We have read the letter «Inclusive use of effect size conversion and Bayes factor in Intensive Care Medicine research»,1 referred to our article recently published in Medicina Intensiva, entitled: «HLA genetic polymorphisms and prognosis of patients with COVID-19», with great interest.2

Our study involved 72 patients with COVID-19 (10 deceased and 62 surviving after 30 days), and reported a statistically significant association between different genetic polymorphisms of the human leukocyte antigen (HLA) system and mortality due to COVID-19 in the regression analysis. One of the mentioned alleles was HLA-A*11.

The comparison of deceased patients and survivors identified a statistical tendency (p = 0.051) towards a greater presence of the HLA-A*11 allele among the patients that died (3/30 [30%]) versus the survivors (4/62 [6.5%]). However, the deceased individuals presented higher APACHE-II (p < 0.001) and SOFA scores (p = 0.001) than the survivors. The regression analysis found the presence of the HLA-A*11 allele to be associated to increased mortality controlling for SOFA (OR: 7.693; 95%CI: 1.063–55.650; p = 0.04) or APACHE-II (OR: 11.858; 95%CI: 1.524–92.273; p = 0.02).

We admitted that our study had limitations, including the fact that we only entered two variables in each regression model, due to the small number of deceased patients in our study, which precluded the inclusion of several variables.3 Nevertheless, we feel that the strengths of the study were the fact that the presence of the HLA-A*11 allele was associated to mortality after controlling for two patient severity scores (SOFA or APACHE-II), and that the presence of this allele has previously been associated to a poor outcome in the context of other infectious diseases.4,5

We agree with the methodological reflection of Ramos-Vera, recommending the confirmation of clinical research in order to validate the results obtained and strengthen their practical credibility. As commented by Ramos-Vera, the option of using Bayesian inference offers a very good approach to the control of error in estimations. Moreover, we are happy to know that the Bayesian analysis made by Ramos-Vera to define the level of evidence of our study using the Bayes factor (BF) method found the results referred to FB10 (in favor of the alternative hypothesis) and FB01 (in favor of the null hypothesis) to indicate extreme evidence of our findings. Nevertheless, we recommend the validation of observational or experimental results to be made using replication with samples external to the study. We therefore consider that it would be interesting to obtain findings similar to our own in future samples involving different COVID-19 patients, in order to determine whether the findings are stable, reducing the uncertainty of possible type 1 errors.

In conclusion, we agree with Ramos-Vera that the confirmation of clinical research is necessary (based on external samples or Bayesian inference) in order to validate the results obtained and strengthen their practical credibility.