- Email: [email protected]
Driver seat belt use indicates decreased risk for child passengers in a motor vehicle crash
Accident Analysis and Prevention 42 (2010) 771–777
Contents lists available at ScienceDirect
Accident Analysis and Prevention journal homepage: www.elsevier.com/locate/aap
Driver seat belt use indicates decreased risk for child passengers in a motor vehicle crash Cody S. Olsen ∗ , Lawrence J. Cook, Heather T. Keenan, Lenora M. Olson Intermountain Injury Control Research Center, Department of Pediatrics, University of Utah School of Medicine, PO Box 581289, Salt Lake City, UT 84158-1289, USA
a r t i c l e
i n f o
Article history: Received 14 April 2009 Received in revised form 18 September 2009 Accepted 16 November 2009 Keywords: Seat belt Child injuries Emergency department Motor vehicle crash Generalized estimating equations Relative risk
a b s t r a c t Study objective: We examined the association between driver restraint use and child emergency department (ED) evaluation following a motor vehicle crash (MVC). Methods: This cohort study included child passengers aged 0–12 years riding with an adult driver aged 21 years or older involved in a MVC in Utah from 1999 to 2004. The 6 years of Utah MVC records were probabilistically linked to statewide Utah ED records. We estimated the relative risk of ED evaluation following a MVC for children riding with restrained versus unrestrained drivers. Generalized estimating equations were used to calculate relative risks adjusted for child, driver, and crash characteristics. Results: Six percent (6%) of children riding with restrained adult drivers were evaluated in the ED compared to twenty-two percent (22%) of children riding with unrestrained adult drivers following a MVC (relative risk 0.29, 95% confidence interval 0.26–0.32). After adjusting for child, vehicle, and crash characteristics, the relative risk of child ED evaluation associated with driver restraint remained significant (relative risk 0.82, 95% confidence interval 0.72–0.94). Driver restraint use was associated with child restraint use, less alcohol/drug involvement, and lower relative risk of severe collision types (head-on, rollover). Conclusions: Driver seat belt use is associated with decreased risk of ED evaluation for child passengers in the event of a MVC. © 2009 Elsevier Ltd. All rights reserved.
1. Introduction 1.1. Background Increasing rates of child restraint use in the U.S. have resulted in fewer child injuries sustained in motor vehicle crashes (MVCs) (NHTSA, 2006). Correct child restraint use has been shown to decrease the risk of injury and death to children involved in MVCs by 30–50% (Berg et al., 2000; Durbin et al., 2005; Smith and Cummings, 2006). Interventions, including implementation of child safety seat laws in all 50 states in the U.S., have been shown to be effective in both increasing child restraint use and decreasing fatal and non-fatal injuries in children involved in MVCs (Zaza et al., 2001). Despite improvements in rates of child restraint use, injuries sustained by children in MVCs remain a problem. MVCs remain the primary cause of morbidity and mortality among children in the United States (Centers for Disease Control and Prevention, 2008), resulting in approximately 304 000 hospital
∗ Corresponding author. Tel.: +1 1 801 581 5755; fax: +1 1 801 581 8686. E-mail addresses: [email protected] (C.S. Olsen), [email protected] (L.J. Cook), [email protected] (H.T. Keenan), [email protected] (L.M. Olson). 0001-4575/$ – see front matter © 2009 Elsevier Ltd. All rights reserved. doi:10.1016/j.aap.2009.11.009
days and 2 billion dollars in hospital charges annually (Gardner et al., 2007). Child passengers are influenced by the adult driver’s behaviors, as they are exposed to the adult’s safety equipment preferences, driving habits, and resultant crash characteristics. A recent study found that 95% of child passengers are restrained when the driver transporting them is also restrained; only 60% of child passengers are restrained when the driver is not (Cody et al., 2002). Another study found that fatally injured children were unrestrained more often when riding with an unrestrained driver than when riding with a restrained driver (NHTSA, 2003). Child passengers involved in a crash may be exposed to the same forces (speed, change in velocity, and collision type) as their drivers. An unrestrained driver involved in a motor vehicle crash (MVC) is more likely to speed or drive under the influence of alcohol or drugs, and more likely to be involved in a severe collision (Preusser et al., 1991; ValenciaMartin et al., 2008). Those results suggest that a child riding with an unrestrained driver is more likely to be at risk for serious injury or death following a MVC. 1.2. Objectives of the study We hypothesized that child passengers, aged 0–12 years, riding with restrained drivers were less likely to be evaluated in the
772
C.S. Olsen et al. / Accident Analysis and Prevention 42 (2010) 771–777
emergency department (ED) following a MVC compared to child passengers riding with unrestrained drivers. We examined this association by estimating the relative risk of ED evaluation for children riding with restrained drivers compared to those riding with unrestrained drivers. We fit two ancillary models. In the first, we adjusted for child characteristics which contribute to the risk of child ED evaluation. In the second, we adjusted for child, driver, vehicle, and crash characteristics. 2. Materials and methods In this cohort study, we retrospectively identified two groups of child passengers aged 0–12 years involved in MVCs: those riding with restrained adult drivers, and those riding with unrestrained adult drivers. We estimated the relative risk of ED evaluation for children riding with restrained drivers compared to children riding with unrestrained drivers in a MVC. The University of Utah Institutional Review Board approved this study. 2.1. Data source MVC data for the years 1999 through 2004 were obtained from the Utah Department of Transportation, Division of Traffic and Safety. These data include all reported crashes on public roads in Utah involving injury or resulting in property damage over $1000. MVC data are collected by local law enforcement officers at the scene of the crash. The MVC database includes information about both injured and uninjured vehicle occupants (age, gender, seating position, injury severity codes, and self-reported use of safety devices) and the crash circumstances (road and weather conditions, vehicle damage descriptions, and driver actions). Statewide ED records were obtained from the Utah Health Data Committee/Office of Healthcare Statistics, Utah Department of Health, to which all licensed EDs in Utah are mandated to submit data. This database contains billing information for all admissions and discharges from the ED, including patient demographics, external cause of injury codes, and diagnosis codes for all Utah ED visits. 2.2. Probabilistic linkage The analysis database was constructed by probabilistically linking the MVC database to the ED database for the 6-year period. Briefly, probabilistic record linkage is a method that uses variables common to two or more databases to determine the probability that two records refer to the same person and/or event. The ability to link specific crash events to ED outcome data allows for a more complete analysis of the event in total. From the linked dataset, comparisons can be made that would otherwise be impossible from examination of either database individually. These methods are described in more detail elsewhere (Cook et al., 2001; Jaro, 1995). 2.3. Inclusion/exclusion criteria We used linked MVC and ED data for child occupants involved in MVCs between 1999 and 2004. In order to study children who were likely riding with parents, guardians, or other adults who would influence their decision to wear a restraint, we included children aged 0–12 years riding with a driver aged 21 years or older whose level of restraint use was known. To focus the analysis on the association between child passenger outcomes and the characteristics of the adult driver who is most likely to influence child restraint use, we excluded vehicles with more than one adult age 21 years or older. Vehicles including non-driver occupants between 13 and 20 years old were not excluded, though data from those specific occupants were not included in analyses. Occupants of vehicles other than passenger cars, light trucks, sport-utility vehicles, and vans
