Table 3.

Performance of 30-day hospital readmission prediction models.

Studies	Methods	Data	Prediction Model Performance
Current study	PA data with a logistic regression ML model	Continuous PA data, hospital medical records	Accuracy of predicted events: 71.43%Precision ($\mathit{T}\mathit{P}$ rate): 70.35%
Lin W.-Y. et al. [25]	PA data with statistical-mathematical model	Continuous PA data, hospital records	Accuracy of predicted events: 52.38%Precision ($\mathit{T}\mathit{P}$ rate): 37.78%
Amalakuhan B. et al. [5]	55 feature variables for COPD exacerbations, random forest ML model	Demographic data, hospital medical records based on ICD-9 codes	Positive predictive value (accuracy in prediction): 70% (0.7)
Chawla H. et al. [23]	Vector magnitude units (VMU), i.e., summed movements in three planes over each minute, logistic regression ML model	PA data recorded with GT3X+ accelerometer, derived indices, hospital medical records	31.58% of patients had all-cause hospital readmissions, patients with lower PA are 6.7 times more likely to be readmitted
Min X. et al. [26]	Traditional and deep learning ML models: logistic regression, support vector machine, random forest, and multilayer perceptron	Knowledge-driven: hospital Score, LACE index, handcrafted features; Data-driven: reshaped data grouped into categories	Prediction performance with data-driven features: 65% Combined (knowledge-driven and data-driven): 65.30%
Goto T. et al. [27]	Recorded PA data used with logistic regression and Lasso regression ML models	Self-reported, manually assessed, static PA data	7% of patients had 30-day readmissions. Prediction classification ability (precision): 61.00%