Evaluation and Comparison of Roadside Crash Injury Metrics (2023)

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

61 4 Building the IAD The purpose of this chapter is to describe the construction of the IAD to evaluate the crash severity metrics. Four available in-depth crash databases, NASS/CDS, CISS, CIREN, and SCI, were examined for cases to be included in the IAD. A high-level summary of these datasets is provided in Table 4-1. A subset of the available cases from each of these four datasets have corresponding vehicle EDR data available from the VT EDR database. A fifth dataset, the NTDB, was also examined for possible inclusion, but the NTDB does not contain the requisite vehicle kinematics data to evaluate the candidate injury metrics. Instead, the NTDB was used to provide a more detailed quantification of the injuries typical of single vehicle crashes. Table 4-2 provides a summary of the cases available from each dataset for the IAD by impact type. Table 4-1. Real-world crash data sources to evaluate candidate injury metrics. Data Source Nationally Representative? Purpose Notes NASS/CDS + EDR Y Training dataset Detailed vehicle and occupant injury data for ~5,000 crashes per year. EDR data available for a subset of cases. NASS/CDS 2015 was last year collected; replaced by CISS. CISS + EDR Y Test dataset Detailed vehicle and occupant injury data for ~2,500 crashes per year. EDR data available for a subset of cases. Available 2017 onward (2019 latest year available, ~7500 total cases). CIREN + EDR N Potential test dataset Detailed injury and injury causation data for ~300 cases/year. EDR data available for a subset of cases. Insufficient cases available for validation. SCI + EDR N Potential test dataset In-depth crash data for special crash situations or vehicle technologies. ~150 cases/year. EDR data available for a subset of cases. Insufficient cases available for validation. NTDB Y Tabulation of roadside crash-related injuries Detailed injury/hospital data but very limited crash data (and no vehicle kinematics data). Table 4-2. Cases available for the IAD by impact type. Dataset Impact Type Front Side Unweighted Weighted Unweighted Weighted Training (NASS/CDS + EDR) 1,579 869,177 335 213,987 Test (CISS + EDR) 824 592,697 385 361,327 All Impact Types Potential Test (SCI + EDR) 113 Potential Test (CIREN + EDR) 225

62 In-Depth Crash Datasets Matched with EDR Data 4.1.1 Background NASS/CDS, CISS, CIREN, and SCI are all in-depth databases, containing detailed occupant injury data as well as EDR data once merged with the VT EDR database. NASS/CDS and CISS are nationally representative databases comprising tow-away crashes in the United States. CISS is the follow-on database to NASS/CDS and has altered primary sampling unit locations to better reflect U.S. population shifts. The last year of NASS/CDS was collected in 2015, and currently CISS exists for 2017, 2018, and 2019. CIREN is a collection of cases investigated by five Level 1 trauma centers. CIREN is a clinical sample of people who enter a Level 1 trauma center and therefore is not nationally representative of the U.S. population. SCI is a collection of special crash investigations also not nationally representative of the U.S. population, as case selection is based on U.S. DOT priorities and may include investigation of urgent crash safety issues. 4.1.2 Case Selection The IAD included NASS/CDS sampled crashes with EDR data from 2006 to 2015 (Radja 2015) and CISS sampled crashes with EDR data from 2017 to 2019. For an occupant to be included in the IAD, they had to meet the following criteria: • The general area of damage (GAD) from the crash was to the front or side of the vehicle to select only frontal, side, and oblique crashes. • The crash must have only one recorded event. In multi-event crashes, it can be difficult to determine which event caused the injuries. • Only passenger vehicles were included. Passenger vehicles include cars, light trucks (includes pick-up trucks and SUVs), and vans. The body type of the other vehicle in the crash was not restricted. • Only drivers and right front passengers were considered. • Occupants needed to be at least 13 years old to restrict the dataset to “adult size” occupants (Bareiss and Gabler 2020). • Vehicle intrusion could not exceed the MASH intrusion limits (AASHTO 2016). The location of the intrusion was used to determine whether a driver or right-front passenger should be removed from the dataset. If intrusion exceeded MASH limits only at the location of the right-front passenger seat, only the passenger was removed from the analysis. If intrusion exceeded MASH limits only at the location of the driver seat, only the driver was removed from the analysis. • Cases were included only if the occupant was not ejected, as ejected occupants do not benefit from the enclosed vehicle compartment. • Cases were included only if the vehicle did not roll over, as occupants in rollovers experience different forces due to the motion of the vehicle.

