Response.

To the Editor:

We thank Dr Dietl and colleagues for their interest in our work and their thoughtful opinions on the predictive model for mortality in patients with coronavirus disease 2019 (COVID-19).

Yes, as Dr Dietl and colleagues mentioned, the different main outcomes (fatal outcome vs a composite outcome, including death, ICU admission, or mechanical ventilation), statistical methods (stepwise selection vs LASSO, Cox regression vs Logistic regression), and coding method of variables (continuous variables vs categorical variables), would contribute to the discrepancy of the final risk model’s variables in two papers. , In the early stage of the pandemic, little was known about the prognosis of hospitalized patients with COVID-19, so it was urgent to explore the risk factors for mortality. The nationwide database was set up by January 31; we then immediately started to construct a predictive model for the fatal outcome, aiming to provide more information for management and prevention as soon as possible.

We agreed with the point by Dr Dietl that performing the external validation is important. Because of the urgent situation in the early peak of the pandemic, it was difficult to recruit other cohorts for external validation at that time. We had mentioned this as a limitation of our study in the discussion part. Alternatively, internal validation could be performed for the development of a prediction model. Some studies had used bootstrap resampling to assess the developed nomogram without the external validation. , We also performed bootstrapping in the paper, and the C-index for prediction was 0.91, which indicated a reliable capacity for predicting. The calibration curves also implied good consistency between the prediction and the observation.

At the early phase of this pandemic, it was reported that a high proportion of critical illness subjects would be deteriorated into fatality. It was also necessary to assess which are at high risk of developing critical illness, which might be useful to aid in delivering proper treatment and optimizing use of resources. Since mid February, the spread of the COVID-19 in China started to decrease with the effective prevention and isolation strategy. Our institute then was able to obtain data from four additional cohorts. These continuous cohorts made it possible to perform the external validation in the companion study finished in late March, which aimed to construct a predictive risk score to estimate the risk of developing critical illness.

Model-based prediction regarding COVID-19 could help physicians identify patients with poor prognosis at an early stage. If possible, performing the complete validation would be better because of the different population with predisposing factors such as race or spectrum of comorbidities. Meanwhile, some other external factors might be relevant to the disease progression. Collapse of medical resources, especially the overload of ICU capacity, might account for a higher case fatality rate in critically ill patients with COVID-19. In the future, with the development of advanced algorithms such as deep learning and artificial intelligence, prognostic prediction models will be more comprehensive and able to take into account different application scenarios.