Peduzzi et al reported an equation for calculating the sample size for a study undergoing a logistic regression analysis. This may be referred to as the "rule of 10". The authors are from Yale University and the West Haven Veterans Affairs Medical Center.
Analysis: logistic regression analysis
Parameters:
(1) number of variables to be tested
(2) incidence of combined adverse outcome
(3) number of events per variable
minimum sample size =
= (number of events per variable) * (number of variables to be tested) / (incidence)
The authors found that the number of events per variable should be >= 10. Below 10 the regression coefficients can be biased in both positive and negative directions, resulting in major problems.
Van Smeden et al felt that there was no rationale for using 10 events per variable for binary logistic regression analysis.
To read more or access our algorithms and calculators, please log in or register.
