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.

76 5 Analyze IAD for Frontal Crashes Introduction The purpose of this chapter is to present the injury models and risk curves associated with each of the candidate injury metrics in frontal crashes. The frontal IAD was used to train MAIS2+F whole body and region-specific logistic models based on eight candidate injury metrics: MDV, OIV, OIV+RA, OIV with a binary RA covariate, RA, OLC, ASI, and VPI. The injury models were evaluated with an equivalent test dataset composed of frontal IAD crashes extracted from CISS. The ability of each candidate injury metric to predict real-world occupant crash injury was then ranked according to various statistical methods. Methods The IAD contained 494 sampled occupants representing 167,330 occupants involved in real- world frontal crashes from NASS/CDS and 351 sampled occupants representing 215,913 occupants involved in real-world frontal crashes from CISS. The training dataset contained occupants in real-world frontal impacts sampled in NASS/CDS with associated EDR data in the VT EDR database. Similarly, the test dataset contained occupants in real-world frontal impacts sampled in CISS with associated EDR data in the VT EDR database. 5.2.1 Metric Computation and Validation MATLAB software (MathWorks 2019) was used to compute the metrics for the cases in the IAD. A set of crash pulses, one for each case in the IAD, were read into the computational MATLAB algorithm. The tool then output a table of the resulting injury metrics for every case included in the IAD. Each metric underwent a separate validation process to ensure the MATLAB computation tool was computing the metrics correctly. The data for the validation come from four sources: (1) NHTSA crash tests, (2) EDR downloads from NASS/CDS, (3) ISO test cases, and (4) artificially generated crash pulses. These data were input to our computational tool to compute injury metrics. Our results were then compared with independently computed metrics either generated from independent injury metric tools or obtained from published reference values. The TRAP Version 2.3.11, developed by TTI, was used to independently compute both OIV and ASI when given acceleration (G) data (Bligh et al. 2000). The remaining metrics used alternate reference computation methods (Table 5-1). The MATLAB algorithm can evaluate both EDR and accelerometer crash pulses. EDR crash pulses must be input as one comma-separated value (CSV) file, whereas accelerometer crash pulses can be input as a series of either CSV or uniform data system (UDS) files, the standard crash pulse format used for NHTSA crash tests. The code “cleans” EDR data by removing trailing zeros and invalid data points. OIV, ASI, and RA were validated by using the TRAP software to compute their values for a dataset comprising NCAP crash tests with available accelerometer data. The research team wanted to evaluate the ability of the in-house MATLAB algorithm to accurately calculate these metrics with time and velocity inputs, the standard format of delta-v EDR data. So, rather than feeding the accelerometer data from crash tests into the algorithm, the crash test acceleration data were

77 integrated first. Then, the two sets of results were compared. MDV, OLC, and VPI were validated using other methods, so this process of transforming the acceleration data to velocity data was not necessary. Table 5-1. Validation method used for each injury metric. Metric Metric Acronym Reference Computation Method Data Source Error Maximum delta-v MDV 1) NHTSA Signal Browser 2) Visual inspection of crash pulse plot NASS/CDS EDR downloads ≤ 0.5% Occupant impact velocity OIV TRAP NHTSA crash tests with EDR downloads 0.0% Acceleration Severity Index ASI TRAP NHTSA crash tests with EDR downloads 0.0% Occupant load criterion OLC Cases generated with known values Artificial crash pulse with known OLC 0.0% Vehicle Pulse Index VPI ISO published values ISO reference crash pulse cases 0.2% Occupant ridedown acceleration RA TRAP NHTSA crash tests with EDR downloads < 1% When building the frontal crash IAD, the occupant had to cross the longitudinal FSM boundary prior to crossing the lateral FSM boundary, or the occupant had to cross only the longitudinal FSM boundary. Since this method was used to select longitudinally dominant crashes, only the longitudinal delta-v crash pulse was used to calculate OIV. Additionally, only the longitudinal delta-v crash pulse was used to compute the remaining metrics: MDV, OLC, ASI, VPI, and RA. 5.2.2 Injury Risk Modeling The injury models were developed using binary logistic regression. The 1998 AIS was used to determine injury severity, since the 1998 injury codes were available for all the case years in the dataset (AAAM 1998). Any occupant with an MAIS2+F (occupants with an MAIS of 2 or greater, including occupants who were fatally injured) rating was considered injured. Occupants with an unknown injury severity were excluded unless their injury was fatal. Fatally injured occupants were included in the MAIS2+F category, regardless of MAIS level. MAIS3+F models could not be built, as this injury category did not contain sufficient cases to build reliable predictive models (Table 4-7). Several predictor variables were used within the models to predict injury: • Crash Severity Metrics. Eight initial injury models were built for frontal crashes. Each model used one (or more) of the five candidate crash severity metrics as an independent predictor variable. The crash severity metrics are each a function of the longitudinal delta- v. The delta-v versus time series data were obtained from the EDR. • Belt Use. Belt use is a categorical variable, where a value of 1 indicates the occupant was using a three-point belt restraint, and 0 indicates the occupant was unbelted. Belt use was determined using the EDR belt status variable. • Age. Age is a categorical variable, where 1 indicates ≥ 65 years old and 0 indicates ≥ 13 years old and < 65 years old.

