7/June – Death regression analysis for each state in Brazil¶
This is a bayesian logistic regression with a bernoulli likelihood for the outcome of the patient (death/cure) based on the reported symptoms covariates.
A HMC MCMC procedure was employed to estimate the model with 2 independent chains. On the left panel of the plots you’ll find the regression coefficients and their uncertainties. On the middle panel you’ll find the Effective Sample Size of the MCMC sampling and on the right panel you’ll find the MCMC Rhat diagnostic metric.
Some important points:
Only states with more than 100 reported deaths on SIVEP system were used in the estimation;
SIVEP data doesn’t include all testing done for all states and all cities;
SIVEP data is observational data and correlation doesn’t imply causation, this analysis is only a predictive model for the outcome being Death/Cure for COVID-19 based on the reported symptoms;
The majority class (cure) was undersampled to match the same number of cases of samples of the minority class (death), in order to deal with an imbalanced dataset;
For more information about what each symptom means, please look at the SIVEP data dictionary;
Note
This plot uses official data from SIVEP system in Brazil, the maximum of the notification date for this data was 06/June.
