Combining Time Series Modalities to Create Endpoint-driven Patient Records
Keywords: ICU, EHR, EMR, Timeseries, PPG, Wearable, Surgical
TL;DR: We combine clinical information with wearable data to predict averse outcomes.
Abstract: A major hurdle for developing effective ML systems in healthcare is access to the right data at the right time. Many hospitals maintain several isolated patient data management systems, often leading to incomplete datasets when developing ML systems, severely impacting the clinical usability of prediction systems. Moreover, Intensive Care Unit (ICU) stays are short due to considerable cost, leading to (premature) transfers to the nursing ward, where real-time monitoring is often non-existent. ML-powered predictive systems here are increasingly ineffective due to data shortage, but patients still risk various complications. Our work addresses this with a framework that combines pre-operative, operational, ICU, and lab-test parameters. Additionally, we include high-resolution continuous vital sign measurements originating from a non-intrusive hybrid nursing ward in our dataset. Using this wearable data, we observe improved prediction accuracy for Surgical Site Infection (SSI) after gastrointestinal surgery. Our work suggests a need for hybrid monitoring after a patient’s ICU stay to further ML modeling in clinical settings and a need for more problem-centric ML.
Primary Subject Area: Domain specific data issues
Paper Type: Extended abstracts: up to 2 pages
DMLR For Good Track: Participate in DMLR for Good Track
Participation Mode: In-person
Confirmation: I have read and agree with the workshop's policy on behalf of myself and my co-authors.
Submission Number: 91
Loading