|Home | About | Journals | Submit | Contact Us | Français|
Abbreviations: CI, confidence interval; ISS, injury severity score; NASS-CDS, National Automotive Sampling System Crashworthiness Data System; OR, odds ratio.
Prior studies on racial and ethnic disparities in survival after motor vehicle crashes have examined only population-based death rates or have been restricted to hospitalized patients. In the current study, we examined 3 components of crash survival by race/ethnicity: survival overall, survival to reach a hospital, and survival among those hospitalized. Nine years of data (from 2000 through 2008) from the National Automotive Sampling System Crashworthiness Data System were used to examine white non-Hispanic, black non-Hispanic, and Hispanic drivers aged ≥15 years with serious injuries (injury severity scores of ≥9). By using multivariable logistic regression, we found that a driver's race/ethnicity was not significantly associated with overall survival after being injured in a crash (for blacks, odds ratio (OR) = 0.69, 95% confidence interval (CI): 0.36, 1.32; for Hispanics, OR = 1.00, 95% CI: 0.59, 1.72), and blacks and Hispanics were equally likely to survive to be treated at a hospital compared with whites (for blacks, OR = 1.00, 95% CI: 0.52, 1.93; for Hispanics, OR = 1.13, 95% CI: 0.71, 1.79). However, among patients who were treated at a hospital, blacks were 50% less likely to survive 30 days compared with whites (OR = 0.50, 95% CI: 0.33, 0.76). The disparity in survival after serious traffic injuries among blacks appears to occur after hospitalization, not in prehospital survival.
There is growing evidence that racial and ethnic disparities exist in outcomes after motor vehicle crashes and trauma care (1–3). Although it is well established (4–6) that blacks and Hispanics have higher crash-related fatality rates in the United States than whites, such studies do not differentiate among the risks of being in a crash, being injured in a crash, or surviving after being injured. Our study attempts to isolate factors affecting survival after being injured in a crash, which may relate to emergency medical services or trauma care.
Four prior studies (1–3, 7) have reported an increased risk of in-hospital death among blacks after injury or motorcycle crash, although one (8) reported no increased risk after vehicular trauma admission. Hospital mortality after injury was found to be 14%–58% higher among blacks than whites; suggested explanations include quality of care, injury severity, and insurance status, among others (1–3). Hospital studies often do not have information on crash-related variables that may affect survival, such as seat belt use, vehicle speed, ejection, or rollover (1–3, 8–10). In addition, although informative for hospitalized patients, such studies do not include the large proportion of deaths that occur before reaching the hospital. To date, no studies have examined whether injured blacks and Hispanics are more or less likely to die at the scene or be transported to a hospital compared with whites. We therefore used national, publicly available data to examine 3 components of crash survival by race/ethnicity: probability of overall survival, probability of transport to a hospital, and probability of survival among those hospitalized.
We used 9 years of data (from 2000 through 2008) from the National Automotive Sampling System Crashworthiness Data System (NASS-CDS) for the analysis. NASS-CDS data are maintained by the National Highway Traffic Safety Administration and are a probability sample of all crashes in the United States involving light motor vehicles (passenger cars, light trucks, or vans) towed due to damage. Field teams in 27 primary sampling units across the country collect detailed information, including vehicle damage and occupant injuries, on selected crashes. Field researchers examine police reports and medical records, inspect the crash site and vehicles, and conduct interviews. Data are weighted to enable national estimates. NASS-CDS data are provided in Statistical Analysis System format on the NASS-CDS website (11). We restricted the data to drivers because NASS-CDS collects race and ethnicity information on the driver only, not on other vehicle occupants.
The race and ethnicity of the driver were classified by the NASS-CDS from 2000 through 2006 as follows: white non-Hispanic, black non-Hispanic, Hispanic, Asian, American Indian, and other races. Starting in 2007, race and ethnicity were collected as separate variables, which we combined to create mutually exclusive groups to align with the previous data. In our discussion, the term “whites” will refer to non-Hispanic whites, and “blacks” will refer to non-Hispanic blacks. Although race and ethnicity are separate constructs, we will use the term “race/ethnicity” as recommended previously (12) to describe the 3 groups in our analysis: non-Hispanic whites, non-Hispanic blacks, and Hispanics. We restricted the analysis to whites, blacks, and Hispanics because the sample was too small to examine other groups.
