- Email: [email protected]
Seat position and the risk of serious thoracoabdominal injury in lateral motor vehicle crashes
Accident Analysis and Prevention 37 (2005) 668–674
Seat position and the risk of serious thoracoabdominal injury in lateral motor vehicle crashes Craig D. Newgard a,∗ , Roger J. Lewis b,c,d , Jess F. Kraus e,f , K. John McConnell a a
Center for Policy and Research in Emergency Medicine, Department of Emergency Medicine, Oregon Health and Science University, 3181 SW Sam Jackson Park Road, Mail Code CR-114, Portland, OR 97239-3098, USA b David Geffen School of Medicine at UCLA, University of California, Los Angeles, CA, USA c Department of Emergency Medicine, Harbor-UCLA Medical Center, Torrance, CA, USA d Los Angeles Biomedical Research Institute, Torrance, CA, USA e Southern California Injury Prevention Research Center, Los Angeles, CA, USA f School of Public Health, University of California, Los Angeles, CA, USA Received 24 February 2005; accepted 2 March 2005
Abstract Context: Previous studies have suggested that motor vehicle occupants seated on the near-side of a lateral impact have a higher proportion of thoracoabdominal injuries. However, due to limitations in previous studies, the true association between seat position, side of lateral impact, and thoracoabdominal injury is unclear. Objective: To assess the relationship between seat position (i.e., near-side, middle-seat, and far-side, regardless of row), side of lateral motor vehicle crash (MVC), and serious thoracoabdominal injury after adjusting for important crash factors. Design: National population-based cohort of adult subjects involved in MVCs and included in the National Automotive Sampling System Crashworthiness Data System database (NASS CDS) from 1995 to 2003. Patients: Occupants aged ≥16 years involved in MVCs where the highest external deformation of the vehicle was located on the right or left side (i.e., lateral). Main outcome measure: Serious thoracic or abdominal injury, defined as an Abbreviated Injury Scale (AIS) ≥3 in the thoracic or abdominal body region. Results: Fifteen thousand, one hundred and sixty persons involved in primary lateral MVCs were represented in the NASS CDS database during the 9-year period. There were 1867 (2%) persons with serious thoracic injuries and 507 persons (0.5%) with serious abdominal injuries. In multivariable logistic regression models that adjusted for important crash factors and the NASS CDS sampling design, seat position was a strong effect modifier of the association between side of lateral impact and serious thoracic (p < 0.0001) and abdominal (p = 0.0009) injury, with the risk of serious thoracic and abdominal injury highest for occupants seated on the near-side of the crash. The mean probability of injury was higher for near-side and middle-seat occupants compared to far-side occupants, and the probability of thoracic injury was approximately four times higher than that of abdominal injury for all seat positions. Conclusions: There is a strong, synergistic relationship between seat position and side of lateral MVC in assessing risk of serious thoracic and abdominal injury among adult occupants. The probability of serious thoracoabdominal injury increases with increasing proximity of seat position to side of the crash and the risk of thoracic injury is higher than abdominal injury for all seat positions. © 2005 Elsevier Ltd. All rights reserved. Keywords: Thoracic injury; Abdominal injury; Motor vehicle crash
1. Introduction ∗
Corresponding author. Tel.: +1 503 494 1668; fax: +1 503 494 4640. E-mail address: [email protected] (C.D. Newgard).
0001-4575/$ – see front matter © 2005 Elsevier Ltd. All rights reserved. doi:10.1016/j.aap.2005.03.008
Motor vehicle crashes (MVCs) have the potential to produce serious thoracic and abdominal injury. In particular,
C.D. Newgard et al. / Accident Analysis and Prevention 37 (2005) 668–674
lateral crashes (i.e., side impacts) have been implicated as a causal factor in thoracoabdominal injury (Siegel et al., 1993; Dischinger et al., 1993; Siegel et al., 2001; Loo et al., 1996; Horton et al., 2000; McLellan et al., 1996; Kearney et al., 1989; Haland et al., 1993; Pattimore and Dave, 1992; Farmer et al., 1997; Gokcen et al., 1994; McGwin et al., 2003; Franklyn et al., 2002). Several studies have further suggested that seat position (near-side versus far-side) must be considered in the assessment of injury risk in occupants involved in lateral crashes (Dischinger et al., 1993; Horton et al., 2000; McLellan et al., 1996; Haland et al., 1993; Farmer et al., 1997; McGwin et al., 2003; Franklyn et al., 2002). Despite the volume of literature on these associations, definitive conclusions and generalizability is tempered by methodologic limitations in the majority of studies, including restriction to severely injured patients or occupants in certain seat positions (e.g. drivers), limited comparison groups (e.g., subjects in frontal collisions), small sample sizes, difficulty controlling for crash severity, missing variables, and non-uniform criteria for assessing severity of injury. Due to the limitations of previous studies, the true association between seat position and serious thoracic and abdominal injury in lateral MVCs remains unclear. It is also unclear whether one injury type (thoracic versus abdominal) is more common than the other in lateral crashes and what the quantifiable risk is for each injury type by seat position. In this study, our objectives were (1) to assess whether seat position modifies the effect of lateral impact on serious thoracic and (separately) abdominal injury, and (2) to quantify the risk of serious thoracic and abdominal injury by seat position among persons involved in lateral MVCs after adjusting for pertinent crash factors in a national population-based sample.
669
2.2. Patients A national, population-based sample of occupants ≥16 years involved in primary lateral MVCs (i.e., vehicles where the highest external deformation location was located on the right or left side of the vehicle) with passenger vehicles and light trucks were included in the analysis. Highest external deformation located on the side of the vehicle correlated very closely with the highest principal direction of force directed laterally for both right and left lateral crashes. We reasoned that external deformation is easier to assess in the field by pre-hospital personnel than principal direction of force, and that using this inclusion criterion may improve the ability to clinically apply study results. The study excluded children because there was a low incidence of serious thoracic and abdominal injuries in children included in the database and because previous studies have demonstrated a higher risk of injury or death in adults compared with children even after controlling for important crash characteristics (Corneli et al., 2000; Robertson, 1999). 2.3. Main outcome measure Serious thoracic and abdominal injury was defined as an Abbreviated Injury Scale (AIS) ≥3 in these body regions. The AIS for a given injury ranges from 1 (minor) to 6 (nonsurvivable). A score of 3 or more represents a “serious” injury (The Abbreviated Injury Scale, 2001). Because the diagnostic assessment and clinical management of serious thoracic and abdominal injuries differ, all analyses were carried out separately on thoracic and abdominal injury as the outcome measure. 2.4. Effect modifier variable
2. Materials and methods 2.1. Study design and setting We used the National Automotive Sampling System Crashworthiness Data System database (NASS CDS) from 1995 to 2003 for the study. In brief, NASS CDS is a probability-sampled, population-based, nationally representative cohort of persons involved in MVCs that is collected using a three-stage sampling of crashes from specific regions throughout the United States to ensure national generalizability of the data without requiring investigation of every crash in the country (NASS CDS, 2001). The NASS CDS database includes crashes where at least one vehicle was towed due to damage (and was thus available for crash investigation). We selected the 9-year time period because data collection changes for NASS CDS were instituted in 1995. The study was approved by the Institutional Review Boards of Oregon Health and Science University and the Los Angeles Biomedical Research Institute of Harbor-UCLA Medical Center.
