delphi.ucsd.edu comprehensive community-level approaches

1
Ted Chan Kevin Patrick Lucila Ohno-Machado Jeannie Huang Bill Griswold Chaitan Baru Yannis Papakonstantinou Sanjoy Dasgupta Claudiu Farcas Fredric Raab Investigators: Health data are multi-layered: An individual’s health is affected by many factors across different layers Problem Goal: Create platform for generating a 360-degree view of an individual’s health by integrating all relevant data. Allow efficient analytics on these data Successful interventions require reasoning across layers Comprehensive Community-Level Approaches to Health Data Yannis Katsis NSF 1237174, Information & Intelligent Systems, 10/2012 – 08/2016 Jessica Block Max Menarini Center for Medicare & Medicaid Innovation Grant DELPHI will serve as the core architecture for a “Care Technology Innovation Platform” to address cardiovascular disease and stroke DELPHI-leveraged Efforts Organization Data Type County of San Diego, Health & Human Services Agency Health Statistics U.S. Census Bureau Demographics San Diego Air Pollution Control District Air Quality SDG&E Weather Caltrans Traffic ARJIS Crime US Customs & Border Protection Border Wait Times SanGIS GIS Beach Quality Report Water Quality Authentication Server OAuth 2.0 Application Server Web Application Smartphone Application Servlets AJAX Library Android Library Database Server Application Server Behavioral & Social Factors Physical activity Social Networks Diet Stress Personal Medical Factors Medical Records BMI Pharmaceutical Environmental Factors Transportation Crime Pollutants Health Determinants Source/Application libraries for Android & Web Applications Allows both native smartphone applications and web applications to act as both data consumers and data providers/sources HIPAA-compliant Backend Protects data behind firewall, filtering them before exposing them to the applications. Satisfies HIPAA requirements System Architecture Protected Backend Open Backend F I R E W A L L Sources/Applications Proof of Concept San Diego County as a test-bed Use DELPHI to build a 360-degree view of health in San Diego by integrating health data from different local companies & agencies Data Acquisition Acquire data of health-related factors and build corresponding wrappers Application Development Develop applications that retrieve integrated health data from DELPHI, showcasing the need for a 360 degree view of an individual’s health Asthma Patient App: Combines asthma info with air quality data from mobile sensors and the San Diego Air Pollution Control District Partnerships Predictive Analytics for Spatiotemporal Sensor Data Health data are often spatiotemporal Air quality data, wearable fitness tracking data, traffic data, cardiograms, etc. DELPHI WHOLE HEALTH INFORMATION MODEL (WHIM) W H A P I I N T E R F A C E WHIM: Models multi-level health data ANALYTICS WHAPI: Allows developers to access all integrated health data and write applications that leverage them INTERFACE: Allows source owners to register their sources DATA SOURCES APPLICATIONS Existing DBMSs are not suitable for spatiotemporal health data Spatiotemporal data are measurements taken at different points in time & space and therefore they cannot be directly integrated and correlated delphi.ucsd.edu GPS Readings Air Quality Readings e.g., Compute average quality of air breathed during a walk Learn from spatiotemporal health data by operating on the underlying models Instead of using the measurements, infer the corresponding model & use it for predictive analytics & already aggregated data from several sources provided by the Healthy Communities Institute (http://www.healthycommunitiesinstitute.com/) GPS Model Air Quality Model e.g., Compute average quality of air breathed during a walk Predicts GPS position at any point in time Predicts air quality at any point in time & space Unlike EMR data, the signals do not have common IDs to facilitate integration, since they are not aligned in space & time Model-aware integration returns the expected result, since the models predict the value of the corresponding signal for each point in time & space, thus facilitating the integration ANALYTICS LAYER: Allows developers to run common analytics efficiently Robert Wood Johnson Foundation Grant DELPHI is being proposed as the core data integration platform for a county-wide exploration of health indicators

Upload: others

Post on 03-Jan-2022

8 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: delphi.ucsd.edu Comprehensive Community-Level Approaches

