract

AbstractProlonged severe acute respiratory syndrome coronavirus 2 (SARS‐CoV‐2) infection has received much attention since it is associated with mortality and is hypothesized as the cause of long COVID‐19 and the emergence of a new variant of concerns. However, a prediction model for the accurate prediction of prolonged infection is still lacking. A total of 2938 confirmed patients with COVID‐19 diagnosed by positive reverse transcriptase‐polymerase chain reaction tests were recruited retrospectively. This study cohort was divided into a training set (70% of study patients; n = 2058) and a validation set (30% of study patients; n = 880). Univariate and multivariate logistic regression analyses were utilized to identify predictors for prolonged infection. Model 1 included only preadmission variables, whereas Model 2 also included after‐admission variables. Nomograms based on variables of Model 1 and Model 2 were built for clinical use. The efficiency of nomograms was evaluated by using the area under the curve, calibration curves, and concordance indexes (C‐index). Independent predictors of prolonged infection included in Model 1 were: age ≥75 years, chronic kidney disease, chronic lung disease, partially or fully vaccinated, and booster. Additional independent predictors in Model 2 were: treated with nirmatrelvir/ritonavir more than 5 days after diagnosis and glucocorticoid. The inclusion of after‐admission variables in the model slightly improved the discriminatory power (C‐index in the training cohort: 0.721 for Model 1 and 0.737 for Model 2; in the validation cohort: 0.699 for Model 1 and 0.719 for Model 2). In our study, we developed and validated predictive models based on readily available variables of preadmission and after‐admission for predicting prolonged SARS‐CoV‐2 infection of patients with COVID‐19.