78 • Sex. Sex is a categorical variable, where 1 indicates male and 0 indicates female. • BMI. Body mass index (kg/m2) is a categorical variable, where 1 indicates obese (BMI ≥ 30 kg/m2) and 0 indicates not obese (BMI < 30 kg/m2). • Occupant Seating Location. The occupant’s seating location is a categorical variable, where a value of 1 indicates the occupant was the driver, and 0 indicates the occupant was in the right front passenger seat. • Vehicle Type. This variable was defined using NHTSA’s Vehicle Body Type Classification (NHTSA 2018). All the vehicles in this dataset fall into the category of either Passenger Car (PC; body types: 1-11, 17) or Light Trucks and Vans (LTV; body types: 14-16, 19-22, 24, 25, 28-41, 45-49). All vehicles in the dataset could be classified into one of these categories. Vehicle type is a categorical variable, where a value of 1 indicates a PC and 0 indicates an LTV. • PDOF. For the analysis, PDOF was redefined from 0° to 180° in 10-degree increments, where 0 corresponds to the front of the vehicle. Previously, a PDOF of 360° and a PDOF of 0° would have been treated as different conditions, even though they share the same angle from the front of the vehicle. This new PDOF definition resolves that issue and additionally makes the PDOF variable independent of the seating location variable. • RA Bin. One of the two OIV models included the occupant RA as a binary covariate. The current maximum RA threshold is 20.49 G; however, none of the occupants in the training dataset exceeded this threshold. Two occupants in the dataset exceeded the 15 G preferred threshold, so this value was used to bin the occupants. A value of 1 indicates RA ≥ 15 G, and 0 indicates < 15 G. First, logistic regression models were developed using the full set of available covariates. These initial regression models were examined to determine which of the covariates were statistically significant. Based on those results, a second set of regression models was developed in which only the statistically significant covariates were included. The logistic regression models for each set were developed using the survey package and SVYDESIGN function in R (Lumley 2020). The R survey package was designed based on SAS and is capable of computing confidence limits that account for the complex NASS-CDS stratified sampling scheme. 5.2.3 Region-Specific Models The body region injury data, from the NASS/CDS occupant injury table, list every injury for each body region for each occupant. Any occupant with unknown body region data was removed from the dataset. Additionally, any occupant with an MAIS2+F injury, regardless of region, when OLC was zero was removed from the dataset. The NASS/CDS regions were combined into the HF, N, and TALT regions (Table 5-2).

