Abstract
BACKGROUND: Majority coronavirus disease 2019 (COVID-19) patients are classified as mild and moderate (non-severe) diseases. We aim to develop a model to predict isolation length for non-severe patients. METHODS: Among 188 non-severe patients, 96 patients were enrolled as training cohort to identify factors associated with isolation length via Cox regression model and develop a nomogram. Other 92 patients formed as validation cohort to validate nomogram. Concordance index (C-index), area under the curve (AUC) and calibration curves were used to evaluated nomogram. RESULTS: Increasing absolute eosinophil count (AEC) after admission was correlated with shorter isolation length (P = 0.02). Baseline activated partial thromboplastin time (APTT) > 30 s was correlated with longer isolation length (P = 0.03). A nomogram to predict isolation probability at 11-, 16- and 21-day was developed and validated. The C-indices of training and validation cohort were 0.604 and 0.682 respectively. Both cohorts showed a good discriminative ability (AUC, 11-day: 0.646 vs 0.730; 16-day: 0.663 vs 0.750; 21-day: 0.711 vs 0.783; respectively) and calibration power. CONCLUSIONS: Baseline APTT and dynamic change of AEC were two significant factors associated with isolation length of non-severe patients. Nomogram could predict isolation probability for each patient to estimate appropriate quarantine length.