were excluded, as were child occupants riding in non-passenger areas such as a truck bed or camper. 2.4. Definitions We used ED evaluation—whether or not a child’s crash record linked to an ED record—as the primary outcome in our analysis. Using International Classification of Diseases, 9th revision codes recorded in the ED record, we found that 83% of linked hospital visits had at least one injury diagnosis and another 14% contained code V71: observation and evaluation for suspected conditions not found. This data shows that ED evaluation is a reasonable surrogate for injury. Child ED evaluation also represents the burden to the ED of crashes involving child occupants. Optimal and sub-optimal child restraint use was defined to correspond as closely as possible to the American Academy of Pediatrics’ guidelines for most appropriate child safety equipment usage (American Academy of Pediatrics, 2002). Optimal restraint included the use of a child safety seat from birth through 4 years old or use of a child safety seat or lap and shoulder belt for children 5 through 12 years old. Sub-optimal usage was defined as a lap belt, shoulder belt, or both in children 0 through 4 years old, or use of a lap or shoulder belt alone for children 5 through 12 years. When no restraint device was used, restraint use was categorized as none. Adult restraint use was categorized as used when a lap belt, shoulder belt, or both were used, and as none when no restraint device was used. Table 1 describes all variables used in data analyses, including ED evaluation, restraint use, age, seating position, alcohol/drug involvement, vehicle ≥10 years old, night, adverse weather, posted speed ≥55 mph, and collision type. These variables were chosen a priori based on previously reported relationships with crash and injury severity (Berg et al., 2000; Durbin et al., 2005; Garcia-Espana and Durbin, 2008; Mao et al., 1997). 2.5. Data analysis Because our dataset consisted of vehicles with only one adult and one or more children, adult driver information was considered vehicle-level data, and child information was considered occupantlevel data. With this hierarchical data structure, children were the main units of analysis. Associations between driver restraint use and child, driver, vehicle, and crash characteristics were examined using generalized estimating equations (GEEs) (Hutchings et al., 2003; Zeger et al., 1988). The relative risk of each characteristic being present for restrained drivers compared to unrestrained drivers was estimated with 95% confidence intervals (CIs). In order to estimate relative risks using GEE models, a binomial error distribution and log link were specified (Spiegelman and Hertzmark, 2005; Wacholder, 1986). For occupant-level outcomes, an exchangeable working correlation structure was specified to account for correlation within a vehicle. We used GEE models again to estimate the relative risk of ED evaluation for children riding with restrained drivers compared to those riding with unrestrained drivers. The relative risk of child ED evaluation was also estimated for the presence versus absence of child, driver, vehicle, and crash characteristics. Additionally, two supplementary multivariable GEE models were fit to adjust for (1) child characteristics, and (2) child, driver, vehicle, and crash characteristics available from MVC records. The outcome used for these GEE models was child ED evaluation (Yes vs. No). A binomial error distribution, log link, and exchangeable correlation were specified for the first multivariable and all unadjusted GEE models. The complexity of the second multivariable model necessitated specifying a Poisson error distribution to obtain convergence. This method has
C.S. Olsen et al. / Accident Analysis and Prevention 42 (2010) 771–777
773
Table 1 Variable definitions. Variable
Description
Child-level variables Age 0–4 years 5–8 years 9–12 years
Infant/toddlers/preschoolers School-aged children Older children
Child restraint usea Optimal Sub-optimal None Seating position Front seat Rear seat Child ED evaluation
Driver and vehicle-level variablesb Driver restrainta Used Not used Alcohol/drug involvementa Driver evaluation in ED Vehicle ≥10 years old Night Adverse weather Posted speed ≥55 Collision type Multiple vehicle other Multiple vehicle head-on Single vehicle fixed object Single vehicle other Rollover a b
Use of car seat for birth through age 12; use of both lap and shoulder restraint for age 5 through age 12 Use of lap and/or shoulder restraint for birth through age 4; use of only one of lap or shoulder belt for age 5 through age 12 No restraint device was used Front-middle, or front-passenger position Any position in 2nd, 3rd, or 4th row of vehicle Child’s MVC record was probabilistically linked to an emergency department record, or that of an inpatient transferred from the emergency department
Use of lap or shoulder restraint, or both No restraint device used Driver was issued an alcohol or drug related citation, or alcohol or drugs were considered a primary or secondary contributing factor Driver’s crash record was probabilistically linked to an ED record Age of vehicle, calculated from the crash year and vehicle model year, ≥10 years old Crash occurred between 8:00 pm and 5:59 am One of the following was present in crash record: rain, snow, fog, dust, mist, sleet/hail, or windstorm Posted speed limit of ≥55 mph according to MVC record Collision involved multiple vehicles, and is not defined below Multiple vehicles collided while traveling straight in opposite directions. Single vehicle collided with fixed object Single vehicle ran off roadway or collided with pedestrian, bicycle, other motorized or non-motorized vehicle, or animal Vehicle overturned
No adjustments were made to account for over-reporting of restraint use or under-reporting of alcohol/drug involvement. All children in one vehicle have driver and vehicle-level variables in common.
been shown to consistently and efficiently estimate the relative risk, and to give robust CIs (Zou, 2004). In all stages, we estimated 95% CIs for relative risk estimates. Subjects with complete (non-missing) data were included in relevant analyses. Probabilistic linkage was performed using CODES2000 software (version 7; Strategic Matching, Inc., Morrisonville, NY). Statistical models were generated using the GENMOD procedure within SAS/STAT software (Version 9; SAS Institute Inc., Cary, NC). 3. Results We identified 40 708 children riding with 25 435 adult drivers involved in a MVC (Table 2). Almost half of the children (45%) were 4 years old or younger. Most child occupants (57%) were optimally restrained and seated in a rear seat. The remaining children were not optimally restrained including: 4% not restrained, 19% suboptimally restrained, and 27% in the front seat. In all, 2872 (7%) child occupants were evaluated in the ED. Mean driver age was 34 years and most were female. Restraint use was reported by 95% of drivers, and 13% of drivers were evaluated in the ED. 3.1. Child characteristics and driver non-restraint Children riding with restrained drivers made up 95% of our sample (Table 2). Children riding with restrained drivers were almost twice as likely to be optimally restrained (relative risk [RR] 1.81; 95% confidence interval [CI] 1.70–1.93) and more likely to be riding
in the back seat (RR 1.21; 95% CI 1.16–1.25) than children riding with unrestrained drivers in unadjusted comparisons. 3.2. Driver characteristics and driver non-restraint Analysis of the characteristics of the 25 435 drivers transporting children in our study showed that restrained drivers were much less likely to be cited for driving under the influence of alcohol or drugs, or have alcohol or drugs listed as contributing circumstance (RR 0.12; 95% CI 0.09–0.17), or to be evaluated in the ED following the MVC (RR 0.39; 95% CI 0.36–0.43) compared to unrestrained drivers according to unadjusted comparisons. Restrained drivers were also less likely to be involved in rollover (RR 0.47; 95% CI 0.37–0.59), multiple vehicle head-on (RR 0.44; 95% CI 0.26–0.77), or single vehicle versus fixed object collisions (RR 0.59; 95% CI 0.49–0.71) than unrestrained drivers. 3.3. Unadjusted relative risk of child ED evaluation Six percent (n = 2448) of children riding with restrained drivers linked to an ED record compared to 22% (n = 424) of children riding with unrestrained drivers. A GEE model estimated a relative risk of ED evaluation of 0.29 (95% CI 0.26–0.32) for children riding with restrained drivers compared to those riding with unrestrained drivers. Table 3 shows the relative risk of child ED evaluation associated with adult driver restraint use and other child, driver, vehicle, and crash characteristics. The relative risk of ED evaluation was greatest among children riding with a driver that was evaluated in
774
C.S. Olsen et al. / Accident Analysis and Prevention 42 (2010) 771–777
Table 2 Cohort description and relative risk (RR) of having characteristic if driver is using safety restraints.
Child characteristics Age 0–4 years old 5–8 years old 9–12 years old
Driver restraint used N (%)a
Driver restraint not used N (%)a
Relative risk
N = 38 821
N = 1887
RR (95% CI)
17 567 (45) 11 681 (30) 9573 (25)
756 (40) 632 (33) 499 (26)
1.12 (1.06, 1.19) 0.90 (0.84, 0.96) 0.93 (0.86, 1.02)
Male
19 639 (51)
989 (53)
0.96 (0.92, 1.00)
Child Optimally restrained Sub-optimally restrained Not restrained
29 773 (79) 7154 (19) 715 (2)
773 (43) 244 (13) 798 (44)
1.81 (1.70, 1.93) 1.44 (1.26, 1.66) 0.04 (0.04, 0.05)
Back seat
28 749 (74)
1163 (62)
1.21 (1.16, 1.25)
Driver restraint used N (%)a
Driver restraint not used N (%)a
Relative risk
N = 24 271
N = 1164
RR (95% CI)
Driver characteristics Age 21–30 years old 31–40 years old 41–50 years old 51 years old or older
10 089 (42) 9164 (38) 3600 (15) 1418 (6)
543 (47) 432 (37) 138 (12) 51 (4)
0.89 (0.84, 0.95) 1.02 (0.94, 1.10) 1.25 (1.07, 1.47) 1.33 (1.01, 1.75)
5920 (24) 147 (1) 2984 (12)
335 (29) 57 (5) 367 (32)
0.85 (0.77, 0.93) 0.12 (0.09, 0.17) 0.39 (0.36, 0.43)
14 194 (58) 6822 (28) 2344 (10) 711 (3) 200 (1)
696 (60) 305 (26) 100 (9) 44 (4) 19 (2)
0.98 (0.93, 1.03) 1.07 (0.97, 1.18) 1.12 (0.93, 1.36) 0.78 (0.57, 1.04) 0.51 (0.32, 0.81)
Vehicle ≥10 years old
5805 (24)
448 (39)
0.62 (0.58, 0.67)
Crash characteristicsb Adverse weather Posted speed ≥55 mph Night
2835 (12) 6858 (28) 2211 (9)
112 (10) 243 (21) 113 (10)
1.22 (1.02,1.46) 1.35 (1.21, 1.52) 0.94 (0.78, 1.12)
21 060 (88) 129 (1) 1352 (6) 780 (3) 744 (3)
923 (80) 14 (1) 111 (10) 36 (3) 77 (7)
1.10 (1.07, 1.13) 0.44 (0.26,0.77) 0.59 (0.49, 0.71) 1.05 (0.75, 1.45) 0.47 (0.37, 0.59)
Male Alcohol/drug involvement Driver ED evaluation Vehicle characteristics Number of children in vehicle One Two children Three children Four children Five or more children
Collision Multiple vehicle other Multiple vehicle head-on Single vehicle fixed object Single vehicle other Rollover a b
Percentages are calculated from different totals due to missing data. Vehicle is the reported observational unit since driver restraint use can differ within a crash.