The NASS-CDS variables of interest included age, sex, body mass index, time of crash, whether the vehicle rolled over or the occupant was ejected from the vehicle, seat belt use, travel speed, adverse weather, type of medical facility, length of stay, alcohol use, occupant injury severity, injured body region, and death. We defined “seat belt use” as either field researcher-reported or police-reported seat belt use. We defined “high speed” as traveling 50 miles per hour or greater (1 mile = 1.6 km), on the basis of information in the police report or, if missing, when the posted speed limit was >50 miles per hour. “Adverse” weather at the time of the crash included rain, sleet, hail, snow, fog, or severe crosswinds. NASS-CDS field researchers classified the hospital where an occupant was treated as either a hospital or trauma center by using established criteria to define “trauma center” (13). Treatment at a hospital included emergency department care, intensive care, or hospital admission (and excluded those who died en route to the hospital). Time of the crash was grouped into 2 categories (midnight to 6 am vs. all other times) to examine overnight times that may elicit slower emergency response. We categorized the driver as having used alcohol if the police report indicated that the driver had consumed alcohol or if the blood alcohol test for the driver was positive. The NASS-CDS defines length of hospital stay based on the number of overnights; however, subjects who die on the day of admission are given a length of stay of 1 day.
Field researchers assigned an Abbreviated Injury Scale score to each occupant injury (14). Injury Severity Scores were calculated in the NASS-CDS by adding the squares of the Abbreviated Injury Scale scores for the 3 most severely injured body regions. We categorized Injury Severity Score (ISS) into established cutpoints (<9, 9–15, 16–24, ≥25) and defined seriously injured subjects as those with an ISS ≥9 (15). We also created variables for injuries to the head, thorax, and abdomen with an Abbreviated Injury Scale score ≥4 to look specifically at severe injuries to those body regions. Death was defined as occurring within 30 days of the crash and was determined by NASS-CDS field researchers on the basis of all available sources (police report, medical records, autopsy report, and other official records).
All analyses were conducted using SAS, version 9.1, software (SAS Institute, Inc., Cary, North Carolina) by using survey logistic procedures. Sampling weights and cluster and strata effects were included to account for the complex sampling design of the NASS-CDS. We excluded drivers with missing information on race/ethnicity as well as those <15 years of age and those who died prior to the crash. Because we were interested in postinjury factors affecting risk of death (and not the risk of injury itself), we included only drivers who were at least seriously injured (ISS ≥9) and were therefore at risk of death. We constructed 3 logistic regression models examining driver race/ethnicity among those seriously injured: 1) overall odds of survival; 2) odds of being treated at a hospital (vs. dying at the scene or en route); and 3) odds of survival among those treated at a hospital. We then repeated the analyses among those with more severe injuries (ISS ≥16) to see if results were consistent. We evaluated covariates for inclusion in the multivariate model if they were associated with either race/ethnicity or survival at P < 0.20 (P value chosen to be inclusive of potential confounders). We then retained all covariates that were either significantly (P < 0.05) associated with the outcome in adjusted models or that produced a 10% or greater change in the association between race/ethnicity and the outcome. For ease of interpretation, if a variable was significant in 1 model, we included it in all 3 models.
We compared the characteristics of those with and without information on race/ethnicity and then conducted a sensitivity analysis using multiple imputation to impute missing race/ethnicity values. To create the imputation model, we included all covariates in Table 1 as well as sampling design variables (cluster and strata) using the MI and MIANALYZE procedures in SAS, version 9.1, software (SAS Institute, Inc.). Finally, to isolate early in-hospital deaths from deaths occurring later, we examined model 3 in 2 parts: survival to day 2 and survival beyond day 2.
Among the 61,057 eligible drivers aged ≥15 years from 2000 through 2008, 9,363 (2.8% weighted) had serious injuries (Figure 1). We excluded subjects with missing race/ethnicity information (n = 2,612) and those not white, black, or Hispanic (n = 275), leaving a total of 6,476 drivers. We excluded drivers with any missing covariates (n = 615), yielding a final sample size of 5,861 (representing 513,662 seriously injured drivers in the weighted analysis).