79 Table 5-2. NASS/CDS body regions used to form the model body regions. NASS/CDS Region NASS/CDS Specific Anatomic Structure Model Region Head All HF Face All Neck All N Thorax All TALT Abdomen All Spine C-Spine N T-Spine TALT L-Spine Upper extremity All Injury risk not modeled Lower extremity All 5.2.4 Predictive Capability After building the models using the NASS/CDS training dataset, the developed models were run on the test dataset from CISS. Precision, recall, accuracy, F2 score, and AUC were computed for the training and test datasets. Precision, also known as the predictive value of the model, is the probability that the occupants predicted to be injured were actually injured (Equation 4). Recall, also known as sensitivity or the true positive rate, is the ability of the model to correctly predict injury for occupants who actually did suffer an injury (Equation 5). The F2 score prioritizes recall and is indicative of the model’s ability to accurately predict outcomes and is a reliable metric for unbalanced datasets (Equation 6), that is, in this case, many more non-injured compared to injured occupants. A higher F2 score indicates better predictive capability of the model and was the primary comparison metric. 𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃 = 𝑇𝑇𝑃𝑃 𝑇𝑇𝑃𝑃 + 𝐹𝐹𝑃𝑃 (4) 𝑅𝑅𝑃𝑃𝑃𝑃𝑅𝑅𝑅𝑅𝑅𝑅 = 𝑇𝑇𝑃𝑃 𝑇𝑇𝑃𝑃 + 𝐹𝐹𝐹𝐹 (5) 𝐹𝐹2 = (5)(𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃)(𝑅𝑅𝑃𝑃𝑃𝑃𝑅𝑅𝑅𝑅𝑅𝑅) (4 ∗ 𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃) + (𝑅𝑅𝑃𝑃𝑃𝑃𝑅𝑅𝑅𝑅𝑅𝑅) (6) 𝐴𝐴𝑃𝑃𝑃𝑃𝐴𝐴𝑃𝑃𝑅𝑅𝑃𝑃𝐴𝐴 = 𝑇𝑇𝑃𝑃 + 𝑇𝑇𝐹𝐹 𝑇𝑇𝑃𝑃 + 𝑇𝑇𝐹𝐹 + 𝐹𝐹𝑃𝑃 + 𝐹𝐹𝐹𝐹 (7) When calculating precision and recall, a decision threshold in the form of a percent injury risk must be chosen. This threshold was determined for each model individually by selecting the percent injury risk value that optimized the F2 score. These thresholds were found using the training dataset and were used throughout the comparison process. When more than one injury risk value was associated with the same precision and recall summation, the lowest injury risk value was chosen. The NASS/CDS and CISS weighted values were used to calculate each of these comparison metrics. Overall Injury Model Results 5.3.1 Initial Injury Risk Models The initial models used seat belt use, sex, age, BMI, seating location, vehicle type, and PDOF as covariates. Each model additionally used one of the five crash severity metrics as a covariate.

80 Table C-1 through Table C-6 show the regression coefficients for each of the injury risk models. The initial model has been shown for the RA metric (Table 5-3). This model has been shown because RA was not significant, so no final model was constructed. Additionally, a model was built to determine the significance of including both RA and OIV in the same model (Table 5-5). This model included an interaction term between the two metrics. Both RA and the interaction were insignificant, while OIV was significant. A corresponding final model was not constructed. A p-value < 0.05 was considered significant and is denoted by ** in the parameter tables. A negative coefficient indicates that, with all other predictors held constant, a decrease in a continuous variable will reduce the injury risk. For a binary covariate, the baseline condition (listed in the model tables) reduces the injury risk. A positive coefficient indicates that, with all other predictors held constant, an increase in a continuous variable will increase the injury risk. For a binary covariate, the non-baseline condition reduces the injury risk. For example, the severity metrics all have positive coefficients, because an increase in any of the metrics will heighten the occupant’s risk of injury. Additionally, age (≥ 65) always has a positive coefficient, because older occupants are more likely to suffer an MAIS2+F injury. The crash severity metric, belt status, and age p-values were significant in every model. PDOF was significant in the OLC, ASI, and VPI models. Occupant sex, seating location, vehicle type, and BMI were not significant in any of the models. Equation 8 is the final form of the model and Equation 9 is the logit expanded. 𝑃𝑃[𝑀𝑀𝐴𝐴𝑀𝑀𝑀𝑀2+ F] = 1 1 + 𝑃𝑃−𝑅𝑅𝑃𝑃𝑙𝑙𝑃𝑃𝑙𝑙 (8) 𝑅𝑅𝑃𝑃𝑙𝑙𝑃𝑃𝑙𝑙 = 𝛽𝛽0 + 𝛽𝛽1 ⋅ (𝑃𝑃𝑃𝑃𝑖𝑖𝐴𝐴𝑃𝑃𝐴𝐴 𝑚𝑚𝑃𝑃𝑙𝑙𝑃𝑃𝑃𝑃𝑃𝑃) + 𝛽𝛽2 ⋅ 𝑏𝑏𝑃𝑃𝑅𝑅𝑙𝑙𝑃𝑃𝑙𝑙𝑅𝑅𝑙𝑙𝐴𝐴𝑃𝑃 + 𝛽𝛽3 ⋅ 𝑃𝑃𝑃𝑃𝑠𝑠 + 𝛽𝛽4 ⋅ 𝑅𝑅𝑙𝑙𝑃𝑃 + 𝛽𝛽5 ⋅ 𝑃𝑃𝑏𝑏𝑃𝑃𝑃𝑃𝑃𝑃 + 𝛽𝛽6 ⋅ 𝑃𝑃𝑃𝑃𝑅𝑅𝑙𝑙𝑃𝑃𝑃𝑃𝑙𝑙_𝑅𝑅𝑃𝑃𝑃𝑃𝑅𝑅𝑙𝑙𝑃𝑃𝑃𝑃𝑃𝑃 + 𝛽𝛽7 ⋅ 𝑣𝑣𝑃𝑃ℎ𝑃𝑃𝑃𝑃𝑅𝑅𝑃𝑃_𝑙𝑙𝐴𝐴𝑡𝑡𝑃𝑃 + 𝛽𝛽8 ⋅ 𝑃𝑃𝑃𝑃𝑃𝑃𝐹𝐹 (9)

