Cummings P, Rivara FP. Car Occupant Death According to the Restraint Use of Other OccupantsA Matched Cohort Study. JAMA. 2004;291(3):343-349. doi:10.1001/jama.291.3.343
Author Affiliations: Harborview Injury Prevention and Research Center (Drs Cummings and Rivara) and Departments of Epidemiology (Drs Cummings and Rivara) and Pediatrics (Dr Rivara), University of Washington, Seattle.
Context A car occupant could be killed if struck by another
occupant who was catapulted forward, backward, or sideways in a crash.
Objective To estimate the association between death of a car
occupant (the target) and restraint use by other occupants.
Design Matched-pair cohort study comparing the outcomes of 2
target occupants in the same passenger car that crashed.
Setting United States traffic crashes in 1988-2000, using data
from the Fatality Analysis Reporting System.
Subjects Target pairs, at least 1 of whom died: 61 834
front-seat pairs, 5278 rear-seat pairs, and 21 127 pairs on the
left or right side.
Main Outcome Measures Adjusted risk ratio (RR) for death within 30
days of a crash.
Results The risk of death was greater for a restrained front
target occupant in front of an unrestrained occupant compared with a
restrained front target in front of a restrained occupant (adjusted RR,
1.20; 95% confidence interval [CI], 1.10-1.31). For a restrained
rear target occupant behind an unrestrained occupant compared with a
restrained rear target occupant behind a restrained occupant, the
adjusted RR was 1.22 (95% CI, 1.10-1.36). For a restrained
side target occupant sitting next to an unrestrained occupant compared
with a restrained side target occupant sitting next to a restrained
occupant, the adjusted RR was 1.15 (95% CI, 1.08-1.22).
Among unrestrained target occupants, the adjusted RRs were, for front
targets, 1.04 (95% CI, 0.97-1.12), rear targets, 1.22 (95%
CI, 1.10-1.36), and side targets, 0.85 (95% CI, 0.80-0.92).
Conclusion Persons who wish to reduce their risk of death in a
crash should wear their own restraint and should ask others in the same
car to use their restraints.
A car occupant could be killed if struck by another occupant who was
catapulted forward, backward, or sideways in a crash.1- 3 We
used a matched cohort study design to estimate the association between the
death of a car occupant and the restraint use of another occupant in the same
Persons entered this cohort study when they crashed in a car. The study
outcome was death within 30 days of the crash. We wished to adjust for potential
confounding factors, such as crash speed and the ability of the vehicle to
protect occupants. It is difficult to measure these variables accurately,
and estimates of speed were often missing in the data we used. We therefore
restricted our study to pairs of occupants who were in the same car when they
crashed; these pairs were therefore matched in regard to all vehicle and crash
characteristics, such as vehicle type, speed, crash location, and crash severity.
A matched-pair cohort design eliminates confounding by all matched variables.4- 8 In
addition, risk ratios (RRs) for death may be estimated for all pairs, including
those in which both survived, using information only from pairs with at least
1 member who died.4- 8
The study subjects who made up the cohort pairs are referred to as targets
in this article to distinguish them from other occupants. The main study exposures
for each target occupant were the presence of another occupant (not the other
target pair member) who might be thrown against the target, as well as the
restraint use of this other occupant. A target might be exposed to no other
occupant, a restrained other occupant, or an unrestrained other occupant.
An occupant was considered restrained if he/she wore a seat belt of any type
or used a child car seat.
For front-seat target pairs, the main study exposure was the presence
and restraint use of a rear occupant directly behind the target. For example,
a driver and a front-seat passenger in the same car constituted a front-seat
target pair. If there was no one in the left rear seat of that car, the target
driver was exposed to no other occupant. If there was an unrestrained right
rear passenger, the target front passenger was exposed to an unrestrained
other occupant (Figure 1).
For rear targets, the exposure studied was the presence and restraint
use of a front occupant directly in front of the target (Figure 1).
A side target pair consisted of 1 occupant in the front seat and 1 in
the back seat, both on the same side of the car (Figure 1). When the target pair was on the left side of the car,
the study exposure was the presence and restraint use of any occupants on
the right side of the car, in the same front or rear seat row as the target.
If the target pair was in the right front and right rear seats, then the study
exposure was the presence and restraint use of an occupant in the left front
(driver) or left rear passenger seats.