Characteristics of the study population by race/ethnic group are presented in Table 1. The majority of injured drivers were male, with significantly more males among blacks (65%) and Hispanics (70%) than among whites (58%). White injured drivers were older than injured drivers in other groups. Significantly more crashes occurred from midnight to 6 am among blacks (27%) and Hispanics (33%) than among whites (20%). Roughly half of all drivers were not wearing seat belts. Blacks and Hispanics were significantly more likely than whites to have severe thorax injuries. Approximately 92% of all 3 groups survived to be treated at a hospital (vs. dying at the scene/in transit).Vehicle rollovers, being ejected from the vehicle, high-speed crashes, alcohol use, and the ISS did not vary significantly by race/ethnicity. Among those treated at a hospital, similar percentages of whites and blacks (78% and 81%, respectively) but fewer Hispanics (58%) were treated at trauma centers. The majority of covariates in Table 1 were significantly associated with mortality.
Overall, 4,539 (87.8%) drivers with serious injuries (ISS ≥9) survived 30 days. Crude weighted survival rates among whites, blacks, and Hispanics were 88.3%, 84.8%, and 88.1%, respectively. Multivariate models of overall survival (model 1), surviving to be treated at a hospital (model 2), and surviving once treated at a hospital (model 3) are shown in Table 2. Blacks and Hispanics did not have significantly lower risk of overall survival compared with whites (model 1: for blacks, odds ratio (OR) = 0.69, 95% confidence interval (CI): 0.36, 1.32; for Hispanics, OR = 1.00, 95% CI: 0.59, 1.72). Drivers who were older, who were ejected from the vehicle, who were involved in high-speed crashes, who had consumed alcohol, who had more severe injuries, or who had severe injuries to the thorax or abdomen had significantly lower odds of overall survival. In model 2, whites, blacks, and Hispanics all had similar odds of being treated at a hospital versus dying at the scene or in transit (for blacks, OR = 1.00, 95% CI: 0.52, 1.93; for Hispanics, OR = 1.13, 95% CI: 0.71, 1.79). Drivers with a body mass index ≥30, those ejected from the vehicle, those whose crashes occurred late at night or at high speed, those who had consumed alcohol, and those with the most severe injuries (ISS ≥25) or with injuries to the thorax or abdomen were all less likely to reach the hospital alive.
In model 3, restricting the analysis to those treated at a hospital, we found that blacks were 50% less likely to survive (OR = 0.50, 95% CI: 0.33, 0.76), compared with whites (Table 2). Hispanics did not have a lower risk of survival. Other factors associated with lower odds of survival once treated at a hospital included age, ejection from vehicle, injury severity, and severe head or abdominal injuries. We repeated the analysis among those with an ISS ≥16 and findings were similar. Finally, we examined model 3 in 2 steps: survival to day 2 in the hospital and 30-day survival among those who lived to day 2. Blacks were significantly less likely to survive at both time points (OR = 0.55, 95% CI: 0.34, 0.88, and OR = 0.42, 95% CI: 0.23, 0.76, respectively). Adjustment by all additional covariates in Table 1 did not alter this association.
When we compared the characteristics of subjects missing information on race/ethnicity with those with complete information, we found that the missing group did not differ significantly by age, seat belt use, ejection from vehicle, crash time, alcohol use, vehicle rollover, high-speed crash, or treatment at a hospital. However, those with missing information on race were significantly more likely to be male and to be more severely injured, and less likely to survive (all P = 0.02). To evaluate potential bias from missing values, we used multiple imputation to impute missing values on race (n = 2,612) and then repeated the analysis. Results were substantively unchanged (Table 3), with blacks still significantly less likely to survive among those treated at a hospital (OR = 0.60, 95% CI: 0.38, 0.95).
Using 9 years of data from a national probability sample of crashes in the United States, we found that a driver's race/ethnicity had no overall association with the odds of survival after being seriously injured in a traffic crash. Blacks and Hispanics were also no less likely to survive to be treated at a hospital versus dying at the scene or in transit compared with whites. However, among those who were treated at a hospital, blacks were 50% less likely to survive 30 days compared with whites.
