Back calculate intercept from multivariable COVID-19 logistic regression model I am interested in creating a web tool to predict the absolute risk of in-hospital death from a published risk model of COVID-19 patients.  
Is it possible to estimate the intercept from the significant, multivariable model coefficients published(Table 3 at https://doi.org/10.1016/S0140-6736(20)30566-3)?
 A: No, for a few reasons.  The most relevant is that, even if you could numerically determine the value of the intercept term in the model, you would not be able to tell what conditions are represented by the intercept term of the model fit by the researchers (unless they provided some very specific details about their methods, e.g., "did they center their predictors", etc.).
The way these models are defined, you need to know what the "reference" point of each coefficient is.  By convention the reference point for continuous variables is often zero, but the choice of reference is completely arbitrary for categorical variables (though it does appear to be stated in the article).
On a more conceptual note, models like this one are often limited in their ability to predict accurate "class probabilities "because their goal is to explain relationships between predictors, or to describe relative risk-reducing or risk-increasing relationships among the predictors in the model.  Often these models use only a limited number of variables to arrive at a parsimonious model, which tries to keep multicollinearity low, so that the individual effects of predictors are easier to interpret.  
Another way of looking at it is this:  The model may have been designed to address research questions  such as: "what is the relative importance of age with respect to the risk of an in-hospital death after controlling for other demographic factors?".  Even if the model does this very well, it may be very bad at answering fundamentally different questions, such as "What is the probability that a female will experience an in-hospital death?"
A: 
the case fatality ratio in our study cannot reflect the true mortality of COVID-19. Last but not least, interpretation of our findings might be limited by the sample size

I would add that this study is limited to the very specific case of Jinyintan Hospital and Wuhan Pulmonary Hospital, and might be completely different elsewhere. So whether they had 191 patients a lot more or a lot less does not matter. It is only a case of two hospitals in a single region and you do not know much about variation from place to place, which makes predictions very difficult.