The National Highway Traffic Safety Administration maintains the Fatality
Analysis Reporting System, which has data since 1975 on all crashes on US
public roads that resulted in a death within 30 days.9 We
selected records for passenger cars of model years 1975 through 2001 that
crashed in 1988 through 2000; convertibles, light trucks, vans, minivans,
and sport utility vehicles were excluded. We also excluded records for any
state and year if more than 25% of records from that state and year were missing
information about restraint use. Cars with more than 2 persons in the front
or the rear were excluded. We only included vehicles with the following target
pairs: (1) driver and front seat passenger; (2) left rear and right rear passenger;
(3) driver and left rear passenger; or (4) right front and left rear passenger
We created a set of data for cars with front-seat target pairs aged
16 years or older, at least 1 of whom died. We excluded 12.1% of eligible
cars because of missing restraint information. Front targets in cars with
complete restraint information were compared with excluded targets; mean age
was 41 years vs 39 years, male proportion was 59% vs 62%, and proportion who
died was 59% vs 57%. We excluded a further 0.9% of cars because a front target
was missing information about age or sex.
We created a second set of data for rear-seat target pairs, at least
1 of whom died. Rear targets of all ages were included. We excluded 17.2%
of eligible cars because of missing restraint information. Rear targets in
cars with restraint information were compared with excluded targets; mean
age was 25 years for both, male proportion was 52% vs 54%, and proportion
who died was 59% vs 59%. We excluded 1.3% of cars because of missing information
about rear target age or sex.
We created a third set of data for left-side and right-side target pairs,
at least 1 of whom died; all ages were included. We excluded 14.7% of eligible
cars because of missing restraint data. Side targets in cars with restraint
data were compared with excluded targets; mean age was 30 years vs 29 years,
male proportion was 55% vs 56%, and proportion who died was 60% vs 60%. We
excluded 1.3% of cars because of missing data about side target age or sex.
We determined front-seat air bag presence using software that can ascertain
air bag presence from the vehicle identification number.10 This
study of public data without identifiers was exempted from human subjects
Risk ratios for the death of a target occupant were estimated using
conditional Poisson regression with Stata software.7,8,11- 13 All
RRs were adjusted for target occupant sex, seat position (left or right, front
or back), restraint use, age in years, and the square of age. Further adjustment
for air bag presence, the sex and age of occupants who might hit the target,
and the interaction of target seat position with principal direction of crash
force had little influence on the RR estimates; therefore, this further adjustment
was not performed, to minimize loss of records due to missing data. We assessed
whether RR estimates varied by the restraint use of targets and other occupants
who might hit a target; we added appropriate interaction terms to the regression
models and reported different estimates if P<.05
from a likelihood ratio test for the interaction terms.14
Conditional Poisson regression can estimate RRs for binary outcomes15- 18;
however, the variance estimates may be too large.7 We
therefore estimated percentile confidence intervals (CIs) using 2000 bootstrap
Our RR estimates were adjusted for the risk of death associated with
each target's own restraint use. But one target could be thrown against the
other. We could not also adjust for the risk of death associated with the
restraint use of the other target with whom each target was paired; within
a target pair, each target's own restraint use determines perfectly the restraint
use to which the other target in the pair is exposed. As a sensitivity analysis,
we recalculated the main RR estimates using only target pairs that had the
same restraint use: both restrained or both unrestrained. Within those pairs,
there can be no confounding either by the target's own restraint use or by
the restraint use of the other target pair member.
Table 1, Table 2, and Table 3 show
the characteristics of car target occupants and other occupants according
to front-seat target pairs, rear-seat target pairs, and side-seat target pairs,
respectively. There were 61 834 cars with 2 front-seat targets; 7969
targets were in front of a left rear occupant and 10 785 were in front
of a right rear occupant. Mean target occupant age was 41 years. Most target
drivers were male (69%) and most target front passengers were female (52%).
Nearly half of the front targets (46%) were restrained and 59% died.
The risk of death for a front target (regardless of his/her own restraint
use) with a rear restrained occupant behind them was nearly the same as that
of a front target with no one behind them (RR, 1.01) (Table 4). The risk of death for a front target with an unrestrained
person behind them was greater compared with a target with no one behind them;
the RR was 1.05 for unrestrained front targets and 1.21 for restrained front
When a front target with an unrestrained rear occupant was compared
with a front target with a restrained rear occupant, the RR for death varied
with the target's own restraint use (RR, 1.04 for an unrestrained front target
and 1.20 for a restrained front target) (Table 4).
We classified crash angle as frontal if the principal impact angle was
11, 12, or 1 o'clock (47% of cars) (Figure
2). The RR for death of a restrained front target with an unrestrained
occupant behind, compared with a restrained target with a restrained occupant
behind, varied with the crash angle: 1.26 (95% CI, 1.13-1.41) in frontal crashes
and 1.12 (95% CI, 1.00-1.25) in other crashes (P =
.006 for a likelihood ratio test that the model with these RRs differed from
a model with no crash angle terms).
There were 5278 cars with 2 rear targets; all left rear targets had
a driver in front of them and 90% of right rear targets had a passenger in
front of them (Table 2). Mean
rear target occupant age was 25 years, 52% were male, 29% were restrained,
and 59% died.
