Predictive Analytics and Implementation Research (PAIR) Circle Poster

Predictive Analytics and Implementation Research (PAIR): Charting a Research Agenda for the 21st Century Workshop - Executive Summary

Two Rockledge Center
NHLBI Conference Room 9100-9104
6701 Rockledge Drive


Background: Predictive analytics in implementation research (PAIR) begins with integration of relevant, diversified data that provide comprehensive profiles of disease burden and community resources. Experiential data of requirements and impact of evidence-based interventions generated previously are applied to the targeted community-specific profiles. Forecasts of the resources needed and anticipated impact of adaptation of evidence-based interventions are generated. Predictive analytical forecasts may include: risk-factor burdens, resources needed to implement the intervention, projections of subsequent improvements in health care delivery and/or outcomes, assessment of wide-scale health policy changes, etc. Through the use of pertinent data and modelling, the critical details and challenges can be identified and addressed prior to implementation or conducted intermittently to refine deployed strategies. Some intervention research projects for evidence-based interventions to address heart, lung, blood, and sleep disorders (HLBS) are now being sub-optimally conducted. The use of PAIR may lead to less-disruptive, more cost-efficient and sustainable intervention strategies.

Executive Summary: PAIR’s ability to assess the complex interrelationships of a community’s state-of-health and state-of-resources, in order to forecast prospective impact of implementation research was well recognized. The term "environment-wide" was offered to describe the appropriate organization and integration of data representing the interrelationships of clinical, social, behavioral, and environmental risk and resource profiles, which are needed to give the full community picture in order to achieve better public health. This definition led to the refined concept that PAIR could be more accurately expressed as implementation research which is driven through the conduct of environment-wide associations (EWAS). Participants emphasized the timeliness of extending on-going advancements across the pre-workshop defined elements of PAIR: data science; artificial intelligence and machine learning; predictive modelling; data visualization; training and implementation research. Other areas which were explored include: data access; need for continual updating of data; ethical use, privacy, and security of the data; importance of shared values of the data and the analyses; need for information driving behavior be readily accessible and actionable; ability to explain EWAS to garner public support and use; critical need for development of EWAS/PAIR implementation frameworks to support hypothesis testing; and, importance for on-going stakeholder engagement.

See PAIR Workshop References


Highlights of prospective elements of 21st Century Agenda for PAIR to address HLBS disorders include:

  • Place matters: Better organization and availability of data resources to assess disease risk in geospatially defined populations and the medical, behavioral, social and environmental factors associated with that risk.
  • Methods to merge domains in order to link and layer clinical, social, behavioral, and environmental risk markers.
  • Identification of high-risk community needs to facilitate community-level interventions that are evidence-based, feasible, and tailored.
  • Definition of the "exposome" at neighborhood or more concise definition of a community.
  • Merging data from multiple biomedical domains; need for imaging and other laboratory results.
  • Integration of environmental, behavioral, and social risk information with biomedical risk to create a global risk score, for heart, lung, blood, and sleep outcomes.
  • Implementation science to assess effectiveness and costs of EWAS and global risk scores in a range of community contexts.
  • Methodological development of environment-wide association studies (EWAS) informed by prior experience with genome-wide association studies (GWAS).
  • Community-based trials of global risk assessment to demonstrate feasibility, cost, effectiveness, and sustainability of risk reduction strategies.
  • Creation of curricula, training programs, and career pathways to assure a scientific workforce supportive of high-quality/high-impact research in data science through the use of predictive analytics.

NHLBI Contact

  • Rebecca A. Roper, MS, MPH
  • Center for Translation Research and Implementation Science
  • National Heart, Lung, and Blood Institute

