Readmission rates are an important factor in assessing hospital performance. To prevent unfair penalization, prediction models are employed by policymakers and administrators from claims data to adjust for the performance of hospitals. Mostly regression models used for such predictions suffer in accuracy. Liu et. al. have applied ICD-9 codes embedding and ANN models to predict risk-standardized hospital readmission rates for 30 day readmission in patients hospitalized for pneumonia, heart failure or acute myocardial infarction. This approach uses a NLP technique called GloVe (Global Vector for Word Representations) for converting medical codes to numerical vectors which are passed as input to the ANN model. These ANN models trained upon diagnosis codes are more accurate than commonly used gradient boosting and hierarchical logistic regression models as code embeddings captures co-occurrence between diagnosis codes and ANN learns non-linear mapping patterns between the embeddings and the outcome.
Input variables : ICD-9 diagnosis and procedure codes of patients from claims data
Output Variables : 30-Day Readmission Prediction
| 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 : journals.plos.org
| Model Category | : | Public |
| Date Published | : | April, 2020 |
| Healthcare Domain | : | Provider |
| Code | : | github.com |
| Utilization Management |
| Readmission |