Northumbria Healthcare NHS Trust Predicts Surgical Risk with Azure Machine Learning

Overview

Northumbria Healthcare NHS Foundation Trust surgeons built a machine learning model using 220 input parameters from historical surgical-outcomes data on Azure Machine Learning and its Responsible AI dashboard to produce individualized, explainable patient risk profiles for joint-replacement surgery, helping patients make informed decisions and the Trust place surgical candidates at appropriately specified facilities.

The challenge

When a patient asks if surgery will go all right, surgeons often end up giving quite general answers because it can be very hard to predict, and the Trust needed a better indication of how patients will fare so it could have more meaningful conversations with them and better plan for their surgery, especially while clearing a COVID-19 surgical backlog.

The solution

The team used Microsoft Azure Machine Learning to build a model using 220 different input parameters, including factors like age, ethnicity, smoking history, body mass index and blood test results such as platelet count, to generate individualized surgical risk profiles. Azure Machine Learning also provides access to the Responsible AI dashboard, which brings together tools for machine learning interpretability, unfairness assessment and mitigation, error analysis, causal inference, data exploration and counterfactual analysis, letting the team understand, refine and explain its outcomes. The team is making its models and methodologies available as open-source software and is piloting the approach with an NHS hospital group in Wales.

Reported business value

With Azure Machine Learning, the team can show patients a risk score highly tailored to their individual circumstances, helping them decide whether to have surgery or make lifestyle changes to decrease risk beforehand. The work also helps the Trust triage patients to doctors and operating facilities with an appropriate level of care, and a pilot with a Welsh NHS hospital group aims to help identify patients who could safely have surgery at a lower-specification hospital, reducing wait times for both lower- and higher-risk patients.