Using the Birth Certificate to Implement the Pregnancy Risk Assessment Monitoring System PRAMS

1
Using the Birth Certificate to Implement the
Pregnancy Risk Assessment Monitoring System
(PRAMS)

• Leslie Lipscomb, MPH
• Chris Johnson, MS
• National Association for Public Health Statistics
  and
• Information Systems and the Vital Statistics
  Cooperative Program Project Directors Joint
  Meeting
• June 6-8, 2005

2
What is PRAMS?

• Population-based surveillance system of woman and
  infants
• State-specific data collection within a
  standardized system
• Information on maternal attitudes, behaviors and
  experiences
• Action oriented

3
States Participating in PRAMS, 2005
WA
ME
VT
MN
OR
NY
MI
RI
NYC
NE
NJ
OH
IL
UT
WV
MD
CO
NC
OK
NM
AR
SC
AL
MS
GA
LA
TX
AK
FL
HI
4
PRAMS Objectives

• To promote collection of population-based data of
  high scientific quality
• To conduct comprehensive analyses
• To translate results into useable information
• To build state capacity for collecting,
  analyzing, and translating data

5
How the Birth Certificate is Used to Implement
PRAMS

• Identifying the PRAMS Sample
• Data Collection
• Data Weighting
• Data Linkage

6
Sampling
7
PRAMS Population of Interest

• Mothers who are residents of state, who delivered
  a live-born infant within state during the
  calendar year.

8
PRAMS Sampling Frame

• List representing the population eligible for
  inclusion in the sample.
• Operational sampling frame is list of all infants
  born alive within state to resident mothers
  during calendar year.
• Vital records birth certificate file serves as
  the source of the sampling frame.

9
Exclusions to the Sampling Frame

• Stillbirths, fetal deaths, induced abortions
• Out-of-state occurrences
• In-state births to nonresidents of the state
• Records missing mothers name
• Records processed too late (gt 6 months from birth)

10
Exclusions to the Sampling Frame (continued)

• Records processed too early (lt 2 months from
  birth -- but these records are included in later
  batches)
• Multiple Gestation Infants
• For twin and triplet sets, only one infant is
  selected to be included in the sampling frame.
• For multiple gestations of order 4 or more, all
  infants are excluded from the sampling frame.

11
Exclusions to the Sampling Frame (continued)

• Adopted Infants
• If identified at the time the sample is drawn,
  adopted infants are excluded.
• If not identified at the time the sample is
  drawn
• If birth mother is listed on certificate, she can
  be contacted.
• If adoptive mother is listed on certificate, she
  should be dropped from further follow-up.

12
Inclusions to the Sampling Frame

• Infants who have died
• Records missing address information or other key
  birth certificate information (other than
  mothers name)

13
Stratified Sampling

• Allows precise estimates for subgroups
• comparisons of greatest interest
• Alternative to proportional sample
• groups not oversampled are represented
  proportionally
• Stratification Variable Choices

Age Area Medicaid status
Birthweight Race and ethnicity Education
14
Frame Construction
15
Sample Selection
16
Data Collection
17
Data Collection Sequence of Events

• Monthly sample drawn from birth certificates (2-6
  months after delivery)
• BCENTRY file created contains personal
  identifiers to assist with contacting mothers
• Data collection period (up to 90 days)
• Mailings
• Search for telephone numbers
• Telephone calls
• Data cleaning and quality control
• Data transmitted to CDC

18
PRAMS Weighting
19
Rationale for weighting
20
Coverage weights

• Frame is constructed during the year.
• Some births dont make it into the frame.
  Examples
• Home births
• Remote/rural hospitals
• We account for missing births using the
  end-of-year official birth file and creating a
  coverage weight.

21
Coverage weight Example
22
Sampling Weights

• To achieve better estimates within small groups,
  we oversample those groups. Examples are
• Minority race groups
• Low birth-weights
• Rural areas or counties
• Since we sample at different rates by group, we
  must weight each observation to represent the
  original sampling frame.

