1. Current Environment/Challenges Many statistical signals (few with meaning) – easy to get lost in data and signals Many statistical signals (few with meaning) – easy to get lost in data and signals Many data sources, many data fields Many data sources, many data fields Many fields untapped by current statistics Many fields untapped by current statistics Data sources with preliminary findings often are missing final outcome status Data sources with preliminary findings often are missing final outcome status Lack of data linkage across data sources Lack of data linkage across data sources Public health resources are stretched thin with many competing needs – can’t sort through it all Public health resources are stretched thin with many competing needs – can’t sort through it all Still - part of the “real story” of disease burden is buried in these data Still - part of the “real story” of disease burden is buried in these data

2. The Question: What approaches can assist in identifying the actionable information that exists within all of these data/signals?

3. Previous Approach Look across “independent” data sources/statistics to corroborate signals Look across “independent” data sources/statistics to corroborate signals Problems with this approach: Problems with this approach: Each data source may represent different underlying populations Each data source may represent different underlying populations Difficult to align signals if data sources on different time tables Difficult to align signals if data sources on different time tables Signals may not agree and that may not be good reason to ignore the alert Signals may not agree and that may not be good reason to ignore the alert

4. Proposed Approach: Decision Support System

5. Decision Support System Incorporated into existing information systems Incorporated into existing information systems Organizes data to inform clinical or public health responses Organizes data to inform clinical or public health responses Enables providers to target clinical management of medical conditions Enables providers to target clinical management of medical conditions Reminders for vaccines (flu vaccine prompts if > 65 yo) Reminders for vaccines (flu vaccine prompts if > 65 yo) Screening recommendations (mammograms) Screening recommendations (mammograms) Chronic disease management (asthma, diabetes) Chronic disease management (asthma, diabetes) Associated with improved outcomes Associated with improved outcomes Immunizations: Up-to-date rates Immunizations: Up-to-date rates Decreased emergency room visits for asthma Decreased emergency room visits for asthma

6. How Might We Use A Decision Support System for Surveillance? It is not uncommon to have many visits that fall into a general GI category It is not uncommon to have many visits that fall into a general GI category Able to parse category into more useful subcategories: Able to parse category into more useful subcategories: GI (common)  “bloody diarrhea” (moderately common)  young people with “bloody diarrhea” (rare) With a little extra effort, we now may propose Shigella or E. coli as a possible differential diagnosis – thereby framing decisions for the public health practitioner With a little extra effort, we now may propose Shigella or E. coli as a possible differential diagnosis – thereby framing decisions for the public health practitioner Different public health actions are warranted for a possible Shigella or E. coli problem vs. a general GI flag Different public health actions are warranted for a possible Shigella or E. coli problem vs. a general GI flag

7. Decision Support Needs Prompts for action Prompts for action Event detection (ALF example to follow) Event detection (ALF example to follow) Event management Event management Situation awareness, hurricane management, influenza onset, disaster management, large public events (inauguration), etc. Situation awareness, hurricane management, influenza onset, disaster management, large public events (inauguration), etc. Decision-making may be affected by level of responsible domain Decision-making may be affected by level of responsible domain Federal Federal State State Local Local How could this work as integrated electronic solution? How could this work as integrated electronic solution?

15. Do De Do... Skip a Few

18. Our Usual Way to Focus on Statistical Flags

20. Rudimentary Example: Alert List Filter (ALF) Automated way to “digest” many red and yellow statistical flags Automated way to “digest” many red and yellow statistical flags Scores each alert based on several criteria (number of visits, “density”, and “recency”) Scores each alert based on several criteria (number of visits, “density”, and “recency”) Ranks top 20 flags with highest scores and presents them to the user in a table Ranks top 20 flags with highest scores and presents them to the user in a table Allows user to visualize the last 7-8 days of syndromic flags with some inferred importance Allows user to visualize the last 7-8 days of syndromic flags with some inferred importance

21. Alert List Filter (ALF) Example (Cont’d) Instead of focusing solely on the red flag with the greatest number of visits, ALF directed the analyst’s attention to an otherwise buried potential cluster of respiratory illness in kids Instead of focusing solely on the red flag with the greatest number of visits, ALF directed the analyst’s attention to an otherwise buried potential cluster of respiratory illness in kids The analyst was then able to visualize this “lost” potential cluster by querying the system using elements proposed by the filter The analyst was then able to visualize this “lost” potential cluster by querying the system using elements proposed by the filter

23. Decision Support Challenges Each level of complexity is a potential barrier to the practitioner’s ability to find the actionable event Each level of complexity is a potential barrier to the practitioner’s ability to find the actionable event Robust logic must be discovered that will propose useful response paths thereby limiting complexity for the surveillance system user Robust logic must be discovered that will propose useful response paths thereby limiting complexity for the surveillance system user

24. Decision Support Challenges (Cont’d) In order to accomplish this, work needs to be done to discover: In order to accomplish this, work needs to be done to discover: the likely events in different data sources the likely events in different data sources how information can be harvested from every available element of the data how information can be harvested from every available element of the data how meaningful differences (not just statistically significant) are detected how meaningful differences (not just statistically significant) are detected the suggested response protocols the suggested response protocols This information needs to inform systems so they can assist practitioners in quickly making sense of masses of data This information needs to inform systems so they can assist practitioners in quickly making sense of masses of data