Seat position was coded as a polytomous categorical variable, with three categories: left-side (i.e., any occupant seated on the left-side of the vehicle), right-side (i.e., any occupant seated on the right-side), and middle-seat position, regardless of row. Seat position was categorized regardless of row because previous literature has shown fatality risk to be the same for front and rear seat positions in lateral crashes (Evans and Frick, 1988). We confirmed the lack of an association between front- versus rear-seated occupants with these injury patterns by testing the contribution of an additional covariate for front versus rear seat position in the models. Seat position was interacted with side of lateral impact (i.e., left versus right) to test the synergy between these two terms in lateral MVCs. In a second analysis, seat position was combined with side of impact to produce three dichotomous terms (dummy variables): (1) occupants seated on the left and involved in a left-sided impact, (2) occupants seated on the right and involved in a right-sided impact, and (3) occupants seated in the middle and involved in a left or right lateral crash. Far-side occupants (i.e., occupants seated on the far-side of a vehicle
670
C.D. Newgard et al. / Accident Analysis and Prevention 37 (2005) 668–674
involved in a left- or right-sided crash) served as the reference group. 2.5. Patient and vehicle confounders Variables were selected based on previous motor vehicle and biomechanical studies demonstrating association with injury and crash severity (Siegel et al., 1993; Dischinger et al., 1993; Siegel et al., 2001; Loo et al., 1996; Horton et al., 2000; McLellan et al., 1996; Farmer et al., 1997; McGwin et al., 2003; Newgard et al., 2005) as well as factors identified a priori as potentially mediating thoracoabdominal injury in these crashes. Eighteen variables were considered in the analysis, including age (years), gender, pregnancy (any duration of gestation), initial Glasgow Coma Scale (GCS) score (GCS scores of 14 and 15 were grouped together due to often subtle clinical distinctions between the two highest scores), police reported alcohol presence, delta V (change in velocity), rollover with collision, passenger space intrusion (<15, 15–29, 30–45, 46–60, and >60 cm), restraint use (manual lap, lap and shoulder belt, or automatic lap and shoulder belt system versus none), entrapment, ejection, air bag deployment, steering wheel deformity (none, 1–5, 6–10, and >10 cm), vehicle weight (<2500, 2500–3000, or >3000 lbs), vehicle model year, lateral seat position (described above), front versus rear seat position, and side of impact (left versus right) (NASS CDS, 1998). Delta V, age, GCS score, passenger space intrusion, and steering wheel deformity were coded as continuous variables, and the additional 13 variables were coded as categorical variables. Non-collision crashes and pure rollovers were excluded to allow for the appropriate use of delta V to control for crash severity. To account for crashes where the vehicle had multiple impacts, we conducted a secondary analysis on MVCs with primary or secondary lateral impacts (i.e., where right or left side vehicular deformation was either the largest or second largest area of external deformity in the crash). In this paper, we use the term “risk” synonymously with odds because the outcomes (thoracic and abdominal injury) were rare (i.e., ≤2%). Throughout the paper, the number of persons noted is the actual number of persons included in the NASS sample during this time period, while all calculations (i.e., proportions, means, and model estimates) are adjusted for NASS sample design (i.e., accounting for clusters, strata, and weights). 2.6. Statistical analysis After excluding non-collision crashes, pure rollovers, occupants missing injury information, occupants missing seat position information, occupants riding on another passenger’s lap, and occupants riding in a non-enclosed area of the vehicle, the 1995–2003 NASS CDS database contained 54,253 adult occupants. Of this sample, 15,160 persons were involved in primary lateral crashes and
formed the cohort for the primary analysis. There were 21,216 persons involved in primary or secondary lateral impacts. To allow the inclusion of all adult occupants contained in the NASS CDS database during this time period and to preserve the original weighting scheme of the NASS CDS database, we used multiple imputation to impute missing values (Rubin and Schenker, 1991; Barnard and Meng, 1999; Schafer, 1997; Allison, 2001; Lee et al., 1989; Raghunathan et al., 2001; Reiter and Raghunathan, 2002). We imputed missing values using parallel chains of multiple imputation, split by side of lateral impact (left or right), to maximize the statistical efficiency of assessing the interaction term (side of lateral impact × seat position) (Allison, 2001). In addition to the variables noted above, eight variables found to be correlated with the variables of interest were also included in the imputation process: energy absorption, highest deformation extent, other collision vehicle weight, posted speed limit, police-reported travel speed, injury severity scale (ISS) score, injury-related fatality, and AIS ≥3 for any body region. Continuous variables with a skewed distribution were log-transformed for the imputation process. Information for clusters, strata within clusters, and year were included as fixed effects in the imputation process to preserve the complex design features of NASS CDS (Schafer, 1997; Reiter and Raghunathan, 2002). Following the parallel chains of multiple imputation the dataset was re-combined. Ten datasets were created using multiple imputation, each of which were analyzed separately and the results combined, accounting for the variance within each dataset and between datasets to adequately represent the uncertainty in imputing missing values. We first investigated the interaction between seat position and side of lateral impact through multivariable logistic regression models (Jaccard, 2001) that accounted for the complex sampling design (strata, clusters, and weighting) of the NASS CDS database. We considered side of lateral impact (left versus right) the focal independent variable in the interaction term, with seat position (left, right, or middle) as the major effect modifier (Jaccard, 2001). Further adjusting estimates based on clustering within vehicles did not change the results (Hutchings et al., 2003). Variables that did not contribute to either model (i.e., assessing thoracic and abdominal injury, separately) at a level of p < 0.10, that did not significantly alter the association between seat position and injury, and that were not felt to be important confounders were removed from the model in a non-automated, stepwise, backward selection process. The final model included 13 variables. In order to determine the joint significance of the interacted variables (seat position × side of lateral impact), we used the likelihood ratio test (Jaccard, 2001). The −2 log likelihood value for the model without the interaction term was subtracted from the same value for the model with the interaction term, and this difference was compared to a critical χ2 value for α = 0.05 and d.f. = 2 (Jaccard, 2001).
C.D. Newgard et al. / Accident Analysis and Prevention 37 (2005) 668–674
671
Table 1 Occupant and vehicle characteristics among occupants involved in primary lateral impact motor vehicle crashes (n = 15,160)
Mean delta V ± S.E.M. (kmph) Mean age ± S.E.M. (years) Passenger space intrusion ≥30 cm GCS ≤12 Ejection Entrapment Rollover Steering wheel deformity Restraint use Police-reported alcohol use Pregnant Left lateral collision Right lateral collision Left-sided seat Right-sided seat Middle-seat Fatalities
All persons
Persons with serious abdominal injury (AIS ≥3)
Persons with serious thoracic injury (AIS ≥3)
N = 15,160
N = 507
(0.5%)
N = 1867
37.9 38.2 372 194 54 96 52 115 257 101 10 284 223 357 143 7 229
±1.8 ±2.3 (68%) (27%) (9%) (15%) (13%) (20%) (57%) (17%) (3%) (60%) (40%) (67%) (32%) (0.9%) (35%)
17.7 35.4 5031 2216 459 716 1271 1456 10,751 1693 160 7992 7168 11,172 3838 150 983
±0.50 ±0.57 (16%) (9%) (0.5%) (1%) (5%) (4%) (83%) (5%) (0.9%) (53%) (47%) (79%) (20%) (0.9%) (0.7%)
38.7 43.5 1291 632 164 313 191 389 982 395 11 1088 779 1391 460 16 718
Percentage missing
(2%) ±1.3 ±1.0 (67%) (25%) (8%) (19%) (13%) (18%) (54%) (18%) (0.7%) (61%) (39%) (77%) (22%) (0.9%) (26%)
31% 0 35% 31% <1% 2% <1% 2% 2% 10% <1% 0 0 0 0 0 0
Number of persons represents the actual number of occupants sampled for the NASS database during this time period, while all proportions, means and standard errors of the mean are sample-adjusted for cluster, strata, and weighting. S.E.M. = standard error of the mean; kmph = kilometers per hour; cm = centimeters; GCS = Glasgow Coma Scale score; AIS = Abbreviated Injury Scale.
To better quantify the risk of serious thoracoabdominal injury in occupants seated in various seat positions in lateral crashes, we analyzed a second multivariable model that included the same covariates identified above plus three dummy variables that combined seat position and side of lateral impact (i.e., near-side left, near-side right, and middle-seat, with a reference group of far-side occupants), rather than the interaction term. To provide another metric for quantifying the probability of a specific injury type (i.e., thoracic or abdominal) by seat position in lateral crashes, we used the adjusted coefficients generated from the multivariable model in each multiply imputed dataset (collapsing the right and left terms for near-side crashes into a single near-side crash term) and calculated the probability of serious thoracic and (separately) abdominal injury for each subject, by dataset, under three scenarios: (1) all occupants seated on the near-side of the lateral crash, (2) all occupants seated in the middle-seat, and (3) all occupants seated on the far-side of the vehicle. We then calculated sample-adjusted means and 95% confidence intervals for the probability of serious thoracic and abdominal injury for each seat position. All analyses were repeated with both primary and secondary lateral impacts included to ensure consistency with a less restricted sample. Database management was performed using SAS (SAS 8.1, SAS Institute, Cary, NC, USA). SAS-callable IVEware (Survey Methodology Program, Survey Research Center, Institute for Social Research, University of Michigan, MI, USA) was used for multiple imputation and multivariable analyses to account for the complex sampling design of the NASS CDS database. The Jackknife Repeated Replication technique was used to estimate logistic regression parameter variances (Kish and Frankel, 1974).