the ED (RR 11.07; 95% CI 10.23–11.98), children involved in headon (RR 4.60; 95% CI 3.61–5.86) or rollover crashes (RR 4.51; 95% CI 4.02–5.05), and children riding with a driver with alcohol or drug involvement (RR 3.59; 95% CI 2.84–4.54). Optimally restrained children (RR 0.28; 95% CI 0.25–0.31) and sub-optimally restrained children (RR 0.30; 95% CI 0.26–0.34) were less likely to be evaluated in the ED than children wearing no safety restraints. 3.4. Multivariable models for relative risk of child ED evaluation To adjust for important MVC characteristics, we fit two models to estimate the adjusted relative risk of child ED evaluation for children riding with restrained drivers (Table 3). The first model adjusted for child restraint use, seating position, and age. All three covariates were significant in estimating the risk of child ED evaluation. Restraint use decreased the risk of child ED evaluation (RR
0.44 optimal restraint use vs. none; 95% CI 0.38–0.52). Riding in the back seat decreased the risk of a child ED evaluation (RR 0.72; 95% CI 0.67–0.78) compared to riding in the front seat. Younger children (ages 0–4) were slightly less at risk for an ED evaluation compared to the older age groups. After adjusting for child restraint use, the relative risk of child ED evaluation for driver restraint reduced in magnitude from the unadjusted estimate of 0.29 to an adjusted estimate of 0.49 (95% CI: 0.41–0.57). The second multivariable model adjusted for driver, vehicle, and crash characteristics in addition to child characteristics. In the second model, estimates for child characteristics did not change substantially from those in the first adjusted model. The risk of a child ED evaluation was greatest when drivers had an ED evaluation (RR 9.49; 95% CI 8.68–10.36), alcohol or drug involvement (RR 1.43; 95% CI 1.12–1.83), and a multiple vehicle head-on collision (RR 2.05; 95% CI 1.61–2.62). Significant effects of driver restraint
C.S. Olsen et al. / Accident Analysis and Prevention 42 (2010) 771–777
775
Table 3 Relative risk (RR) of child emergency department (ED) evaluation for driver restraint and other child, driver, vehicle, and crash characteristics. Characteristic of interest
Unadjusted RR (95% CI)
Adjusted for child-level factorsa RR (95% CI)
Adjusted for child, driver, crash-level factorsb RR (95% CI)
Driver restraint used
0.29 (0.26, 0.32)
0.49 (0.41, 0.57)
0.82 (0.72, 0.94)
Child-level characteristics Child optimally restrained Sub-optimally restrained Not restrained
0.28 (0.25, 0.31) 0.30 (0.26, 0.34) Reference
0.44 (0.38, 0.52) 0.52 (0.43, 0.62) Reference
0.52 (0.45, 0.60) 0.61 (0.52, 0.71) Reference
Back seat
0.63 (0.59, 0.68)
0.72 (0.67, 0.78)
0.71 (0.67, 0.77)
Age 0–4 years old 5–8 years old 9–12 years old
Reference 1.34 (1.25, 1.44) 1.50 (1.39, 1.62)
Reference 1.23 (1.14, 1.33) 1.25 (1.14, 1.37)
Reference 1.27 (1.18, 1.37) 1.37 (1.25, 1.49)
Vehicle characteristics Driver evaluated in ED Alcohol/drug involvement Vehicle ≥10 years old
11.07 (10.23, 11.98) 3.59 (2.84, 4.54) 1.33 (1.21, 1.45)
9.49 (8.68, 10.36) 1.43 (1.12, 1.83) 1.04 (0.96, 1.13)
Crash characteristics Adverse weather Posted speed ≥55 mph Night
0.90 (0.79, 1.03) 1.21 (1.11, 1.32) 1.27 (1.11, 1.44)
0.90 (0.80, 1.02) 1.05 (0.96, 1.14) 1.11 (0.98, 1.26)
Reference 4.60 (3.61, 5.86) 1.85 (1.61, 2.14) 0.63 (0.46, 0.86) 4.51 (4.02, 5.05)
Reference 2.05 (1.61, 2.62) 1.26 (1.10, 1.44) 0.90 (0.66, 1.22) 1.85 (1.64, 2.09)
Collision Multiple vehicle other Multiple vehicle head-on Single vehicle fixed object Single vehicle other Rollover a b
1251 (3%) children were excluded due to missing data. 3187 (8%) children were excluded due to missing data.
use persisted, though the relative risk decreased in magnitude to 0.82 (95% CI 0.72–0.94). 4. Discussion This study has three main findings that highlight the association between adult restraint use and risk of ED evaluation for child passengers in MVCs. First, our descriptive analysis shows that child passengers of restrained drivers are more likely to be restrained themselves, and that restrained drivers are less likely to be in severe crashes than unrestrained drivers. Second, our statistical models show a decreased risk of being evaluated in the ED for children riding with restrained drivers compared to children riding with unrestrained drivers. Third, through adjusted models, we identify contributors to the association between driver restraint use and a decreased risk of ED evaluation for child passengers. These findings suggest that children are at a decreased risk of requiring a medical evaluation when riding with a restrained driver in a MVC. Our analysis adds to literature characterizing restrained drivers and describes the environment to which child passengers are exposed when they are involved in a MVC with a restrained driver. Restrained drivers are less likely to exhibit risky behaviors including speeding, intoxication, prior traffic violations, and involvement in more injury crashes than restrained drivers (Fernandez et al., 2006; Preusser et al., 1991; Valencia-Martin et al., 2008). Similar characteristics have been observed specifically in drivers of restrained child passengers. Drivers transporting restrained child passengers injured in a MVC are more likely to be restrained and less likely to be cited for moving or alcohol/drug related violations (Miller et al., 1998). We confirm associations between driver restraint use and child restraint use, and driver restraint use and lower risk of alcohol/drug involvement. Moreover, our analysis shows an association between driver restraint use and lower risk of
severe collision types, further suggesting a decreased risk of injury to child passengers riding with restrained drivers. While other studies have linked driver restraint use and child restraint use (Agran et al., 1998; Cody et al., 2002; Decina and Knoebel, 1997; Miller et al., 1998; Williams, 1976) and child restraint non-use and injury (Berg et al., 2000; Durbin et al., 2005; Johnston et al., 1994; Smith and Cummings, 2006) our results bridge the gap, linking driver restraint use and a decreased risk of ED evaluation for child MVC passengers aged 0–12 years. In our study, child passengers riding with restrained drivers were less likely to be evaluated in the ED following a MVC than were child passengers riding with unrestrained drivers. While the decreased risk for child ED evaluation does not represent a causal link with driver restraint use, per se, it does suggest that the more favorable environment provided to children in vehicles where drivers use restraints is a contributing factor. As a result, legislation and enforcement interventions which have consistently shown to increase motor vehicle safety practices, including enforcement of speed limits, DUI/DWI laws, and primary seat belt laws, would likely contribute to a safer child passenger environment and fewer child ED visits (Carpenter and Stehr, 2008; Cohen and Einav, 2003; Davis et al., 2006; Houston and Richardson, 2006a,b; Pressley et al., 2009; Snowdon et al., 2009). We identified child characteristics which contribute to the overall association between driver restraint use and lower risk of child ED evaluation. After adjusting for child age, seating position, and restraint use, the relative risk of child passenger ED evaluation for children riding with restrained drivers remained significantly lower than the risk for children riding with unrestrained drivers. Though some child passengers may be able to secure themselves in safety restraints and may choose to ride in the rear seat, the behaviors and characteristics of the driver transporting them significantly influence a child’s risk of ED evaluation in a MVC.