Meeting Participants

  • Carmela Alcántara Ph.D.
    Director, Sleep, Mind, and Health Research Program
    Columbia School of Social Work
  • Craig Blakely, Ph.D., M.P.H.,
    Dean, School of Public Health and Information Science
    University of Louisville
  • Robert M. Califf, M.D., MACC
    Professor of Medicine, Duke Clinical Research Institute
    Duke University School of Medicine
  • Barbara A. Downs, Ph.D.,
    Director, Federal Statistical Research Data Center
    United States Census Bureau
  • John L. Eltinge, Ph.D.,
    Assistant Director, Research and Methodology
    United States Census Bureau
  • Andrew Hamilton, M.S., B.S.N.,
    Chief Informatics Officer and Deputy Director
  • Lorens Helmchen, Ph.D.,
    Associate Professor, Health Policy and Management
    Milken Institute School of Public Health
    George Washington University
  • David M. Kent, M.D., M.Sc.
    Director, Predictive Analytics and Comparative Effectiveness (PACE) Center
    Sackler School of Graduate Biomedical Sciences
    Tufts University, Tufts Medical Center
  • Amy Kind, Ph.D., M.D.
    Director, Department of Medicine Health Services and Care Research Program
    Associate Professor, Division of Geriatrics
    University of Wisconsin School of Medicine and Public Health
  • John Kravitz, M.H.A., CHCIO, SVP
    Chief Information Officer
    Geisinger Health System
  • Thomas A. Pearson, Ph.D., M.D., M.P.H.
    Executive Vice President for Research and Education
    University of Florida Health Sciences Center
  • Mattia Prosperi, Ph.D., M.Eng.
    Associate Professor, Department of Epidemiology
    College of Public Health and Health Professions & College of Medicine
    University of Florida
  • Matthew Quinn, M.B.A. Senior Advisor for Health Information Technology
    Health Resources and Services Administration (HRSA)
  • Paula K. Shireman, M.D., M.S., M.B.A.
    Professor, Dielmann Chair of Surgery
    School of Medicine
    University of Texas Health Science Center at San Antonio
  • Rhonda Szczesniak, Ph.D.
    Associate Professor, Division of Biostatistics & Epidemiology
    Division of Pulmonary Medicine
    Cincinnati Children’s Hospital
    Department of Pediatrics
    University of Cincinnati

CTRIS Meeting Attendees

  • Cheryl Anne Boyce, Ph.D.
    Chief, Implementation Science Branch
  • Kathleen Fenton, M.D.
    AAAS Fellow
    HIGH Branch
  • Melissa Green-Parker, Ph.D.
    Health Scientist Administrator
    HIGH Branch
  • George A. Mensah, M.D., FACC
    Director, Center for Translation Research and Implementation Science
  • Mark Parker
    Program Support Assistant
  • LeShawndra N. Price, Ph.D.
    Chief, Health Inequities and Global Health Branch
  • Rebecca Roper, M.S., M.P.H.
    Health Science Policy Analyst
    Implementation Science Branch

NHLBI Meeting Attendees

  • Marishka K. Brown, Ph.D.
    Program Director
    National Center on Sleep Disorders Research
  • Rebecca Ann Campo, Ph.D.
    Health Scientist Administrator
    Clinical Applications and Prevention Branch
  • Thomas L. Croxton, Ph.D., M.D.
    Branch Chief, Airway Biology and Disease Program
    Division of Lung Diseases
  • David Calvin Goff, Jr., Ph.D., M.D.
    Division Director, Office of the Director
    Division of Cardiovascular Sciences
  • Lucy Leemeng Hsu, M.P.H.
    Scientific Program Specialist
    Division of Cardiovascular Sciences
  • Sharon Michelle Smith, Ph.D.
    Health Scientist Administrator
    Translational Blood Science and Resources Branch
    Division of Blood Diseases and Resources
  • George John Papanicolaou, Ph.D.
    Research Geneticist, Epidemiologist
    Division of Cardiovascular Sciences
  • Xin Tian, Ph.D.
    Statistician, Office of Biostatistics Research
    Division of Cardiovascular Sciences
  • Colin O. Wu, Ph.D
    Statistician, Office of Biostatistics Research
    Division of Cardiovascular Sciences


8:00 a.m.

8:30 a.m.
Welcome and Workshop Objectives
Rebecca Roper, M.S., M.P.H. and George A. Mensah, M.D., FACC

9:00 a.m.
Progress and Challenges in Predictive Analytics
Thomas A. Pearson, Ph.D., M.D., M.P.H.

9:30 a.m.
Predictive Analytics in Cardiovascular Disease Prevention and Treatment
Robert M. Califf, M.D., MACC

10:00 a.m.
Panel Discussion

Panel: Dr. Thomas A. Pearson and Dr. Robert M. Califf

Moderator: LeShawndra N. Price, Ph.D.

10:30 a.m.

10:45 a.m.
Panel Session: Challenges and Opportunities in Data Science

  • Using what we already have on hand: What are barriers to and strategies for affordable and efficient access to recurring extant data sets?
  • Categories of data and tradeoff considerations for data timeliness, completeness, and parsimony: When is the ability to follow timely patterns more important than precision of large scale legacy data?
    • Moderator: John Eltinge, Ph.D.
    • Summary: Sharon Smith, Ph.D.
    • Discussants: Amy Kind, Ph.D., M.D.; Andrew Hamilton, M.S., B.S.N.; John Kravitz, M.H.A., CHCIO, SVP; Ali Mokdad, Ph.D.; and Matt Quinn, M.B.A.