3. Results There were 15,160 adult occupants (range 16–97 years) involved in primary lateral impacts during the 9-year period and available for analysis. There were 1867 persons (2%) with serious thoracic injuries, including 718 fatalities (26% of patients with serious thoracic injuries), and 507 persons (0.5%) with serious abdominal injuries, including 229 fatalities (35% of patients with serious abdominal injuries). Three hundred and eighty-five persons had both severe thoracic and abdominal injuries. The mean delta V for left-sided versus right-sided crashes was not significantly different (left = 17.1 km/h (kmph), 95% confidence interval (CI) 15.9–18.3; right = 18.5 kmph, 95% CI 17.7–19.3). Patient and vehicle demographics are listed in Table 1. In multivariable logistic regression models that adjusted for important crash factors and the NASS CDS sampling design, seat position was a strong effect modifier of the association between side of lateral impact and serious thoracoabdominal injury (interaction term for near-side lateral crash: for thoracic injury, OR 52.8, 95% CI 9.7–286; for abdominal injury, OR 8.3, 95% CI 2.2–32) (data not shown). The likelihood ratio test also demonstrated the interaction between seat position and side of impact to be highly significant for both injury types (p < 0.0001 and p = 0.0009 for thoracic and abdominal injury, respectively). When seat position and side of impact were combined as covariates, occupants seated on the near-side of the crash were at high risk for both serious thoracic and abdominal injury compared to far-side occupants (Table 2). The mean probabilities of serious thoracic and abdominal injury in nearside and middle-seated occupants were higher than far-side occupants. The probability of serious thoracic injury in lat-
672
C.D. Newgard et al. / Accident Analysis and Prevention 37 (2005) 668–674
Table 2 Odds of injury outcome for selected factors using multivariable logistic regression models for occupants involved in primary lateral impacts (n = 15,160) Outcome
Delta V (per kmph) Passenger space intrusion (per 15 cm) Age (per decade) GCS (per unit decline) Ejection Entrapment Rollover Steering wheel deformity (per 5 cm) Restraint use Police-reported alcohol use Pregnant Far-side seat—left or right lateral crash (reference) Left-side seat—left lateral crash Right-side seat—right lateral crash Middle-seat—left or right lateral crash
Serious abdominal injury (AIS ≥3)
Serious thoracic injury (AIS ≥3)
OR
OR
95%CI
1.06 1.6 1.2 1.1 2.7 1.4 1.8 1.2 0.59 1.4 4.0 1.0 3.2 4.7 2.6
eral crashes was approximately four times higher than serious abdominal injury at each seat position (Fig. 1). These results did not qualitatively change when both primary and secondary lateral impacts were included in the analysis (data not shown). Several additional variables were associated with one or both of the injuries in multivariable models. Each 15 cm increase in passenger space intrusion was associated with an incremental increase in risk of thoracic and abdominal injury, and increasing age was also associated with both thoracic and abdominal injury. Ejection was associated with both injury types, but entrapment was only associated with thoracic in-
Fig. 1. Adjusted mean probability of serious thoracic and abdominal injury by seat position for occupants involved in primary lateral motor vehicle crashes (n = 15,160). Error bars represent the 95% confidence interval around each estimate of the mean probability of serious injury. While the mean probability of injury in the NASS CDS sample is low due to the populationbased sampling and nationally representative injury rates, the probability of injury approaches 1 for persons involved in high energy crashes with many risk factors. The importance of seat position in lateral crashes is demonstrated in the relative difference (rather than absolute difference) in mean probability of injury by seat position.
(1.04–1.08) (1.4–1.9) (1.03–1.4) (1.1–1.3) (1.2–5.9) (0.7–2.5) (0.95–3.3) (0.86–1.7) (0.34–1.03) (0.83–2.2) (0.87–18.5) – (1.8–5.8) (2.9–7.8) (0.54–12.6)
1.09 1.7 1.5 1.2 5.1 3.6 1.9 1.1 0.45 2.0 1.1 1.0 4.6 3.8 2.9
95%CI (1.07–1.11) (1.4–2.0) (1.4–1.7) (1.1–1.4) (2.8–9.2) (1.4–8.9) (0.83–4.2) (0.85–1.5) (0.28–0.73) (1.1–3.7) (0.28–4.3) – (2.8–7.8) (2.2–6.6) (0.86–9.9)
jury. Restraint use was protective against both injury types, but more so for thoracic injury. A lower GCS was also a marker for serious thoracic and abdominal injury in lateral crashes. Vehicle weight, vehicle model year, gender, and front versus rear seat position did not contribute significantly to the models.
4. Discussion In this study, we demonstrate that seat position significantly modifies the association between side of impact and serious thoracic and abdominal injury in lateral crashes. That is, there is a strong, synergistic relationship between seat position and side of lateral MVC in assessing risk of serious thoracic and abdominal injury among adult occupants. This relationship persisted after adjusting for several measures of crash severity (delta V, passenger space intrusion, and steering wheel deformity) and multiple other important crash factors, suggesting that the relationship between seat position and side of impact is not solely explained by crash severity or passenger compartment intrusion. Occupants seated on the near-side or middle-seat of a lateral crash had a higher probability of serious thoracoabdominal injury compared to far-side occupants, with the probability of thoracic injury being higher than that of abdominal injury at all seat positions. While the total number of occupants seated in the middle-seat was small (with consequent loss of precision for estimates of injury risk in multivariable models), the mean probability of injury by seat position suggests that occupants seated in the middle-seat may still be at risk for serious thoracoabdominal injury in lateral crashes. This finding supports a previous study that demonstrated occupants in the middle-seat to have an increased fatality risk relative to far-side occupants in lateral impacts (Evans and Frick, 1988).
C.D. Newgard et al. / Accident Analysis and Prevention 37 (2005) 668–674
Our results may have clinical relevance in the risk assessment of serious thoracic or abdominal injury in persons involved in lateral crashes, as out-of-hospital personnel routinely provide information on pertinent variables from the scene of MVCs to the clinician. In addition to the seat position and side of impact findings described above, each incremental decrease in the GCS score (below a baseline GCS 14–15) was associated with a 24% increase in the odds of serious thoracic injury and a 15% increase in the odds of serious abdominal injury. While the GCS is typically associated with traumatic brain injury or physiologic derangement from injury, it may also serve as a surrogate marker of thoracic and abdominal injury for occupants involved in lateral crashes and may encourage diagnostic imaging of the thoracoabdominal region for a patient involved in a lateral crash who has a depressed mental status. Other readily available variables from the crash scene that may assist in the risk assessment of serious thoracoabdominal injury include degree of passenger space intrusion, increasing age, ejection, entrapment, and lack of restraint use. Clinical intoxication was an independent risk factor for serious thoracic injury, which is consistent with prior studies that have shown alcohol to increase the odds of serious injury and fatality in MVCs (Waller et al., 2003; Bedard et al., 2002). Restraints were protective against both injury types, though this relationship was stronger for thoracic injuries. The finding that restraints were protective in lateral crashes (at least for these two injury types) is in conflict with the belief that seat belts are not effective in this type of crash, as suggested by a previous study (Haland et al., 1993). Restraints may prevent contact with other structures in the occupant compartment (including other occupants) and raise the tolerance to thoracic and abdominal injury in lateral crashes. There may also be engineering applications using these results. As we demonstrate the probability of serious thoracic injury to be approximately four times higher than that of abdominal injury at every seat position in lateral crashes, side impact air bags should be designed accordant with this differential risk. Until the number of vehicles equipped with side impact air bags increases, it will remain unclear whether these devices reduce the risk of serious thoracic or abdominal injury in lateral crashes. There are limitations in our analysis. The NASS CDS dataset contains missing values, requiring the use of multiple imputation to allow the analysis of 100% of the data. To further explore the effects of multiple imputation on the results, we conducted sensitivity analyses for the imputation models using different patterns of missing data (missing completely at random, missing at random, and missing in non-random patterns) for the variables with large percentages of missing data (i.e., delta V, passenger space intrusion, and GCS) in a similar hypothetical dataset, assessing for changes in the results secondary to these missing data patterns. The results remained qualitatively stable, with no indication of systematic bias secondary to using multiple imputation with differing patterns of missingness.