776
C.S. Olsen et al. / Accident Analysis and Prevention 42 (2010) 771–777
Finally, we identified driver, vehicle, and crash characteristics associated with more severe outcomes in children: alcohol/drug involvement, collision type and driver ED evaluation—an indicator of crash severity. After accounting for these characteristics, the remaining adjusted relative risk of child ED evaluation associated with driver restraint use remains, though is less significant. These results are similar to those from a study of 8–12 year old child occupants, who were shown to have increased odds of injury when riding with an unrestrained driver compared to a restrained driver, after adjusting for child, vehicle, and crash-level factors (GarciaEspana and Durbin, 2008). Our study is both strengthened and limited by the use of linked statewide administrative datasets. The nature of the datasets precluded us from analyzing the data per driver-year or per vehiclemile-travelled, thus limiting inference to relative risks per crash. Additionally, the databases do not contain all desired covariates and may contain inaccuracies. For example, restraint use was selfreported. Differential reporting of restraint use by self-report has been shown to inflate estimates of the effectiveness of restraint use (Escobedo et al., 1992). Our risk estimate of child ED evaluation would be inflated if drivers transporting children who were not evaluated in the ED over-reported restraint use more often than drivers transporting children who were evaluated in the ED. This inflation, however, has been shown to be most important when there is a large difference between actual and reported restraintuse rates (Dean et al., 1995). Observations by the Utah Highway Safety Office of relatively high restraint-use rates (67–86% adults, 94–97% for children) during the study period suggest minimal potential bias from differential over-reporting (Utah Department of Public Safety, 2007). Another limitation of using the MVC and ED administrative databases is that some potentially important child, driver, vehicle, and crash characteristics were unavailable, not recorded, or not recorded consistently. It could be useful, for example, to adjust for the speed of the vehicle, direction of impact, severity of structural damage to the vehicle, driver blood alcohol content, or driver’s relationship to child passengers. Additionally, some variables did not include useful specifics. For example, we could not distinguish correct booster seat usage, and therefore chose to consider children over age 5 years, who reported using a lap and shoulder belt, optimally restrained. This choice may have resulted in more children being categorized as optimally restrained than truly were, and therefore likely resulted in a conservative estimate of the effect of child restraint use. Though more detailed information or additional covariates would be useful for adjusted relative risk estimates, the reported unadjusted estimate would not change. Using probabilistic linkage to identify which children were evaluated in the ED likely missed some children who were evaluated in the ED and misclassified some children not evaluated in the ED. It is unlikely that the probabilistic linkage would introduce differential misclassification; thus, misclassification would tend to bias toward the null. We used high-probability (p > 0.90) matches for analysis resulting in an estimated 31 false matches out of the 2872 record matches. Our study included child MVC occupants riding with a one adult aged 21 years or older in Utah. The characteristics of this sample may differ from the entire population of child MVC occupants. However, drivers in this sample are likely to have had a strong influence on whether child restraints were used, allowing us to make associative inferences between the adult in the vehicle and child occupant characteristics and outcomes. Despite the limitations, our study has several strengths. The MVC and ED population databases allowed us to analyze a large cohort of children involved in MVCs. Using probabilistic linkage methods to combine the MVC and ED data allowed us to analyze crash information, including our primary characteristic of inter-
est, driver restraint use, and a clinical outcome, ED evaluation. ED evaluation represents the burden to the ED of crashes involving children, and is a reasonable surrogate for injury. Finally, we were able to model the relative risk of ED evaluation using GEEs, while accounting for the hierarchical data structure and correlation within vehicles. 5. Conclusion Our results identify an association between driver restraint use and a decreased risk of ED evaluation for child MVC passengers, emphasizing the influence that choices of adult drivers have on child passenger safety. Mechanisms to increase safe driving practices of drivers transporting child passengers have the potential to decrease the number of child passenger injuries from MVCs in the U.S. Acknowledgments The authors gratefully acknowledge Amy Donaldson for her review of the manuscript and insightful feedback prior to submission and Andrea Thomas for her careful review of the analysis data. This work was partially supported by funding from the National Highway Traffic Safety Administration (cooperative agreement number DTNH22-03-H-27207). References Agran, P.F., Anderson, C.L., Winn, D.G., 1998. Factors associated with restraint use of children in fatal crashes. Pediatrics 102 (3), E39. American Academy of Pediatrics, 2002. Selecting and using the most appropriate car safety seats for growing children: guidelines for counseling parents. Pediatrics 109 (3), 550–553. Berg, M.D., Cook, L., Corneli, H.M., Vernon, D.D., Dean, J.M., 2000. Effect of seating position and restraint use on injuries to children in motor vehicle crashes. Pediatrics 105 (4 Pt 1), 831–835. Carpenter, C.S., Stehr, M., 2008. The effects of mandatory seatbelt laws on seatbelt use, motor vehicle fatalities, and crash-related injuries among youths. J. Health Econ. 27 (3), 642–662. Centers for Disease Control and Prevention, 2008. Web-based injury statistics query and reporting system (WISQARS). National Center for Injury Prevention and Control. Website: http://www.cdc.gov/ncipc/wisqars (accessed March 2008). Cody, B.E., Mickalide, A.D., Paul, H.P., Colella, J.M., 2002. Child passengers at risk in America: a national study of restraint use National SAFE KIDS Campaign. Washington, DC. Website: http://www.usa.safekids.org/ content documents/ACFD6C.pdf (accessed April 2008). Cohen, A., Einav, L., 2003. The effects of mandatory seat belt laws on driving behavior and traffic fatalities. Rev. Econ. Stat. 85 (4), 828–843. Cook, L.J., Olson, L.M., Dean, J.M., 2001. Probabilistic record linkage: relationships between file sizes, identifiers and match weights. Methods Inf. Med. 40 (3), 196–203. Davis, J.W., Bennink, L.D., Pepper, D.R., Parks, S.N., Lemaster, D.M., Townsend, R.N., 2006. Aggressive traffic enforcement: a simple and effective injury prevention program. J. Trauma 60 (5), 972–976, discussion 976–977. Dean, J.M., Reading, J.C., Nechodom, P.J., 1995. Overreporting and measured effectiveness of seat belts in motor vehicle crashes in Utah. Transp. Res. Rec. (1485), 186–191. Decina, L.E., Knoebel, K.Y., 1997. Child safety seat misuse patterns in four states. Accid. Anal. Prev. 29 (1), 125–132. Durbin, D.R., Chen, I., Smith, R., Elliott, M.R., Winston, F.K., 2005. Effects of seating position and appropriate restraint use on the risk of injury to children in motor vehicle crashes. Pediatrics 115 (3), e305–e309. Escobedo, L.G., Chorba, T.L., Remington, P.L., Anda, R.F., Sanderson, L., Zaidi, A.A., 1992. The influence of safety belt laws on self-reported safety belt use in the United States. Accid. Anal. Prev. 24 (6), 643–653. Fernandez, W.G., Mehta, S.D., Coles, T., Feldman, J.A., Mitchell, P., Olshaker, J., 2006. Self-reported safety belt use among emergency department patients in Boston, Massachusetts. BMC Public Health 6, 111. Garcia-Espana, J.F., Durbin, D.R., 2008. Injuries to belted older children in motor vehicle crashes. Accid. Anal. Prev. 40 (6), 2024–2028. Gardner, R., Smith, G.A., Chany, A.M., Fernandez, S.A., Mckenzie, L.B., 2007. Factors associated with hospital length of stay and hospital charges of motor vehicle crash related hospitalizations among children in the United States. Arch. Pediatr. Adolesc. Med., 889–895. Houston, D.J., Richardson Jr., L.E., 2006a. Reducing traffic fatalities in the American States by upgrading to primary enforcement. J. Policy Anal. Manage. 25 (3), 645–659.
C.S. Olsen et al. / Accident Analysis and Prevention 42 (2010) 771–777 Houston, D.J., Richardson Jr., L.E., 2006b. Safety belt use and the switch to primary enforcement, 1991–2003. Am. J. Public Health 96 (11), 1949–1954. Hutchings, C.B., Knight, S., Reading, J.C., 2003. The use of generalized estimating equations in the analysis of motor vehicle crash data. Accid. Anal. Prev. 35 (1), 3–8. Jaro, M.A., 1995. Probabilistic linkage of large public health data files. Stat. Med. 14 (5–7), 491–498. Johnston, C., Rivara, F.P., Soderberg, R., 1994. Children in car crashes: analysis of data for injury and use of restraints. Pediatrics 93 (6 Pt 1), 960–965. Mao, Y., Zhang, J., Robbins, G., Clarke, K., Lam, M., Pickett, W., 1997. Factors affecting the severity of motor vehicle traffic crashes involving young drivers in Ontario. Inj. Prev. 3 (3), 183–189. Miller, T.R., Spicer, R.S., Lestina, D.C., 1998. Who is driving when unrestrained children and teenagers are hurt? Accid. Anal. Prev. 30 (6), 839–849. National Highway Traffic Safety Administration (NHTSA), 2003. Research Note: The Relationship Between Driver and Child Passenger Restraint Use Among Fatally Injured Child Passengers Ages 0–15. DOT 809 558, Washington, DC. National Highway Traffic Safety Administration (NHTSA), 2006. Traffic Safety Facts 2005 Data: Children. DOT/HS 810 618, Washington, DC. Pressley, J.C., Trieu, L., Barlow, B., Kendig, T., 2009. Motor vehicle occupant injury and related hospital expenditures in children aged 3 years to 8 years covered versus uncovered by booster seat legislation. J. Trauma 67 (Suppl. 1), S20–S29. Preusser, D.F., Williams, A.F., Lund, A.K., 1991. Characteristics of belted and unbelted drivers. Accid. Anal. Prev. 23 (6), 475–482.
777
Smith, K.M., Cummings, P., 2006. Passenger seating position and the risk of passenger death in traffic crashes: a matched cohort study. Inj. Prev. 12 (2), 83–86. Snowdon, A.W., Hussein, A., Purc-Stevenson, R., Follo, G., Ahmed, E., 2009. A longitudinal study of the effectiveness of a multi-media intervention on parents’ knowledge and use of vehicle safety systems for children. Accid. Anal. Prev. 41 (3), 498–505. Spiegelman, D., Hertzmark, E., 2005. Easy SAS calculations for risk or prevalence ratios and differences. Am. J. Epidemiol. 162 (3), 199–200. Utah Department of Public Safety, 2007. Utah Safety Belt Observational Survey: July 2007 Report. Salt Lake City, UT. Valencia-Martin, J.L., Galan, I., Rodriguez-Artalejo, F., 2008. The joint association of average volume of alcohol and binge drinking with hazardous driving behaviour and traffic crashes. Addiction 103 (5), 749–757. Wacholder, S., 1986. Binomial regression in GLIM: estimating risk ratios and risk differences. Am. J. Epidemiol. 123 (1), 174–184. Williams, A.F., 1976. Observed child restraint use in automobiles. Am. J. Dis. Child 130 (12), 1311–1317. Zaza, S., Sleet, D.A., Thompson, R.S., Sosin, D.M., Bolen, J.C., 2001. Reviews of evidence regarding interventions to increase use of child safety seats. Am. J. Prev. Med. 21 (Suppl. 4), 31–47. Zeger, S.L., Liang, K.Y., Albert, P.S., 1988. Models for longitudinal data: a generalized estimating equation approach. Biometrics 44 (4), 1049–1060. Zou, G., 2004. A modified Poisson regression approach to prospective studies with binary data. Am. J. Epidemiol. 159 (7), 702–706.