Since it was reported in 2003 that certain racial and ethnic groups receive lower-quality health care in the United States (16), multiple studies have examined potential racial/ethnic disparities in motor vehicle and traumatic injury mortality. Although motor vehicle–related death rates per 100,000 people are similar among whites, blacks, and Hispanics overall (17), studies that account for miles traveled have found that blacks and Hispanics generally have higher crash-related mortality than whites. Baker et al. (4) found that death rates per billion vehicle miles traveled among 13- to 19-year-olds were highest among Hispanics, followed by blacks and whites. In a study using the Fatality Analysis Reporting System, Braver et al. (5) found that both blacks (relative risk = 1.34) and Hispanics (relative risk = 1.21) were more likely than whites to die when traveling in a motor vehicle. Although such studies suggest a racial disparity in the risk of dying in a motor vehicle crash, they do not differentiate among the risk of being in a crash, being injured in a crash, or surviving after being injured.
Our results are consistent with those of 4 prior studies that found increased risk of in-hospital death among blacks after injury (1, 3, 7) or after motorcycle crash (2) but differ from 1 study that found no such association (8). Analyzing injury diagnoses from the Healthcare Cost and Utilization Project Nationwide Inpatient Sample, Arthur et al. (1) found that blacks had a significantly (14%) higher risk of in-hospital death compared with whites after adjustment for injury severity, hospital type, insurance, comorbidity, and income. Similarly, using hospital registry data from the National Trauma Data Bank, Haider et al. (3) observed a higher risk of hospital mortality among injured (ISS ≥9) blacks and Hispanics (ORs = 1.17 and 1.47, respectively) after adjustment for demographic, insurance, injury, and comorbidity data. Also using the National Trauma Data Bank, Crompton et al. (2) found that among injured motorcyclists, blacks were 58% more likely than whites to die during hospitalization. Finally, among patients with traumatic brain injury, Bowman et al. (7) found that blacks had significantly higher in-hospital mortality. One study examining vehicular trauma by using Florida hospital discharge data (8) found that race/ethnicity was not associated with higher in-hospital mortality, although the inclusion of only a single state prevents generalizing results to the entire country.
All 4 hospital studies share the same limitation: Analyses are restricted to patients who are treated at a hospital and therefore do not include any information on prehospital mortality or crash-related factors. Because a high proportion of crash-related deaths occur at the scene or in transit (63% in our data), it is important to examine the full timeline when deaths may occur. We did so in the current study and found no disparity in survival to reach a hospital, in contrast with survival among those who were treated at a hospital.
A number of factors have been suggested that could explain the disparity among blacks in injury outcome, including comorbidities, socioeconomic status, insurance, seat belt use, and hospital location/effects. We will briefly discuss each of these factors and how it relates to our analysis. First, blacks have higher rates of comorbidities than whites, including hypertension, diabetes, cardiovascular disease, and obesity (18, 19), and preexisting comorbidities increase the risk of death after injury (20). Tepas et al. (8) found that having ≥2 comorbidities was significantly associated with higher mortality after a crash, and race was not independently significant in their data. Like 2 of the hospital-based studies (2, 3) reporting an increased risk of death among blacks, our study was unable to assess comorbidity. If blacks had significantly worse health at baseline, this could potentially explain our observed result among hospitalized patients.
Second, socioeconomic status and insurance status are complicated and potentially related variables affecting mortality risk. Low-income (1) and uninsured (3, 8) patients have a higher risk of in-hospital death after injury, by which mechanism it is still unclear. In the study by Haider et al. (3), uninsured status was a stronger predictor of death than was race and reduced the race association considerably, but not fully. Crompton et al. (2) found that the uninsured were significantly more likely to die than insured patients, and race was not independently significant. This raises questions about reduced access to care, quality of care, and, possibly, risk-taking behavior among these groups. Although we could not assess socioeconomic status or insurance directly, we examined a number of “risk-taking” factors such as alcohol involvement and seat belt use, and although alcohol use was negatively associated with overall survival, adjusting for these variables did not alter the principal results. Still, it must be considered that residual confounding by socioeconomic status and insurance status may play a role in our observed result, and that these factors may be more important than race as independent risk factors for mortality following motor vehicle crashes.
