Patient Ambulations Predict Hospital Readmission.

Area of Science:

• Gerontology
• Biomedical Engineering
• Health Informatics

Background:

• Improved functional ability and physical activity correlate with better health outcomes, including reduced hospital readmission.
• Remote monitoring of patients at home offers a way to track physical activity and functional status.

Purpose of the Study:

• To develop an algorithm for detecting patient ambulations from time-resolved step counts.
• To assess the predictive power of ambulation metrics for hospital readmission within 30 days of discharge.

Main Methods:

• A retrospective analysis of 233 patients (average age 70.5 years) using time-resolved step count data.
• Derivation of eleven statistical features from time series data and assessment of their F-statistics for discriminating readmitted vs. non-readmitted patients.
• Training logistic regression models to predict readmission using derived features, with evaluation via 5-fold cross-validation.

Main Results:

• The fraction of days with at least one ambulation was the strongest predictive feature (F-statistic=17.2).
• Logistic regression models achieved Area Under the Receiver Operating Characteristic Curve (AUROC) performances of 0.741 (all patients), 0.766 (congestive heart failure subgroup), and 0.769 (non-congestive heart failure subgroup).

Conclusions:

• Patient ambulation metrics derived from wearable sensors are powerful predictors of hospital readmission.
• These metrics can predict adverse clinical outcomes even without incorporating physiological vital signs.