The risk of death for a rear target (regardless of the target's own
restraint use) with a restrained occupant in front was nearly the same as
that of a rear target with no one in front (RR, 1.01; 95% CI, 0.86-1.18) (Table 4). The risk of death for a rear
target with an unrestrained occupant in front was greater than that of a rear
target with no one in front (RR, 1.24; 95% CI, 1.06-1.46).
The risk of death was greater if a rear target was behind an unrestrained
front occupant compared with a rear target behind a restrained front occupant
(RR, 1.22; 95% CI, 1.10-1.36) (Table 4).
We classified crashes as rear-angle impact if the principal impact was
at 4 to 8 o'clock (19% of cars). The RR for a rear target with an unrestrained
front occupant compared with a rear target with a restrained front occupant
was 1.45 (95% CI, 1.13-1.88) in rear-angle crashes and 1.17 (95% CI, 1.04-1.31)
in other crashes (P = .12 for a likelihood ratio
test that these RRs differed).
There were 9648 cars with 2 left-side targets and 11 479 cars with
2 right-side targets (Table 3).
Mean target occupant age was 30 years, 55% were male, 38% were restrained,
and 59% died.
The risk of death was greater for a side target (regardless of the target's
own restraint use) if there was an unrestrained occupant seated in the same
seat row beside the target compared with a side target with no one beside
them (RR, 1.07; 95% CI, 1.02-1.12) (Table
4). For restrained side targets, the risk of death was less for
a target with a restrained occupant beside him/her compared with a target
with no one beside them (RR, 0.94). This RR was 1.25 for unrestrained side
For a restrained side target, the risk of death was greater if the target
had an unrestrained occupant beside him/her compared with a target next to
a restrained occupant (RR, 1.15; 95% CI, 1.08-1.22) (Table 4). For an unrestrained side target, the risk of death was
less if the target was beside an unrestrained occupant compared with a target
beside a restrained occupant (RR, 0.85; 95% CI, 0.80-0.92).
We classified crashes as near-side impacts if the principal impact was
2, 3, or 4 o'clock for targets on the right or 8, 9, or 10 o'clock for targets
on the left (Figure 2); 16% of crashes
were near-side impact. The RR for death of an unrestrained target with an
unrestrained occupant beside them compared with an unrestrained target with
a restrained occupant beside was 0.85 (95% CI, 0.80-0.92), regardless of crash
angle. The RR for death of a restrained target with an unrestrained occupant
beside compared with a restrained target with a restrained occupant beside
them was greater in a near-side impact (RR, 1.47; 95% CI, 1.28-1.67) compared
with other impacts (RR, 1.14; 95% CI, 1.07-1.21) (P<.001
for a test that these 2 RRs differed).
We reestimated the RR estimates in Table 4 without the target pairs that were discordant on restraint
use: 16% of front target pairs, 12% of rear target pairs, and 29% of side
target pairs. For the front and rear target pairs, no RR estimate in Table 4 changed by more than 0.02. For
unrestrained side targets, the RR comparing a target next to a restrained
occupant with a target next to an unrestrained occupant changed from 0.85
(Table 4) to 0.94 (95% CI, 0.87-1.02).
For restrained side targets, the RR comparing a target next to a restrained
occupant with a target next to an unrestrained occupant changed from 1.15
(Table 4) to 1.21 (95% CI, 1.12-1.30).
The same regression models were used to estimate the RR for death of
each of 3 categories of target compared with a baseline category of targets
who were unrestrained and exposed to an unrestrained other occupant (Table 5): (1) unrestrained targets exposed
to a restrained other occupant; (2) restrained targets exposed to an unrestrained
other occupant; and (3) restrained targets exposed to a restrained other occupant.
Restrained targets had a lower risk of death compared with unrestrained targets.
The risk of target occupant death was lowest when the target and the other
occupant were both restrained.
We found evidence that a person's risk of death in a crash is associated
with the restraint use of other occupants. In this study, the risk was lowest
when all occupants were restrained.
Restraint use by rear occupants might influence the risk of death of
a front target if some rear occupants were catapulted forward. This mechanism
is consistent with our finding that the risk of front target occupant death
associated with a rear occupant was greater when the rear occupant was not
restrained. The increase in risk was confined chiefly to restrained front
target occupants (RR, 1.21); these targets remain fixed in the path of a rear
occupant who is thrown forward.
Front occupants might be hurled against a target in the rear seat, especially
when the impact is from the rear. In a rear-angle impact, rear targets may
be relatively fixed in place by the seat back; this may explain why the RR
for death of a rear target (RR, 1.22) behind an unrestrained front occupant
compared with a rear target behind a restrained front occupant varied little
with restraint use of the rear target.
A side target may be easily hit by an occupant beside them in the same
seat row; the target and the other occupant are not separated by a seat back.
