Title: Potential Infant Mortality Analysis Tools: Linked Data Sets
1Potential Infant Mortality Analysis Tools
Linked Data Sets
- Milton Kotelchuck, PhD MPH
- Boston University School of Public Health
- State Infant Mortality Collaborative Meeting
- September 20, 2004
2(No Transcript)
3(No Transcript)
4Goals of Presentation
- Describe PELL (a longitudinal linked data set)
- 2. Suggest possible new opportunities for
analysis of infant mortality and other
reproductive outcome trends using linked data sets
5The Massachusetts Pregnancy to Early Life
Longitudinal (PELL) Project
Maternal and Child Health Department
Boston University School of Public
Health Massachusetts Department of Public Health
Centers for Disease Control Prevention
6PELL TeamA Public-Private Partnership
BUSPH Mary Barger, CNM, MPH Gene Declercq,
PhD Stephen Evans, MPH Milton Kotelchuck, PhD,
MPH Judy Weiss, PhD Mass DPH Wanda Barfield,
MD, MPH Angela Nannini, NP, PhD CDC Kay Tomashek,
MD, MPH
7PELL Overview
- Public/Private partnership
- Funded to assess impact of prenatal environment
on subsequent child (and maternal) health - Utilizes a broad range of public health data
- Longitudinal data system (developmentally based)
- Allows for multiple kinds of linkages and
analysis, including dyadic, maternal, child,
multiple sibling and family - Breadth of data system is expandable
- Offers a conceptual and practical model for other
states
8Core Data Sets
Birth Certificate
Hospital Discharge (HD) Mothers Birth
Hospital Discharge (HD) Childs Birth
Fetal Death
(Core Linkage)
- Core linkage of Birth Certificate and Hospital
Discharge Data - Dyadic Linkage Mother and Child
- Includes Fetal Deaths
- Started in 1998 - 400,000 Infant Records
320,000 Maternal Records, and 80,000
Maternally-linked Records to Date
9PELL Data Sets
- Vital and Health Status Data Sets
- Clinical Data Sets
- Health Services Usage Data Sets
- Programmatic Data Sets
- Contextual Data Sets
10Health Status Health Services/Programmatic
Data Linkages
WIC
Birth Defects Registry
Child and Maternal Death Files
Birth Certificate
(HD) Birth Mothers
Early Intervention
Linked Birth-Infant Death
MassPro Quality Data
Fetal Death
(HD) Birth Child
Healthy Start
HD/Observational Stays/Emergency Room Data
(Core)
Child Care Coordination
Future Data Sets (NICU, ART, ...)
Future Contextual Data Sets (Area Resource File,
Census, )
11Maternal or Child Linked Longitudinal Capacity
Subsequent Records Childs Hospital Discharges,
Observational Stay, ER Usage Mothers Hospital
Discharges, Observational Stays, ER Usage 1999
Birth Certificate
Hospital Discharge (HD) Mothers Birth
Prior Information Mothers Pregnancy Hospital
Discharges, Observational Stays, Emergency Room
Usage 1998, 1999
Hospital Discharge (HD) Childs Birth
Fetal Death
Subsequent Records Childs Hospital Discharges,
Observational Stay, ER Usage Mothers Hospital
Discharges, Observational Stays, ER Usage 2000
(Core, 1999)
Program Participation Data, PNC Usage (1998, 1999)
Program Participation Data 1999, 2000,
12Maternally Linked Data Base
Birth Certificate
Hospital Discharge (HD) Mothers Birth 1
Birth Certificate
Hospital Discharge (HD) Mothers Birth 2
Hospital Discharge (HD) Childs Birth
Fetal Death
Hospital Discharge (HD) Childs Birth
Fetal Death
(Core, 1999)
(Core, 2000)
13Delivery
One Year
Five Years
Conception
Intrauterine (Fetal) Pregnancy (Mother)
Infancy/Mother
Postpartum
Early Childhood/ Interconceptual (Mother)
Birth Certificate/ Fetal Death
Hospital Discharge
Observational Stay/ER
Healthy Start
WIC
Early Intervention
MassHealth
Infant/Child Death
Maternal/Mother Death
Future Data Sets
PELL Longitudinal Data
14Issues to be Addressed in Implementing a Linked
Data System
- Conceptualization of Linked Data Systems
- Technical Linkage Issues
- Data Access/Confidentiality Issues
- Data Utilization/Practice Issues
- Statistical Issues
- Finance/Management/Training Issues
- Analytic Uses and Results
15Visions/Conceptualizations of the Linked Data
Systems
- No Linked Data Sets
- Contextually Augmented or Geographically
Co-occurring Data Sets - Specific Paired Linked Data Sets
- Medical Record Model
- Data Warehouse Model
- Virtual Linked Data Systems