673
We used delta V, passenger space intrusion, and steering wheel deformity as surrogate measures of crash severity. Both delta V and passenger space intrusion had strong, independent associations with serious injury in these analyses. While these measures offer an improved means of adjusting for crash severity, they may not completely control for severity of the crash and their use necessitates the exclusion of subjects involved in crashes for which delta V cannot be calculated (i.e., pure rollovers and non-collision crashes). Due to the construct of our analysis, we were unable to include a covariate for an adjacent, unrestrained occupant in the vehicle, which has been suggested to further increase an occupant’s risk of death in lateral crashes (Cummings and Rivara, 2004). Inclusion of this factor in near-side lateral crashes may further increase an occupant’s risk of injury, though it is unclear whether thoracoabdominal injuries in particular would be increased in this setting. Finally, there were a small number of occupants seated in the middle-seat, which reduced the precision of estimates for injury risk in this group.
5. Conclusions Seat position, as a measure of proximity to the side of lateral impact, is a strong effect modifier in assessing the risk of serious thoracoabdominal injury in lateral MVCs. The synergy between seat position and side of lateral impact persisted after adjusting for crash severity, restraint use, and multiple other important crash factors. Occupants seated on the near-side or middle-seat of vehicles involved in lateral impacts have an increased probability of serious thoracic and abdominal injury when compared to far-side occupants, and the probability of serious thoracic injury is approximately four times that of abdominal injury for all seat positions.
Acknowledgements This project was supported by grant number F32 HS00148 from the Agency for Healthcare Research and Quality and the Society for Academic Emergency Medicine Research Training Grant.
References Allison, P.D., 2001. Missing Data. Sage University Papers Series on Quantitative Applications in the Social Sciences, 07-136. Thousand Oaks, Sage, CA. Barnard, J., Meng, X.L., 1999. Application of multiple imputation in medical studies: from AIDS to NHANES. Stat. Methods Med. Res. 8, 17–36. Bedard, M., Guyatt, G.H., Stones, M.J., Hirdes, J.P., 2002. The independent contribution of driver, crash, and vehicle characteristics to driver fatalities. Accid. Anal. Prev. 34, 717–727.
674
C.D. Newgard et al. / Accident Analysis and Prevention 37 (2005) 668–674
Corneli, H.M., Cook, L.J., Dean, J.M., 2000. Adults and children in severe motor vehicle crashes: a matched-pairs study. Ann. Emerg. Med. 36, 340–345. Cummings, P., Rivara, F.P., 2004. Car occupant death according to the restraint use of other occupants – a matched cohort study. JAMA 291, 343–349. Dischinger, P.C., Cushing, B.M., Kerns, T.J., 1993. Injury patterns associated with direction of impact: drivers admitted to trauma centers. J. Trauma 35, 454–459. Evans, L., Frick, M.C., 1988. Seat position in cars and fatality risk. Am. J. Public Health 78, 1456–1458. Farmer, C.M., Braver, E.R., Mitter, E.L., 1997. Two-vehicle side impact crashes: the relationship of vehicle and crash characteristics to injury severity. Accid. Anal. Prev. 29, 399–406. Franklyn, M., Fitxharris, M., Yang, K., Frampton, R., Morris, A., Fildes, B., 2002. Aortic injuries in side impacts: a preliminary analysis. Annu. Proc. Assoc. Adv. Automot. Med. 46, 113–124. Gokcen, E.C., Burgess, A.R., Siegel, J.H., Mason-Gonzalez, S., Dischinger, P.C., Ho, S.M., 1994. Pelvic fracture mechanism of injury in vehicular trauma patients. J. Trauma 36, 789–795. Haland, Y., Lovsund, P., Nygren, A., 1993. Life-threatening and disabling injuries in car-to-car side impacts—implications for development of protection systems. Accid. Anal. Prev. 25, 199–205. Horton, T.G., Cohn, S.M., Heid, M.P., Augenstein, J.S., Bowen, J.C., McKenney, M.G., Duncan, R.C., 2000. Identification of trauma patients at risk of thoracic aortic tear by mechanism of injury. J. Trauma 48, 1008–1014. 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, 3–8. Jaccard, J., 2001. Interaction Effects in Logistic Regression. Sage University Papers Series on Quantitative Applications in the Social Sciences, 07-135. Thousand Oaks, Sage, CA. Kearney, P.A., Rouhana, S.W., Burney, R.E., 1989. Blunt rupture of the diaphragm: mechanism, diagnosis, and treatment. Ann. Emerg. Med. 18, 1326–1330. Kish, L., Frankel, M., 1974. Inference from complex samples (with discussion). J. R. Stat. Soc. 36, 1–37. Lee, E.S., Forthofer, R.N., Lorimor, R.J., 1989. Analyzing Complex Survey Data. Sage University Papers Series on Quantitative Applications in the Social Sciences, 07-071. Thousand Oaks, Sage, CA. Loo, G.T., Siegel, J.H., Dischinger, P.C., Rixen, D., Burgess, A.R., Addis, M.D., O’Quinn, T., McCammon, L., Schmidhauser, C.B., Marsh, P., Hodge, P.A., Bents, F., 1996. Airbag protection versus compartment intrusion effect determines the pattern of injuries in multiple trauma motor vehicle crashes. J. Trauma 41, 935–951. McGwin, G., Metzger, J., Moran, S.G., Rue, L.W., 2003. Occupant- and collision-related risk factors for blunt thoracic aorta injury. J. Trauma 54, 655–662.
McLellan, B.A., Rizoli, S.B., Brenneman, F.D., Boulanger, B.R., Sharkey, P.W., Szalai, J.P., 1996. Injury pattern and severity in lateral motor vehicle collisions: a Canadian experience. J. Trauma 41, 708– 713. U.S. Department of Transportation, 1998. 1988–1996 NASS CDS Variable-Attribute Structure Manual. U.S. Department of Transportation; National Highway Traffic Safety Administration; National Center for Statistics and Analysis, Crash Investigation Division; Washington, DC. U.S. Department of Transportation, 2001. National Automotive Sampling System (NASS) Crashworthiness Data System Analytical User’s Manual, 2001 File. U.S. Department of Transportation; National Highway Traffic Safety Administration; National Center for Statistics and Analysis; Washington, DC. Newgard, C.N., Lewis, R.J., Kraus, J.F., 2005. Steering wheel deformity and serious thoracic or abdominal injury among drivers and passengers involved in motor vehicle crashes. Ann. Emerg. Med. 45, 43–50. Pattimore, D., Dave, T.P., 1992. Torso injury patterns and mechanisms in car crashes: an additional diagnostic tool. Injury 23, 123–126. Raghunathan, T.E., Lepkowski, Van Hoewyk, J., Solenberger, P.W., 2001. A multivariate technique for multiply imputing missing values using a sequence of regression models. Survey Methodol. 27, 85–95. Reiter, J.P., Raghunathan, T.E., 2002. Multiple imputation for missing data in surveys with complex sample designs. Technical Report. Institute for Social Research, University of Michigan, Ann Arbor, MI. Robertson, L.S., 1999. Reduced fatalities related to rear seat shoulder belts. Injury Prev. 5, 62–64. Rubin, D.B., Schenker, N., 1991. Multiple imputation in health-care databases: an overview and some applications. Stat. Med. 10, 585–598. Schafer, J.L., 1997. Analysis of Incomplete Multivariate Data. Chapman & Hall, New York. Siegel, J.H., Loo, G., Dischinger, P.C., Burgess, A.R., Wang, S.C., Schneider, L.W., Grossman, D., Rivara, F., Mock, C., Natarajan, G.A., Hutchins, K.D., Bents, F.D., McCammon, L., Leibovich, E., Tenenbaum, N., 2001. Factors influencing the patterns of injuries and outcomes in car versus car crashes compared to sport utility, van, or pick-up truck versus car crashes: crash injury research engineering network study. J. Trauma 51, 975–990. Siegel, J.H., Mason-Gonzalez, S., Dischinger, P., Cushing, B., Read, K.R., Robinson, R., Smialek, J., Heatfield, B., Hill, W., Bents, F., Jackson, J., Livingston, D., Clark, C.C., 1993. Safety belt restraints and compartment intrusions in frontal and lateral motor vehicle crashes: mechanisms of injuries, complications, and acute care costs. J. Trauma 5, 736–759. The Abbreviated Injury Scale, 1990 Revision, Update 98, 2001. Association for the Advancement of Automotive Medicine. Waller, P.F., Hill, E.M., Maio, R.F., Blow, F.C., 2003. Alcohol effects on motor vehicle crash injury. Alcohol Clin. Exp. Res. 27, 695–703.
