DynaHealth Backend (Data Collection, Processing, Training, and Deployment)

Authors: Duy Ho, Hieu Trinh, Nhung Tran

Award: 1st place

Award Page: https://info.umkc.edu/hack-a-roo/spring-2021-it-track/

Note: For frond-end, please visit: https://github.com/ngchieu857529/hackaroo_spring_2021

Motivation

Health insurance is one of the most important protections for an individual. Having a health insurance plan not only helps ease the mind but also covers a significant amount of cost in critical or unexpected conditions such as illness and accident. However, not everyone has easy access to and knowledge about health insurance in a convenient and informative way. Thus, users may not have sufficient awareness about the topic to make a decision. Moreover, data from many sources are not represented in a transparent or meaningful manner, leading to uncertainty and confusion.

Overview

Main Features

Two domains:

COVID-19:

Statistics

• Number of cases
• National
• Statewide
• County

Distribution of Cases

• Geography

Vaccination

• Total first/second doses
• Geographic Distribution of doses

Health Insurance

Socioeconomic Visualizations:

• Employment status
• Insurance status
• Geography
• Coverage type
• Family status
• Education
• Income
• Poverty

Smart Insurance:

• Cost Estimator
  • Chatbot
  • Plan Recommendations
• Neural Network

Implementation

Front End

• React:
  • Leading platform for quick, reliable, and scalable web applications.
  • Mobile-friendly
  • Maintainability and support
• Vercel: deployment platform. Integrated with GitHub with automatic deployment

Backend

• Python: best framework for data science and deep learning
• PyTorch: one of the most popular frameworks for building DL models
• Jupyter Notebook: an interactive tool to work and share Python code efficiently
• Jupyter Dash + Plotly: powerful & interactive visualization tool for Python

Datasets

• COVID-19 Dataset: https://data.cdc.gov/Case-Surveillance/COVID-19-Case-Surveillance-Public-Use-Data-with-Ge/n8mc-b4w4 - COVID-19 Dataset (updated April 13, 2021) ~ 22.5 million records

• Vaccination Datasets: https://data.cdc.gov/Vaccinations/COVID-19-Vaccine-Distribution-Allocations-by-Juris/saz5-9hgg/data - Vaccination Data - Moderna (April 2021) https://data.cdc.gov/Vaccinations/COVID-19-Vaccine-Distribution-Allocations-by-Juris/b7pe-5nws - Vaccination Data - Pfizer (April 2021)

• Health Insurance Datasets: https://www.census.gov/data/datasets/time-series/demo/cps/cps-asec.html - Census Health Insurance Data and Demographics (updated April 13, 2021) https://www2.census.gov/programs-surveys/cps/datasets/2020/march/ASEC2020ddl_pub_full.pdf - Census Health Insurance Data and Demographics Data Dictionary (updated April 13, 2021) https://www.healthcare.gov/health-and-dental-plan-datasets-for-researchers-and-issuers/ - Health Insurance Plans and Prices for Individuals and Families (2017-2021) https://www.kaggle.com/teertha/ushealthinsurancedataset - Simplified Health Insurance

Approach

• Data Research/Collection

• Data Curation

  • Data simplification
  • Eliminate invalid values

• Data Integration

  • Time-series Analysis
  • Boundary Integration

• Data Visualization

• Multiple charts/graphs

• Model Training

  • Data Normalization
  • Neural Network architecture: 9 layers

• Model Inference

  • Input: Age, BMI, Sex, Children, Smoke, Location
  • Output: estimated yearly

• Model Deployment

  • API endpoints: hackaroo.ngrok.io/api/model/quick
  • Parameters: age, height, weight, children, smoker, state
  • https://hackaroo.ngrok.io/api/model/quick?height=150&weight=55&age=65&children=1&smoker=0&sex=1&state=MO [Deprecated]

• Application Deployment

  • Vercel: https://hackaroo-spring-2021.vercel.app/ [Deprecated]
  • Automatic deployment upon commit.

Sample Server JSON

Input : http://hackaroo.ngrok.io/api/model/quick?height=150&weight=55&age=65&children=1&smoker=0&sex=1&state=MO Output:

{
  "plans": {
    "silver": {
      "4748A0001A002008": {
        "State Code": "MO",
        "region": "southeast",
        "Metal Level": "silver",
        "Issuer Name": "WellFirst Health",
        "Plan ID (Standard Component)": "4748A0001A002008",
        "Plan Marketing Name": "WellFirst Silver HSA-E 4500X",
        "Plan Type": "EPO",
        "Customer Service Phone Number Local": "1-866-514-4194",
        "Customer Service Phone Number Toll Free": "1-866-514-4194",
        "Customer Service Phone Number TTY": "1-866-514-4194",
        "Network URL": "http://www.wellfirstbenefits.com/find-a-doc/marketplace-epo-plan-providers/",
        "Plan Brochure URL": "http://www.wellfirstbenefits.com/Individuals-and-Families",
        "Summary of Benefits URL": "https://sbc.wellfirstbenefits.com/api/GetPdf?wellFirst_WellFirst%20MO%20Silver%20HSA-E%204500X01_0121.PDF&true",
        "Individual+1 child, Age 50+": "888.99"
      },
      "99723M0009011": {
        "State Code": "MO",
        "region": "southeast",
        "Metal Level": "silver",
        "Issuer Name": "Ambetter from Home State Health",
        "Plan ID (Standard Component)": "99723M0009011",
        "Plan Marketing Name": "Ambetter Balanced Care 11 (2021)",
        "Plan Type": "EPO",
        "Customer Service Phone Number Local": "1-855-650-3789",
        "Customer Service Phone Number Toll Free": "1-855-650-3789",
        "Customer Service Phone Number TTY": "N/A",
        "Network URL": "https://ambetter.homestatehealth.com/findadoc",
        "Plan Brochure URL": "https://www.ambetterhealth.com/content/dam/centene/ambetter/brochures/MO-2021.pdf",
        "Summary of Benefits URL": "https://api.centene.com/SBC/2021/99723M0009011-01.pdf",
        "Individual+1 child, Age 50+": "901.26"
      },
      {...},
       ...
    }
  }
}

Embedded Interactive Visualizations

Future Scope

• Human in the loop: Real-time feedback​
• Multiple simultaneous messages from each side.​
• More knowledge domains​
• UI improvement​
• Model deployment on the Cloud (Amazon AWS, Microsoft Azure, …)​
• Application Deployment in Google Playstore​ and App Store
• Multi-modality Analysis
• Natural Language Processing​
• Object Detection (to interpret user’s uploaded images)​
• User Sentiment Analysis​
• Diverse text-to-speech models (gender-based, age-sensitive, and more natural-sounding)​
• Speech-to-text features (to simulate phone calls or Facetiming)​