Comprehensive Summary
This prospective study explored whether wearable inertial measurement units (IMU) sensors combined with deep learning could be predictive of functional recovery in older adults, post-surgery. They recruited 39 surgical patients who were an average age of 79.3 years and 43.6% women, from the SURGE-Ahead project. They wore lumbar IMUs for 5 days after their operations. They assessed their mobility, using the Charite Mobility Index (CHARMI), activities of daily living (ADL) using the Barthel index and discharge destination (e.g., home, geriatric care or rehabilitation facilities). They ultimately found that the deep learning models had strong predictive abilities with high accuracy. For the CHARMI, the model showed great reliability with a R² value of 0.65. The Barthel index achieved a higher R² value of 0.70. Finally, the model correctly predicted whether patients would return home, require rehabilitation, or need geriatric care with 82% accuracy. Predictions showed significant agreement with clinical assessments (weighted kappa ≥ 0.80). The key predictors were the IMU z-channel which captured vertical/lumbar motion and recumbency and walking bout parameters. This highlights that IMU sensors and AI can successfully predict functional assessments and aid in planning for geriatric patients.
Outcomes and Implications
This study shows that the use of wearable IMUs with deep learning has significant clinical relevance in geriatric surgery and recovery care for elderly patients. The continuous and objective monitoring of mobility and functional ability be integrated with traditional clinical assessments, which are often subjective and occasional. Additionally, the prediction of discharge destinations supports clinicians and patients in making efficient care plans by separating those who can safely return home and those who require rehabilitation or bariatric care, which can potentially reduce readmission rates. Additionally, it encourages optimized resource allocation and prevents early, unsafe discharges. Furthermore, this model identifies limited mobility of ADL impairments, allowing for early interventions such as physical therapy or extending inpatient rehabilitation, which can improve overall postoperative outcomes for these vulnerable patients and promote personalized rehabilitation strategies. However, this model requires further research because the current performance is promising but is still lower than the agreement that is typically seen by clinicians (kappa of 0.90 - 0.096). Therefore, this model should currently be used as a supportive clinical decision tool instead of the sole diagnostic system. Future research should focus on validating these findings with a larger and diverse sample to increase generalizability. Additionally, a standardized protocol should be developed before it can be implemented in clinical environments.