Contents lists available at ScienceDirect
Accident Analysis and Prevention journal homepage: www.elsevier.com/locate/aap
Driver seat belt use indicates decreased risk for child passengers in a motor vehicle crash Cody S. Olsen ∗ , Lawrence J. Cook, Heather T. Keenan, Lenora M. Olson Intermountain Injury Control Research Center, Department of Pediatrics, University of Utah School of Medicine, PO Box 581289, Salt Lake City, UT 84158-1289, USA
a r t i c l e
i n f o
Article history: Received 14 April 2009 Received in revised form 18 September 2009 Accepted 16 November 2009 Keywords: Seat belt Child injuries Emergency department Motor vehicle crash Generalized estimating equations Relative risk
a b s t r a c t Study objective: We examined the association between driver restraint use and child emergency department (ED) evaluation following a motor vehicle crash (MVC). Methods: This cohort study included child passengers aged 0–12 years riding with an adult driver aged 21 years or older involved in a MVC in Utah from 1999 to 2004. The 6 years of Utah MVC records were probabilistically linked to statewide Utah ED records. We estimated the relative risk of ED evaluation following a MVC for children riding with restrained versus unrestrained drivers. Generalized estimating equations were used to calculate relative risks adjusted for child, driver, and crash characteristics. Results: Six percent (6%) of children riding with restrained adult drivers were evaluated in the ED compared to twenty-two percent (22%) of children riding with unrestrained adult drivers following a MVC (relative risk 0.29, 95% confidence interval 0.26–0.32). After adjusting for child, vehicle, and crash characteristics, the relative risk of child ED evaluation associated with driver restraint remained significant (relative risk 0.82, 95% confidence interval 0.72–0.94). Driver restraint use was associated with child restraint use, less alcohol/drug involvement, and lower relative risk of severe collision types (head-on, rollover). Conclusions: Driver seat belt use is associated with decreased risk of ED evaluation for child passengers in the event of a MVC. © 2009 Elsevier Ltd. All rights reserved.
1. Introduction 1.1. Background Increasing rates of child restraint use in the U.S. have resulted in fewer child injuries sustained in motor vehicle crashes (MVCs) (NHTSA, 2006). Correct child restraint use has been shown to decrease the risk of injury and death to children involved in MVCs by 30–50% (Berg et al., 2000; Durbin et al., 2005; Smith and Cummings, 2006). Interventions, including implementation of child safety seat laws in all 50 states in the U.S., have been shown to be effective in both increasing child restraint use and decreasing fatal and non-fatal injuries in children involved in MVCs (Zaza et al., 2001). Despite improvements in rates of child restraint use, injuries sustained by children in MVCs remain a problem. MVCs remain the primary cause of morbidity and mortality among children in the United States (Centers for Disease Control and Prevention, 2008), resulting in approximately 304 000 hospital
∗ Corresponding author. Tel.: +1 1 801 581 5755; fax: +1 1 801 581 8686. E-mail addresses: [email protected] (C.S. Olsen), [email protected] (L.J. Cook), [email protected] (H.T. Keenan), [email protected] (L.M. Olson). 0001-4575/$ – see front matter © 2009 Elsevier Ltd. All rights reserved. doi:10.1016/j.aap.2009.11.009
days and 2 billion dollars in hospital charges annually (Gardner et al., 2007). Child passengers are influenced by the adult driver’s behaviors, as they are exposed to the adult’s safety equipment preferences, driving habits, and resultant crash characteristics. A recent study found that 95% of child passengers are restrained when the driver transporting them is also restrained; only 60% of child passengers are restrained when the driver is not (Cody et al., 2002). Another study found that fatally injured children were unrestrained more often when riding with an unrestrained driver than when riding with a restrained driver (NHTSA, 2003). Child passengers involved in a crash may be exposed to the same forces (speed, change in velocity, and collision type) as their drivers. An unrestrained driver involved in a motor vehicle crash (MVC) is more likely to speed or drive under the influence of alcohol or drugs, and more likely to be involved in a severe collision (Preusser et al., 1991; ValenciaMartin et al., 2008). Those results suggest that a child riding with an unrestrained driver is more likely to be at risk for serious injury or death following a MVC. 1.2. Objectives of the study We hypothesized that child passengers, aged 0–12 years, riding with restrained drivers were less likely to be evaluated in the
772
C.S. Olsen et al. / Accident Analysis and Prevention 42 (2010) 771–777
emergency department (ED) following a MVC compared to child passengers riding with unrestrained drivers. We examined this association by estimating the relative risk of ED evaluation for children riding with restrained drivers compared to those riding with unrestrained drivers. We fit two ancillary models. In the first, we adjusted for child characteristics which contribute to the risk of child ED evaluation. In the second, we adjusted for child, driver, vehicle, and crash characteristics. 2. Materials and methods In this cohort study, we retrospectively identified two groups of child passengers aged 0–12 years involved in MVCs: those riding with restrained adult drivers, and those riding with unrestrained adult drivers. We estimated the relative risk of ED evaluation for children riding with restrained drivers compared to children riding with unrestrained drivers in a MVC. The University of Utah Institutional Review Board approved this study. 2.1. Data source MVC data for the years 1999 through 2004 were obtained from the Utah Department of Transportation, Division of Traffic and Safety. These data include all reported crashes on public roads in Utah involving injury or resulting in property damage over $1000. MVC data are collected by local law enforcement officers at the scene of the crash. The MVC database includes information about both injured and uninjured vehicle occupants (age, gender, seating position, injury severity codes, and self-reported use of safety devices) and the crash circumstances (road and weather conditions, vehicle damage descriptions, and driver actions). Statewide ED records were obtained from the Utah Health Data Committee/Office of Healthcare Statistics, Utah Department of Health, to which all licensed EDs in Utah are mandated to submit data. This database contains billing information for all admissions and discharges from the ED, including patient demographics, external cause of injury codes, and diagnosis codes for all Utah ED visits. 2.2. Probabilistic linkage The analysis database was constructed by probabilistically linking the MVC database to the ED database for the 6-year period. Briefly, probabilistic record linkage is a method that uses variables common to two or more databases to determine the probability that two records refer to the same person and/or event. The ability to link specific crash events to ED outcome data allows for a more complete analysis of the event in total. From the linked dataset, comparisons can be made that would otherwise be impossible from examination of either database individually. These methods are described in more detail elsewhere (Cook et al., 2001; Jaro, 1995). 2.3. Inclusion/exclusion criteria We used linked MVC and ED data for child occupants involved in MVCs between 1999 and 2004. In order to study children who were likely riding with parents, guardians, or other adults who would influence their decision to wear a restraint, we included children aged 0–12 years riding with a driver aged 21 years or older whose level of restraint use was known. To focus the analysis on the association between child passenger outcomes and the characteristics of the adult driver who is most likely to influence child restraint use, we excluded vehicles with more than one adult age 21 years or older. Vehicles including non-driver occupants between 13 and 20 years old were not excluded, though data from those specific occupants were not included in analyses. Occupants of vehicles other than passenger cars, light trucks, sport-utility vehicles, and vans
