Algorithmic Medical Decision Making: Leveraging Heterogeneous Data Sources for Quantifying Risk
Targeted Dates:
- Finalize invited speakers by first week of May
- Extend general call for participation by mid-May
Call for Participation
Overview
As more health data becomes available to support health care decision making, algorithmic decision theory will be needed to quantitatively process, analyze and communicate health risks and to understand their relationships to lifestyle choice, clinical and insurance decisions and health policy. The ability to quantify risks in this context is a fundamental requirement for effective decision making. We envision development and integration of risk analysis and decision making through: Algorithms for risk assessment, including those based on combinations of structured and unstructured information; methods for reconciliation of conflicting opinions and communication of risk results; algorithms for integrating multiple knowledge sources to provide coherent and customized point-of-care decision support; algorithms for combining evidence in useful ways; and algorithms for extracting structured objects such as drugs and medical conditions from free text.
This workshop will explore how data from heterogeneous sources (e.g., social media, blogs, wireless health monitors, electronic health records, claims data) can be used to assess and quantify risk in health care decision making.
Topics
We are interested in presentations relevant to the following topics:
- Algorithms for computing optimal strategies for patient care and policy decision making
- Understanding individual and group decision-making processes in the presence of risk
- Risk communication, risk attitudes, and judgment
- Algorithms for estimating clinical risks from large and/or heterogeneous data sets
- Insight to practical challenges affecting risks to patient health, patient satisfaction and quality of life
The duration of contributed presentation is expected to be approximately 30 minutes.
Submission and Registration Process
To participate in this workshop as a presenter, please:
- Submit an abstract (200 word limit) and title for your presentation to moc.mbi.su|oztirnehc#moc.mbi.su|oztirnehc no later than August 15.
- Notification of acceptance will be provided by August 23.
- Register and submit registration fees using the web registration form no later than August 31.
To participate in this workshop as an attendee, please:
- Register and submit registration fees using the web registration form no later than August 31.
Tentative Agenda
Thursday, Oct. 28
(Morning): Risk and the multi-perspective decision making process
- Confirmed invited speakers:
- Wojtek Michalowski, U. Ottawa, Telfer School of Management
- Gretchen Chapman, Rutgers, School of Psychology and Center for Medical Decision Making
(Afternoon): Risk communication
- Confirmed invited speakers:
- Valerie Reyna, Cornell, College of Human Ecology
Friday, Oct. 29
(Morning): Risk estimation from large data sets
- Confirmed invited speakers:
- George Hripcsak, Columbia University, Biomedical Informatics
- David Madigan, Columbia University, Statistics
(Afternoon): Risk estimation from heterogeneous data sources
- Confirmed invited speakers:
- Sid Dalal, RAND
- John Brownstein, Children's Hospital Informatics Program at the Harvard-MIT Division of Health Sciences and Technology
Agenda
October 28
The first day emphasizes the importance of risk analysis for medical decision making by focusing on the human-algorithm interface.
Session 1: Risk and the multi-perspective decision making process
Questions:
- What role does risk play in medical decision making?
- What algorithms are available to aid in decision making?
- How does bounded rationality impact decision making?
- What are the technical/organizational/social barriers for translating new insights on decision making into clinical practice?
- How do we integrate the data from diverse sources (both subjective and objective) to help make informed decisions on risk for different treatment protocols.
Suggested invited speakers:
- Peter Ubel, U. Michigan, Center for Behavioral and Decision Sciences in Medicine
- Gordon Hazen, Northwestern, Industrial Engineering
- Louis Eeckhoudt, Facultés universitaires catholiques de Mons
- Wojtek Michalowski, U. Ottawa, Telfer School of Management (accepted)
- Gretchen Chapman, Prof. of Psychology, Rutgers, and head of Center for Medical Decision Making (accepted)
Session 2: Risk communication
Questions:
- How do decision makers interpret data representing risk/uncertainty?
- What methods are effective for communicating risk?
Suggested invited speakers:
- Valerie Reyna, Cornell, College of Human Ecology (accepted)
- David Spiegelhalter, U. of Cambridge, Centre for Mathematical Sciences (declined)
- Angie Fagerlin, U. Michigan, Center for Behavioral and Decision Sciences in Medicine (declined)
- Erika A. Waters, Asst. Prof. of Surgery, Washington U. (declined)
October 29
The second day focuses on the algorithmic challenges associated with working with large data sets and estimation of risk from heterogeneous data sources.
Session 3: Risk estimation from large data sets
Questions:
- What are some of the data quality issues that risk estimation methods are particularly sensitive to?
- How have quantitative researchers modified their methodologies or otherwise adapted to data quality issues?
- What are the challenges in large scale data collection, annotation, management and analysis that will influence accurate and rapid risk assessment.
Suggested invited speakers:
- George Hripcsak, Columbia University, Biomedical Informatics (accepted)
- Jennifer Nelson at Group Health (Vaccine Safety Datalink project - data quality)
- Wen Ye, Biostatistics, University of Michigan
- Sharon-Lise Normand, Harvard Medical School
- David Madigan (accepted)
Relevant Links/Documents:
- Large Datasets in Biomedicine - A Discussion of Salient
- Methodologies for Data Quality Assessment and Improvement
- 2009 Workshop on Large-scale Data Mining
Session 4: Risk estimation from heterogeneous data sources
- Examples of data sources: social media, surveys, claims, EHR, …
Questions:
- How can evidence regarding medical safety/health risks that are derived from social media be used to enhance knowledge derived from more traditional databases to provide decision support at the point of care and in policy creation?
- How effectively can we derive or elicit patient values and risk preferences (i.e., utility functions) through social media?
- What algorithms are needed to derive quantifiable risk metrics from unstructured social media text?
- What algorithms are most effective for reconciling risk assessments from multiple sources?
Suggested invited speakers:
- Sid Dalal, Senior Advisor to the RAND President for Technology (accepted)
- John Brownstein, Childrens Hospital (invitation extended)
Relevant Links/Documents:
