Combining Hospital-grade Clinical Data and Wearable Vital Sign Monitoring to Predict Surgical Complications

Access to the right data at the right time is a substantial hurdle for developing effective machine learning (ML) in healthcare. Many hospitals maintain several isolated patient databases, often leading to incomplete datasets that severely impact the clinical usability of ML prediction systems. Moreover, in visceral surgery, postoperative complications often occur in the nursing ward, where real-time monitoring is non-existent. ML-powered predictive systems are increasingly ineffective as they leave the Intensive Care Unit (ICU) due to data shortage. However, the patient is still at risk of various complications. Our work addresses both issues using a collection framework combining pre-operative, intra-operative, ICU, and general patient parameters. We add a high-resolution continuous vital sign measurement modality collected on a non-intrusive hybrid nursing ward. Using the wearable data, we observe improved prediction accuracy for Surgical Site Infection (SSI). Our work suggests a need for hybrid monitoring after a patient’s ICU stay to further ML modeling in clinical settings.