  1. How do you identify the best strategies to locate and to evaluate data suitable for analyses across the individual, system, and society level?
  2. What type of resources (e.g. financial, infrastructure platforms, etc.) are needed to import and manipulate big data to conduct predictive analytics?
  3. What factors should you consider for manipulation and analyses of a project?
  4. Who are the key partners and stakeholders who hold data for predictive analytics? Are there strategies to promote data use and analyses for environmental, economic, social determinants?

At 11:20, Dr. Smith gives five-minute summary of highlights.

11:30 a.m.
Panel Session: Analytical Approaches: Tools, Techniques, Training

Intersection of data sources and methods in conducting public health informatics modelling

  • Are different approaches needed for modelling prevention versus control or containment of a clinical or public health threat?
  • How can we engage schools of public health and medicine to support training and use of extant data sets to explore multi-level strategies for public health impact?
    • Moderator: Craig Blakely, Ph.D., M.P.H.
    • Summary: Melissa Green Parker, Ph.D.
    • Discussants: Camela Alcántara, Ph.D.; Lorens Helmchen, Ph.D.; Mattia Prosperi, Ph.D.; Paula Shireman, M.D., M.S., M.B.A.; and, Rhonda Szczesniak, Ph.D.


  1. Which available methods and tools can yield quick returns on implementation research questions?
  2. Who are the key partners needed to move the Predictive Analytics methodology forward in the short-term (i.e., two to three years)? In the medium term (i.e., four to seven years)?
  3. What are the key strategies for training the workforce?
  4. What are research examples that address compelling scientific questions and challenges in NHLBI mission areas that can be supported by current analytical approaches and techniques?

At 12:15, Dr. Green Parker gives five-minute summary of highlights.

12:25 p.m.

1:00 p.m.
Predictive Analytics Use Cases

Sparked by summaries of four use cases, discussants will join the conversation as everyone compares and contrasts their experience in use of predictive analytics for implementation research. We foresee a lively dialogue of practical insights on data sources, data linkages, strategies for stakeholder partnerships, analytical methods, and scope of prospective application in the conduct of implementation research, across the spectrum of heart, lung, blood and sleep conditions.

  • Moderator: David Goff, Ph.D., M.D.
  • Summary: Marishka K. Brown, Ph.D.; George Papanicolaou, Ph.D.
  • Discussants: Andrew Hamilton, M.S., B.S.N.; Ali Mokdad, Ph.D.; Camela Alcántara, Ph.D.; Amy Kind, Ph.D., M.D.
  • Use Case Presenters:
    1. Mattia Prosperi, Ph.D.: "Technical and societal hurdles to use predictive analytics modelling for implementation research"
    2. Paula Shireman, M.D., M.S., M.B.A.: "Use of predictive analytics to study the effect of social risk factors and frailty on health outcomes after surgery (including cardiovascular surgery) to identify patients that would benefit from care pathways that help poor and/or frail people."
    3. Lorens Helmchen, Ph.D.: "Use of economic analyses in predictive analytics"
    4. Rhonda Szczesniak, Ph.D.: "Mapping environmental contributions to rapid lung disease progression in Cystic Fibrosis"

At 2:20, Drs. Brown and Papanicolaou give eight-minute summary of highlights.

2:30 p.m.
Charting the Research Agenda: What Role for NIH, Other Research Funding Agencies, Private Sector, Public-Private Partnerships?

Discussion of suggestions and questions collected over the course of the previous session, including obstacles and incentives for integration and use of existing data.

  • Moderator: Cheryl Anne Boyce, Ph.D.
  • Summary: Rebecca Campo, Ph.D., Tom Croxton, Ph.D., M.D.


  1. Which use cases of predictive analytical modelling are most relevant for implementation research strategies? Could they be considered proof-of-concept projects or do they need to be tested for implementation research?
  2. Based on your case study examples, how should studies be configured, results reported and interpreted to facilitate broad application of the findings?
  3. What were the biggest challenges in conducting these case studies? What are the limitations of the findings?
  4. How can economics enhance predictive analytics in implementation research?

At 3:50, Drs. Campo and Croxton give eight-minute summary of highlights.

4:00 p.m.
Final Summary and Next Steps
Dr. George A. Mensah and Dr. Thomas A. Pearson

4:30 p.m.
Ms. Rebecca Roper