Improving Data Timeliness
Through Provisional Vital
Statistics
Today’s Presentations
• Provisional Vital Statistics: Performance and Potential
Mark Flotow, Acting Division Chief, Illinois Center for Health Statistics
Illinois Department of Public Health
• Michigan’s Experience in Producing Provisional Infant Mortality Rates
Glenn Copeland, State Registrar & Director, Div. of Vital Records & Health Statistics
Michigan Department of Community Health
• Questions/Discussion
Provisional Vital Statistics:
Performance and Potential
Mark Flotow
Illinois Center for Health Statistics
5 August 2014
On behalf of NAPHSIS
Emerging Electronic Systems
- Goodbye to paper-based systems
- Hello to near real-time, Web-based vital
certificate intake systems
→
→
→
→
a decade of building among the states
national coordination of efforts
legal angle helps the public health angle
births (95+%), death (80%), fetal death
(45+%), marriage and divorce (coming)
Output Harvest (so far . . .)
- Timeliness: originators do data entry; record-level
edits; simultaneous record access; process
monitoring and “preliminary records”; electronic
querying
- Improvements (both quality and flexibility): GISready; meta-data; variable sorting; aggregate data
error checks; near real-time surveillance; custom
products; linkages; populating other databases;
Provisional vital Statistics (PvS)
Integration of
Provisional vital Statistics (PvS)
- vital statisticians have used PvS for decades to
monitor data quality (but often well after the
fact)
- new generation of vital records intake/output
systems allows for “surveillance-level” PvS for
data quality
- an increasing number of PvS products are being
shared with PH agency programs
- building experience: “noise” vs. “notion”
(More) Integration of
Provisional vital Statistics (PvS)
- on Web sites and via other public distribution
- PvS timeliness versus accuracy: when does it
matter (most)?
- managing expectations: what any user should be
aware of when using PvS . . . the caveats
- custom products for targeted audiences
- “pure surveillance” versus “value added” PvS
Concept Topic:
PvS as Surveillance vs. Estimate
Pros of “pure surveillance” . . .
- quick, easy, “as is, right now” system extracts
- very timely
- useful for experienced users or programs
Cons of “pure surveillance” . . .
- very much caveat-dependent; “user beware”
- can be misleading when irregularities occur
- is it useful for the average user, beyond satisfying
curiosity, or as a “clarion call” data are out?
Concept Topic (continued):
PvS as Surveillance vs. Estimate
Pros of “estimate or projection” . . .
- incorporates experience or “value added”
- ease of comparison for user
- can be periodically revised
Cons of “estimate or projection” . . .
- can be complicated to systematically produce
- becomes more problematic for small populations
- projection as forecast; varying elements of doubt
Concept Topic (continued):
PvS as Surveillance vs. Estimate
Discussion Points:
-Surveillance vs. Projection is not a dichotomy but
includes a spectrum of in-between possibilities
- tailoring products to specific users/audience
segmentation/which measures to which
programs
- how often to update? “Latest and greatest” only?
- data quality – sharing the experience
PvS 3-5 Years From Now
- routine use of PvS as an integral part of VR/VS
production (i.e., a best practice)
- enhanced PH surveillance (e.g., maternal mortality)
- matched PvS products with PH purpose/use
- PvS will cover mainly the latest year or two, as the
production of final vital statistics also has
timeliness improvements (more of
More/Better/Faster)
- “demise & rebirth” of PvS → less surveillance and
all jurisdictions participating (for both birth and
deaths) more estimates/projections on the public side
of dissemination
PvS 3-5 Years From Now (cont.)
- PvS will cover more variables, not just the most
popular ones (mainly due to broadening
quality control)
- state PvS will (continue to) enhance NVSS PvS
- possibilities not yet realized; potentials not yet
tapped . . .
- (PvS and) VS will continue to be “real data” as 100%
sample and as “ground truthing” (vs. “big
data”)
Provisional Infant Death Data
Recent Experiences and Accomplishments
in Michigan
Glenn Copeland, State Registrar and Director
Division for Vital Records and Health Statistics
Michigan Department of Community Health
Michigan’s Status Quo
• Files finalized in June
• Statistics developed and released in August
• Timeliness took several hits
• Vital records forms revisions
• Debugging new birth and death systems
• Automated systems initially trapped data
• Files and data release pushed into fall
• 9 months from year end
• Birth cohort infant death files delayed as well
• trailed out to 20 months from birth year.
Impetus for Change
• Governor’s score cards
• Using national estimates based on projections
• Strong interest in more timely data
• Key health priority
• Infant mortality reduction/disparity reduction
• CoIIN
•
•
•
•
HRSA funds to support comprehensive efforts
Builds on multiple public and private investments
Expects timely data on progress
MCH program sees as key opportunity
• Addresses program goals and compliments Michigan approach
What does timely data look like?
• Minimal needs
• Rapid estimates of infant mortality
• numbers and rates
• Data by county/local health department
• Need information by race/ethnicity
• Additional information
• Information on birth characteristics
• Birth weight, gestational age, prenatal care, maternal age
• Requires linked birth and death files
• Cause of death information
Existing Barriers
• Delays associated with paper filings
• Inability to extract death data from new EDR
• Delays birth/death matching
• Slows location of infant deaths
• Lack of timely interstate exchange
• STEVE implementation is pending
• Problems with cause of death
• Assignment errors and delays through NCHS
How to get from Here to There
• Birth system is very reliable and timely
• Files nearly complete within 60 days
• Data very clean
• Lacking only out of state resident deliveries
• SUID Surveillance
• Already manually locating all infant deaths
• EDR continues to accelerate data
• Reducing paper filings
• Trained nosologist on staff
Approach
• Leverage birth data
• Estimate out of state resident delivery numbers
• Use prior year(s) data
• Get denominator data in 70 days by county
• Use brute force on infant deaths
• Route infant deaths for manual coding/keying
• Capture subset of death data to meet specific needs
• Add manually coded cause of death
• Estimate out of state resident deaths
• Use prior year(s) data
Status
• State/County infant mortality rates for 2013
• Released in April 2014
• Preparing for quarterly release of fresh data
• Data would be out at 90 day intervals
• Plan to provide 12 month cumulative figures
• Plan to provide more detail at 6 months
• State and county level data
• Birth characteristics for the death cohort
• Death characteristics data
• Age at death numbers and rates
• Rates by birth weight/prematurity
• Cause of death – focus on key conditions
Issues
• Some effort involved
• Pay back is prompt feedback to program staff
• Need to solve data exports from EDR
• Extract preprocessed data
• Anticipate resolution to NCHS cause coding
• Need to leverage automation
• Matching of births to deaths
• STEVE implementation
• Seamless feeds of STEVE data into EBC/EDR
Summary
• You don’t need a final file
• If you can just focus
• Isolate what pieces of information are critical
• Don’t need to wait on the last variable in the last record
• You can provide decision support much sooner
• Leverage what you have in place as best you can
• Introduces error and a grain of salt
• Much more valuable to be close than to be precise
• Potential for greater efficiencies are there
• Improvements in the pipe to make this very practical