|
| Source | Incident type | Purpose | Best learner | Performance | Data source |
|
| Ong et al. [23] | All | To automatically detect extreme-risk events in clinical incident reports | Support vector machine | AUC = 0.92, F-measures = 0.86, precision = 0.88, and recall = 0.83 for incident types | Clinical incident reports |
| Marschollek et al. [24] | Fall events | To derive comprehensible fall risk classification models | C4.5 | Accuracy = 0.66, sensitivity = 0.55, specificity = 0.67, positive/negative predictive values = 0.15/0.94 | Fall incident reports |
| Cheng and Zhao [25] | Medication | To predict drug-drug interaction | Support vector machine | AUROC = 0.67 | DrugBank |
| Wang et al. [13] | All | To automate the identification of patient safety incidents in hospitals | Support vector machine | F-score = 0.78 for incident type and F-score = 0.87 for severity level | Incident reporting systems |
| Marella et al. [12] | All | To screen cases associated with the electronic health record | Naive Bayes | AUROC = 0.93, accuracy = 0.86, and F-score = 0.88 | Patient safety reporting system and electronic health records |
| Fong et al. [26] | All | To identify health information technology-related events | Logistic regression | AUC = 0.93 and F1 score = 0.77 | Patient safety event report |
| Comfort et al. [27] | Medication | To classify individual case safety reports within social digital media | Support vector machine | Accuracy = 0.78 and gKappa = 0.83 | Individual case safety reports and social digital media |
| Liu et al. [28] | Fall events | To explore potential fall incident clusters | Clustering | N/A | Incident reporting systems |
| Evans et al. [29] | All | To determine the incident type and the severity of harm outcome | Support vector machine | AUROC = 0.89 for incident types and AUROC = 0.71 for severity of harm | Incident reporting systems |
| Wang et al. [14] | Fall events | To predict the severity of inpatient falls | Multi-view ensemble learning with missing values | AUC = 0.81 | Incident reports |
| Wang et al. [15] | All | To identify incident types and severity levels | Convolutional neural network | F-scores >0.85 | Incident reporting systems |
| Liu et al. [16] | All | To improve the classification of the fall incident severity level | Random forest | Macro-F1 = 0.73 | Incident reporting systems |
|
