AI for Health - Predicting Deterioration
Adapting and evaluating the deterioration index (DI) model to better predict the risk of deterioration of patients in the hospital beds.
Premium participation: Get 1-on-1 Mentoring & GuidanceTry a free mentoring session
The need for a modern prediction model for patient deterioration
In many hospitals, a significant portion of overall resources is devoted to identifying and treating patients who are clinically deteriorating. The current model, developed to predict which patients have a high risk of deteriorating, is the Modified Early Warning Score (MEWS), which has been in use for around 20 years in hospitals worldwide. Measurements for this model are taken at least once per day and more often if needed.
The EPD publisher (EPIC) has created a new module that calculates a deterioration index (DI) from data available in the electronic patient files (EPD). The DI model is an ordinal logistic regression model, where the DI is auto-calculated every 15 minutes and outputs a percentage that could be directly interpreted. It is currently not set up to directly initiate action.
Our challenge partner
GOAL: New prediction model validation and improvement
To find whether this model could perform better than the MEWS we need you! Firstly model validation and adaptation of the DI model will be done to evaluate the output probabilities with outcomes of patients. Secondly the model parameters can be re-estimated to optimize the probability output for the Elisabeth-TweeSteden Ziekenhuis (ETZ) population or hospital sub-populations. Validating and improving these models will help to better predict the risk of deterioration of patients in the hospital beds.
Who are we looking for?
Anyone with an interest in learning about a model that can predict patient deterioration can apply. We expect some experience with programming languages and an interest in machine learning. A background experience with linear regression or logistic regression is preferred, but anyone with the right motivation and ‘proof’ of understanding of the concepts discussed in this proposal can sign up!
You can join as a contributor (12 hours per week commitment for 2 months), coach (2-4 hours per week, only for experienced ML professionals) and teacher (give one relevant ML / domain masterclass).
The challenge will run from the 16. November 2021 until the 15. February 2022. To achieve this goal, our project will be guided by an expert from the ETZ hospital and FruitPunch AI for Health. The project will kick off in the second week of November and the team will consist of 5 people that will work on the project for about 12 hours per week for 3 months.
Did you know
- Today, patients that are at risk of deteriorating have direct daily measurements to predict the standard Modified Early Warning Score (MEWS) model, which has been in use for around 20 years in hospitals worldwide.
- Around 80% of patients have physiological parameters that are outside normal ranges within the 24 hours before intensive care unit (ICU) admission.
- Models like MEWS and DI could predict the risk of deterioration earlier using measurements from the electronic patient files, which could lead to better patient care and outcomes.
Info Session: 28 October 2021
Presenter: Inge Tamminga, Senior Business Intelligence Analyst at ETZ
Application Deadline: 12 November 2021
Kickoff meeting: 19 November 2021
Midterm presentation: 14 December 2021
Final presentation: 15 February 2021