pregnancy-associated crashes and birth outcomes: linking birth/fetal death records to motor vehicle...

Pregnancy-associated Crashes and Birth Outcomes: Linking

birth/fetal death records to motor vehicle crash data

Lisa Hyde, Larry Cook

Lenora Olson, Hank Weiss,

J. Michael Dean

Prior Fetal Injury Research

Research on the effects of motor vehicle crashes on fetal outcomes is limited

– Lack of pregnancy information on crash records

– Lack of motor vehicle crash history on birth certificates

Study Objective

Assess the effect of involvement in a motor vehicle crash on the likelihood of adverse events for the fetus

Use probabilistic linkage to combine motor vehicle crash and birth/fetal death records

Probabilistic Linkage Basics

Probabilistic Linkage Theory

Crash Record

Ambulance Record

Mary Smith F 05/05/45 07/15/96 11:47 Albany US5 Seat=1 Belt=N

Mary Smith Sanchez F 05/05/44 07/15/96 11:55 Albany Fracture Mem Hosp

Probabilistic record linkage is a method of using statistical properties of variables common to a pair of records to calculate the probability that the records apply to the same person and event...

Probabilistic record linkage is a method of using statistical properties of variables common to a pair of records to calculate the probability that the records apply to the same person and event...


Briefly, two statistical properties of each common variable -- reliability and discriminating power -- determine the odds ratio for a true match. The odds ratio is the uniformly most powerful test statistic for discriminating between matched and unmatched record pairs.

Briefly, two statistical properties of each common variable -- reliability and discriminating power -- determine the odds ratio for a true match. The odds ratio is the uniformly most powerful test statistic for discriminating between matched and unmatched record pairs.


Probability that a common variable agrees on a matched pair. Approximately 1 - error rate.

Probability that a common variable agrees on an unmatched pair. Approximately 1 / number of values.

Reliability (m)

Discriminating Power (u)

Record Linkage with Imperfect Data

Let us choose a pair of imperfect records and try to decide if they are a match. That is, do they refer to the same individual and event?

CrashRecords

CrashRecords

HealthRecords

HealthRecords

Probabilistic Record Linkage

If each ambulance record matches to one crash record in a file of 100,000 crashes then the odds for a match at random are

1:99,999

CrashRecords

CrashRecords

AmbulanceRecords

AmbulanceRecords


Mary Smith F 05/05/45 07/15/96 11:47 Albany US5 Seat=1 Belt=NMary Smith F 05/05/45 07/15/96 11:47 Albany US5 Seat=1 Belt=N

Mary Smith Sanchez F 05/05/44 07/15/96 11:55 Albany Fracture Mem HospMary Smith Sanchez F 05/05/44 07/15/96 11:55 Albany Fracture Mem Hosp

Crash Record

Ambulance Record

First name agrees...

m = 0.90 u = 0.01 ratio = 90:1

Agreement on first name improves the odds for a match:

1:99,999 x 90:1 = 1:1,111




Crash Record

Ambulance Record

Last or middle name agrees with last or middle...

m = 0.90 u = 0.04 ratio = 22:1

Agreement on last name improves the odds for a match:

1:1,111 x 22:1 = 1:51




Sex agrees...

m = 0.99 u = 0.50 ratio = 2:1

Crash Record

Ambulance Record

Agreement on sex improves the odds for a match:

1:51 x 2:1 = 1:25



Crash Record

Ambulance Record


Date of BirthMonth agrees...m = 0.99 u = 0.08 ratio = 12:1Day agrees...m = 0.99 u = 0.03 ratio = 30:1Year disagreesm = 0.99 u = 0.01 ratio = 1:99

Agreement on birth date improves the odds for a match:

1:25 x 4:1 = 1:6



Crash Record

Ambulance Record


Date of CrashMonth agrees...m = 0.99 u = 0.08 ratio = 12:1Day agrees...m = 0.99 u = 0.03 ratio = 30:1Year agreesm = 1.00 u = 1.00 ratio = 1:1

Agreement on crash date improves the odds for a match:

1:6 x 360:1 = 60:1



Crash Record

Ambulance Record

Agreement on crash time improves the odds for a match:

60:1 x 12:1 = 1,699:1


Time of CrashHour agrees...m = 0.90 u = 0.04 ratio = 23:1Minute disagrees...m = 0.50 u = 0.02 ratio = 1:2



Crash Record

Ambulance Record

Agreement on crash location improves the odds for a match:

1,699:1 x 10:1 = 16,990:1


Place of crash agrees...

m = 0.99 u = 0.10 ratio = 10:1




Crash Record

Ambulance Record

This pair of records has both agreements and disagreements. Our calculations say that the odds are 16,990:1 that the records refer to the same individual and crash event.