81 Table 5-3. Parameters for the ORA initial frontal logistic regression model used to predict occupant MAIS2+F injuries. ** indicates statistical significance (p-value < 0.05). Predictor Variable Parameter Coefficient Std. Error p-Value --- β0, Intercept -3.019 0.685 < 0.001** Longitudinal ORA β1, ORA (G) 0.285 0.153 0.064 Belt Use β2, Belted -0.597 1.036 0.566 Sex β3, Male -0.253 0.495 0.610 Age β4, Age ≥ 65 2.911 0.567 < 0.001** BMI β5, BMI < 30 kg/m2 -0.513 0.662 0.440 Seating Location β6, Driver -0.980 0.461 0.036** Vehicle Body Type β7, Passenger Car 1.244 0.601 0.041** PDOF β8, PDOF -0.034 0.020 0.086 Table 5-4. Parameters for the ORA initial frontal logistic regression model, accounting for OIV*ORA interaction, used to predict occupant MAIS2+F injuries. ** indicates statistical significance (p-value < 0.05). Predictor Variable Parameter Coefficient Std. Error p-Value --- β0, Intercept -8.115 1.467 < 0.001** Longitudinal ORA β1, ORA (G) 0.044 0.411 0.915 Belt Use β2, Belted 1.848 0.550 0.001** Sex β3, Male 0.019 0.381 0.961 Age β4, Age ≥ 65 3.380 0.633 < 0.001** BMI β5, BMI < 30 kg/m2 0.267 0.522 0.609 Seating Location β6, Driver -0.134 0.690 0.846 Vehicle Body Type β7, Passenger Car -0.291 0.432 0.503 PDOF β8, PDOF -0.032 0.022 0.148 Longitudinal OIV β9, OIV (m/s) -0.752 0.129 < 0.001** ORA*OIV Interaction β10, Interaction -0.004 0.026 0.865 5.3.2 Final Injury Risk Models Based on the initial regression models, a second set of injury models was developed. These models use only the covariates that were statistically significant in the initial model: one of the five crash severity metrics, belt use, and age. Belt use was not significant in the final ASI and VPI models, so it was removed. Table 5-5 through Table 5-10 show the regression coefficients for each of the models. All covariates and the intercept were significant in every model.

82 Table 5-5. Parameters for the MDV final frontal logistic regression model used to predict occupant MAIS2+F injuries. ** indicates statistical significance (p-value < 0.05). Predictor Variable Parameter Coefficient Std. Error p-Value --- β0, Intercept -8.445 1.525 < 0.001** Longitudinal Delta-v β1, Delta-v (m/s) 0.674 0.119 < 0.001** Belt Use β2, Belted -1.812 0.594 0.003** Age β4, Age ≥ 65 3.469 0.610 < 0.001** Table 5-6. Parameters for the OIV final frontal logistic regression model used to predict occupant MAIS2+F injuries. ** indicates statistical significance (p-value < 0.05). Predictor Variable Parameter Coefficient Std. Error p-Value --- β0, Intercept -8.663 1.500 < 0.001** Longitudinal OIV β1, OIV (m/s) 0.726 0.124 < 0.001** Belt Use β2, Belted -1.864 0.592 0.002** Age β4, Age ≥ 65 3.400 0.586 < 0.001** Table 5-7. Parameters for the OIV+RA final frontal logistic regression model used to predict occupant MAIS2+F injuries. ** indicates statistical significance (p-value < 0.05). Predictor Variable Parameter Coefficient Std. Error p-Value --- β0, Intercept -8.651 1.502 < 0.001** Longitudinal OIV β1, OIV (m/s) 0.723 0.125 < 0.001** Belt Use β2, Belted -1.845 0.600 0.003** Age β4, Age ≥ 65 3.399 0.586 < 0.001** ORA Bin β11, ORA ≥ 15 G 11.931 1.255 < 0.001**

