Ashfaq et. al. have developed a deep learning framework for predicting 30-day unscheduled readmission. This model incorporates both: (a) human derived features(HDF): age at the time of visit, Gender, Medication compliance, Total procedures performed, Duration of stay, Duration of all stays, Type of visit, Charlson comorbidity score, Number of prior emergency care visits, Number of prior admissions and Number of prior outpatient visits and (b) machine-derived features (MDF): all clinical codes related to diagnoses, procedures medications and lab tests. The HDF and MDF are fed as input to a sequential Long Short-Term Memory (LSTM) neural network. The resulting class imbalance is addressed by cost-sensitive classification and is used to predict 30-day readmission risk at each visit of the patient. It outperforms models where any of these features are ignored and possess a high discrimination ability F1-score of 0.51 and AUC 0.77.
Input variables : Features from EHR data-Patient Demographics, Visit details, Clinical Codes
Output Variables : 30-day readmission risk at each visit
| Statistical | : | Somers D | Accuracy | Precision and Recall | Confusion Matrix | F1 Score | Roc and Auc | Prevalence | Detection Rate | Balanced Accuracy | Cohen's Kappa | Concordance | Gini Coefficent | KS Statistic | Youden's J Index |
| Business | : | Readmission Rate | Avg Hospital LOS |
| Infrastructure | : | Log Bytes | Logging/User/IAMPolicy | Logging/User/VPN | CPU Utilization | Memory Usage | Error Count | Prediction Count | Prediction Latencies | Private Endpoint Prediction Latencies | Private Endpoint Response Count |
Visit Model : github.com
Additional links : sciencedirect.com
| Model Category | : | Public |
| Date Published | : | July, 2019 |
| Healthcare Domain | : | Provider |
| Code | : | github.com |
| Utilization Management |
| Readmission |