Since LR modelling is far less data hungry than alternative modelling techniques, this technique may especially be useful in relatively small data sets. With very small data sets, any modelling technique will lead to poorly performing models. Our results confirm the generally accepted rule that reasonable predictive modelling requires at least 10 events per variable, even with a robust technique such as LR [11, 12, 15]. We note that larger numbers of events per variable are desirable to achieve better stability and higher expected performance.