ML Ops Case Study - Preventing Premature Discharge of Patients from the ICU


Details
Diabetes Mellitus has been identified as one of the diseases leading to unfavorable health outcomes for patients with COVID-19. It is imperative that a patient diagnosed with Diabetes Mellitus admitted into an intensive care unit (ICU) should be allowed a stay of greater than 24 hours to improve outcomes. Given that the ICU may not readily have access to the electronic health records of these patients, being able to detect diabetes mellitus using information from within the last 24 hours of admission would be valuable.
Women in Data Science (WiDS) in conjunction with several health practitioners were able to curate datasets from de-identified ICU patients with information from the last 24 hours of admission to detect diabetes mellitus.
We will review and discuss a binary classification (ML) that can be re-used and adapted using ML Ops in the intensive care unit of a hospital facility to determine if a patient has diabetes mellitus and should be approved for a stay of longer than 24 hours.
During this interactive session, we will
- Review the dataset and discuss an end-to-end process for detecting diabetes using the CRISP-DM process model
- Discuss technology alternatives for the end-to-end process
- Discuss shortcomings of the CRISP-DM process model in the context of ML Ops
- Review the newer CRISP-ML(Q): A Machine Learning Process Model with Quality Assurance Methodology
- Optional: 15mins Guide to the data source and how to set up for individual learning
The agenda is as follows:
12 noon to 12:10 pm Opening Keynote, announcements, and other items of interest to meetup members
12:10 pm to 12:50 pm Interactive Session and Demo by Bola Adesanya
12:50 pm to 1:00 pm Closing Remarks
1:00 pm to 1:15 pm Optional: How to access the data and set up for individual learning

ML Ops Case Study - Preventing Premature Discharge of Patients from the ICU