Thanks for sharing the article, which addressed at an important topic, Bacteremic pneumococcal pneumonia in the elderly. I agree with your approach and findings. However, the statistical analysis is flawed. This is because you reported the effect measure as Odds Ratio (OR). However, you and I know that Odds Ratio is applied only in case-control or retrospective studies. For your article, which was a prospective study, Relative Risk (RR) should have been calculated.
What are your thoughts?
Article Bacteremic Pneumococcal Pneumonia in Elderly and Very Elderl...
Thanks for the comments you make regarding the article. Referring to the statistical issue on OR and RR, it is true that there is considerable controversy on the use of OR and RR depending on the study design. When the study is case-control or retrospective there is agreement in the use of ORs. However, in prospective studies, we can use both the OR and the RR, but we must take into account the problems or limitations we have. The use of ORs in prospective studies is appropriate for studies of rare outcome but may be misleading when the outcome is common. Specifically, when event rates are high (commonly the case of trials and systematic reviews) the OR overestimates the RR. When this happens, we must use another type of analysis, alternative to the ordinary multivariate logistic regression models, such as log-binomial, binomial, Poisson or modified models. However, these models often present problems of convergence, or predict probabilities greater than 1. Therefore, focusing on our article, in which our mortality rates are not high, and so, we do not have the problem of a highly prevalent outcome, we believe that the use of multivariate logistic regression models providing ORs is correct, and it allows us to avoid the problems of nonconvergence of the alternative models (Steyerberg EW, Clinical Prediction Models, 2010; Bieler GS, 2010; Deeks J, 1998).
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Sampling Design
- Sample Size