Explainable AI ICU nurse
Explainable AI ICU nurse

Explainable AI ICU nurse

What is it

  • A showcase of how of prediction of ICU survival with machine learning can be made suitable for a clinical setting.
  • The explainable AI methodology in this dashboard is trained on data from the first 24 hours of hospital Intensive Care Unit visit of 130,000 patients provided by MIT.

How we built it

  • A scalable feature pipeline transform the relevant variables to proper features
  • A gradient boosted random forest classifier is optimised using cross-validated Bayesian genetic optimisation.
  • The features contribution to patient survival on an individual and population level are interactively visualised using Shapley values.

Explainable AI for healthcare enables

  1. Improved triage and decision making through explainable prediction of (ICU) patient survival where the personalised summary of patient state offers guidance to clinicians.
  2. Population and subgroup level insights into how patient characteristics and treatments interact and impact survival.
  3. Leveraging real world clinical experience of thousands of patients to make healthcare more transparent for patients, policymakers and stakeholders.