Seat position and the risk of serious thoracoabdominal injury in lateral motor vehicle crashes Craig D. Newgard a,∗ , Roger J. Lewis b,c,d , Jess F. Kraus e,f , K. John McConnell a a
Center for Policy and Research in Emergency Medicine, Department of Emergency Medicine, Oregon Health and Science University, 3181 SW Sam Jackson Park Road, Mail Code CR-114, Portland, OR 97239-3098, USA b David Geffen School of Medicine at UCLA, University of California, Los Angeles, CA, USA c Department of Emergency Medicine, Harbor-UCLA Medical Center, Torrance, CA, USA d Los Angeles Biomedical Research Institute, Torrance, CA, USA e Southern California Injury Prevention Research Center, Los Angeles, CA, USA f School of Public Health, University of California, Los Angeles, CA, USA Received 24 February 2005; accepted 2 March 2005
Abstract Context: Previous studies have suggested that motor vehicle occupants seated on the near-side of a lateral impact have a higher proportion of thoracoabdominal injuries. However, due to limitations in previous studies, the true association between seat position, side of lateral impact, and thoracoabdominal injury is unclear. Objective: To assess the relationship between seat position (i.e., near-side, middle-seat, and far-side, regardless of row), side of lateral motor vehicle crash (MVC), and serious thoracoabdominal injury after adjusting for important crash factors. Design: National population-based cohort of adult subjects involved in MVCs and included in the National Automotive Sampling System Crashworthiness Data System database (NASS CDS) from 1995 to 2003. Patients: Occupants aged ≥16 years involved in MVCs where the highest external deformation of the vehicle was located on the right or left side (i.e., lateral). Main outcome measure: Serious thoracic or abdominal injury, defined as an Abbreviated Injury Scale (AIS) ≥3 in the thoracic or abdominal body region. Results: Fifteen thousand, one hundred and sixty persons involved in primary lateral MVCs were represented in the NASS CDS database during the 9-year period. There were 1867 (2%) persons with serious thoracic injuries and 507 persons (0.5%) with serious abdominal injuries. In multivariable logistic regression models that adjusted for important crash factors and the NASS CDS sampling design, seat position was a strong effect modifier of the association between side of lateral impact and serious thoracic (p < 0.0001) and abdominal (p = 0.0009) injury, with the risk of serious thoracic and abdominal injury highest for occupants seated on the near-side of the crash. The mean probability of injury was higher for near-side and middle-seat occupants compared to far-side occupants, and the probability of thoracic injury was approximately four times higher than that of abdominal injury for all seat positions. Conclusions: There is a strong, synergistic relationship between seat position and side of lateral MVC in assessing risk of serious thoracic and abdominal injury among adult occupants. The probability of serious thoracoabdominal injury increases with increasing proximity of seat position to side of the crash and the risk of thoracic injury is higher than abdominal injury for all seat positions. © 2005 Elsevier Ltd. All rights reserved. Keywords: Thoracic injury; Abdominal injury; Motor vehicle crash
1. Introduction ∗
Corresponding author. Tel.: +1 503 494 1668; fax: +1 503 494 4640. E-mail address: [email protected] (C.D. Newgard).
0001-4575/$ – see front matter © 2005 Elsevier Ltd. All rights reserved. doi:10.1016/j.aap.2005.03.008
Motor vehicle crashes (MVCs) have the potential to produce serious thoracic and abdominal injury. In particular,
C.D. Newgard et al. / Accident Analysis and Prevention 37 (2005) 668–674
lateral crashes (i.e., side impacts) have been implicated as a causal factor in thoracoabdominal injury (Siegel et al., 1993; Dischinger et al., 1993; Siegel et al., 2001; Loo et al., 1996; Horton et al., 2000; McLellan et al., 1996; Kearney et al., 1989; Haland et al., 1993; Pattimore and Dave, 1992; Farmer et al., 1997; Gokcen et al., 1994; McGwin et al., 2003; Franklyn et al., 2002). Several studies have further suggested that seat position (near-side versus far-side) must be considered in the assessment of injury risk in occupants involved in lateral crashes (Dischinger et al., 1993; Horton et al., 2000; McLellan et al., 1996; Haland et al., 1993; Farmer et al., 1997; McGwin et al., 2003; Franklyn et al., 2002). Despite the volume of literature on these associations, definitive conclusions and generalizability is tempered by methodologic limitations in the majority of studies, including restriction to severely injured patients or occupants in certain seat positions (e.g. drivers), limited comparison groups (e.g., subjects in frontal collisions), small sample sizes, difficulty controlling for crash severity, missing variables, and non-uniform criteria for assessing severity of injury. Due to the limitations of previous studies, the true association between seat position and serious thoracic and abdominal injury in lateral MVCs remains unclear. It is also unclear whether one injury type (thoracic versus abdominal) is more common than the other in lateral crashes and what the quantifiable risk is for each injury type by seat position. In this study, our objectives were (1) to assess whether seat position modifies the effect of lateral impact on serious thoracic and (separately) abdominal injury, and (2) to quantify the risk of serious thoracic and abdominal injury by seat position among persons involved in lateral MVCs after adjusting for pertinent crash factors in a national population-based sample.
669
2.2. Patients A national, population-based sample of occupants ≥16 years involved in primary lateral MVCs (i.e., vehicles where the highest external deformation location was located on the right or left side of the vehicle) with passenger vehicles and light trucks were included in the analysis. Highest external deformation located on the side of the vehicle correlated very closely with the highest principal direction of force directed laterally for both right and left lateral crashes. We reasoned that external deformation is easier to assess in the field by pre-hospital personnel than principal direction of force, and that using this inclusion criterion may improve the ability to clinically apply study results. The study excluded children because there was a low incidence of serious thoracic and abdominal injuries in children included in the database and because previous studies have demonstrated a higher risk of injury or death in adults compared with children even after controlling for important crash characteristics (Corneli et al., 2000; Robertson, 1999). 2.3. Main outcome measure Serious thoracic and abdominal injury was defined as an Abbreviated Injury Scale (AIS) ≥3 in these body regions. The AIS for a given injury ranges from 1 (minor) to 6 (nonsurvivable). A score of 3 or more represents a “serious” injury (The Abbreviated Injury Scale, 2001). Because the diagnostic assessment and clinical management of serious thoracic and abdominal injuries differ, all analyses were carried out separately on thoracic and abdominal injury as the outcome measure. 2.4. Effect modifier variable
2. Materials and methods 2.1. Study design and setting We used the National Automotive Sampling System Crashworthiness Data System database (NASS CDS) from 1995 to 2003 for the study. In brief, NASS CDS is a probability-sampled, population-based, nationally representative cohort of persons involved in MVCs that is collected using a three-stage sampling of crashes from specific regions throughout the United States to ensure national generalizability of the data without requiring investigation of every crash in the country (NASS CDS, 2001). The NASS CDS database includes crashes where at least one vehicle was towed due to damage (and was thus available for crash investigation). We selected the 9-year time period because data collection changes for NASS CDS were instituted in 1995. The study was approved by the Institutional Review Boards of Oregon Health and Science University and the Los Angeles Biomedical Research Institute of Harbor-UCLA Medical Center.
