Patient Event Prediction & Explanation
Tech Stack: Databricks, Python, SQL, PySpark, Pandas, Tensorflow, PyTorch, Scikit-learn, SHAP
- Created a patient event prediction Python module to predict and explain the likelihood of events, such as disease diagnosis and drug switches for patients, based on their medical history.
- Analyzed patient journeys to create cohorts and establish a framework for applying machine learning to real-world patient-level data.
- Investigated various approaches for encoding a patient's medical history (structured EHRs) into a tensor for input to predictive models.
- Integrated novel deep learning architectures like Deepr, PatImg2d, and PatImg3d, along with a SHAP interpreter to achieve benchmarking scores and perform rapid testing.