Ted Chan Kevin Patrick

Lucila Ohno-Machado Jeannie Huang

Bill Griswold Chaitan Baru

Yannis Papakonstantinou

Sanjoy Dasgupta Claudiu Farcas

Fredric Raab

Investigators:

•  Health data are multi-layered: An individual’s health is affected by many factors across different layers

Problem

•  Goal: Create platform for generating a 360-degree view of an individual’s health by integrating all relevant data. Allow efficient analytics on these data

•  Successful interventions require reasoning across layers

Comprehensive Community-Level Approaches to Health Data

Yannis Katsis

NSF 1237174, Information & Intelligent Systems, 10/2012 – 08/2016

Jessica Block

Max Menarini

Center for Medicare & Medicaid Innovation Grant DELPHI will serve as the core architecture for a “Care Technology Innovation Platform” to address cardiovascular disease and stroke

DELPHI-leveraged Efforts

Organization Data Type County of San Diego, Health & Human Services Agency

Health Statistics

U.S. Census Bureau Demographics San Diego Air Pollution Control District

Air Quality

SDG&E Weather Caltrans Traffic ARJIS Crime US Customs & Border Protection

Border Wait Times

SanGIS GIS Beach Quality Report Water Quality

Authentication Server

OAuth 2.0

Application Server

Web Application

Smartphone Application

Servlets

AJAX Library

Android Library

Database Server

Application Server

Behavioral & Social Factors

Physical activity

Social Networks

Diet

Stress

Personal Medical Factors

Medical Records

BMI

Pharmaceutical

Environmental Factors

Transportation

Crime Pollutants

Health Determinants

•  Source/Application libraries for Android & Web Applications Allows both native smartphone applications and web applications to act as both data consumers and data providers/sources

•  HIPAA-compliant Backend Protects data behind firewall, filtering them before exposing them to the applications. Satisfies HIPAA requirements

System Architecture

Protected Backend Open Backend

F I R E W A L L

Sources/Applications

Proof of Concept

•  San Diego County as a test-bed Use DELPHI to build a 360-degree view of health in San Diego by integrating health data from different local companies & agencies

Data Acquisition Acquire data of health-related factors and build corresponding wrappers

Application Development Develop applications that retrieve integrated health data from DELPHI, showcasing the need for a 360 degree view of an individual’s health

Asthma Patient App: Combines asthma info with air quality data from mobile sensors and the San Diego Air Pollution Control District

Partnerships

Predictive Analytics for Spatiotemporal Sensor Data

•  Health data are often spatiotemporal Air quality data, wearable fitness tracking data, traffic data, cardiograms, etc.

DELPHI

WHOLE HEALTH

INFORMATION MODEL (WHIM)

W H A P I

I N T E R F A C E

WHIM: Models multi-level health data

ANALYTICS

WHAPI: Allows developers to access all integrated health data and write applications that leverage them

INTERFACE: Allows source owners to register their sources

DATA SOURCES APPLICATIONS

Existing DBMSs are not suitable for spatiotemporal health data Spatiotemporal data are measurements taken at different points in time & space and therefore they cannot be directly integrated and correlated

delphi.ucsd.edu

GPS Readings

Air Quality Readings

e.g., Compute average quality of air breathed during a walk

Learn from spatiotemporal health data by operating on the underlying models Instead of using the measurements, infer the corresponding model & use it for predictive analytics

& already aggregated data from several sources provided by the Healthy Communities Institute (http://www.healthycommunitiesinstitute.com/)

GPS Model

Air Quality Model

e.g., Compute average quality of air breathed during a walk

Predicts GPS position at any point in time

Predicts air quality at any point in time

& space

Unlike EMR data, the signals do not have common IDs to facilitate integration, since they are not aligned in space & time

Model-aware integration returns the expected result, since the models predict the value of the corresponding signal for each point in time & space, thus facilitating the integration

ANALYTICS LAYER: Allows developers to run common analytics efficiently

Robert Wood Johnson Foundation Grant DELPHI is being proposed as the core data integration platform for a county-wide exploration of health indicators