Linkage of Motor Vehicle Crash Records with Birth and Fetal

Death Certificates

Study Databases (1992-1999) Utah Motor Vehicle Crash Data

– All reported motor vehicle crashes– Collected by police officers at the crash scene– Only drivers

Utah Birth Certificate Data– All single live births in Utah

Utah Fetal Death Certificate Data– All reported fetal deaths after 20 weeks gestation– Excludes elective abortions

Linkage Variables

Mother’s first and last name Mother’s date of birth Date of infant birth or fetal death Date of crash

– Compared gestational age / date of last menses with date of crash to ensure the crash occurred during pregnancy

Statistical Analysis Descriptive statistics and logistic regression were

used to assess the impact of having an MVC during pregnancy and wearing a seatbelt on adverse outcomes

Adverse outcomes included:– Low birth weight (<2500 grams)– Excessive maternal bleeding– Fetal distress – Placental abruption

Covariates in Logistic Model Age of the mother Race Weight gain Education level Smoking Alcohol Month of first prenatal visit Number of previous births Medical risk factors Seatbelt use Crash severity (KABCO) Trimester of crash

Birth certificate

Crash database

Birth Certificate Results

Crash Records to Birth Certificates

(1992-1999)

Number of Motor Vehicle CrashesDuring Pregnancy

n = 322,704

8,938 births with a crash (2.8%)

• 322,704 single live births in Utah, 1992 – 1997

• 8,938 (< 3%) were involved in an MVC during pregnancy

Crash

n = 8,938

No Crash

n = 313,766

Age 25.6 years 26.3 years*

Smoking 11.5% 9.3%*

Alcohol 1.7% 1.5%

Number of Previous Births

1.2 1.3*

Completed High School

85.9% 85.6%

Received Care 1st Trimester

84.6% 83.5%** Significant at 0.05 level

Trimester of Crash

0%

20%

40%

1st Trimester 2nd Trimester 3rd Trimester

Trimester of Crash

Per

cent

of

Cra

shes

n = 8,938

Logistic Regression Results for Crash vs. No Crash

Odds Ratio 95% CI

Low birth weight 1.0 (0.9, 1.1)

Excessive bleeding 1.0 (0.7, 1.3)

Fetal distress 1.1 (1.0, 1.2)

Placental abruption 1.0 (0.8, 1.2)

n = 322,704

Seatbelt

n = 7,143

No Seatbelt

n = 1,099

Age 25.8 years 23.9 years*

Smoking 9.8% 21.2%*

Alcohol 1.6% 2.6%*

Number of Previous Births

1.2 1.3

Completed High School

88.1% 73.2%*

Received Care 1st Trimester

85.5% 78.2%** Significant at 0.05 level

Logistic Regression Results forNo Seatbelt vs. No Crash

Odds Ratio 95% CI

Low birth weight 1.3 (1.0, 1.6)*

Excessive bleeding 1.6 (0.9, 2.9)



* Significant at 0.05 leveln = 322,704

Logistic Regression Results forNo Seatbelt vs. Seatbelt

Effect Odds Ratio 95% CI

Low birth weight 1.2 (0.9, 1.6)

Excessive bleeding 2.1 (1.0, 4.2)*



* Significant at 0.05 leveln = 8,938

Fetal DeathCertificate Results

Crash Records to Fetal

Death Certificates

(1992-1999)

Fetal Death Results

2,645 fetal deaths recorded during study period

45 (1.7%) linked to a motor vehicle crash record

No CrashCrash

45 fetal deaths involved in a crash (1.7%)

n = 2,645

Pregnancies Resulting in Fetal Death

Belted crash

Unbelted crash

Unknown

belt use

Total pregnancies

7,143 1,099 696

Fetal deaths

(Percent)

28

(0.4%)

12

(1.2%)

5

(0.7%)

Unbelted pregnant women were 2.8 (95% CI 1.4, 5.6) times more likely to experience a fetal death than belted pregnant women

Conclusions

Probabilistic linkage is a feasible method to combine crash and birth records

– Comparison group of women not in crashes– No recall bias / loss to follow-up

Failure to to wear a seatbelt may increase the likelihood of adverse fetal events

Questions?

Larry Cook

615 Arapeen Dr., Suite 202

Salt Lake City, UT 84108

801.585.9760

[email protected]

pregnancy-associated crashes and birth outcomes: linking birth/fetal death records to motor vehicle...

Documents

ambulance record matches

pair of records

probabilistic linkage

use probabilistic linkage

birthfetal death records

unmatched record pairs

n mary smith sanchez

pair of imperfect records