Assumptions. You have already had extensive coursework in your public health graduate programYou are starting to get your feet wet in your state assignmentsWhat will be of interest here is information specific to working in state public health agenciesOr, here's some stuff other people might not
1. Data collection and analysis CSTE/CDC Epidemiology Fellows Orientation, Atlanta, GA,
Sept. 1, 2009
Richard S. Hopkins, MD, MSPH
2. Assumptions You have already had extensive coursework in your public health graduate program
You are starting to get your feet wet in your state assignments
What will be of interest here is information specific to working in state public health agencies
Or, here’s some stuff other people might not have told you
3. Topics to be addressed Data sources
Every state is different
Working with other agencies
Identifying interesting questions
Choice of analytic techniques
Public health surveillance principles
+ random comments and tidbits
4. Commonly available data sources Reportable diseases
Infectious and non-infectious
Electronic laboratory reporting
Don’t forget infectious disease items
5. More commonly available sources Hospital discharge data
May be maintained by state hospital association
Access is often quite restricted
Some include data from other clinical settings (FL has records from outpatient surgical clinics, EDs, and radiation therapy centers)
Death, birth, and fetal death certificate data
Linked infant death-birth files are powerful
Youth risk behavior survey
Substance use surveys
6. Sidebar: cause of death Who assigns the cause of death?
Hint: It’s not the doctor
What are those lines on the certificate?
Immediate cause of death
Due to, or as a consequence of.….
Other conditions present but not contributing to death
ICD 10 used in death classification since 1999
ICD-9 CM, now ICD-10 CM
ICD9-CM is NOT identical to ICD9 out at the third and fourth digit. Take care!
7. Sidebar (2): cause of death Underlying cause is assigned by an internationally-agreed upon algorithm
Used to be done by people called nosologists
Now is mostly done by machine
Nosologists now make sure there is enough data on certificate to allow classification
Any mention, ACME, multiple cause
8. More on cause of death Which of these is the underlying cause?
Measles and malnutrition
Pneumonia and leukemia
Heart disease and diabetes
What does it mean to say that deaths are attributable to a particular cause? (e.g. “smoking-attributable deaths”)
Population attributable risk percent
What is a hurricane-associated death?
9. Some other data sources Public health program service data
Patients seen in public health clinics (STD, HIV/AIDS, TB, prenatal, well child, WIC, special needs children, primary care)
Usually not population-based
Who does NOT come to public health clinics?
Some immunization registries are close to population-based
Usually readily accessible
10. More data sources School entry (kindergarten) screening results
Other school screenings (vision, hearing, height, weight, scoliosis, etc)
Special education student records
Visits to school health and mental health facilities
Driver’s license files (height and weight)
Highway crash data files
Driver arrest and conviction records
Calls to suicide hot lines
11. More data sources ED visits
Part of hospital discharge data collection system in some states
Syndromic surveillance systems (RODS, ESSENCE etc)
Definition of trauma?
EMS run reports
Spinal cord injury registries
12. Data sources of the future Electronic health records
Hospitals (admission, lab, pharmacy, radiology records)
Outpatient encounter records
Electronic laboratory reporting is the leading edge
Emerging ‘electronic case detection’
Using combinations of results and observations in EHRs to identify possible cases
Candidate cases still have to be evaluated against surveillance case definition
Health Information Exchanges and RHIOs
Aggregate and exchange EHRs in a community
Potentially extremely rich surveillance data source
13. Some things to ask about data sources of interest Timeliness – how old are the data when you can get access?
Population coverage – whole state, selected populations?
Sensitivity – what fraction of cases you are interested in are in the data set?
Positive predictive value – closely related to other aspects of data quality – what fraction of ‘cases’ in the data set are actually ‘cases’?
Completeness of key fields in the data set (e.g. race, home address)
15. Working with partners Consider the different motivations and reward systems of various partners.
Academics need grants and publications
Some partners have an ax to grind
Make the ground rules clear before you start:
Who will author report, and who will be first?
Who has to approve document before it is released, and who just gets to comment?
Who is responsible for which part of the work?
16. Interesting questions If you are working in a small state, urgent public health issues will not come up all that often
You won’t make your reputation by working on one of these
The trick is to see what is interesting about an apparently routine situation
Ask people who are experts about this routine problem what the important unanswered questions are
The additional effort is often quite small
17. Choice of analytic techniques You have learned a lot of sophisticated techniques in school
Sophisticated is not always better, if you consider the purpose of the analysis
To publish a paper?
To provide guidance to local action?
To persuade administrators and policy-makers?
Sometimes the point of doing a careful multivariate analysis is to figure out which univariate or stratified analyses to include in your report.
18. Program evaluation As you look around for projects, you are likely to be asked to help with program evaluations
Epidemiologists have real skills and insights to contribute to program evaluations
Well-done program evaluation is bigger than epidemiology
Epidemiologists can help most with impact or outcome assessment.
19. Program evaluation (2) Formative evaluation:
Could a program achieve its goals if carried out as designed?
Is staffing and structure appropriate to achieve the goals?
Clear statement of intermediate steps between program activities and desired outcomes
With a model, you can identify intermediate effects to measure
Some of these effects or end-points are of a kind that epidemiologists can help with
You will often be the best person around to help with evaluation data analysis and study design
20. Buy-in to evaluation activities This is the most important reason for failure of evaluations.
Phrased as “engage the stakeholders in the design of the evaluation’
What this really means is: If people who are invested in the program don’t get a real chance to influence the shape of the evaluation, they will at best ignore it and at worst undermine and attack it.
21. Surveillance principles (1) 1. Have clear objectives for your surveillance system. Design it to meet those objectives.
2. Collect only the data needed to meet those objectives. More is wasteful, and it rapidly erodes cooperation.
3. Show those providing reports or data how the health department is using these to improve community health status.
4. Value and build on personal relationships as well as laws and rules.
5. Identify and remove barriers to rapid reporting of those events that are put under surveillance.
22. Surveillance principles (2) 6. Provide authoritative consultation to clinicians and laboratorians on clinical and public health issues. Reporting will follow.
7. Have redundant systems for surveillance to minimize damage from gaps or interruptions
8. Routinely and frequently analyze and interpret the surveillance data by person, place, and time.
9. Assure that information from all sources received by all parts of a health department is reviewed and interpreted together
10. Convey your confidence about the value of surveillance, epidemiology, disease control, and public health to those who have data or cases to be reported.
23. A recent analysis in FL
24. Graphing for persuasion