Comment on: Long-term risk prediction after major lower limb amputation: 1-year results of the PERCEIVE study

BJS Open, https://doi.org/10.1093/bjsopen/zrad135, published 24 January 2024

Dear Editor

We would like to commend Gwilym and colleagues for their commitment to amputation level shared decision-making. The PERCEIVE trial was a major undertaking requiring a large collaboration¹.

One of their study goals was to validate existing amputation outcome prediction models and to compare model performance to clinicians’ ability to predict outcomes. While an intriguing comparison, this requires careful adherence to the methods used in the original studies. External validation is critical in facilitating the generalizability of models so that clinicians can apply them confidently to clinical settings. 

Unfortunately, there were several significant discrepancies between how the models were applied in the study versus in the model development. The most prominent concerns were the replication of predictors and outcomes. In the AMPREDICT Mobility and Reamputation models, it does not appear that the following predictors were included or measured as in the original models: functional independence, alcohol misuse, highest education level, and patient self-rated health. 

The AMPREDICT Mobility model predicted the Locomotor Capability Index-5 and the authors used the Special Interest Group in Amputee Medicine grades. These are very different measures; the LCI-5 includes a much more granular level of functional activities. The AMPREDICT model predicted 12-month outcomes among those who survived one year. The authors reported that 355 of their patients survived a year; however, 489 had mobility outcome data. It is unclear if this represents measurements taken before 12 months. For the reamputation outcome, it appears approximately 50% (190 of 355) had available data.  It is unclear what happened with the other 50% that survived.

We acknowledge the importance of this research.  However, unless the methods applied directly replicate the predictors and outcome measures, the interpretation of the results will be limited. Hopefully the work of these authors will encourage ongoing investigation addressing this challenging subject.