CASE STUDY

Anomaly Detection in EMR

Company Overview

The company provides real-time predictive services of erroneous medical prescriptions for healthcare providers, pharmacy benefit management (PBM) companies, and pharmacy chains.

Business Case

Real-time predictions of wrongly prescribed medical treatments.
 

Data

Over eight years of medical prescriptions, admissions, diagnoses, and clinical measurements. High value and confidential data.
 

Tools

Production-grade deep learning models implemented in TensorFlow.
 

Business Impact and Insights

Shifting categorical data representations to embeddings and continuous data representation, enabling use of advanced tools, such as recurrent neural networks.

These representations also provided a way to semantically categorize similar and contradicting drugs derived from the data alone, without domain expert interference.

Building upon these representations, anomaly detection models are used to predict erroneous prescriptions which outclassed previous category-based models.

< BACK TO projects portfolio