Therefore, we were not surprised that the RR for death of a side target exposed
to a restrained occupant beside them compared with a side target exposed to
no one was not 1; targets and occupants who sit beside each other can collide
even when both are restrained. We were surprised, however, to find that for
an unrestrained side target, exposure to a restrained occupant may be more
hazardous than exposure to an unrestrained occupant (Table 4). Perhaps when both the side target and the other occupant
are unrestrained, collisions between them are uncommon. Both may often be
thrown forward, not toward each other, even if the principal impact is from
the near side, as the car will often be moving forward when hit. Or perhaps
our estimate of this association was in error due to confounding by the restraint
use of the other target; this possibility was suggested by our sensitivity
The matched cohort design allowed us to estimate RRs for targets in
the same car5- 8;
therefore, confounding by car or crash characteristics, including crash severity,
should be minimized. We expected that the risk of front (or rear) target death
would not be much affected by the presence of a restrained rear (or front)
occupant, as most restrained occupants could not be thrown far from their
own seat. Finding RRs of 1.01 for these associations, close to the expected
value of 1, offers some reassurance that our estimates were not seriously
biased by confounding.
A limitation of our study design was that we could estimate RRs6- 8 but not absolute risks
because most crash survivors were not in the Fatality Analysis Reporting System
data. A second limitation was that we could not simultaneously estimate the
RR associated with an occupant seated diagonally across the vehicle from a
target; for example, if a right rear occupant hit a target driver. Within
a target pair, each target's position in front of, behind, or to the side
of another occupant determines perfectly the other target's relationship to
a diagonal occupant. A third limitation was that it was not possible to adjust
RR estimates simultaneously for the restraint use of each target (the adjustment
we did make) and for the restraint use of the other target in each matched
target pair. However, we recomputed our analyses without the target pairs
that were discordant on restraint use, eliminating target restraint use as
a confounder. For the front and rear target pairs, these restricted analyses
yielded RRs quite close to those in Table
4. For side target pairs, the RR estimates from the pairs concordant
on restraint use differed somewhat from the estimates in Table 4. These differences might be due to some confounding bias
in the estimates in Table 4 or
bias in the restricted analysis due to selection of only 71% of the available
pairs. Regardless of which estimates are more accurate, the risk of death
for a restrained side target was greater if an occupant seated beside him
or her was unrestrained compared with restrained: one RR estimate was 1.15
and the other was 1.21.
Our matched cohort design estimated average RRs for all pairs similar
to those that were studied (Figure 1),
including all pairs without a death.5- 8 This
is a useful feature of the matched-pair design, as only a small fraction of
the pairs who both survived a crash were in the data source we used. Our results
have no application to vehicles with just a driver, as there is no other occupant
in such a vehicle. Our results may not apply to vehicles with 3 front or 3
rear occupants; we did not study these vehicles. Two common target configurations
could not be included in our matched design: rear targets alone in the rear
seat and side targets in the front seat with no side target in the rear seat.
Some data were missing for 13% to 19% of the target pairs in each analysis;
this might have biased our RR estimates. However, targets with complete information
about restraint status were similar in regard to age, sex, and death compared
with targets excluded by missing restraint data.
Information about restraint use may be misclassified in some police
crash reports. But crashes with a death are investigated more intensely than
minor crashes. In a study of front occupant pairs, at least 1 of whom died,
the police report regarding restraint use agreed well with the restraint use
determination made by special crash investigators.20 This
suggests that our present study, which used data from target pairs with at
least 1 death, may not have serious bias due to misclassification of restraint
Park21 studied Fatality Analysis Reporting
System data from 1978 through 1983 using a matched-pair cohort method and
estimated that the risk of death was greater for an unrestrained target in
front of an unrestrained rear occupant compared with an unrestrained target
in front of an empty rear seat (RR, 1.04). This is similar to our estimate
of 1.05. Park's study was limited to unrestrained subjects, as restraint use
was rare in crash data during 1978-1983.
Ichikawa et al22 studied crashes in Japan
and estimated that the risk of death was greater for a restrained target in
front of an unrestrained rear occupant compared with a restrained target in
front of a restrained rear occupant (RR, 4.58; 95% CI, 2.55-8.22). This does
not agree well with our estimate of 1.20. Their estimate suggests that a front
target gets more benefit from having a rear occupant buckle up than from wearing
his/her own restraint. Ichikawa et al selected only cars with 2 injured rear
occupants who had the same restraint use. This selection process may have
biased their estimates. Furthermore, they did not adjust for any possible
If the associations we have estimated are causal, use of restraints
by rear seat occupants may prevent about 1 in 6 deaths of restrained front
targets that would otherwise occur when a rear occupant is present. Use of
restraints by front-seat occupants may similarly reduce the risk of death
for all rear targets. Persons who wish to reduce their risk of death in a
crash should wear a restraint and should ask others in the same car to use