63 • Vehicles needed to have a recorded airbag system deployment or a longitudinal delta-v of at least 5 mph (Scanlon 2016). These are the criteria that cause the EDR to lock in the data, preventing them from being overwritten (Gabler et al. 2020). • The EDR recorded the occupant belt status. • The longitudinal and lateral crash pulse data were visually inspected for every case to ensure the pulses were complete. For example, cases were excluded if one of the crash pulses did not appear to capture the entire crash event. Additionally, time points before time zero and trailing zero points at the end of the crash pulse were removed to accurately calculate the crash severity metrics. Cases were also excluded if the crash pulse appeared to not accurately capture the crash event. For example, if the recorded delta-v crash pulse did not show a general trend of reduction in vehicle velocity, it was assumed to be a faulty recording. • Only cases with a case weight less than 5,000 were included to remove unreasonable case weights (Kononen 2011). In the test dataset, three unweighted cases, comprising 0.5% of the unweighted dataset, were removed because their case weights comprised over 10% of the weighted data. There were no cases removed in the training dataset. First, driver and right-front passenger NASS/CDS, CISS, and EDR data were extracted from frontal and side single-event passenger vehicle crashes. This yielded 1,914 sampled occupants. Following this initial selection, the remaining inclusion criteria were applied. Table 4-3 tabulates the number of occupants remaining as each inclusion criterion was applied to the dataset. A total of 924 cases were removed. All vehicles in the final dataset were equipped with frontal airbags, as only vehicles with frontal airbags have EDRs (airbag control modules). Table 4-3. Inclusion criteria for the final IAD. Inclusion Criterion Occupants Remaining NASS/CDS (Training) CISS (Test) Sampled Weighted Sampled Weighted Drivers and right-front passengers in single-event crashes with GAD at the front or side of the vehicle 1,914 1,083,164 1,209 954,024 Occupant age is at least 13 years old 1,867 1,060,998 1,199 946,645 Occupant was not ejected 1,747 950,951 1,198 946,555 Vehicle did not roll over 1,747 950,951 1,198 946,555 MASH intrusion limits were not exceeded 1,719 945,903 1,102 894,205 Complete lateral and longitudinal delta-v crash pulses 1,066 517,605 994 820,798 EDR belt status recorded 1,030 498,218 983 812,132 Airbag system deployed or vehicle had a longitudinal or lateral MDV of at least 5 mph 990 348,416 959 795,274 Case weight is less than 5,000 990 348,416 932 573,223 Injury Assessment Dataset (IAD) 990 348,416 932 573,223 4.1.3 Classifying Crash Type Subsets The OIV portion of the FSM was used to assess occupant motion in each case to classify the crash-involved occupant as a frontal, side, or oblique impact. OIV is the velocity of an occupant