83 Table 5-8. Parameters for the OLC final frontal logistic regression model used to predict occupant MAIS2+F injuries. ** indicates statistical significance (p-value < 0.05). Predictor Variable Parameter Coefficient Std. Error p-Value --- β0, Intercept -5.531 1.1120 < 0.001** Longitudinal OLC β1, OLC (G) 0.326 0.0520 < 0.001** Belt Use β2, Belted -1.622 0.6448 0.013** Age β4, Age ≥ 65 3.252 0.5473 < 0.001** PDOF β8, PDOF -0.041 0.0202 0.043** Table 5-9. Parameters for the ASI final frontal logistic regression model used to predict occupant MAIS2+F injuries. ** indicates statistical significance (p-value < 0.05). Predictor Variable Parameter Coefficient Std. Error p-Value --- β0, Intercept -7.839 1.162 < 0.001** Longitudinal ASI β1, ASI (G) 4.529 0.707 < 0.001** Age β4, Age ≥ 65 3.247 0.615 < 0.001** PDOF β8, PDOF -0.041 0.019 0.036** Table 5-10. Parameters for the VPI final frontal logistic regression model used to predict occupant MAIS2+F injuries. ** indicates statistical significance (p-value < 0.05). Predictor Variable Parameter Coefficient Std. Error p-Value --- β0, Intercept -8.362 1.224 < 0.001** Longitudinal VPI β1, VPI (m/s2) 0.019 0.003 < 0.001** Age β4, Age ≥ 65 3.146 0.590 < 0.001** PDOF β8, PDOF -0.046 0.020 0.024** 5.3.3 Injury Risk Curves The predict function in R was used to obtain injury probability values for each of the six final models. To generate the risk curves, the probability of injury was evaluated for belted and unbelted occupants as the crash severity metric increased by 0.01 from zero to a resulting probability greater than 99%. Age was held constant as at least 13 years old and under 65 years old. PDOF was held constant at zero. This process was performed once for a belted occupant and again for an unbelted occupant. Figure 5-1 through Figure 5-6 plot the risk curves for a belted and unbelted occupant for all six models. Because these severity metrics, besides VPI, are reported as negative values, they were multiplied by -1 prior to plotting to maintain a positive x-axis. The shaded regions represent 95% confidence intervals. Black vertical lines have been used to show the minimum and maximum metric values within the side crash training dataset. The data outside of this range were extrapolated by the model.

84 Figure 5-1. Frontal impact MDV injury risk curve for an occupant older than 12 and younger than 65. Figure 5-2. Frontal impact OIV injury risk curve for an occupant at least 13 and younger than 65 years.

85 Figure 5-3. Frontal impact OIV+RA injury risk curve for an occupant at least 13, younger than 65 years, and with an RA less than 15 G. Figure 5-4. Frontal impact OLC injury risk curve for an occupant at least 13 and younger than 65 with a PDOF of zero.

86 Figure 5-5. Frontal impact ASI injury risk curve for an occupant at least 13 and younger than 65, regardless of belt status, with a PDOF of zero. Figure 5-6. Frontal impact VPI injury risk curve for an occupant at least 13 and younger than 65 years, regardless of belt status, with a PDOF of zero.