were excluded, as were child occupants riding in non-passenger areas such as a truck bed or camper. 2.4. Definitions We used ED evaluation—whether or not a child’s crash record linked to an ED record—as the primary outcome in our analysis. Using International Classification of Diseases, 9th revision codes recorded in the ED record, we found that 83% of linked hospital visits had at least one injury diagnosis and another 14% contained code V71: observation and evaluation for suspected conditions not found. This data shows that ED evaluation is a reasonable surrogate for injury. Child ED evaluation also represents the burden to the ED of crashes involving child occupants. Optimal and sub-optimal child restraint use was defined to correspond as closely as possible to the American Academy of Pediatrics’ guidelines for most appropriate child safety equipment usage (American Academy of Pediatrics, 2002). Optimal restraint included the use of a child safety seat from birth through 4 years old or use of a child safety seat or lap and shoulder belt for children 5 through 12 years old. Sub-optimal usage was defined as a lap belt, shoulder belt, or both in children 0 through 4 years old, or use of a lap or shoulder belt alone for children 5 through 12 years. When no restraint device was used, restraint use was categorized as none. Adult restraint use was categorized as used when a lap belt, shoulder belt, or both were used, and as none when no restraint device was used. Table 1 describes all variables used in data analyses, including ED evaluation, restraint use, age, seating position, alcohol/drug involvement, vehicle ≥10 years old, night, adverse weather, posted speed ≥55 mph, and collision type. These variables were chosen a priori based on previously reported relationships with crash and injury severity (Berg et al., 2000; Durbin et al., 2005; Garcia-Espana and Durbin, 2008; Mao et al., 1997). 2.5. Data analysis Because our dataset consisted of vehicles with only one adult and one or more children, adult driver information was considered vehicle-level data, and child information was considered occupantlevel data. With this hierarchical data structure, children were the main units of analysis. Associations between driver restraint use and child, driver, vehicle, and crash characteristics were examined using generalized estimating equations (GEEs) (Hutchings et al., 2003; Zeger et al., 1988). The relative risk of each characteristic being present for restrained drivers compared to unrestrained drivers was estimated with 95% confidence intervals (CIs). In order to estimate relative risks using GEE models, a binomial error distribution and log link were specified (Spiegelman and Hertzmark, 2005; Wacholder, 1986). For occupant-level outcomes, an exchangeable working correlation structure was specified to account for correlation within a vehicle. We used GEE models again to estimate the relative risk of ED evaluation for children riding with restrained drivers compared to those riding with unrestrained drivers. The relative risk of child ED evaluation was also estimated for the presence versus absence of child, driver, vehicle, and crash characteristics. Additionally, two supplementary multivariable GEE models were fit to adjust for (1) child characteristics, and (2) child, driver, vehicle, and crash characteristics available from MVC records. The outcome used for these GEE models was child ED evaluation (Yes vs. No). A binomial error distribution, log link, and exchangeable correlation were specified for the first multivariable and all unadjusted GEE models. The complexity of the second multivariable model necessitated specifying a Poisson error distribution to obtain convergence. This method has
C.S. Olsen et al. / Accident Analysis and Prevention 42 (2010) 771–777
773
Table 1 Variable definitions. Variable
Description
Child-level variables Age 0–4 years 5–8 years 9–12 years
Infant/toddlers/preschoolers School-aged children Older children
Child restraint usea Optimal Sub-optimal None Seating position Front seat Rear seat Child ED evaluation
Driver and vehicle-level variablesb Driver restrainta Used Not used Alcohol/drug involvementa Driver evaluation in ED Vehicle ≥10 years old Night Adverse weather Posted speed ≥55 Collision type Multiple vehicle other Multiple vehicle head-on Single vehicle fixed object Single vehicle other Rollover a b
Use of car seat for birth through age 12; use of both lap and shoulder restraint for age 5 through age 12 Use of lap and/or shoulder restraint for birth through age 4; use of only one of lap or shoulder belt for age 5 through age 12 No restraint device was used Front-middle, or front-passenger position Any position in 2nd, 3rd, or 4th row of vehicle Child’s MVC record was probabilistically linked to an emergency department record, or that of an inpatient transferred from the emergency department
Use of lap or shoulder restraint, or both No restraint device used Driver was issued an alcohol or drug related citation, or alcohol or drugs were considered a primary or secondary contributing factor Driver’s crash record was probabilistically linked to an ED record Age of vehicle, calculated from the crash year and vehicle model year, ≥10 years old Crash occurred between 8:00 pm and 5:59 am One of the following was present in crash record: rain, snow, fog, dust, mist, sleet/hail, or windstorm Posted speed limit of ≥55 mph according to MVC record Collision involved multiple vehicles, and is not defined below Multiple vehicles collided while traveling straight in opposite directions. Single vehicle collided with fixed object Single vehicle ran off roadway or collided with pedestrian, bicycle, other motorized or non-motorized vehicle, or animal Vehicle overturned
No adjustments were made to account for over-reporting of restraint use or under-reporting of alcohol/drug involvement. All children in one vehicle have driver and vehicle-level variables in common.
been shown to consistently and efficiently estimate the relative risk, and to give robust CIs (Zou, 2004). In all stages, we estimated 95% CIs for relative risk estimates. Subjects with complete (non-missing) data were included in relevant analyses. Probabilistic linkage was performed using CODES2000 software (version 7; Strategic Matching, Inc., Morrisonville, NY). Statistical models were generated using the GENMOD procedure within SAS/STAT software (Version 9; SAS Institute Inc., Cary, NC). 3. Results We identified 40 708 children riding with 25 435 adult drivers involved in a MVC (Table 2). Almost half of the children (45%) were 4 years old or younger. Most child occupants (57%) were optimally restrained and seated in a rear seat. The remaining children were not optimally restrained including: 4% not restrained, 19% suboptimally restrained, and 27% in the front seat. In all, 2872 (7%) child occupants were evaluated in the ED. Mean driver age was 34 years and most were female. Restraint use was reported by 95% of drivers, and 13% of drivers were evaluated in the ED. 3.1. Child characteristics and driver non-restraint Children riding with restrained drivers made up 95% of our sample (Table 2). Children riding with restrained drivers were almost twice as likely to be optimally restrained (relative risk [RR] 1.81; 95% confidence interval [CI] 1.70–1.93) and more likely to be riding
in the back seat (RR 1.21; 95% CI 1.16–1.25) than children riding with unrestrained drivers in unadjusted comparisons. 3.2. Driver characteristics and driver non-restraint Analysis of the characteristics of the 25 435 drivers transporting children in our study showed that restrained drivers were much less likely to be cited for driving under the influence of alcohol or drugs, or have alcohol or drugs listed as contributing circumstance (RR 0.12; 95% CI 0.09–0.17), or to be evaluated in the ED following the MVC (RR 0.39; 95% CI 0.36–0.43) compared to unrestrained drivers according to unadjusted comparisons. Restrained drivers were also less likely to be involved in rollover (RR 0.47; 95% CI 0.37–0.59), multiple vehicle head-on (RR 0.44; 95% CI 0.26–0.77), or single vehicle versus fixed object collisions (RR 0.59; 95% CI 0.49–0.71) than unrestrained drivers. 3.3. Unadjusted relative risk of child ED evaluation Six percent (n = 2448) of children riding with restrained drivers linked to an ED record compared to 22% (n = 424) of children riding with unrestrained drivers. A GEE model estimated a relative risk of ED evaluation of 0.29 (95% CI 0.26–0.32) for children riding with restrained drivers compared to those riding with unrestrained drivers. Table 3 shows the relative risk of child ED evaluation associated with adult driver restraint use and other child, driver, vehicle, and crash characteristics. The relative risk of ED evaluation was greatest among children riding with a driver that was evaluated in
774
C.S. Olsen et al. / Accident Analysis and Prevention 42 (2010) 771–777
Table 2 Cohort description and relative risk (RR) of having characteristic if driver is using safety restraints.
Child characteristics Age 0–4 years old 5–8 years old 9–12 years old
Driver restraint used N (%)a
Driver restraint not used N (%)a
Relative risk
N = 38 821
N = 1887
RR (95% CI)
17 567 (45) 11 681 (30) 9573 (25)
756 (40) 632 (33) 499 (26)
1.12 (1.06, 1.19) 0.90 (0.84, 0.96) 0.93 (0.86, 1.02)
Male
19 639 (51)
989 (53)
0.96 (0.92, 1.00)
Child Optimally restrained Sub-optimally restrained Not restrained
29 773 (79) 7154 (19) 715 (2)
773 (43) 244 (13) 798 (44)
1.81 (1.70, 1.93) 1.44 (1.26, 1.66) 0.04 (0.04, 0.05)
Back seat
28 749 (74)
1163 (62)
1.21 (1.16, 1.25)
Driver restraint used N (%)a
Driver restraint not used N (%)a
Relative risk
N = 24 271
N = 1164
RR (95% CI)
Driver characteristics Age 21–30 years old 31–40 years old 41–50 years old 51 years old or older
10 089 (42) 9164 (38) 3600 (15) 1418 (6)
543 (47) 432 (37) 138 (12) 51 (4)
0.89 (0.84, 0.95) 1.02 (0.94, 1.10) 1.25 (1.07, 1.47) 1.33 (1.01, 1.75)
5920 (24) 147 (1) 2984 (12)
335 (29) 57 (5) 367 (32)
0.85 (0.77, 0.93) 0.12 (0.09, 0.17) 0.39 (0.36, 0.43)
14 194 (58) 6822 (28) 2344 (10) 711 (3) 200 (1)
696 (60) 305 (26) 100 (9) 44 (4) 19 (2)
0.98 (0.93, 1.03) 1.07 (0.97, 1.18) 1.12 (0.93, 1.36) 0.78 (0.57, 1.04) 0.51 (0.32, 0.81)
Vehicle ≥10 years old
5805 (24)
448 (39)
0.62 (0.58, 0.67)
Crash characteristicsb Adverse weather Posted speed ≥55 mph Night
2835 (12) 6858 (28) 2211 (9)
112 (10) 243 (21) 113 (10)
1.22 (1.02,1.46) 1.35 (1.21, 1.52) 0.94 (0.78, 1.12)
21 060 (88) 129 (1) 1352 (6) 780 (3) 744 (3)
923 (80) 14 (1) 111 (10) 36 (3) 77 (7)
1.10 (1.07, 1.13) 0.44 (0.26,0.77) 0.59 (0.49, 0.71) 1.05 (0.75, 1.45) 0.47 (0.37, 0.59)
Male Alcohol/drug involvement Driver ED evaluation Vehicle characteristics Number of children in vehicle One Two children Three children Four children Five or more children
Collision Multiple vehicle other Multiple vehicle head-on Single vehicle fixed object Single vehicle other Rollover a b
Percentages are calculated from different totals due to missing data. Vehicle is the reported observational unit since driver restraint use can differ within a crash.
