Imperial researchers have created a new model that highlights uncertainty when predicting the risk of death in emergency bowel surgery.

Prediction models are ubiquitous in modern medicine, and are often used to predict risk of death after surgery. However, current models output a single number (e.g. “there is a 21.3% risk of death”), which can give the misleading impression that we are certain about the risk for individual patients, leading to over-confident decision making.

A new paper from researchers in the Department of Surgery and Cancer and University College London, published in npjDigitalMedicine, proposes a new model that predicts the risk of death in emergency bowel surgery but highlights the uncertainty around the prediction with an intuitive visualisation.

The paper shows that this new model, based on data from 127,134 surgeries across 186 hospitals, makes better predictions than those currently used in clinical practice and could be used to predict the risk of death before some blood tests have been taken. This could give insight into the patient’s prognosis earlier in their care.

By highlighting when a patient's risk is uncertain, the new model could guide doctors to run rapid extra tests, giving more information about their patients and reducing the uncertainty. If these tests suggest that the patient is at higher risk of death, this can trigger the involvement of extra teams and senior doctors to optimise the patient before surgery, supervise their care during surgery and admit them to the Intensive Care Unit afterward.

The new model also has the potential to assist discussions with patients and their families. When current models predict a very high risk of death, this can lead doctors to give a pessimistic assessment to the patient, who could refuse surgery as a result.

Speaking about the research, joint first author, Dr Finn Catling said: "Our new model might show that the prediction is very uncertain - a patient's risk of death could be anything from moderate to very high. This is very important information for the patient to bear in mind, and might alter their life-changing decision."

This research was funded in part by the NIHR Imperial Biomedical Research Centre. 

Highlighting uncertainty in clinical risk prediction using a model of emergency laparotomy mortality risk. Jakob F. Mathiszig-Lee, Finneas J. R. Catling, S. Ramani Moonesinghe & Stephen J. Brett. npj Digital Medicine volume 5, Article number: 70 (2022)