64 as they exit the simplified vehicle occupant compartment boundaries as defined by the FSM (Mitchie 1981a). These boundaries are 0.6 m in the longitudinal direction and 0.3 m in the lateral direction. Both the longitudinal and lateral EDR crash pulses were examined to determine when/if the occupant crossed an FSM boundary. The following paragraphs describe how this information is used to classify impact type. The cases with frontal damage were categorized into one of five frontal categories (FC) based on the occupant’s motion relative to the FSM boundaries (Table 4-4). Cases in categories FC1, FC4, and FC5 have been included in the final frontal model analysis. FC2 and FC3 cases were not included because the lateral motion as the result of the crash dominated the motion of the occupant relative to the FSM boundaries. FC3 cases were not included in the side crash analysis due to the GAD being at the front of the vehicle. The FC1 cases have also been included in the oblique crash model. The FC5 cases were included in the oblique crash model depending on the specific horizontal location (SHL) of the damage. In these cases, the SHL needed to be at the front-left, front-right, or side-front of the vehicle. The sum of the cases in categories FC1, FC4, and FC5 was 717. At this point in the analysis, the remaining inclusion criteria were applied. There were 494 sampled occupants, representing 167,330 real-world frontal crashes, used to construct the models. Of these 494 occupants, seven of them (1.4%) had a lateral MDV larger than the longitudinal delta-v. Table 4-4. Five possible categories for the cases with frontal damage. Category Description OIV Definition Occupants Models FC1 Crossed longitudinal boundary before lateral boundary Longitudinal velocity differential at time of crossing longitudinal boundary 66 Frontal; Oblique FC2 Crossed lateral boundary before longitudinal boundary Longitudinal velocity differential at time of crossing lateral boundary 29 Oblique FC3 Did not cross longitudinal boundary; crossed lateral boundary Longitudinal velocity differential at time of crossing lateral boundary 14 -- FC4 Did not cross lateral boundary; crossed longitudinal boundary Longitudinal velocity differential at time of cross longitudinal boundary 433 Frontal FC5 Did not cross either boundary Longitudinal maximum delta-v 175 Frontal; Oblique (depending on SHL) The cases with side damage were categorized into one of five side categories (SC) based on the occupant’s motion relative to the FSM boundaries (Table 4-5). Cases in categories SC1, SC4, and SC5 have been included in the final side model analysis. SC2 and SC3 cases were not included because the longitudinal motion as the result of the crash dominated the motion of the occupant relative to the FSM boundaries. SC3 cases were not included in the side crash analysis due to the GAD being at the side of the vehicle. The SC1 cases have also been included in the oblique crash model. The SC5 cases were included in the oblique crash model depending on the SHL of the damage. The sum of the cases in categories SC1, SC4, and SC5 was 273. At this point in the analysis, the remaining inclusion criteria were applied. There were 183 sampled occupants, representing 66,124 real-world side crashes, used to construct the models. Of these 183 occupants, 41 of them (22%) had a longitudinal MDV larger than the lateral delta-v.

65 Table 4-5. Five possible categories for the cases with side damage. Category Description OIV Definition Occupants Models SC1 Crossed lateral boundary before longitudinal boundary Lateral velocity differential at time of crossing lateral boundary 32 Side; Oblique SC2 Crossed longitudinal boundary before lateral boundary Lateral velocity differential at time of crossing longitudinal boundary 2 Oblique SC3 Did not cross lateral boundary; crossed longitudinal boundary Lateral velocity differential at time of crossing longitudinal boundary 49 -- SC4 Did not cross longitudinal boundary; crossed lateral boundary Lateral velocity differential at time of cross lateral boundary 73 Side SC5 Did not cross either boundary Lateral maximum delta-v 117 Side; Oblique (depending on SHL) Following the classification of each case, the frontal, side, and oblique crash IADs were each filtered through a unique set of inclusion criteria (Table 4-6). Frontal crashes contained frontal damage and the occupant motion was indicative of a frontal crash (i.e., dominated by the longitudinal motion of the vehicle). Only cases where the occupant crossed the longitudinal FSM boundary before the lateral boundary, crossed the longitudinal boundary only, or crossed neither boundary were included in this frontal model. The principal direction of force (PDOF) must have been between 0° +/- 40°, including 0 and 40. 40° was the chosen threshold because the majority of the cases fall within this range. The few cases with a PDOF greater than 40° are likely sideswipe crashes. Cases with OLC = 0 were excluded if the occupant suffered a serious (MAIS2+F) injury. When OLC = 0, it is often because the occupant did not translate far enough forward in their seat for OLC to be calculated. This should indicate a low severity crash. Therefore, OLC = 0 cases that resulted in MAIS2+F injuries may be due to a faulty case. All predictor variables were also required to be known. Side crashes contained side damage and the occupant motion was indicative of a side crash (i.e., dominated by the lateral motion of the vehicle). Only cases where the occupant crossed the lateral FSM boundary before the longitudinal boundary, crossed the lateral boundary only, or crossed neither boundary were included in this side model. All predictor variables were also required to be known. Oblique crashes contained either frontal or side damage and the occupant motion was indicative of an oblique crash. An occupant must have crossed both FSM or neither FSM boundary to be included in the analysis.

