Probability versus Causal Inference in Observational Studies

pends on the research question and proper selection of the confounders used to adjust for the analysis. Their study was designed to test the probability for the outcomes specified while having hemoglobin excursions outside the target range and not to test the association between them [3] , because the analysis was adjusted for hypertension and coronary artery disease which were by definition not confounders in this design being influenced by the hemoglobin level [4–6] . Thus, if one adjusts the analysis for them, the link between hemoglobin excursions and the outcomes tested may be reduced or even canceled out, thereby not revealing the possible effect of an increased hemoglobin level on the outcomes. In this study, hypertension and coronary artery disease were related to the effect of hemoglobin level [5, 6] . This allowed for only prognostic and not etiological inferences to be made because, in the prognostic model, pathophysiology was not an issue but only probabilities for the outcomes specified [3] . Dear Editor We appreciate the reply of Dr. Handelman to our comment [1] on the paper ‘Hospitalization and mortality in hemodialysis patients: association with hemoglobin variability’, published in your journal [2] . In his reply, Dr. Handelman compares the indication of impending problems secondary to having a hemoglobin level below the target range with the fever and elevated white count as indicators of infection as well as with the association between severe edema and congestive heart failure. However, the inferences made quite often in the paper by using wording such as ‘association’, ‘influence’, etc., did not correspond with the methodology applied in their study, which was the major point of our critique, so that no relation exists between the examples specified above by Dr. Handelman. Even infection may not necessarily be associated with an elevated white count, but with a low white count if it is a septic one and severe edema may not be associated with congestive heart failure, but with liver cirrhosis, so that every conclusion dePublished online: June 5, 2014