YSYi-En Su

Yi-En Su

*This is an algorithmically generated author record based on publicly available information.

Yi-En Su has a total of 7 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Continuous Mattress-Derived Passive Sensing as a Low-Burden Monitoring Layer Supporting Comprehensive Geriatric Assessment in Long-Term Care: A Feasibility and Clinical-Utility Study (Preprint)

    by Chih-Yang Cheng, Chen-Yu Hsieh, Ming-Yueh Chou, Yi-En Su, I-Chiu Chang

    BACKGROUND Comprehensive geriatric assessment (CGA) is performed intermittently, leaving long intervals during which a resident's status may change unobserved. Continuous, low-burden passive sensing could supplement CGA, but the field has tended to overstate predictive value using single-split designs and imbalance-sensitive metrics. The realistic, deployable value of such sensing must be established with rigorous validation. OBJECTIVE To evaluate (1) the feasibility of long-term mattress-based passive sensing in long-term care and (2) whether the resulting behavioral phenotype is clinically useful as a between-assessment monitoring and triage layer for key CGA domains — explicitly not as a replacement for, or direct predictor of, CGA scores. METHODS Retrospective analysis of 192 long-term care residents aged ≥65 years at baseline, continuously monitored with the WhizPad mattress sensor. We quantified data coverage, then built daily-resolution behavioral-phenotype features over 7/14/30-day windows preceding each CGA assessment. For a primary domain (pressure-injury risk, Braden Scale) and a secondary domain (activities of daily living, ADL, Barthel Index) we evaluated discrimination of the higher-need band using unseen-resident (subject-wise) cross-validation (GroupKFold) with resident-cluster bootstrap 95% confidence intervals, calibration (Brier score and reliability), and clinical utility (decision-curve net benefit versus reassess-all and reassess-none strategies). A validation-methodology analysis quantified how naive designs overstate performance. RESULTS Sensing was feasible at scale: median monitoring span 1320 days (~3.6 years), median 1196 resident-days with data, 91% daily coverage, and 226,239 total resident-days. Under unseen-resident validation the sensor phenotype discriminated the higher-need band with an AUROC of 0.654 (95% CI 0.595–0.706) for pressure-injury risk and 0.713 (0.664–0.758) for ADL. Decision-curve analysis showed positive net benefit over both default strategies at decision thresholds of 0.30–0.60 (pressure injury) and 0.20–0.60 (ADL). Feature importance was mechanistically coherent, dominated by in-bed posture/position and movement signals. The methodology analysis showed that an imbalanced three-class macro-F1 endpoint and same-resident temporal validation substantially overstated apparent value: a status-proxy AUROC of ~0.76 under same-resident temporal splitting fell to ~0.62 under unseen-resident validation. CONCLUSIONS Continuous mattress sensing is feasible at scale and provides a modest but genuinely useful low-burden monitoring/triage signal for selected CGA domains, adding net clinical benefit at realistic decision thresholds despite modest discrimination. It is best positioned as a surveillance aid between assessments, not a CGA substitute. We provide a validation protocol to curb the overoptimistic claims common in this field and define the design changes required to strengthen the signal.