66 Table 4-6. Additional inclusion criteria for the final frontal, side, and oblique crash datasets. Restriction Frontal Side Oblique GAD Front Side Front or Side FSM Boundary Crossed Longitudinal Boundary Only or First Crossed Lateral Boundary Only or First Crossed Both or Neither Boundaries PDOF < ± 40° - - Minor Impacts If OLC =0, then MAIS < 2 - - SHL - - If neither FSM boundary was crossed, SHL was front left, front right, or at the front side of the vehicle Predictor Variables Known Age, PDOF Impact Type GAD 4.1.4 Results 4.1.4.1 Composition of the Frontal Crash IAD Subset After filtering the data in accordance with the inclusion criteria, 494 sampled occupants (167,330 weighted occupants) comprised the dataset. This dataset was used to build the injury risk models for each candidate injury metric. Table 4-7 summarizes the occupants who comprise the dataset by their predictor variable categories. There were 489 occupants with region-specific injury information available to train the logistic regression models. A total of 490 cases were available to train the linear regression Harm models. The test dataset did not require available data for every initial covariate, as they were not all included in the final models. As a result, data for some of the initial covariates are unknown for occupants in the test dataset. The associated cumulative distribution function plots for PDOF and the metrics can be found in Appendix C.

67 Table 4-7. Training and test dataset case composition for frontal injury models. Training Dataset (NASS/CDS) Test Dataset (CISS) Sampled Weighted Sampled Weighted Occupants % Occupants % Occupants % Occupants % All Occupants All Occupants 494 100% 167,330 100% 351 100% 215,913 100% Injury MAIS 0, 1 424 86% 150,233 90% 312 89% 201,611 93% MAIS2+F 70 14% 17,097 10% 39 11% 14,302 7% MAIS3+F 29 6% 6,964 4% 13 4% 2,869 1% Sex Male 254 51% 89,691 54% 146 42% 105,554 49% Female 240 49% 77,639 46% 200 57% 105,516 49% Unknown 0 0% 0 0% 5 1% 4,843 2% Belt Status Belted 383 78% 140,741 84% 281 80% 181,886 84% Unbelted 111 22% 26,589 16% 70 20% 34,027 16% Age Group < 65 404 82% 123,459 74% 292 83% 33,123 15% ≥ 65 90 18% 43,871 26% 59 17% 182,790 85% Seating Location Driver Seat 401 81% 136,868 82% 293 83% 177,738 82% Right Front 93 19% 30,462 18% 58 17% 38,175 18% BMI Obese 147 30% 43,389 26% 62 18% 24,999 12% Not obese 347 70% 123,941 74% 139 40% 64,322 30% Unknown 0 0% 0 0% 150 42% 126,592 58% Vehicle Type Passenger Car 333 67% 116,574 70% 218 62% 127,515 59% Light Truck or Van 161 33% 50756 30% 133 38% 88,399 41% H/F Region MAIS 0,1 467 95% 164,375 98.2% 323 92% 200,174 93% MAIS2+F 22 4% 2,402 1.5% 21 6% 12,517 5.5% Unknown 5 1% 553 0.3% 7 2% 3,222 1.5% N Region MAIS 0,1 482 98% 161,192 96.7% 323 92% 200,254 93% MAIS2+F 7 1% 5,585 3% 21 6% 12,437 5.5% Unknown 5 1% 553 0.3% 7 2% 3,222 1.5% TALT Region MAIS 0,1 461 93% 154,855 92.7% 320 91% 204,897 95% MAIS2+F 28 6% 11,922 7% 24 7% 7,794 3.5% Unknown 5 1% 553 0.3% 7 2% 3,222 1.5% Harm H ≤ $10,000 346 70% 133,013 79% 244 70% 171,467 79% $10,000 < H ≤ $100,000 119 24% 27,237 16% 83 28% 33,938 16% H > $100,000 25 5% 6,611 4% 12 1% 1,016 0.5% Unknown 4 1% 469 1% 12 1% 8,065 4.5% 4.1.4.2 Composition of the Side Crash IAD Subset After filtering the data in accordance with the inclusion criteria, 183 sampled occupants, representing 66,124 weighted occupants, comprised the dataset. This dataset was used to build the

