Detecting Postoperative Complications at Home Using AI and Wearables: New Study Published at IEEE MeMeA 2025
Postoperative complications remain a major concern after cancer surgery, especially once patients are discharged and no longer under direct clinical supervision. With the increasing availability of wearable sensors, continuous remote monitoring is emerging as a promising approach to support recovery and ensure timely intervention. However, interpreting these rich physiological data streams remains a challenge—particularly in the absence of clearly labeled deterioration events.
In our recent study, presented at the IEEE MeMeA 2025 conference, we explored how artificial intelligence could help address this challenge. The paper, titled “Unsupervised Detection of Postoperative Complications in Home-Monitored Patients: Preliminary Results,” investigates whether deep learning can support earlier identification of complications in patients recovering at home after major abdominal oncological surgery.
We applied an unsupervised LSTM autoencoder to continuous heart rate and respiratory rate data collected using a wearable patch sensor (Philips HealthDot). The model was trained on data from seven patients and then tested on three additional patients to detect deviations that could indicate potential clinical deterioration.
The findings are promising: our model detected physiological changes an average of 6.6 hours earlier than the standard Remote Early Warning Score (REWS) system. These earlier alerts could provide critical lead time for clinical teams to follow up and intervene if necessary.
While based on a small cohort of ten patients, the results highlight the feasibility of using unsupervised machine learning for postoperative monitoring in real-world settings. The use of patient-specific dynamic thresholds also shows potential for reducing false alarms while maintaining sensitivity to early warning signs.
This preliminary work is part of our ongoing research to explore data-driven tools that enhance postoperative care—especially in the transition from hospital to home.
📄 Read the full paper in IEEE Xplore: https://ieeexplore.ieee.org/document/11068044