the ED (RR 11.07; 95% CI 10.23–11.98), children involved in headon (RR 4.60; 95% CI 3.61–5.86) or rollover crashes (RR 4.51; 95% CI 4.02–5.05), and children riding with a driver with alcohol or drug involvement (RR 3.59; 95% CI 2.84–4.54). Optimally restrained children (RR 0.28; 95% CI 0.25–0.31) and sub-optimally restrained children (RR 0.30; 95% CI 0.26–0.34) were less likely to be evaluated in the ED than children wearing no safety restraints. 3.4. Multivariable models for relative risk of child ED evaluation To adjust for important MVC characteristics, we fit two models to estimate the adjusted relative risk of child ED evaluation for children riding with restrained drivers (Table 3). The first model adjusted for child restraint use, seating position, and age. All three covariates were significant in estimating the risk of child ED evaluation. Restraint use decreased the risk of child ED evaluation (RR
0.44 optimal restraint use vs. none; 95% CI 0.38–0.52). Riding in the back seat decreased the risk of a child ED evaluation (RR 0.72; 95% CI 0.67–0.78) compared to riding in the front seat. Younger children (ages 0–4) were slightly less at risk for an ED evaluation compared to the older age groups. After adjusting for child restraint use, the relative risk of child ED evaluation for driver restraint reduced in magnitude from the unadjusted estimate of 0.29 to an adjusted estimate of 0.49 (95% CI: 0.41–0.57). The second multivariable model adjusted for driver, vehicle, and crash characteristics in addition to child characteristics. In the second model, estimates for child characteristics did not change substantially from those in the first adjusted model. The risk of a child ED evaluation was greatest when drivers had an ED evaluation (RR 9.49; 95% CI 8.68–10.36), alcohol or drug involvement (RR 1.43; 95% CI 1.12–1.83), and a multiple vehicle head-on collision (RR 2.05; 95% CI 1.61–2.62). Significant effects of driver restraint
C.S. Olsen et al. / Accident Analysis and Prevention 42 (2010) 771–777
775
Table 3 Relative risk (RR) of child emergency department (ED) evaluation for driver restraint and other child, driver, vehicle, and crash characteristics. Characteristic of interest
Unadjusted RR (95% CI)
Adjusted for child-level factorsa RR (95% CI)
Adjusted for child, driver, crash-level factorsb RR (95% CI)
Driver restraint used
0.29 (0.26, 0.32)
0.49 (0.41, 0.57)
0.82 (0.72, 0.94)
Child-level characteristics Child optimally restrained Sub-optimally restrained Not restrained
0.28 (0.25, 0.31) 0.30 (0.26, 0.34) Reference
0.44 (0.38, 0.52) 0.52 (0.43, 0.62) Reference
0.52 (0.45, 0.60) 0.61 (0.52, 0.71) Reference
Back seat
0.63 (0.59, 0.68)
0.72 (0.67, 0.78)
0.71 (0.67, 0.77)
Age 0–4 years old 5–8 years old 9–12 years old
Reference 1.34 (1.25, 1.44) 1.50 (1.39, 1.62)
Reference 1.23 (1.14, 1.33) 1.25 (1.14, 1.37)
Reference 1.27 (1.18, 1.37) 1.37 (1.25, 1.49)
Vehicle characteristics Driver evaluated in ED Alcohol/drug involvement Vehicle ≥10 years old
11.07 (10.23, 11.98) 3.59 (2.84, 4.54) 1.33 (1.21, 1.45)
9.49 (8.68, 10.36) 1.43 (1.12, 1.83) 1.04 (0.96, 1.13)
Crash characteristics Adverse weather Posted speed ≥55 mph Night
0.90 (0.79, 1.03) 1.21 (1.11, 1.32) 1.27 (1.11, 1.44)
0.90 (0.80, 1.02) 1.05 (0.96, 1.14) 1.11 (0.98, 1.26)
Reference 4.60 (3.61, 5.86) 1.85 (1.61, 2.14) 0.63 (0.46, 0.86) 4.51 (4.02, 5.05)
Reference 2.05 (1.61, 2.62) 1.26 (1.10, 1.44) 0.90 (0.66, 1.22) 1.85 (1.64, 2.09)
Collision Multiple vehicle other Multiple vehicle head-on Single vehicle fixed object Single vehicle other Rollover a b
1251 (3%) children were excluded due to missing data. 3187 (8%) children were excluded due to missing data.
use persisted, though the relative risk decreased in magnitude to 0.82 (95% CI 0.72–0.94). 4. Discussion This study has three main findings that highlight the association between adult restraint use and risk of ED evaluation for child passengers in MVCs. First, our descriptive analysis shows that child passengers of restrained drivers are more likely to be restrained themselves, and that restrained drivers are less likely to be in severe crashes than unrestrained drivers. Second, our statistical models show a decreased risk of being evaluated in the ED for children riding with restrained drivers compared to children riding with unrestrained drivers. Third, through adjusted models, we identify contributors to the association between driver restraint use and a decreased risk of ED evaluation for child passengers. These findings suggest that children are at a decreased risk of requiring a medical evaluation when riding with a restrained driver in a MVC. Our analysis adds to literature characterizing restrained drivers and describes the environment to which child passengers are exposed when they are involved in a MVC with a restrained driver. Restrained drivers are less likely to exhibit risky behaviors including speeding, intoxication, prior traffic violations, and involvement in more injury crashes than restrained drivers (Fernandez et al., 2006; Preusser et al., 1991; Valencia-Martin et al., 2008). Similar characteristics have been observed specifically in drivers of restrained child passengers. Drivers transporting restrained child passengers injured in a MVC are more likely to be restrained and less likely to be cited for moving or alcohol/drug related violations (Miller et al., 1998). We confirm associations between driver restraint use and child restraint use, and driver restraint use and lower risk of alcohol/drug involvement. Moreover, our analysis shows an association between driver restraint use and lower risk of
severe collision types, further suggesting a decreased risk of injury to child passengers riding with restrained drivers. While other studies have linked driver restraint use and child restraint use (Agran et al., 1998; Cody et al., 2002; Decina and Knoebel, 1997; Miller et al., 1998; Williams, 1976) and child restraint non-use and injury (Berg et al., 2000; Durbin et al., 2005; Johnston et al., 1994; Smith and Cummings, 2006) our results bridge the gap, linking driver restraint use and a decreased risk of ED evaluation for child MVC passengers aged 0–12 years. In our study, child passengers riding with restrained drivers were less likely to be evaluated in the ED following a MVC than were child passengers riding with unrestrained drivers. While the decreased risk for child ED evaluation does not represent a causal link with driver restraint use, per se, it does suggest that the more favorable environment provided to children in vehicles where drivers use restraints is a contributing factor. As a result, legislation and enforcement interventions which have consistently shown to increase motor vehicle safety practices, including enforcement of speed limits, DUI/DWI laws, and primary seat belt laws, would likely contribute to a safer child passenger environment and fewer child ED visits (Carpenter and Stehr, 2008; Cohen and Einav, 2003; Davis et al., 2006; Houston and Richardson, 2006a,b; Pressley et al., 2009; Snowdon et al., 2009). We identified child characteristics which contribute to the overall association between driver restraint use and lower risk of child ED evaluation. After adjusting for child age, seating position, and restraint use, the relative risk of child passenger ED evaluation for children riding with restrained drivers remained significantly lower than the risk for children riding with unrestrained drivers. Though some child passengers may be able to secure themselves in safety restraints and may choose to ride in the rear seat, the behaviors and characteristics of the driver transporting them significantly influence a child’s risk of ED evaluation in a MVC.
