Kisielewicz, C., Self, I. & Bell, R. Assessment of Clinical and Laboratory Variables as a Guide to Packed Red Blood Cell Transfusion of Euvolemic Anemic Dogs. J. Vet. Intern. Med. 28, 576–582 (2014).
We thus strongly recommend that basic data numbers within treatment or exposure and outcome categories be examined and presented, and that adjustment methods such as penalisation be applied whenever the numbers of events per covariate fall below four or five. The weighting (degree of penalisation) for each variable is best determined so that the implied prior interval encompasses the full range of reasonable possibilities for the effect of the variable.
Pavlou, M., Ambler, G., Seaman, S., Iorio, M. D. & Omar, R. Z. Review and evaluation of penalised regression methods for risk prediction in low-dimensional data with few events. Stat. Med. 35, 1159–1177 (2016).
While unbiased estimation of coefficients is important when the aim is to investigate associations, bias is considered to be a less important issue for risk prediction studies where the predictive performance of the model is of main interest.