Seat position was coded as a polytomous categorical variable, with three categories: left-side (i.e., any occupant seated on the left-side of the vehicle), right-side (i.e., any occupant seated on the right-side), and middle-seat position, regardless of row. Seat position was categorized regardless of row because previous literature has shown fatality risk to be the same for front and rear seat positions in lateral crashes (Evans and Frick, 1988). We confirmed the lack of an association between front- versus rear-seated occupants with these injury patterns by testing the contribution of an additional covariate for front versus rear seat position in the models. Seat position was interacted with side of lateral impact (i.e., left versus right) to test the synergy between these two terms in lateral MVCs. In a second analysis, seat position was combined with side of impact to produce three dichotomous terms (dummy variables): (1) occupants seated on the left and involved in a left-sided impact, (2) occupants seated on the right and involved in a right-sided impact, and (3) occupants seated in the middle and involved in a left or right lateral crash. Far-side occupants (i.e., occupants seated on the far-side of a vehicle
670
C.D. Newgard et al. / Accident Analysis and Prevention 37 (2005) 668–674
involved in a left- or right-sided crash) served as the reference group. 2.5. Patient and vehicle confounders Variables were selected based on previous motor vehicle and biomechanical studies demonstrating association with injury and crash severity (Siegel et al., 1993; Dischinger et al., 1993; Siegel et al., 2001; Loo et al., 1996; Horton et al., 2000; McLellan et al., 1996; Farmer et al., 1997; McGwin et al., 2003; Newgard et al., 2005) as well as factors identified a priori as potentially mediating thoracoabdominal injury in these crashes. Eighteen variables were considered in the analysis, including age (years), gender, pregnancy (any duration of gestation), initial Glasgow Coma Scale (GCS) score (GCS scores of 14 and 15 were grouped together due to often subtle clinical distinctions between the two highest scores), police reported alcohol presence, delta V (change in velocity), rollover with collision, passenger space intrusion (<15, 15–29, 30–45, 46–60, and >60 cm), restraint use (manual lap, lap and shoulder belt, or automatic lap and shoulder belt system versus none), entrapment, ejection, air bag deployment, steering wheel deformity (none, 1–5, 6–10, and >10 cm), vehicle weight (<2500, 2500–3000, or >3000 lbs), vehicle model year, lateral seat position (described above), front versus rear seat position, and side of impact (left versus right) (NASS CDS, 1998). Delta V, age, GCS score, passenger space intrusion, and steering wheel deformity were coded as continuous variables, and the additional 13 variables were coded as categorical variables. Non-collision crashes and pure rollovers were excluded to allow for the appropriate use of delta V to control for crash severity. To account for crashes where the vehicle had multiple impacts, we conducted a secondary analysis on MVCs with primary or secondary lateral impacts (i.e., where right or left side vehicular deformation was either the largest or second largest area of external deformity in the crash). In this paper, we use the term “risk” synonymously with odds because the outcomes (thoracic and abdominal injury) were rare (i.e., ≤2%). Throughout the paper, the number of persons noted is the actual number of persons included in the NASS sample during this time period, while all calculations (i.e., proportions, means, and model estimates) are adjusted for NASS sample design (i.e., accounting for clusters, strata, and weights). 2.6. Statistical analysis After excluding non-collision crashes, pure rollovers, occupants missing injury information, occupants missing seat position information, occupants riding on another passenger’s lap, and occupants riding in a non-enclosed area of the vehicle, the 1995–2003 NASS CDS database contained 54,253 adult occupants. Of this sample, 15,160 persons were involved in primary lateral crashes and
formed the cohort for the primary analysis. There were 21,216 persons involved in primary or secondary lateral impacts. To allow the inclusion of all adult occupants contained in the NASS CDS database during this time period and to preserve the original weighting scheme of the NASS CDS database, we used multiple imputation to impute missing values (Rubin and Schenker, 1991; Barnard and Meng, 1999; Schafer, 1997; Allison, 2001; Lee et al., 1989; Raghunathan et al., 2001; Reiter and Raghunathan, 2002). We imputed missing values using parallel chains of multiple imputation, split by side of lateral impact (left or right), to maximize the statistical efficiency of assessing the interaction term (side of lateral impact × seat position) (Allison, 2001). In addition to the variables noted above, eight variables found to be correlated with the variables of interest were also included in the imputation process: energy absorption, highest deformation extent, other collision vehicle weight, posted speed limit, police-reported travel speed, injury severity scale (ISS) score, injury-related fatality, and AIS ≥3 for any body region. Continuous variables with a skewed distribution were log-transformed for the imputation process. Information for clusters, strata within clusters, and year were included as fixed effects in the imputation process to preserve the complex design features of NASS CDS (Schafer, 1997; Reiter and Raghunathan, 2002). Following the parallel chains of multiple imputation the dataset was re-combined. Ten datasets were created using multiple imputation, each of which were analyzed separately and the results combined, accounting for the variance within each dataset and between datasets to adequately represent the uncertainty in imputing missing values. We first investigated the interaction between seat position and side of lateral impact through multivariable logistic regression models (Jaccard, 2001) that accounted for the complex sampling design (strata, clusters, and weighting) of the NASS CDS database. We considered side of lateral impact (left versus right) the focal independent variable in the interaction term, with seat position (left, right, or middle) as the major effect modifier (Jaccard, 2001). Further adjusting estimates based on clustering within vehicles did not change the results (Hutchings et al., 2003). Variables that did not contribute to either model (i.e., assessing thoracic and abdominal injury, separately) at a level of p < 0.10, that did not significantly alter the association between seat position and injury, and that were not felt to be important confounders were removed from the model in a non-automated, stepwise, backward selection process. The final model included 13 variables. In order to determine the joint significance of the interacted variables (seat position × side of lateral impact), we used the likelihood ratio test (Jaccard, 2001). The −2 log likelihood value for the model without the interaction term was subtracted from the same value for the model with the interaction term, and this difference was compared to a critical χ2 value for α = 0.05 and d.f. = 2 (Jaccard, 2001).
C.D. Newgard et al. / Accident Analysis and Prevention 37 (2005) 668–674
671
Table 1 Occupant and vehicle characteristics among occupants involved in primary lateral impact motor vehicle crashes (n = 15,160)
Mean delta V ± S.E.M. (kmph) Mean age ± S.E.M. (years) Passenger space intrusion ≥30 cm GCS ≤12 Ejection Entrapment Rollover Steering wheel deformity Restraint use Police-reported alcohol use Pregnant Left lateral collision Right lateral collision Left-sided seat Right-sided seat Middle-seat Fatalities
All persons
Persons with serious abdominal injury (AIS ≥3)
Persons with serious thoracic injury (AIS ≥3)
N = 15,160
N = 507
(0.5%)
N = 1867
37.9 38.2 372 194 54 96 52 115 257 101 10 284 223 357 143 7 229
±1.8 ±2.3 (68%) (27%) (9%) (15%) (13%) (20%) (57%) (17%) (3%) (60%) (40%) (67%) (32%) (0.9%) (35%)
17.7 35.4 5031 2216 459 716 1271 1456 10,751 1693 160 7992 7168 11,172 3838 150 983
±0.50 ±0.57 (16%) (9%) (0.5%) (1%) (5%) (4%) (83%) (5%) (0.9%) (53%) (47%) (79%) (20%) (0.9%) (0.7%)
38.7 43.5 1291 632 164 313 191 389 982 395 11 1088 779 1391 460 16 718
Percentage missing
(2%) ±1.3 ±1.0 (67%) (25%) (8%) (19%) (13%) (18%) (54%) (18%) (0.7%) (61%) (39%) (77%) (22%) (0.9%) (26%)
31% 0 35% 31% <1% 2% <1% 2% 2% 10% <1% 0 0 0 0 0 0
Number of persons represents the actual number of occupants sampled for the NASS database during this time period, while all proportions, means and standard errors of the mean are sample-adjusted for cluster, strata, and weighting. S.E.M. = standard error of the mean; kmph = kilometers per hour; cm = centimeters; GCS = Glasgow Coma Scale score; AIS = Abbreviated Injury Scale.
To better quantify the risk of serious thoracoabdominal injury in occupants seated in various seat positions in lateral crashes, we analyzed a second multivariable model that included the same covariates identified above plus three dummy variables that combined seat position and side of lateral impact (i.e., near-side left, near-side right, and middle-seat, with a reference group of far-side occupants), rather than the interaction term. To provide another metric for quantifying the probability of a specific injury type (i.e., thoracic or abdominal) by seat position in lateral crashes, we used the adjusted coefficients generated from the multivariable model in each multiply imputed dataset (collapsing the right and left terms for near-side crashes into a single near-side crash term) and calculated the probability of serious thoracic and (separately) abdominal injury for each subject, by dataset, under three scenarios: (1) all occupants seated on the near-side of the lateral crash, (2) all occupants seated in the middle-seat, and (3) all occupants seated on the far-side of the vehicle. We then calculated sample-adjusted means and 95% confidence intervals for the probability of serious thoracic and abdominal injury for each seat position. All analyses were repeated with both primary and secondary lateral impacts included to ensure consistency with a less restricted sample. Database management was performed using SAS (SAS 8.1, SAS Institute, Cary, NC, USA). SAS-callable IVEware (Survey Methodology Program, Survey Research Center, Institute for Social Research, University of Michigan, MI, USA) was used for multiple imputation and multivariable analyses to account for the complex sampling design of the NASS CDS database. The Jackknife Repeated Replication technique was used to estimate logistic regression parameter variances (Kish and Frankel, 1974).