68 injury risk models. Table 4-8 summarizes the occupants who comprise the dataset by their predictor variable categories. There were 183 occupants with region-specific injury information available to train the logistic regression models. A total of 183 cases were available to train the linear regression Harm models. The test dataset did not require available data for every initial covariate, as they were not all included in the final models. As a result, data for some of the initial covariates is unknown for occupants in the test dataset. The associated cumulative distribution function plots for PDOF and the metrics can be found in Appendix C.

69 Table 4-8. Dataset case composition for the side crash injury models. Training Dataset (NASS/CDS) Test Dataset (CISS) Sampled Weighted Sampled Weighted Occupants % Occupants % Occupants % Occupants % All Occupants All Occupants 183 100% 66,124 100% 101 100% 72,372 100% Injury MAIS 0, 1 154 84% 55,189 83% 96 95% 70,589 98% MAIS2+F 29 16% 10,935 17% 5 5% 1,783 2% MAIS3+F 9 5% 3,565 5% 1 1% 231 0.3% Sex Male 82 45% 27,592 42% 37 37% 28,416 39% Female 101 55% 38,532 58% 64 63% 43,957 61% Belt Status Belted 150 82% 51,708 78% 87 86% 65,035 90% Unbelted 33 18% 14,416 22% 14 14% 7,337 10% Impact Type Nearside 94 51% 37,041 56% 36 36% 29,776 41% Far-side 89 49% 29,083 44% 65 44% 42,596 59% Age Group < 65 149 81% 50,218 76% 79 78% 14,814 20% ≥ 65 34 19% 15,906 24% 22 22% 57,558 80 Seating Location Driver Seat 143 78% 54,396 82% 81 80% 53,255 74% Right Front 40 22% 11,729 18% 20 20% 19,118 26% BMI Obese 55 30% 24,916 38% 13 13% 6,156 9% Not obese 128 70% 41,208 62% 16 16% 11,477 16% Unknown 0 0% 0 0% 72 71% 54,739 75% Vehicle Type Passenger Car 149 81% 55,313 84% 58 57% 39,997 55% Light Truck or Van 34 19% 10,811 16% 43 43% 32,375 45% HF Region MAIS 0, 1 174 95% 58,841 89% 99 98% 71,899 98% MAIS2+F 9 5% 7,283 11% 2 2% 473 2% Unknown 0 0% 0 0% 0 0% 0 0% N Region MAIS 0, 1 179 98% 65,406 99% 101 100% 72,372 100% MAIS2+F 4 2% 718 1% 0 0% 0 0% Unknown 0 0% 0 0% 0 0% 0 0% TALT Region MAIS 0, 1 175 96% 63,914 97% 100 99% 72,141 99.5% MAIS2+F 8 4% 2,210 3% 1 1% 231 0.5% Unknown 0 0% 0 0% 0 0% 0 0% Harm H ≤ $10,000 142 78% 53,012 80% 88 87% 67,100 93% $10,000 < H ≤ $100,000 32 17% 9,342 14% 12 12% 5,041 6.5% H > $100,000 9 5% 3,770 6% 1 1% 231 0.5% Unknown 0 0% 0 0% 0 0% 0 0%

