The model includes improved aggregation methods for a flexible deep learning architecture which learns a joint representation of patient chart, lab and output events with no pre-processing or variable selection. The model has an embedding layer followed by an aggregation function to reduce an arbitrary number of inputs to a fixed size representation and then it uses a recurrent neural network to classify patient mortality. It has been trained to take entire patient timeseries as input, regardless of event type, frequency or cardinality and to discretize each event. MIMIC-III dataset is used to obtain Intensive Care Unit (ICU) data which has high longitudinal density. Dynamic classification at each time step is achieved by a dense layer with sigmoid activation applied to each hidden state. AUROC score is used to evaluate model performance.
Input variables : EHR Data
Output Variables : Mortality Prediction
| Statistical | : | Mallow's CP | R Squared | Mean Square Error | Adjusted R Squared | Mean Absolute Error | Huber Loss |
| Business | : | Population at High Risk of Disease | Risk by Geography | Risk by Demographics | Risk by Clinical Parameters | Optimized Hospital Resource Utilization | Decreased Cost of Care | Decreased Patient Visits |
| 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 : arxiv.org
| Model Category | : | Public |
| Date Published | : | September, 2019 |
| Healthcare Domain | : |
Payer
Provider |
| Code | : | github.com |
| Health Risk Management |
| Health Risk Prediction |