Discussion about this post

User's avatar
Edward  H Livingston, MD, FACS's avatar

The problem here is relying on the dichotomous yes/no that comes from hypothesis testing-P values. Although the P value was not significant because the CI crossed 1.0, there was a 95% chance that the true HR was somewhere between 1.0 and 1.35, favoring liberal blood transfusion. Almost all of the effect embodied in the CI favored liberal transfusion. To me, the conclusion of this study is that there is reasonable evidence that liberal transfusion is beneficial in this population and that more evidence is needed to use the results of this trial in clinical care. If the HRQOL outcomes (yet to be reported) favor blood transfusion, I would conclude that when the totality of effects are accounted for, a liberal transfusion should be considered for these patients.

Expand full comment
Sander Greenland's avatar

I come from a statistics and research methodology background rather than a clinical background, although I've been involved in a hundred clinical studies. As such, I cannot fathom why and how researchers still obsess about whether p is above or below 0.05 as if the latter is some natural constant, or equivalently whether the 95% CI contains the null value. The founders of modern statistics including Fisher himself advised flexibility with the cutoff depending on the circumstances, while Neyman & Pearson and their successors emphasized that the cutoff should be chosen based on the costs of false positive (Type 1) vs false negative (Type 2) errors. And that is leaving aside the many (including me) who advise that such cutoffs should not be the basis for decisions, and should serve only as convenient reference points much like labeled points on a graphical axis. Conclusions about treatment effects need other information, including P-values for alternatives of clinical importance such as a minimal clinically important difference. The CI shows quickly where those alternative P-values are relative to 0.05, but better still is to look at them directly.

Here are a few of the many open-access articles my colleagues and I have written recently trying to stem this unhealthy compulsion to treat 0.05 as some magic number or universal constant of science; the first lists common mistakes traceable to that compulsion, the others detail how to reorient one's thinking to get a valid picture of the statistical information in a trial, including information about possible effect sizes other than the null:

Greenland, S., Senn, S.J., Rothman, K.J., Carlin, J.C., Poole, C., Goodman, S.N., Altman, D.G. (2016). Statistical tests, confidence intervals, and power: A guide to misinterpretations. The American Statistician, 70, online supplement 1 at https://amstat.tandfonline.com/doi/suppl/10.1080/00031305.2016.1154108/suppl_file/utas_a_1154108_sm5368.pdf, https://www.jstor.org/stable/44851769

Rafi, Z., Greenland, S. (2020). Semantic and cognitive tools to aid statistical science: Replace confidence and significance by compatibility and surprise. BMC Medical Research Methodology, 20, 244. doi: 10.1186/s12874-020-01105-9, https://bmcmedresmethodol.biomedcentral.com/articles/10.1186/s12874-020-01105-9, updates at http://arxiv.org/abs/1909.08579

Greenland, S., Mansournia, M., Joffe, M. (2022). To curb research misreporting, replace significance and confidence by compatibility. Preventive Medicine, 164, https://www.sciencedirect.com/science/article/pii/S0091743522001761

Expand full comment
24 more comments...

No posts