Finally, it should be considered that worse outcomes among blacks may be related to the hospitals where they receive care. Using Medicare data, Lucas et al. (21) found that black patients were more likely to undergo surgery in hospitals with higher mortality rates, low procedure volume, and high proportions of black patients. Adjusting for the specific hospital attenuated the increased risk of postsurgical mortality after most surgical procedures. A recent study (22) also found that trauma patients treated at hospitals where >25% of patients are black and Hispanic had an increased risk of death. Bazarian et al. (23) found that after a mild traumatic brain injury, blacks were 3 times as likely to receive emergency department care from a resident (vs. a staff physician), a potential marker of quality of care. MacKenzie et al. (24) found that injured patients treated at trauma centers have significantly better in-hospital and 1-year outcomes than those treated at hospitals without trauma centers. We did not have information on specific hospitals in the current study, although trauma center status was evaluated as a potential covariate and did not alter results in the final model, and blacks were treated at trauma centers at the same rate as whites. Further study is needed to examine if the type and quality of hospitals where certain racial groups are treated affect injury outcome.
Our study has several strengths. We were able to analyze all drivers who were involved in a crash, not just those who were admitted to a hospital, allowing us to estimate the odds of being transferred to a hospital versus dying at the scene. We were able to analyze a large number of crash-related factors collected by trained field researchers in a standardized manner. The NASS-CDS also includes all deaths within 30 days of injury, which helps to reduce bias that may result from counting only deaths occurring during the hospital stay. Finally, the NASS-CDS is a national probability sample and should be representative of the US population.
Several limitations should be considered. We did not have information on comorbidity, insurance, or socioeconomic status of drivers and may have misclassified alcohol use if an alcohol test was not conducted on the driver. In addition, the concepts of race and ethnicity are complex and inherently imprecise, and assigning categories obscures substantial heterogeneity within groups (12). This imprecision can lead to nondifferential misclassification of exposure and attenuation of results. Race/ethnicity information was assigned by NASS-CDS field researchers from official records, and we are limited to its quality and completeness. Indeed, the most challenging aspect of our analysis is the high rate of missing race/ethnicity data (28%). Although drivers with missing race/ethnicity data did not differ on the majority of covariates, they were more likely to be male and more severely injured and less likely to survive. Our results should therefore be interpreted with caution. It is reassuring that, when we conducted a sensitivity analysis using multiple imputation of race/ethnicity values, results were consistent with the primary results. Ultimately, however, our analyses should be repeated with more complete data evaluating the full timeline from crash to hospital. It should also be noted that our analysis was restricted to those with serious injuries and should not be generalized to those with minor injuries (ISS <9).
Other limitations of the NASS-CDS data should be highlighted briefly. Delta-V, a measure often used as a marker for crash severity, is missing for 49% of cases (25). There is a high degree of variation in the sampling weights in the NASS-CDS, so concerns may arise when analyses with small subpopulations are conducted (26); 1 case with a large sampling weight may unduly influence the results. Even with 9 years of data, sample sizes are somewhat small, and standard errors can be large (27). Our small sample size precluded the ability to analyze groups other than whites, blacks, and Hispanics.
In conclusion, among those seriously injured in a motor vehicle crash, blacks and Hispanics were equally likely as whites to survive to reach a hospital, but blacks were significantly less likely to survive among those treated at a hospital. To our knowledge, this is the first study examining the odds of prehospital survival among whites, blacks, and Hispanics. We observed that the disparity in crash outcomes occurs among those treated at a hospital, not among those who die before reaching a hospital. Further work is needed to examine injury outcomes among hospitalized patients with a full spectrum of covariate data, such as comorbidity, socioeconomic status, and hospital quality characteristics.
Author affiliations: Center for Outcomes Research and Evaluation, Maine Medical Center Research Institute, Maine Medical Center, Portland, Maine (Amy Haskins, David Clark, Lori Travis).
This study was supported by grant R21HD061318 from the National Institutes of Health.
Conflict of interest: none declared.