16Technical Issues PELL Linkage Methodology
- Linkage without common unique identifiers
- Combination of Deterministic and Probabilistic
matching - LinkPro 2.0 (inexpensive, easy to use, effective)
- Set of linkage programs, with a linker file
(similar to a relational
database) - Unique identifier randomly generated to
facilitate linkage and assure confidentiality - Dyadic linkage flexibility (via linker file)
17Core Linkage Variables
- Hospital Number Medical Record Number
- Babys Date of Birth / Mothers Data of Birth
- Sex
- Zipcode
- Date of Delivery (from hospital discharge file)
- Child/Mothers Name (in admin. databases)
18Data Base Structure
Hosp Disch Mother data
MOMUID HDMOMUID HD MOM - Matched HD MOM -
Weighted All other vars
Hosp Disch Child data
KIDUID HDKIDUID HDKID - Matched HDKID -
Weighted All other vars
Linker File
ALLMOMUID MOMUID KIDUID
Healthy Start Mother data
MOMUID HSUID HS - Matched HS - Weighted All other
vars
EI data
KIDUID EIUID EI - Matched EI - Weighted All other
vars
BC/FD data
ALLMOMUID MOMUID KIDUID All other vars
19Technical Issues Summary
- Technical issues are no longer the principal
barrier for linked data systems - Growing project experience and newer computer
programs facilitates linking data - However, life would be much easier with a common
and universal identifier
20Confidentiality/Access to Database Issues
- Major issue in constructing linked databases
- Linked databases involve/require confidential
data - Access involves both political and professional
concerns - More complex in Public/Private partnership
- HIPAA further complicated access to health
databases
21Key PELL Confidentiality Decisions
- All data linkages using confidential data done
and held at Mass. Dept. of Public Health (MDPH) - There is no permanent linked database
- It is a virtual database, using a linker
program with generated ID numbers, and a series
of computer programs that can extract linked data
for subsequent analyses - Only de-identified or now limited use data
sets used for data analyses at MDPH or offsite
22Special procedures for working in partnership
with MDPH
- Massachusetts DPH has extensive confidentiality
concerns and requirements - MDPH RADAR (Research and Data Access Review)
Committee provides an institutional mechanism to
review research/confidentiality requests using
state public health data bases - PELL and RADAR committee developed and codified a
very extensive set of procedures to facilitate
PELL implementation and analysis
23Advancing the Science of Using Linked Data Systems
- Advancing the Science by Using Linked Data Systems
24Uses of Linked Data System
- Policy Evaluation - Healthy Start
- Program Evaluation - Early Intervention WIC
- Improved Surveillance - Multiple birth sibling
groups, maternal deaths - Health Services Research - Costs of multiple
pregnancies impact of observational stays - Clinical Research - No Indicated Risk Cesareans
- Risk Factor Analysis - ART etiology of multiple
births, twin discordancy impact - Longitudinal Analyses - Perinatal events and
subsequent early intervention program
participation - Monitoring Birth Trends
25Possible Infant Mortality Trend Analysis
Enhancements
- Monitoring Birth Trends using the PELL Database
26Enhanced and longitudinal assessment of
- a. Maternal Risk Characteristics
- b. Health Related Behaviors
- c. Reporting Issues
- d. Health Care Access
- e. ART/Multiple Births
- f. OB/Neonatal Technology and Practice
27Improve Quality of Epidemiologic Data(Reporting
Issues)
- Validation/accuracy of
- maternal linkages
- method of delivery
- multiple status
-
28Linking Mothers Across Time Comparing Two
Methods (MA 1998-2002)
- BC/FD Matches 81,594
- HospDD Matches 78,646
- Both Matches 74,463
- BC/FD Unique Matches 6,905
- HospDD Unique Matches 3,916
Lazar et al, 2003
29Linking Mothers Across Time Comparing Two
Methods
- 6,905 Unique BC/FD Records
- 65.8 Inconsistency of SSN
- 24.9 Missing SSN
- 9.3 Unexplained
- 3,916 Unique HospDD Records
- 56.5 Maiden Name Different
- 34.4 First Name Different
- 16.0 MDOB Different
- Unique HospDD Records more likely to be
foreign-born, lacking PNC, fetal deaths,
Hispanic, other race, publically insurance and
