PURPOSE: This study tests an existing Vascular Biochemistry and Haematology Outcome Model (VBHOM) on independent data and presents further refinements to the model. METHODS: Data from 306 patients who underwent lower limb amputation over a 4-year period were collated. Urea, creatinine, sodium, potassium, hemoglobin, white cell count, albumin, age, gender, mode-of-admission, and short-term mortality events were extracted from the database. This study tests an existing model and trains a new model for predicting mortality using forward stepwise logistic regression. RESULTS: The existing model suggests a significant lack of fit (c-index = 0.665, P = .04). For the exception of gender and mode-of-admission, all predictor variables had significant univariate associations with short-term mortality (P < .05). The refined model included age, sodium, potassium, creatinine, and albumin and had good discriminatory power (c-index = 0.8, no evidence of lack of fit, P = .616). CONCLUSIONS: Our simplified model had good predictive ability and suggests redundancy in input variables used by the existing models.
PURPOSE: This study tests an existing Vascular Biochemistry and Haematology Outcome Model (VBHOM) on independent data and presents further refinements to the model. METHODS: Data from 306 patients who underwent lower limb amputation over a 4-year period were collated. Urea, creatinine, sodium, potassium, hemoglobin, white cell count, albumin, age, gender, mode-of-admission, and short-term mortality events were extracted from the database. This study tests an existing model and trains a new model for predicting mortality using forward stepwise logistic regression. RESULTS: The existing model suggests a significant lack of fit (c-index = 0.665, P = .04). For the exception of gender and mode-of-admission, all predictor variables had significant univariate associations with short-term mortality (P < .05). The refined model included age, sodium, potassium, creatinine, and albumin and had good discriminatory power (c-index = 0.8, no evidence of lack of fit, P = .616). CONCLUSIONS: Our simplified model had good predictive ability and suggests redundancy in input variables used by the existing models.
Authors: Brenig L Gwilym; Cherry-Ann Waldron; Emma Thomas-Jones; Ryan Preece; Sarah Milosevic; Lucy Brookes-Howell; Philip Pallmann; Debbie Harris; Ian Massey; Jo Burton; Philippa Stewart; Katie Samuel; Sian Jones; David Cox; Adrian Edwards; Chris Twine; David C Bosanquet Journal: BJS Open Date: 2021-11-09