Table 1.
Descriptive table of ML-based predictive works on discharge disposition
| Author | Setting | Discharge Outcome | Summary | Models Used |
|---|---|---|---|---|
| Goto et al. [9] | Asthma/COPD patients in the emergency dept | Binary (ICU vs. non-ICU hospitalization) | Compares the four models’ predictive capability to a baseline logistic regression concluding that the ML models markedly improved prediction capability | Lasso regression (LR), Randon forest (RF), Boosted decision tree (BDT), Artificial Neural network (ANN) |
| Karhade et al. [10] | Elective inpatient surgery for lumbar degenerative disc disorders | Binary (routine vs non-routine postoperative discharge) | Created an open-access web application for healthcare professionals that showed promising results for preoperative prediction of non-routine discharge | ANN, Support vector machine (SVM), Bayes point machine, BDT |
| Greenstein et al. [11] | Post-operative discharge after total joint arthroplasty (TJA) | Binary (skilled nursing facility vs. elsewhere) | Developed an EMR-integrated prediction tool to predict discharge disposition after TJA | ANN |
| Ogink et al. [12] | Post-operative discharge after degenerative spondylolisthesis | Binary (home vs. non-home) | Similar to [5], compares a set of predictive model’s performance after elective spinal surgery | ANN, SVM, Bayes point machine, BDT |
| Cho et al. [13] | Post-stroke acute care | Binary (home vs. facility) | Compares the performance of four interpretable ML models on post-stroke discharge prediction | LR, RF, AdaBoost, multi-layer perceptron |
| Muhlestein et al. [14] | Post-craniotomy | Binary (home vs. non-home) | Uses 26 ML algorithms to combine the best performers into ensemble model investigate the impact of race on discharge disposition | Ensemble (various) |
| Muhlestein et al. [15] | Post-meningioma resection | Binary (home vs. non-home) | Similar to [10], creates an ensemble model showing significantly improved accuracy compared to traditional logistic regression | Ensemble (various) |
| Abad et al. [16] | ICU critical care | Multi-class (home, nursing facility, rehab, death) | Investigates the impacts of APACHE IV scores on patient discharge via an array of different ML models | LR, XGBC, RF |
| This research study | Post-stroke acute care | Binary (home vs. non-home) and Multi-class (home, nursing facility, rehab, death) | Compares the performance of 5 ML models in both a binary and multi-class experiment and investigates the explainability of the best-performing models | RF, XGBC, KNN, SVM, LR |