less than HS education -
30Method of Delivery Accuracy VBAC Status(MA
1998-2001Maternally Linked Birth Files Only)
- Maternally Linked Gold Standard 7.0
- Birth Cert/Fetal Death Cert Only 5.0
- Hosp Discharge Data Only 6.2
- BC/FD or HospDD 6.8
- Misclassified Disproportionately LBW and Premature
31- Creation of Multiple Sibling Groups
- Wombmate Linkages
32Improved Coding of Multiple Births and Creation
of Sibling Groupings1(Mass, 1998-2000)
Plurality from MA Birth or Fetal Death
Certificates
New Plurality1
1 Lazar et al 2004
33Assess the Total Spectrum of Infant/fetal
Mortality
-
- Improve Analysis of Fetal Mortality
34Status of twin pairs at delivery and one year
later(Mass, 1998-2000)
Pair status one year after delivery
Total Fetal Deaths 98 Fetal Death
Rate 15.1/1000 Total Infant Deaths 163
IMR 25.5/1000 IMRFD 360/1000 IMRLB
22.9/100
35Enhance Measurement of Maternal Morbidity
-
- a. Prior to Delivery (antenatal
hospitalizations) - b. Delivery Experiences
- (No Indicated Risk C-Sections)
- c. Subsequent to the Delivery
36Maternal Antenatal Hospitalizations and
Observational Stays by Plurality (Mass,
1998-2001)
37Timing and Duration of Mothers Delivery
Hospitalization by Plurality (Mass, 1998-2001)
38- Allows for measurement of Near-Miss Maternal
Deaths - Provides new possibilities for analysis of
Near-Miss Infant Deaths - - moves beyond VLBW/VPT analyses only
- - offers links to NICU diagnoses
- -provides possible assessment for
preventable infant deaths
39Longitudinal Analysis of Infant Survival
Characteristics
-
- a. subsequent use of health services
- b. re-hospitalizations
- c. costs
- d. involvement with public health programs,
especially early intervention programs -
-
40Infants Mean Days of Infant Hospitalization by
Plurality 1,2(Mass, 1998-2001)
1 Hospitalizations include overnight
hospitalizations and observational stay days 2All
hospitalizations through one year of age
41Infant Post-Natal Observational Stays and
Hospital Days by Plurality (Mass, 1998-2001)
42Early Intervention Program Participation by
Plurality (Mass, 1998-2001)
43Characteristics associated with Early
Intervention referral for infants 1998-1999)
- Variable 0-12 mo 12-36 mo Not referred
- OR (95 CI) OR (95 CI) OR (95 CI)
- Gestational age (0.5-2.6) 0.6 (0.4-1.1)
- EI established condition 1.4 (0.7-2.9) 0.5
(0.2-1.3) 1.0 (0.4-2.4) - Male 1.3 (0.9-1.8) 0.8 (0.5-1.3) 0.8 (0.6-1.2)
- Mother (0.4-1.7)
- Mother black 0.5 (0.3-0.7) 2.3 (1.3-4.4) 1.7
(1.1-2.7) - Mother Hispanic 0.7 (0.5-1.2) 1.9 (1.0-3.6) 0.9
(0.6-1.6) - No insurance 0.3 (0.1-0.7) 3.4 (1.3-8.9)
2.2(0.9-5.3) - Of 893 infants (0.8 of births) weighing 85 were referred to EI (78 within 12 mo of
birth, 7 12-23 mo from birth).
Barfield et al, 2003
44Enhancement of ART/Multiple Birth Status Analysis
Possibilities
- Original CDC MA Birth Certificate ART Linkage
(and new Pell Opportunities) - MA Birth Certificate PELL Analyses
- Improved Understanding of the Contributions of
Multiple Birth Status to Infant Mortality
45Plurality by ART Status (MA Birth Certificate)
(Mass, 1998-2000)
ART is associated with 11.5 of all MA twins and
41.6 of all higher order multiples
46Twin Birthweight Discordancy(Mass, 1998-2001)
Twin Discordancy
47Twin Discordancy by Birth Outcomes( Mass,
1998-2001)
1restricted to twins 28-41 weeks gestation, to
parallel study design of Branum and Schoendorf
(2003) 2below 10th percentile for the smaller
twin, twin SGA percentiles based on Alexander,
et. al. 2003
48Twin Discordancy by Hospitalization(Mass,
1998-2001)
49- Assessment of Regionalization and
Maternal/Neonatal Transfer
50Possible OB/Neonatal Practice Linkages
- Link clinical to public health data bases
- Provides public health denominator for
clinical care - Facilitates NICU discharge follow-up (health
status and PH program involvement) - NeoQIC/Gould California Neonatal Quality
Improvement Project - Allows differentiation by diagnoses among NICU
utilization (e.g., isolate macrosomia)
51 Longitudinally linked data bases, like PELL,
provides States with a platform for expanded
etiologic, longitudinal, and programmatic
research on maternal and infant health and health
trends.
52(No Transcript)