Background Despite ISS being a widely accepted tool for measuring injury severity, many researchers and practitioners use different partition of ISS into severity groups. The lack of uniformity in ISS use inhibits proper comparisons between different studies. Creation of ISS group boundaries based on single AIS value squares and their sums was proposed in 1988 during Major Trauma Study (MTOS) in the USA, but was not validated by analysis of large databases.

Methods A validation study analysing 316 944 patients in the Israeli National Trauma registry (INTR) and 249 150 patients in the American National Trauma Data Bases (NTDB). A binary algorithm (Classification and Regression Trees (CART)) was used to detect the most significantly different ISS groups and was also applied to original MTOS data.

Results The division of ISS into groups by the CART algorithm was identical in both Trauma Registries and very similar to original division in the MTOS. For most samples, the recommended groups are 1–8, 9–14, 16–24 and 25–75, while in very large samples or in studies specifically targeting critical patients there is a possibility to divide the last group into 25–48 and 50–75 groups, with an option for further division into 50–66 and 75 groups.

Conclusions Using a statistical analysis of two very large databases of trauma patients, we have found that partitioning of ISS into groups based on their association with patient mortality enables us to establish clear cut-off points for these groups. We propose that the suggested partition of ISS into severity groups would be adopted as a standard in order to have a common language when discussing injury severity.