3. Results There were 15,160 adult occupants (range 16–97 years) involved in primary lateral impacts during the 9-year period and available for analysis. There were 1867 persons (2%) with serious thoracic injuries, including 718 fatalities (26% of patients with serious thoracic injuries), and 507 persons (0.5%) with serious abdominal injuries, including 229 fatalities (35% of patients with serious abdominal injuries). Three hundred and eighty-five persons had both severe thoracic and abdominal injuries. The mean delta V for left-sided versus right-sided crashes was not significantly different (left = 17.1 km/h (kmph), 95% confidence interval (CI) 15.9–18.3; right = 18.5 kmph, 95% CI 17.7–19.3). Patient and vehicle demographics are listed in Table 1. In multivariable logistic regression models that adjusted for important crash factors and the NASS CDS sampling design, seat position was a strong effect modifier of the association between side of lateral impact and serious thoracoabdominal injury (interaction term for near-side lateral crash: for thoracic injury, OR 52.8, 95% CI 9.7–286; for abdominal injury, OR 8.3, 95% CI 2.2–32) (data not shown). The likelihood ratio test also demonstrated the interaction between seat position and side of impact to be highly significant for both injury types (p < 0.0001 and p = 0.0009 for thoracic and abdominal injury, respectively). When seat position and side of impact were combined as covariates, occupants seated on the near-side of the crash were at high risk for both serious thoracic and abdominal injury compared to far-side occupants (Table 2). The mean probabilities of serious thoracic and abdominal injury in nearside and middle-seated occupants were higher than far-side occupants. The probability of serious thoracic injury in lat-
672
C.D. Newgard et al. / Accident Analysis and Prevention 37 (2005) 668–674
Table 2 Odds of injury outcome for selected factors using multivariable logistic regression models for occupants involved in primary lateral impacts (n = 15,160) Outcome
Delta V (per kmph) Passenger space intrusion (per 15 cm) Age (per decade) GCS (per unit decline) Ejection Entrapment Rollover Steering wheel deformity (per 5 cm) Restraint use Police-reported alcohol use Pregnant Far-side seat—left or right lateral crash (reference) Left-side seat—left lateral crash Right-side seat—right lateral crash Middle-seat—left or right lateral crash
Serious abdominal injury (AIS ≥3)
Serious thoracic injury (AIS ≥3)
OR
OR
95%CI
1.06 1.6 1.2 1.1 2.7 1.4 1.8 1.2 0.59 1.4 4.0 1.0 3.2 4.7 2.6
eral crashes was approximately four times higher than serious abdominal injury at each seat position (Fig. 1). These results did not qualitatively change when both primary and secondary lateral impacts were included in the analysis (data not shown). Several additional variables were associated with one or both of the injuries in multivariable models. Each 15 cm increase in passenger space intrusion was associated with an incremental increase in risk of thoracic and abdominal injury, and increasing age was also associated with both thoracic and abdominal injury. Ejection was associated with both injury types, but entrapment was only associated with thoracic in-
Fig. 1. Adjusted mean probability of serious thoracic and abdominal injury by seat position for occupants involved in primary lateral motor vehicle crashes (n = 15,160). Error bars represent the 95% confidence interval around each estimate of the mean probability of serious injury. While the mean probability of injury in the NASS CDS sample is low due to the populationbased sampling and nationally representative injury rates, the probability of injury approaches 1 for persons involved in high energy crashes with many risk factors. The importance of seat position in lateral crashes is demonstrated in the relative difference (rather than absolute difference) in mean probability of injury by seat position.
(1.04–1.08) (1.4–1.9) (1.03–1.4) (1.1–1.3) (1.2–5.9) (0.7–2.5) (0.95–3.3) (0.86–1.7) (0.34–1.03) (0.83–2.2) (0.87–18.5) – (1.8–5.8) (2.9–7.8) (0.54–12.6)
1.09 1.7 1.5 1.2 5.1 3.6 1.9 1.1 0.45 2.0 1.1 1.0 4.6 3.8 2.9
95%CI (1.07–1.11) (1.4–2.0) (1.4–1.7) (1.1–1.4) (2.8–9.2) (1.4–8.9) (0.83–4.2) (0.85–1.5) (0.28–0.73) (1.1–3.7) (0.28–4.3) – (2.8–7.8) (2.2–6.6) (0.86–9.9)
jury. Restraint use was protective against both injury types, but more so for thoracic injury. A lower GCS was also a marker for serious thoracic and abdominal injury in lateral crashes. Vehicle weight, vehicle model year, gender, and front versus rear seat position did not contribute significantly to the models.
4. Discussion In this study, we demonstrate that seat position significantly modifies the association between side of impact and serious thoracic and abdominal injury in lateral crashes. That is, there is a strong, synergistic relationship between seat position and side of lateral MVC in assessing risk of serious thoracic and abdominal injury among adult occupants. This relationship persisted after adjusting for several measures of crash severity (delta V, passenger space intrusion, and steering wheel deformity) and multiple other important crash factors, suggesting that the relationship between seat position and side of impact is not solely explained by crash severity or passenger compartment intrusion. Occupants seated on the near-side or middle-seat of a lateral crash had a higher probability of serious thoracoabdominal injury compared to far-side occupants, with the probability of thoracic injury being higher than that of abdominal injury at all seat positions. While the total number of occupants seated in the middle-seat was small (with consequent loss of precision for estimates of injury risk in multivariable models), the mean probability of injury by seat position suggests that occupants seated in the middle-seat may still be at risk for serious thoracoabdominal injury in lateral crashes. This finding supports a previous study that demonstrated occupants in the middle-seat to have an increased fatality risk relative to far-side occupants in lateral impacts (Evans and Frick, 1988).
C.D. Newgard et al. / Accident Analysis and Prevention 37 (2005) 668–674
Our results may have clinical relevance in the risk assessment of serious thoracic or abdominal injury in persons involved in lateral crashes, as out-of-hospital personnel routinely provide information on pertinent variables from the scene of MVCs to the clinician. In addition to the seat position and side of impact findings described above, each incremental decrease in the GCS score (below a baseline GCS 14–15) was associated with a 24% increase in the odds of serious thoracic injury and a 15% increase in the odds of serious abdominal injury. While the GCS is typically associated with traumatic brain injury or physiologic derangement from injury, it may also serve as a surrogate marker of thoracic and abdominal injury for occupants involved in lateral crashes and may encourage diagnostic imaging of the thoracoabdominal region for a patient involved in a lateral crash who has a depressed mental status. Other readily available variables from the crash scene that may assist in the risk assessment of serious thoracoabdominal injury include degree of passenger space intrusion, increasing age, ejection, entrapment, and lack of restraint use. Clinical intoxication was an independent risk factor for serious thoracic injury, which is consistent with prior studies that have shown alcohol to increase the odds of serious injury and fatality in MVCs (Waller et al., 2003; Bedard et al., 2002). Restraints were protective against both injury types, though this relationship was stronger for thoracic injuries. The finding that restraints were protective in lateral crashes (at least for these two injury types) is in conflict with the belief that seat belts are not effective in this type of crash, as suggested by a previous study (Haland et al., 1993). Restraints may prevent contact with other structures in the occupant compartment (including other occupants) and raise the tolerance to thoracic and abdominal injury in lateral crashes. There may also be engineering applications using these results. As we demonstrate the probability of serious thoracic injury to be approximately four times higher than that of abdominal injury at every seat position in lateral crashes, side impact air bags should be designed accordant with this differential risk. Until the number of vehicles equipped with side impact air bags increases, it will remain unclear whether these devices reduce the risk of serious thoracic or abdominal injury in lateral crashes. There are limitations in our analysis. The NASS CDS dataset contains missing values, requiring the use of multiple imputation to allow the analysis of 100% of the data. To further explore the effects of multiple imputation on the results, we conducted sensitivity analyses for the imputation models using different patterns of missing data (missing completely at random, missing at random, and missing in non-random patterns) for the variables with large percentages of missing data (i.e., delta V, passenger space intrusion, and GCS) in a similar hypothetical dataset, assessing for changes in the results secondary to these missing data patterns. The results remained qualitatively stable, with no indication of systematic bias secondary to using multiple imputation with differing patterns of missingness.
