Michigan Newborn Screening

1
Michigan Newborn Screening Live Births Records
Linkage and Follow-Up of Potentially Un-Screened
Infants

• Steven J. Korzeniewski, MA, MSc,
• Maternal Child Health Epidemiology Section
  Manager
• Violanda Grigorescu, MD, MSPH, Glenn E. Copeland,
  BS, William I. Young, Ph.D.,
• Michigan Department of Community Health

2
Outline

• Background/Intent of linkage
• Methods (software, data management, algorithm)
• Results
• Discussion
• Public Health Implications

3
Background

• Linkages re-initiated to identify live births
  potentially unscreened.
• Initial efforts were not sustainable
• Intended to be mutually beneficial to newborn
  screening follow-up program and vital records.
• Means to assess data quality

4
Methods

• Data received from November 2007 through March
  2008 were used for this study
• 2008 transitioned from DOS based to Web based
  electronic birth certificate system
• Newborn screening card number included on birth
  record
• SAS v9.1 (Cary, N.C.) used to create text files
• Record linkage and follow-up conducted by NBS
  Follow-up Program
• Newborn screening Michigan Care Improvement
  Registry (MCIR) data used for follow-up

5
Methods

• Linkage via Link Plus
• A probabilistic record linkage program
• Developed for cancer registries at the Centers
  for Disease Control and Preventions (CDC)
  Division of Cancer Prevention and Control in
  support of CDC's National Program of Cancer
  Registries (NCPR).
• Can be used with any type of data in fixed width
  or delimited format
• Free

6
Methods

• Linkage score (probabilistic linkage)
• based on the theoretical frame work developed by
  Fellegi and Sunter (1969)
• sum of the logarithm of odds across all matching
  variables, based on the probability that a
  matching variable agrees given that a comparison
  pair is a match and the probability that a
  matching variable agrees given that a comparison
  pair is not a match

7
Methods

• Blocking variables - used to block (or
  partition) the two files
• Matching variables are compared between records
  matching on the blocking variable.

8
Methods
9
Methods

• Follow-up
• Unmatched records sent to follow-up staff
• Staff search NBS data
• Contact birth hospital
• Send certified letter to parent requesting screen
  or signed refusal letter

10
Results
11
Results
12
Discussion

• Probabilistic linkage is subjective. useful
• Linkage success is a function of
• Cutoff selection
• Data quality
• Strategy
• An ability to deal with discordance
• Match rates change over time and may require
  alternations in linkage algorithms
• Manual checking of initial linkage results and
  follow-up results must be used to determine
  algorithm changes and avoid false matches.
• Live Births to NBS data matching is a best case
  scenario given data are collected at virtually
  the same time, same place, and often by the same
  person.

13
Discussion

• Significant investment of time for
• Initial data management programming
• Understanding how to select algorithm
• Determination of borderline matches
• Assessment of follow-up results
• Benefits beyond identification of potentially
  unscreened children include
• Data quality check
• Link to various datasets through vital records
  (i.e.- birth defects, EHDI, etc.)

14
Public Health Implications

• Link Plus is applicable to MCH databases
• Linkage facilitates data usage with minimal cost
  investments
• Linkage is successful at detecting unscreened
  infants (we have identified several)
• However, linkage should be handled with caution

15
Acknowledgements

• Co-investigators Violanda Grigorescu, MD, MSPH,
  Glenn E. Copeland, BS, William I. Young, Ph.D.,
• NBS Follow-up Staff

16
Contact

• KorzeniewskiS_at_Michigan.gov
• Thank You