stroke-readmission-model

Capstone project for Data Science Masters program at the University of Wisconsin.

Overview

Contributors: Samuel Edeh and Halee Mason
Capstone: Predicting Risk of Re-Admission for Stroke Patients
Completion Date: August 2017
Data Source: Nationwide Re-admissions Database (NRD)

Description

Built a predictive model using TensorFlow that was optimized to forecast the risk of rehospitalization in the year following stroke. Model was trained on data from the Nationwide Re-admissions Database (NRD) from the Healthcare Cost and Utilization Project (HCUP) State Inpatient Databases (SID). Deployed the model using the Google Cloud Platform here.

Motivation

Collaborated with the client to understand the business value that successful data science application would have for this project and focused on model explainability by using techniques such as LIME. Implemented model to drive patient outcomes to help the client segment high risk patients and engage them more closely. The final deliverable was an API endpoint which allowed for the client to input patient specific parameters (input features) and for the model to output a personalized risk score of rehospitalization.

Demo

call_cloud_service.ipynb provides a demonstration of how to connect to the Google Cloud ML Service to make a prediction from a trained model stored in Google Cloud.

Accessing the App

The app is live here.

Follow the text prompt in the fields and enter information correctly. For e.g., when the prompt asks you to enter "y/n", you could enter "y" or "yes", and it will consider it as "yes" that the patient has the condition. Entering "n" or "no" indicates that the patient does not have the condition.

A patient will have 13 input features. All are yes or no features except the 8th, with is the patients age. Example patient feature vector: patient_1 = [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 86, 0.0, 0.0, 0.0, 0.0, 1.0] correspond to these inputs:

1. pay1_private PrivateInsurance: Does the patient have private insurance? (y/n)
2. metro Metro: Does the patient live in a metro area? (y/n)
3. diabetes Diabetes: Is the patient diabetic? (y/n)
4. copd COPD: Does the patient have COPD? (y/n)
5. ckd CKD: Does the patient have CKD? (y/n)
6. chf CHF: Does the patient have CHF? (y/n)
7. atrial_fib AFib: Does the patient have AFib? (y/n)
8. age Age: Enter the patient's age.
9. hyperlipidemia Hyperlipidemia: Does the patient have hyperlipidemia? (y/n)
10. sex Sex: Is the patient male or female? (m/f)
11. nicotine Nicotine: Is the patient a smoker? (y/n)
12. obesity Obesity: Is the patient obese? (y/n)
13. hypertension Hypertension: Is the patient hypertensive? (y/n)