70 4.1.4.3 Composition of the Oblique Crash IAD Subset After filtering the data in accordance with the inclusion criteria, 149 sampled occupants comprised the dataset. This is a smaller sample size than what was available for the frontal and side impact model datasets. To improve our case count, we built the initial models using the 149 cases. Then, we reapplied our inclusion criteria without requiring data for the variables that were insignificant in the initial models. Once we did this, our dataset comprised 176 sampled occupants (67,107 weighted occupants). The final models were generated using this dataset. Table 4-9 summarizes the occupants who comprise this dataset by their predictor variable categories. The test dataset did not require available data for every initial covariate, as they were not all included in the final models. As a result, some of the data for some of the initial covariates are unknown for occupants in the test dataset. The associated cumulative distribution function plots for PDOF and the metrics can be found in Appendix C. There were 168 sampled occupants (65,418 weighted occupants) with region-specific data that could be used to train body region-specific injury models. Of the 168 cases available, only four, one, and three sampled occupants suffered MAIS2+F injuries in the HF, N, and TALT regions, respectively. Due to so few injury cases available to train the models, oblique crash body region injury models and associated risk curves were not constructed. A total of 175 cases were available to train the linear regression Harm models.

71 Table 4-9. Dataset case composition for oblique injury model. Training Dataset (NASS/CDS) Test Dataset (CISS) Sampled Weighted Sampled Weighted Occupants % Occupants % Occupants % Occupants % All Occupants All Occupants 176 100% 67,107 100% 179 100% 103,839 100% Injury MAIS 0, 1 150 85% 58,648 87% 157 88% 98,237 95% MAIS2+F 26 15% 8,460 13% 22 12% 5,602 5% MAIS3+F 6 3.5% 361 0.5% 9 5% 1,606 1.5% Sex Male 80 45% 33,499 50% 80 45% 42,907 41% Female 94 53% 32,861 49% 96 54% 58,572 56% Unknown 2 2% 747 1% 3 1% 2,360 3% Belt Status Belted 144 82% 54,867 82% 147 82% 88,960 86% Unbelted 32 18% 12,240 18% 32 18% 14,879 14% Age Group < 65 146 83% 49,742 74% 155 87% 90,034 87% ≥ 65 30 17% 17,366 26% 24 13% 13,805 13% Seating Location Driver Seat 142 81% 56,789 85% 157 88% 97,291 94% Right Front 34 19% 10,318 15% 22 12% 6,548 6% BMI Obese 55 31% 24,050 36% 34 19% 13,655 13% Not obese 96 55% 37,330 56% 63 35% 22,306 21% Unknown 25 14% 5,727 8% 82 46% 67,878 65% GAD Front Damage 131 74% 45,609 68% 140 78% 80,521 78% Side Damage 45 16% 21,498 32% 39 22% 23,318 22% Vehicle Type Passenger Car 125 71% 46,557 69% 121 68% 66,345 64% Light Truck or Van 51 29% 20,550 31% 58 32% 37,494 36% HF Region MAIS 0, 1 164 93% 60,845 90.5% 169 94% 101,141 97% MAIS2+F 4 2.5% 4,573 7% 6 3% 2,184 2.5% Unknown 8 4.5% 1,689 2.5% 4 2% 514 0.5% N Region MAIS 0, 1 167 95% 64,691 96.5% 172 96% 102,740 99% MAIS2+F 1 0.5% 727 1% 3 2% 585 0.5% Unknown 8 4.5% 1,689 2.5% 4 2% 514 0.5% TALT Region MAIS 0, 1 165 94% 64,994 97% 166 93% 102,484 99% MAIS2+F 3 2.5% 424 0.5% 9 5% 841 0.5% Unknown 8 4.5% 1,689 2.5% 4 2% 514 0.5% Harm H ≤ $10,000 124 70% 50,712 75.5% 128 72% 86,853 84% $10,000 < H ≤ $100,000 47 27% 16,091 24% 38 21% 14,861 14% H > $100,000 4 2.5% 187 0.3% 9 5% 1,610 1.5% Unknown 1 0.5% 117 0.2% 4 2% 515 0.5%