87 Body Region Model Results 5.4.1 Initial Injury Risk Models The initial models used seat belt use, sex, age, BMI, seating location, vehicle type, and PDOF as covariates, in addition to one of the crash severity metrics. A p-value < 0.05 was considered significant. Equation 8 is the initial form of the model and Equation 10 is the logit expanded. For the HF models, the metric, belt status, and vehicle body type were significant. For the N and TALT models, the metric and age were significant. The full models are not shown in table form. 𝑃𝑃[𝑀𝑀𝐴𝐴𝑀𝑀𝑀𝑀2+ F] = 1 1 + 𝑃𝑃−𝑅𝑅𝑃𝑃𝑙𝑙𝑃𝑃𝑙𝑙 (8) 𝑅𝑅𝑃𝑃𝑙𝑙𝑃𝑃𝑙𝑙 = 𝛽𝛽0 + 𝛽𝛽1 ⋅ (𝑃𝑃𝑃𝑃𝑖𝑖𝐴𝐴𝑃𝑃𝐴𝐴 𝑚𝑚𝑃𝑃𝑙𝑙𝑃𝑃𝑃𝑃𝑃𝑃) + 𝛽𝛽2 ⋅ 𝑏𝑏𝑃𝑃𝑅𝑅𝑙𝑙𝑃𝑃𝑙𝑙𝑅𝑅𝑙𝑙𝐴𝐴𝑃𝑃 + 𝛽𝛽3 ⋅ 𝑃𝑃𝑃𝑃𝑠𝑠 + 𝛽𝛽4 ⋅ 𝑅𝑅𝑙𝑙𝑃𝑃 + 𝛽𝛽5 ⋅ 𝑃𝑃𝑏𝑏𝑃𝑃𝑃𝑃𝑃𝑃 + 𝛽𝛽6 ⋅ 𝑃𝑃𝑃𝑃𝑅𝑅𝑙𝑙𝑃𝑃𝑃𝑃𝑙𝑙_𝑅𝑅𝑃𝑃𝑃𝑃𝑅𝑅𝑙𝑙𝑃𝑃𝑃𝑃𝑃𝑃 + 𝛽𝛽7 ⋅ 𝑣𝑣𝑃𝑃ℎ𝑃𝑃𝑃𝑃𝑅𝑅𝑃𝑃_𝑙𝑙𝐴𝐴𝑡𝑡𝑃𝑃 + 𝛽𝛽8 ⋅ 𝑃𝑃𝑃𝑃𝑃𝑃𝐹𝐹 (9) 5.4.2 Final Injury Risk Models Based on the initial regression models, a second set of injury models was developed. These models use only the covariates that were statistically significant in the initial models. 5.4.2.1 Body Region 1: HF Table 5-11 through Table 5-15 show the regression coefficients for each of the final head and face frontal body region models. A p-value < 0.05 was considered significant and is denoted by ** in the parameter tables. All covariates were significant in every model.

88 Table 5-11. Parameters for the MDV logistic regression frontal HF model used to predict occupant MAIS2+F injuries. ** indicates statistical significance (p-value < 0.05). Predictor Variable Parameter Coefficient Std. Error p-Value --- β0, Intercept -7.662 1.156 < 0.001** Longitudinal Delta-v β1, Delta-v (m/s) 0.337 0.089 < 0.001** Belt Status β2, Belted -3.128 0.787 < 0.001** Vehicle Type β7, Passenger car 2.453 0.885 0.006** Table 5-12. Parameters for the OIV logistic regression frontal HF model used to predict occupant MAIS2+F injuries. ** indicates statistical significance (p-value < 0.05). Predictor Variable Parameter Coefficient Std. Error p-Value --- β0, Intercept -8.054 1.362 < 0.001** Longitudinal OIV β1, OIV (m/s) 0.387 0.109 0.001** Belt Status β2, Belted -3.348 0.834 < 0.001** Vehicle Type β7, Passenger car 2.602 0.908 0.005** Table 5-13. Parameters for the OLC logistic regression frontal HF model used to predict occupant MAIS2+F injuries. ** indicates statistical significance (p-value < 0.05). Predictor Variable Parameter Coefficient Std. Error p-Value --- β0, Intercept -6.543 1.075 < 0.001** Longitudinal OLC β1, OLC (G) 0.162 0.052 0.002** Belt Status β2, Belted -3.301 0.892 < 0.001** Vehicle Type β7, Passenger car 2.758 0.919 0.003** Table 5-14. Parameters for the ASI logistic regression frontal HF model used to predict occupant MAIS2+F injuries. ** indicates statistical significance (p-value < 0.05). Predictor Variable Parameter Coefficient Std. Error p-Value --- β0, Intercept -7.246 1.278 < 0.001** Longitudinal ASI β1, ASI 2.409 0.790 0.003** Belt Status β2, Belted -3.268 0.866 < 0.001** Vehicle Type β7, Passenger car 2.726 0.940 0.004** Table 5-15. Parameters for the VPI logistic regression frontal HF model used to predict occupant MAIS2+F injuries. ** indicates statistical significance (p-value < 0.05). Predictor Variable Parameter Coefficient Std. Error p-Value --- β0, Intercept - 7.528 1.361 < 0.001** Longitudinal VPI β1, VPI (m/s2) 0.010 0.003 0.005** Belt Status β2, Belted - 3.285 0.896 < 0.001** Vehicle Type β7, Passenger car 2.761 0.929 0.004** 5.4.2.2 Body Region 2: N Table 5-16 through Table 5-20 show the regression coefficients for each of the final neck and c-spine frontal body region models. A p-value < 0.05 was considered significant and is denoted by ** in the parameter tables. All covariates were significant in every model.

