Diagnosis and prognosis stratification model for asymptomatic and presymptomatic patients
To ascertain the differences between severe and non-severe COVID-19 patients who were asymptomatic at the time of admission, data of the 63 indicators listed in Table S2 and Table S3 were retrieved and systematically compared. We found systematic differences in laboratory indicators between severe and non-severe COVID-19 patients who were asymptomatic on admission (Fig. S1 and Table S3). To optimize precise risk-stratification management and improve the prognosis of patients, we constructed a two-step risk-stratification model with the data obtained from 39 asymptomatic, 34 presymptomatic, and 1,751 symptomatic COVID-19 patients. The initial diagnostic model developed with the top 10 laboratory indicators could accurately classify asymptomatic and presymptomatic patients (area under the curve [AUC]=0.89, 5-fold cross-validation 95% confidence interval [CI] 0.81–0.98; Fig. 4A), and the degrees of contribution of the relevant 10 indicators are shown in Fig. 4B. It is noteworthy that the top five indicators (brain natriuretic peptide (BNP), IgG, IgM, glucose (GLU), and DD) might have similar stratification efficiencies (AUC=0.86, 5-fold cross-validation 95% CI, 0.73–1.00, Fig. S2A).
To stratify the non-severe and severe presymptomatic patients, we included the non-severe and severe symptomatic patients in a training dataset, and the constructed model with the top 10 laboratory indicators was then used to predict the severity of disease in presymptomatic patients with an AUC of 0.82 (Fig. 4C). The degrees of contribution of the 10 indicators included in this model are shown in Fig. 4D. The top five indicators were BNP, IgG, IgM, procalcitonin (PCT), and albumin/globulin ratio (A/G ratio), which could accurately predict illness severity (AUC=0.75, Fig. S2B), and their degrees of contribution were 26%, 16%, 11%, 7%, and 4%, respectively (Fig. 4D). Severe presymptomatic patients had significantly higher levels of BNP, PCT, magnesium, etc. than non-severe presymptomatic patients (Fig. 4E) . Moreover, we obtained the validation data of 27 presymptomatic patients (non-severe, n=25; severe, n=2) from another study center (the First People’s Hospital, Jiangxia District, Wuhan, China) and found that the levels of N-Terminal-proBNP (NT-proBNP), PCT, magnesium, etc., were high in severely ill COVID-19 patients (Fig. 4F). These results suggest that the application of laboratory indicators to stratify asymptomatic and presymptomatic COVID-19 patients is feasible, and such a risk-stratification model could be used to screen non-severe and severe presymptomatic patients on admission. The package-based stratification calculator was provided online (https://github.com/liangyuan-njmu/SeverePredictModel) for clinicians to assess whether presymptomatic patients will show disease progression.