72 4.1.5 Conclusions An IAD was assembled that comprised observed injury to real-world crash-involved occupants matched with corresponding EDR-recorded vehicle crash pulses. IAD cases were sampled from NASS/CDS and CISS, which are both nationally representative sources when the associated case weighting factors are applied. SCI and CIREN cases were examined for potential inclusion in the IAD, but neither data source contained sufficient cases for a meaningful analysis, and thus these cases were not included in the IAD. Due to very few occupants sustaining MAIS3+F injury in the IAD, the MAIS2+F injury threshold was selected for use in the subsequent model development. In the future chapters, the cases from NASS/CDS were used to train logistic regression models to predict the probability of MAIS2+F occupant injury and the cases from CISS were used to validate the models. The IAD used for training contained 494 sampled occupants in frontal crashes, which was enough to allow construction of whole body and region-specific injury models. The IAD used for training the side and oblique whole-body crash models contained 183 and 176 sampled occupants, respectively. This prevented region-specific models for these crash modes except for a side impact head-face injury model. NTDB 4.2.1 Background The NTDB is the largest collection of trauma data in the United States. The NTDB contains the detailed ICD code for injuries and the cause of injury. The ICD codes provide a specific value to essentially every known human disease, including physical trauma from vehicle crashes. All patients in this database received treatment at a trauma center. For traffic related injuries, this means that the occupant was treated at a hospital, therefore representing more severe crashes. The purpose of this analysis was to determine if the content of the NTDB was suitable for roadside crash analysis due to the in-depth injury information. 4.2.2 Methods This analysis of the NTDB utilized cases from 2012, the most recent case year. The 2012 NTDB codes injuries according to the ICD-9 diagnosis codes. These codes specify specific injuries for insurance claims. For example, ICD-9-CM 807.02 indicates a closed fracture of two ribs. The ICD- 9 system contains E codes that describe the cause of injury. This study is concerned with injuries resulting from a vehicle crash, which corresponds to codes E810 to E819 (Table 4-10). There are additional subclassifications that indicate the collision partner (Table 4-11).

73 Table 4-10. IDC E codes indicating a vehicle crash was the cause of injury. E Code Description E810 Motor vehicle traffic accident involving collision with train E811 Motor vehicle traffic accident involving re-entrant collision with another motor vehicle E812 Other motor vehicle traffic accident involving collision with motor vehicle E813 Motor vehicle traffic accident involving collision with other vehicle E814 Motor vehicle traffic accident involving collision with pedestrian E815 Other motor vehicle traffic accident involving collision on the highway E816 Motor vehicle traffic accident due to loss of control without collision on the highway E817 Noncollision motor vehicle traffic accident while boarding or alighting E818 Other noncollision motor vehicle traffic accident E819 Motor vehicle traffic accident of unspecified nature Table 4-11. ICD E sub codes indicating the injured individual. E Code Description E###.0 Injuring driver of motor vehicle other than motorcycle E###.1 Injuring passenger in motor vehicle other than motorcycle E###.2 Injuring motorcyclist E###.3 Injuring passenger on motorcycle E###.4 Injuring occupant of streetcar E###.5 Injuring rider of animal; occupant of animal-drawn vehicle E###.6 Injuring pedal cyclist E###.7 Injuring pedestrian E###.8 Injuring other specified person E###.9 Injuring unspecified person To understand the content of the NTDB, the E codes were ranked according to frequency within the dataset to identify the common types of crashes that result in an individual receiving treatment from a trauma center. The dataset was further filtered to focus on single vehicle crashes (E816) to understand the distribution of injury severities in the target crash mode. 4.2.3 Results The crash occupants to be treated at a trauma center were drivers involved in a collision with another motor vehicle. Vehicle drivers, motorcyclists, and vehicle passengers in crashes without a collision partner ranked second, fifth, and eighth, respectively (Table 4-12).