89 Table 5-16. Parameters for the MDV logistic regression frontal N model used to predict occupant MAIS2+F injuries. ** indicates statistical significance (p-value < 0.05). Predictor Variable Parameter Coefficient Std. Error p-Value --- β0, Intercept -10.0266 1.593 < 0.001** Longitudinal Delta-v β1, Delta-v (m/s) 0.4481 0.119 < 0.001** Age β4, Age ≥ 65 3.6816 0.746 < 0.001** Table 5-17. Parameters for the OIV logistic regression frontal N model used to predict occupant MAIS2+F injuries. ** indicates statistical significance (p-value < 0.05). Predictor Variable Parameter Coefficient Std. Error p-Value --- β0, Intercept -10.362 1.609 < 0.001** Longitudinal OIV β1, OIV (m/s) 0.505 0.132 < 0.001** Age β4, Age ≥ 65 3.453 0.739 < 0.001** Table 5-18. Parameters for the OLC logistic regression frontal N model used to predict occupant MAIS2+F injuries. ** indicates statistical significance (p-value < 0.05). Predictor Variable Parameter Coefficient Std. Error p-Value --- β0, Intercept -8.573 1.242 < 0.001** Longitudinal OLC β1, OLC (G) 0.223 0.050 < 0.001** Age β4, Age ≥ 65 3.695 0.942 < 0.001** Table 5-19. Parameters for the ASI logistic regression frontal N model used to predict occupant MAIS2+F injuries. ** indicates statistical significance (p-value < 0.05). Predictor Variable Parameter Coefficient Std. Error p-Value --- β0, Intercept -9.726 1.383 < 0.001** Longitudinal ASI β1, ASI 3.433 0.719 < 0.001** Age β4, Age ≥ 65 3.795 0.926 < 0.001** Table 5-20. Parameters for the VPI logistic regression frontal N model used to predict occupant MAIS2+F injuries. ** indicates statistical significance (p-value < 0.05). Predictor Variable Parameter Coefficient Std. Error p-Value --- β0, Intercept -10.149 1.557 < 0.001** Longitudinal VPI β1, VPI (m/s2) 0.014 0.003 < 0.001** Age β4, Age ≥ 65 3.822 1.005 < 0.001** 5.4.2.3 Body Region 3: TALT Table 5-21 through Table 5-25 show the regression coefficients for each of the final TALT frontal body region models. A p-value < 0.05 was considered significant and is denoted by ** in the parameter tables. All covariates were significant in every model.