23
Sampling weights Example
24
Response weights

• Survey data are analyzed using a CART analysis to
  determine which variables predict response.
  Examples
• Education
• Marital status
• Prenatal care
• Observations are assigned to response groups and
  weighted by their response rates.

25
Response weights Example
Response Rate 80
Sample
Respondents
26
Exceptions

• Twins on frame
• Duplicate records
• Plurality errors

27
Data Linkage
28
PRAMS Weighting Linkage

• Must match twins and triplets to calculate proper
  sampling weight (internal linking)
• Algorithm
• Loose linkage (plural births)
• DOB MDOB Hospital County of Birth
• 01212000-03281968-792-125
• Strict linkage (singletons)
• LOOSE Race Education County of Residence
• 01212000-03281968-792-125 3-5-124

29
PRAMS Weighting Linkage

• Matches are counted and ordered
• What goes wrong
• DOB is different midnight babies
• Data entry errors, etc.
• Find near matches, make decision, code by hand

30
The Value of Data Linkage

• Reduces respondent burden
• Improves accuracy (better detection
  measurement)
• Reduces follow-up costs
• The last PRAMS RFA invited data linkage
  activities as examples of enhanced projects
  (beyond the basics)

31
Examples

• Washington States First Steps
• PRAMS, BC, Medicaid records
• DRH Massachusetts DOH linkage of birth
  certificate records with records from Assisted
  Reproductive Technologies registry
• Utah
• Colorado

32
Validation of Self-Report of Medicaid Utilization
and Differences Between Hispanic and Non-Hispanic
Women

• Utah Department of Health
• Laurie Baksh, MPH
• Shaheen Hossain, PhD
• Lois Bloebaum, BSN
• Sharon Clark, MPH
• Brenda Ralls, PhD
• Gulzar Shah, PhD

33
Validating Data

• To assess agreements between self-reported
  Medicaid coverage and actual Medicaid coverage,
  the Utah Department of Health linked the 2000
  PRAMS data set with a linked data set of birth
  certificates and Medicaid eligibility.

34
Methodology

• Phase I
• Linking vital records birth data with Medicaid
  eligibility data.
• Phase II
• Linking the 2000 PRAMS data set with the existing
  linked birth - eligibility file.

35
Methodology - Linking VR and Medicaid data.

• The initial matching process was completed using
  Automatch software.
• Both the birth file and Medicaid eligibility file
  were converted to an ASCII text format
• Variables were re-coded to be consistent across
  data sets.
• Probabilistic matching was conducted.

36
Linking Medicaid and Birth Records in Colorado

• Alyson Shupe, Ph.D.
• Section Chief, Health Statistics
• Colorado Department of Public Health
• and Environment
• PRAMS National Meeting w December 2002

37
Previously

• No access to Medicaid records
• No indication of payer source on BC
• Limited sense of SES of women giving birth in CO
• No idea whether programs aimed at reducing poor
  birth outcomes and costs work

38
Now

• Unrestricted access to Medicaid database
• Linked birth and Medicaid records from state FY
  1998-2000
• Ability to analyze birth files by Medicaid status
• Cost/benefit analysis of Prenatal Plus program

39
How

• Obtained access to and training on using STARS
  database
• Matched on mothers, fathers and infants names,
  and mothers DOB SSN
• Claims data were matched using mothers Medicaid
  ID
• STARS searched for infant DOB, then period of
  service eligibility
• Infants matched on DOB, first and last name, and
  mothers names

40
Challenges

• Completeness of data set
• Other DRGS
• Claims not yet filed
• HMO clients
• Out of state births
• Reliability of claims data
• Inter-rater reliability on matching
• Latino population

41
Future

• More quality control
• Link Medicaid/birth data set with PRAMS data
• Do PRAMS respondents report of Medicaid status
  match Medicaid claims data?
• Add variable Medicaid at time of delivery for
  analysis of birth record data
• Further exploration of claims data
• Revised birth certificate
• HIPAA

42
Conclusions

• BC records are vital to PRAMS
• Completeness is reason for population-based
  results
• Linking data from various systems holds great
  promise