Risk factors and interventions with statistically significant tiny effects


Risk factors and interventions with statistically significant tiny effects
George CM Siontis1 and John PA Ioannidis1,2,*
+ Author Affiliations


1Clinical Trials and Evidence-Based Medicine Unit and the Clinical and Molecular Epidemiology Unit, Department of Hygiene and Epidemiology, University of Ioannina School of Medicine, Ioannina, Greece and 2Stanford Prevention Research Center, Department of Medicine, Stanford University School of Medicine, Stanford, USA
↵*Corresponding author. Stanford Prevention Research Center, Department of Medicine, Stanford University School of Medicine, Stanford, CA 94305, USA. E-mail: jioannid@stanford.edu
Accepted May 19, 2011.
Abstract


Background Large studies may identify postulated risk factors and interventions with very small effect sizes. We aimed to assess empirically a large number of statistically significant relative risks (RRs) of tiny magnitude and their interpretation by investigators.


Methods RRs in the range between 0.95 and 1.05 were identified in abstracts of articles of cohort studies; articles published in NEJM, JAMA or Lancet; and Cochrane reviews. For each eligible tiny effect and the respective study, we recorded information on study design, participants, risk factor/intervention, outcome, effect estimates, P-values and interpretation by study investigators. We also calculated the probability that each effect lies outside specific intervals around the null (RR interval 0.97–1.03, 0.95–1.05, 0.90–1.10).


Results We evaluated 51 eligible tiny effects (median sample size 112 786 for risk factors and 36 021 for interventions). Most (37/51) appeared in articles published in 2006–10. The effects pertained to nutrition (n = 19), genetic and other biomarkers (n = 8), correlates of health care (n = 8) and diverse other topics (n = 16) of clinical or public health importance and mostly referred to major clinical outcomes. A total of 15 of the 51 effects were >80% likely to lie outside the RR interval 0.97–1.03, but only 8 were >40% likely to lie outside the RR interval 0.95–1.05 and none was >1.7% likely to lie outside the RR interval 0.90–1.10. The authors discussed at least one concern for 23 effects (small magnitude n = 19, residual confounding n = 11, selection bias n = 1). No concerns were expressed for 28 effects.


Conclusions Statistically significant tiny effects for risk factors and interventions of clinical or public health importance become more common in the literature. Cautious interpretation is warranted, since most of these effects could be eliminated with even minimal biases and their importance is uncertain.

Enhanced by Zemanta
This entry was posted in Clinical trial, Cochrane Library, Effect size, Evidence-Based-Medicine, Ioannina, Stanford, Stanford University School of Medicine, Statistical significance. Bookmark the permalink.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s