Statistics and the Great Persky

Abstract

If every person with acute myeloid leukemia (AML) age >60 to 65 years in the United States and Europe received a reduced-intensity conditioning hematopoietic cell transplant, we would know precisely how they fared. Forget statistics, confidence intervals, P values, meta-analyses and the like: the outcome is the outcome. The problem is that we do not have these data, much less data on everyone age >60 to 65 years who could have received a transplant. The sad fact is we have data on a very few persons who did, but we do not know exactly why these persons received a transplant and why so many others did not. So we need statistics applied to a small and selected sample to try deduce a larger truth: What would be the outcome if everyone with AML age >60 to 65 years who could receive a reduced-intensity conditioning transplantation received one? With this approach come many assumptions, limitations, and substantial uncertainty. One demon confounding our estimates of outcomes and applicability of conclusions from a small sample to a wider population is selection bias. Selection bias sounds terrible, politically incorrect, like racial profiling. Perhaps something Donald Trump might suggest. However, selection biases operate in every aspect of our lives. For example, our old