I would like to nominate for the Methods Publication Award my 2019 paper "Minimum sample size for developing a multivariable prediction model: Part II - binary and time-to-event outcomes." which appeared in Statistics in Medicine in March of 2019.  A significant portion of the methods in the paper are based on methods first proposed in the second edition of my book Regression Modeling Strategies (Springer, 2015).  The paper has already received 32 citations (54 according to Google Scholar).  Its impact on clinical research is best judged by the fact that it caught the attention of an editor of the British Medical Journal who invited us to write a new version of the paper for a clinical audience.  That paper appeared in BMJ in 2020.

The paper is important to both the statistical community (developers of many predictive models) and to the health sciences, because prevailing approaches to estimating the sample size needed to develop a clinical prediction model are ad hoc, with many statisticians and subject matter researchers relying on oversimplified rules of thumb such as 20 events per candidate predictor.  The paper shows that these rules of thumb need to be abandoned in favor of formulas that are much more tailored to the problem at hand, especially considering the signal:noise ratio of the data.  Many studies have been undertaken with inadequate numbers of patients and a moderate number of studies have spent resources on collecting more patients than needed for the predictive model to be reliable.  Methodologically, the paper develops a multiple-criteria approach to satisfy reliability concerns, minimize overfitting, and obtain a precise estimate of the intercept (or overall average risk or underlying survival curve).  Many published clinical prediction papers did not have a sample size that was adequate even for estimating a model with zero predictors, i.e., could not reliability estimate the model's intercept.