776
C.S. Olsen et al. / Accident Analysis and Prevention 42 (2010) 771–777
Finally, we identified driver, vehicle, and crash characteristics associated with more severe outcomes in children: alcohol/drug involvement, collision type and driver ED evaluation—an indicator of crash severity. After accounting for these characteristics, the remaining adjusted relative risk of child ED evaluation associated with driver restraint use remains, though is less significant. These results are similar to those from a study of 8–12 year old child occupants, who were shown to have increased odds of injury when riding with an unrestrained driver compared to a restrained driver, after adjusting for child, vehicle, and crash-level factors (GarciaEspana and Durbin, 2008). Our study is both strengthened and limited by the use of linked statewide administrative datasets. The nature of the datasets precluded us from analyzing the data per driver-year or per vehiclemile-travelled, thus limiting inference to relative risks per crash. Additionally, the databases do not contain all desired covariates and may contain inaccuracies. For example, restraint use was selfreported. Differential reporting of restraint use by self-report has been shown to inflate estimates of the effectiveness of restraint use (Escobedo et al., 1992). Our risk estimate of child ED evaluation would be inflated if drivers transporting children who were not evaluated in the ED over-reported restraint use more often than drivers transporting children who were evaluated in the ED. This inflation, however, has been shown to be most important when there is a large difference between actual and reported restraintuse rates (Dean et al., 1995). Observations by the Utah Highway Safety Office of relatively high restraint-use rates (67–86% adults, 94–97% for children) during the study period suggest minimal potential bias from differential over-reporting (Utah Department of Public Safety, 2007). Another limitation of using the MVC and ED administrative databases is that some potentially important child, driver, vehicle, and crash characteristics were unavailable, not recorded, or not recorded consistently. It could be useful, for example, to adjust for the speed of the vehicle, direction of impact, severity of structural damage to the vehicle, driver blood alcohol content, or driver’s relationship to child passengers. Additionally, some variables did not include useful specifics. For example, we could not distinguish correct booster seat usage, and therefore chose to consider children over age 5 years, who reported using a lap and shoulder belt, optimally restrained. This choice may have resulted in more children being categorized as optimally restrained than truly were, and therefore likely resulted in a conservative estimate of the effect of child restraint use. Though more detailed information or additional covariates would be useful for adjusted relative risk estimates, the reported unadjusted estimate would not change. Using probabilistic linkage to identify which children were evaluated in the ED likely missed some children who were evaluated in the ED and misclassified some children not evaluated in the ED. It is unlikely that the probabilistic linkage would introduce differential misclassification; thus, misclassification would tend to bias toward the null. We used high-probability (p > 0.90) matches for analysis resulting in an estimated 31 false matches out of the 2872 record matches. Our study included child MVC occupants riding with a one adult aged 21 years or older in Utah. The characteristics of this sample may differ from the entire population of child MVC occupants. However, drivers in this sample are likely to have had a strong influence on whether child restraints were used, allowing us to make associative inferences between the adult in the vehicle and child occupant characteristics and outcomes. Despite the limitations, our study has several strengths. The MVC and ED population databases allowed us to analyze a large cohort of children involved in MVCs. Using probabilistic linkage methods to combine the MVC and ED data allowed us to analyze crash information, including our primary characteristic of inter-
est, driver restraint use, and a clinical outcome, ED evaluation. ED evaluation represents the burden to the ED of crashes involving children, and is a reasonable surrogate for injury. Finally, we were able to model the relative risk of ED evaluation using GEEs, while accounting for the hierarchical data structure and correlation within vehicles. 5. Conclusion Our results identify an association between driver restraint use and a decreased risk of ED evaluation for child MVC passengers, emphasizing the influence that choices of adult drivers have on child passenger safety. Mechanisms to increase safe driving practices of drivers transporting child passengers have the potential to decrease the number of child passenger injuries from MVCs in the U.S. Acknowledgments The authors gratefully acknowledge Amy Donaldson for her review of the manuscript and insightful feedback prior to submission and Andrea Thomas for her careful review of the analysis data. This work was partially supported by funding from the National Highway Traffic Safety Administration (cooperative agreement number DTNH22-03-H-27207). References Agran, P.F., Anderson, C.L., Winn, D.G., 1998. Factors associated with restraint use of children in fatal crashes. Pediatrics 102 (3), E39. American Academy of Pediatrics, 2002. Selecting and using the most appropriate car safety seats for growing children: guidelines for counseling parents. Pediatrics 109 (3), 550–553. Berg, M.D., Cook, L., Corneli, H.M., Vernon, D.D., Dean, J.M., 2000. Effect of seating position and restraint use on injuries to children in motor vehicle crashes. Pediatrics 105 (4 Pt 1), 831–835. Carpenter, C.S., Stehr, M., 2008. The effects of mandatory seatbelt laws on seatbelt use, motor vehicle fatalities, and crash-related injuries among youths. J. Health Econ. 27 (3), 642–662. Centers for Disease Control and Prevention, 2008. Web-based injury statistics query and reporting system (WISQARS). National Center for Injury Prevention and Control. Website: http://www.cdc.gov/ncipc/wisqars (accessed March 2008). Cody, B.E., Mickalide, A.D., Paul, H.P., Colella, J.M., 2002. Child passengers at risk in America: a national study of restraint use National SAFE KIDS Campaign. Washington, DC. Website: http://www.usa.safekids.org/ content documents/ACFD6C.pdf (accessed April 2008). Cohen, A., Einav, L., 2003. The effects of mandatory seat belt laws on driving behavior and traffic fatalities. Rev. Econ. Stat. 85 (4), 828–843. Cook, L.J., Olson, L.M., Dean, J.M., 2001. Probabilistic record linkage: relationships between file sizes, identifiers and match weights. Methods Inf. Med. 40 (3), 196–203. Davis, J.W., Bennink, L.D., Pepper, D.R., Parks, S.N., Lemaster, D.M., Townsend, R.N., 2006. Aggressive traffic enforcement: a simple and effective injury prevention program. J. Trauma 60 (5), 972–976, discussion 976–977. Dean, J.M., Reading, J.C., Nechodom, P.J., 1995. Overreporting and measured effectiveness of seat belts in motor vehicle crashes in Utah. Transp. Res. Rec. (1485), 186–191. Decina, L.E., Knoebel, K.Y., 1997. Child safety seat misuse patterns in four states. Accid. Anal. Prev. 29 (1), 125–132. Durbin, D.R., Chen, I., Smith, R., Elliott, M.R., Winston, F.K., 2005. Effects of seating position and appropriate restraint use on the risk of injury to children in motor vehicle crashes. Pediatrics 115 (3), e305–e309. Escobedo, L.G., Chorba, T.L., Remington, P.L., Anda, R.F., Sanderson, L., Zaidi, A.A., 1992. The influence of safety belt laws on self-reported safety belt use in the United States. Accid. Anal. Prev. 24 (6), 643–653. Fernandez, W.G., Mehta, S.D., Coles, T., Feldman, J.A., Mitchell, P., Olshaker, J., 2006. Self-reported safety belt use among emergency department patients in Boston, Massachusetts. BMC Public Health 6, 111. Garcia-Espana, J.F., Durbin, D.R., 2008. Injuries to belted older children in motor vehicle crashes. Accid. Anal. Prev. 40 (6), 2024–2028. Gardner, R., Smith, G.A., Chany, A.M., Fernandez, S.A., Mckenzie, L.B., 2007. Factors associated with hospital length of stay and hospital charges of motor vehicle crash related hospitalizations among children in the United States. Arch. Pediatr. Adolesc. Med., 889–895. Houston, D.J., Richardson Jr., L.E., 2006a. Reducing traffic fatalities in the American States by upgrading to primary enforcement. J. Policy Anal. Manage. 25 (3), 645–659.
C.S. Olsen et al. / Accident Analysis and Prevention 42 (2010) 771–777 Houston, D.J., Richardson Jr., L.E., 2006b. Safety belt use and the switch to primary enforcement, 1991–2003. Am. J. Public Health 96 (11), 1949–1954. Hutchings, C.B., Knight, S., Reading, J.C., 2003. The use of generalized estimating equations in the analysis of motor vehicle crash data. Accid. Anal. Prev. 35 (1), 3–8. Jaro, M.A., 1995. Probabilistic linkage of large public health data files. Stat. Med. 14 (5–7), 491–498. Johnston, C., Rivara, F.P., Soderberg, R., 1994. Children in car crashes: analysis of data for injury and use of restraints. Pediatrics 93 (6 Pt 1), 960–965. Mao, Y., Zhang, J., Robbins, G., Clarke, K., Lam, M., Pickett, W., 1997. Factors affecting the severity of motor vehicle traffic crashes involving young drivers in Ontario. Inj. Prev. 3 (3), 183–189. Miller, T.R., Spicer, R.S., Lestina, D.C., 1998. Who is driving when unrestrained children and teenagers are hurt? Accid. Anal. Prev. 30 (6), 839–849. National Highway Traffic Safety Administration (NHTSA), 2003. Research Note: The Relationship Between Driver and Child Passenger Restraint Use Among Fatally Injured Child Passengers Ages 0–15. DOT 809 558, Washington, DC. National Highway Traffic Safety Administration (NHTSA), 2006. Traffic Safety Facts 2005 Data: Children. DOT/HS 810 618, Washington, DC. Pressley, J.C., Trieu, L., Barlow, B., Kendig, T., 2009. Motor vehicle occupant injury and related hospital expenditures in children aged 3 years to 8 years covered versus uncovered by booster seat legislation. J. Trauma 67 (Suppl. 1), S20–S29. Preusser, D.F., Williams, A.F., Lund, A.K., 1991. Characteristics of belted and unbelted drivers. Accid. Anal. Prev. 23 (6), 475–482.
777
Smith, K.M., Cummings, P., 2006. Passenger seating position and the risk of passenger death in traffic crashes: a matched cohort study. Inj. Prev. 12 (2), 83–86. Snowdon, A.W., Hussein, A., Purc-Stevenson, R., Follo, G., Ahmed, E., 2009. A longitudinal study of the effectiveness of a multi-media intervention on parents’ knowledge and use of vehicle safety systems for children. Accid. Anal. Prev. 41 (3), 498–505. Spiegelman, D., Hertzmark, E., 2005. Easy SAS calculations for risk or prevalence ratios and differences. Am. J. Epidemiol. 162 (3), 199–200. Utah Department of Public Safety, 2007. Utah Safety Belt Observational Survey: July 2007 Report. Salt Lake City, UT. Valencia-Martin, J.L., Galan, I., Rodriguez-Artalejo, F., 2008. The joint association of average volume of alcohol and binge drinking with hazardous driving behaviour and traffic crashes. Addiction 103 (5), 749–757. Wacholder, S., 1986. Binomial regression in GLIM: estimating risk ratios and risk differences. Am. J. Epidemiol. 123 (1), 174–184. Williams, A.F., 1976. Observed child restraint use in automobiles. Am. J. Dis. Child 130 (12), 1311–1317. Zaza, S., Sleet, D.A., Thompson, R.S., Sosin, D.M., Bolen, J.C., 2001. Reviews of evidence regarding interventions to increase use of child safety seats. Am. J. Prev. Med. 21 (Suppl. 4), 31–47. Zeger, S.L., Liang, K.Y., Albert, P.S., 1988. Models for longitudinal data: a generalized estimating equation approach. Biometrics 44 (4), 1049–1060. Zou, G., 2004. A modified Poisson regression approach to prospective studies with binary data. Am. J. Epidemiol. 159 (7), 702–706.