74 Table 4-12. The top 10 most common traffic crashes to result in an occupant treated at a trauma center. Rank E Code Year Crash Description Number of Cases 1 812 2012 Collision w/ MV – Driver of MV 49,188 2 816 2012 Loss Control-No Collision – Driver of MV 36,237 3 814.7 2012 Collision w/ Pedestrian – Pedestrian 25,580 4 812.1 2012 Collision w/ MV – Passenger in MV 22,314 5 816.2 2012 Loss Control-No Collision – Motorcyclist 14,987 6 812.2 2012 Collision w/ MV – Motorcyclist 14,058 7 815 2012 Highway Collision – Driver of MV 12,892 8 816.1 2012 Loss Control-No Collision – Passenger in MV 12,661 9 813.6 2012 Collision w/ Oth Veh – Pedal Cyclist 5,125 10 813 2012 Collision w/ Oth Veh – Driver of MV 4,741 Among occupants involved in single vehicle crashes (E816), drivers of a motor vehicle were the most commonly injured occupants, followed by motorcyclists (Table 4-13). Table 4-13. Distribution of injury severity for occupants at trauma centers after a single vehicle collision (E816). AIS Severity Driver Passenger Motorcyclist Motorcycle Passenger 1 13,455 5,259 5,749 357 2 11,052 3,908 5,166 313 3 6,611 2,403 3,342 172 4 2,080 752 1,042 48 5 550 205 251 9 6 20 7 6 0 Unknown 349 153 109 10 Total Injuries 34,117 12,687 15,665 909 Among AIS 3+ injuries to occupants involved in single vehicle crashes (E816), the thorax was the most commonly injured region, followed by the head. Table 4-14. AIS 3+ injury distribution by body region for occupants at trauma centers after a single vehicle collision (E816). Body Region Driver Passenger Motorcyclist Motorcycle Passenger Abdomen and Pelvic Contents 675 268 327 15 External, Burns, Other Trauma 7 0 3 0 Face 134 56 68 4 Head 2,325 836 1,081 70 Lower Extremity 1,051 454 640 40 Neck 69 22 17 2 Spine 1,043 326 227 12 Thorax 3,477 1,219 1,960 65 Upper Extremity 480 186 318 21 Total AIS 3+ Injuries 9,261 3,367 4,641 229

75 4.2.4 Discussion Crash analysis using the NTDB is challenging due to the lack of information regarding the crash itself. For example, single vehicle drift-out-of-lane crashes do not have a clear E code. Since the crash does not involve a collision partner (another vehicle, pedestrian, etc.), and the trauma center does not investigate the crash, then single vehicle crashes could be classified in 816 despite not involving control loss. The injury mechanism coding is completed by the trauma center, who is not familiar with the details of the crash. Unlike other crash datasets used in this project, the NTDB does not have any on-scene investigation. This prevents any knowledge of critical information such as the crash configuration (frontal, side, or oblique) or the crash severity (delta-v, OIV, etc.). 4.2.5 Conclusions The NTDB is a highly detailed injury dataset focused specifically on occupant injury. While the dataset contains a large number of individuals treated at a trauma center for injuries sustained in a vehicle crash, the dataset has very limited crash configuration and crash severity information. Due to the lack of critical information for this project, the research team did not include the NTDB in the IAD.