673
We used delta V, passenger space intrusion, and steering wheel deformity as surrogate measures of crash severity. Both delta V and passenger space intrusion had strong, independent associations with serious injury in these analyses. While these measures offer an improved means of adjusting for crash severity, they may not completely control for severity of the crash and their use necessitates the exclusion of subjects involved in crashes for which delta V cannot be calculated (i.e., pure rollovers and non-collision crashes). Due to the construct of our analysis, we were unable to include a covariate for an adjacent, unrestrained occupant in the vehicle, which has been suggested to further increase an occupant’s risk of death in lateral crashes (Cummings and Rivara, 2004). Inclusion of this factor in near-side lateral crashes may further increase an occupant’s risk of injury, though it is unclear whether thoracoabdominal injuries in particular would be increased in this setting. Finally, there were a small number of occupants seated in the middle-seat, which reduced the precision of estimates for injury risk in this group.
5. Conclusions Seat position, as a measure of proximity to the side of lateral impact, is a strong effect modifier in assessing the risk of serious thoracoabdominal injury in lateral MVCs. The synergy between seat position and side of lateral impact persisted after adjusting for crash severity, restraint use, and multiple other important crash factors. Occupants seated on the near-side or middle-seat of vehicles involved in lateral impacts have an increased probability of serious thoracic and abdominal injury when compared to far-side occupants, and the probability of serious thoracic injury is approximately four times that of abdominal injury for all seat positions.
Acknowledgements This project was supported by grant number F32 HS00148 from the Agency for Healthcare Research and Quality and the Society for Academic Emergency Medicine Research Training Grant.
References Allison, P.D., 2001. Missing Data. Sage University Papers Series on Quantitative Applications in the Social Sciences, 07-136. Thousand Oaks, Sage, CA. Barnard, J., Meng, X.L., 1999. Application of multiple imputation in medical studies: from AIDS to NHANES. Stat. Methods Med. Res. 8, 17–36. Bedard, M., Guyatt, G.H., Stones, M.J., Hirdes, J.P., 2002. The independent contribution of driver, crash, and vehicle characteristics to driver fatalities. Accid. Anal. Prev. 34, 717–727.
674
C.D. Newgard et al. / Accident Analysis and Prevention 37 (2005) 668–674
Corneli, H.M., Cook, L.J., Dean, J.M., 2000. Adults and children in severe motor vehicle crashes: a matched-pairs study. Ann. Emerg. Med. 36, 340–345. Cummings, P., Rivara, F.P., 2004. Car occupant death according to the restraint use of other occupants – a matched cohort study. JAMA 291, 343–349. Dischinger, P.C., Cushing, B.M., Kerns, T.J., 1993. Injury patterns associated with direction of impact: drivers admitted to trauma centers. J. Trauma 35, 454–459. Evans, L., Frick, M.C., 1988. Seat position in cars and fatality risk. Am. J. Public Health 78, 1456–1458. Farmer, C.M., Braver, E.R., Mitter, E.L., 1997. Two-vehicle side impact crashes: the relationship of vehicle and crash characteristics to injury severity. Accid. Anal. Prev. 29, 399–406. Franklyn, M., Fitxharris, M., Yang, K., Frampton, R., Morris, A., Fildes, B., 2002. Aortic injuries in side impacts: a preliminary analysis. Annu. Proc. Assoc. Adv. Automot. Med. 46, 113–124. Gokcen, E.C., Burgess, A.R., Siegel, J.H., Mason-Gonzalez, S., Dischinger, P.C., Ho, S.M., 1994. Pelvic fracture mechanism of injury in vehicular trauma patients. J. Trauma 36, 789–795. Haland, Y., Lovsund, P., Nygren, A., 1993. Life-threatening and disabling injuries in car-to-car side impacts—implications for development of protection systems. Accid. Anal. Prev. 25, 199–205. Horton, T.G., Cohn, S.M., Heid, M.P., Augenstein, J.S., Bowen, J.C., McKenney, M.G., Duncan, R.C., 2000. Identification of trauma patients at risk of thoracic aortic tear by mechanism of injury. J. Trauma 48, 1008–1014. 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, 3–8. Jaccard, J., 2001. Interaction Effects in Logistic Regression. Sage University Papers Series on Quantitative Applications in the Social Sciences, 07-135. Thousand Oaks, Sage, CA. Kearney, P.A., Rouhana, S.W., Burney, R.E., 1989. Blunt rupture of the diaphragm: mechanism, diagnosis, and treatment. Ann. Emerg. Med. 18, 1326–1330. Kish, L., Frankel, M., 1974. Inference from complex samples (with discussion). J. R. Stat. Soc. 36, 1–37. Lee, E.S., Forthofer, R.N., Lorimor, R.J., 1989. Analyzing Complex Survey Data. Sage University Papers Series on Quantitative Applications in the Social Sciences, 07-071. Thousand Oaks, Sage, CA. Loo, G.T., Siegel, J.H., Dischinger, P.C., Rixen, D., Burgess, A.R., Addis, M.D., O’Quinn, T., McCammon, L., Schmidhauser, C.B., Marsh, P., Hodge, P.A., Bents, F., 1996. Airbag protection versus compartment intrusion effect determines the pattern of injuries in multiple trauma motor vehicle crashes. J. Trauma 41, 935–951. McGwin, G., Metzger, J., Moran, S.G., Rue, L.W., 2003. Occupant- and collision-related risk factors for blunt thoracic aorta injury. J. Trauma 54, 655–662.
McLellan, B.A., Rizoli, S.B., Brenneman, F.D., Boulanger, B.R., Sharkey, P.W., Szalai, J.P., 1996. Injury pattern and severity in lateral motor vehicle collisions: a Canadian experience. J. Trauma 41, 708– 713. U.S. Department of Transportation, 1998. 1988–1996 NASS CDS Variable-Attribute Structure Manual. U.S. Department of Transportation; National Highway Traffic Safety Administration; National Center for Statistics and Analysis, Crash Investigation Division; Washington, DC. U.S. Department of Transportation, 2001. National Automotive Sampling System (NASS) Crashworthiness Data System Analytical User’s Manual, 2001 File. U.S. Department of Transportation; National Highway Traffic Safety Administration; National Center for Statistics and Analysis; Washington, DC. Newgard, C.N., Lewis, R.J., Kraus, J.F., 2005. Steering wheel deformity and serious thoracic or abdominal injury among drivers and passengers involved in motor vehicle crashes. Ann. Emerg. Med. 45, 43–50. Pattimore, D., Dave, T.P., 1992. Torso injury patterns and mechanisms in car crashes: an additional diagnostic tool. Injury 23, 123–126. Raghunathan, T.E., Lepkowski, Van Hoewyk, J., Solenberger, P.W., 2001. A multivariate technique for multiply imputing missing values using a sequence of regression models. Survey Methodol. 27, 85–95. Reiter, J.P., Raghunathan, T.E., 2002. Multiple imputation for missing data in surveys with complex sample designs. Technical Report. Institute for Social Research, University of Michigan, Ann Arbor, MI. Robertson, L.S., 1999. Reduced fatalities related to rear seat shoulder belts. Injury Prev. 5, 62–64. Rubin, D.B., Schenker, N., 1991. Multiple imputation in health-care databases: an overview and some applications. Stat. Med. 10, 585–598. Schafer, J.L., 1997. Analysis of Incomplete Multivariate Data. Chapman & Hall, New York. Siegel, J.H., Loo, G., Dischinger, P.C., Burgess, A.R., Wang, S.C., Schneider, L.W., Grossman, D., Rivara, F., Mock, C., Natarajan, G.A., Hutchins, K.D., Bents, F.D., McCammon, L., Leibovich, E., Tenenbaum, N., 2001. Factors influencing the patterns of injuries and outcomes in car versus car crashes compared to sport utility, van, or pick-up truck versus car crashes: crash injury research engineering network study. J. Trauma 51, 975–990. Siegel, J.H., Mason-Gonzalez, S., Dischinger, P., Cushing, B., Read, K.R., Robinson, R., Smialek, J., Heatfield, B., Hill, W., Bents, F., Jackson, J., Livingston, D., Clark, C.C., 1993. Safety belt restraints and compartment intrusions in frontal and lateral motor vehicle crashes: mechanisms of injuries, complications, and acute care costs. J. Trauma 5, 736–759. The Abbreviated Injury Scale, 1990 Revision, Update 98, 2001. Association for the Advancement of Automotive Medicine. Waller, P.F., Hill, E.M., Maio, R.F., Blow, F.C., 2003. Alcohol effects on motor vehicle crash injury. Alcohol Clin. Exp. Res. 27, 695–703.