90 Table 5-21. Parameters for the MDV logistic regression frontal TALT model used to predict occupant MAIS2+F injuries. ** indicates statistical significance (p-value < 0.05). Predictor Variable Parameter Coefficient Std. Error p-Value --- β0, Intercept -16.961 2.8464 < 0.001** Longitudinal Delta-v β1, Delta-v (m/s) 0.951 0.1816 < 0.001** Age β4, Age ≥ 65 7.052 1.4588 < 0.001** Table 5-22. Parameters for the OIV logistic regression frontal TALT model used to predict occupant MAIS2+F injuries. ** indicates statistical significance (p-value < 0.05). Predictor Variable Parameter Coefficient Std. Error p-Value --- β0, Intercept -17.189 2.743 < 0.001** Longitudinal OIV β1, OIV (m/s) 1.048 0.202 < 0.001** Age β4, Age ≥ 65 6.530 1.200 < 0.001** Table 5-23. Parameters for the OLC logistic regression frontal TALT model used to predict occupant MAIS2+F injuries. ** indicates statistical significance (p-value < 0.05). Predictor Variable Parameter Coefficient Std. Error p-Value --- β0, Intercept -13.397 2.232 < 0.001** Longitudinal OLC β1, OLC (G) 0.466 0.089 < 0.001** Age β4, Age ≥ 65 6.901 1.580 < 0.001** Table 5-24. Parameters for the ASI logistic regression frontal TALT model used to predict occupant MAIS2+F injuries. ** indicates statistical significance (p-value < 0.05). Predictor Variable Parameter Coefficient Std. Error p-Value --- β0, Intercept -15.419 2.539 < 0.001** Longitudinal ASI β1, ASI 6.958 1.319 < 0.001** Age β4, Age ≥ 65 7.009 1.554 < 0.001** Table 5-25. Parameters for the VPI logistic regression frontal TALT model used to predict occupant MAIS2+F injuries. ** indicates statistical significance (p-value < 0.05). Predictor Variable Parameter Coefficient Std. Error p-Value --- β0, Intercept -15.922 2.936 < 0.001** Longitudinal VPI β1, VPI (m/s2) 0.028 0.005 < 0.001** Age β4, Age ≥ 65 6.792 1.727 < 0.001** 5.4.3 Injury Risk Curves The predict function in R was used to obtain injury probability values for each of the five final models. To generate the risk curves, the probability of injury was evaluated for occupants as the crash severity metric increased by 0.01 from zero to a resulting probability greater than 99%. The vertical black lines on each plot represent the minimum and maximum metric values in the original dataset. Portions of the curves outside these bounds have been extrapolated. For the HF curves, vehicle type was held constant as a passenger car. For the N and TALT curves, age was held constant as at least 13 years old and under 65 years old. For the HF curves, this process was performed for belted drivers and unbelted drivers. Figure 5-7 through Figure 5-11 plot the risk curves for a belted and unbelted occupant for all five models. Because the original values for these metrics were either zero or negative, they were multiplied by -1 prior to plotting to maintain a positive x-axis.

91 Figure 5-7. Frontal impact MDV injury risk curves for all three regions. The HF curves apply to all front- seated occupants in passenger cars. The N and TALT curves apply to occupants older than 12 and younger than 65. Figure 5-8. Frontal impact OIV injury risk curves for all three regions. The HF curves apply to all front- seated occupants in passenger cars. The N and TALT curves apply to occupants older than 12 and younger than 65.

92 Figure 5-9. Frontal impact OLC injury risk curves for all three regions. The HF curves apply to all front- seated occupants in passenger cars. The N and TALT curves apply to occupants older than 12 and younger than 65. Figure 5-10. Frontal impact ASI injury risk curves for all three regions. The HF curves apply to all front- seated occupants in passenger cars. The N and TALT curves apply to occupants older than 12 and younger than 65.

93 Figure 5-11. Frontal impact VPI injury risk curves for all three regions. The HF curves apply to all front- seated occupants in passenger cars. The N and TALT curves apply to occupants older than 12 and younger than 65. Model Validation and Comparison The training data were used to determine a decision threshold for each model and to compare the predictive capability of the models on training data by calculating the F2 scores. The same decision thresholds were used to calculate precision, recall, and the resulting F2 score after the models were applied to the CISS test dataset (Table 5-26). All the final models yield similar decision thresholds, except the OLC model, which has a slightly higher threshold of 32%. Additionally, the F2 scores for each model are very similar. OIV+RA bin performed better on the test dataset than the rest of the models with an F2 score of 0.60. The accuracy of each model on the test dataset was also calculated. All models yielded the high accuracy values; however, it is important to note why accuracy is not a reliable metric for an unbalanced dataset. Accuracy is a measure of how many predictions were correct, and because the majority of the predictions are true negative injury cases, it is easy for the models to have a high accuracy but poor precision, and therefore a low F2 score. Additionally, every model yields an AUC value of at least 0.81 on the training data, indicating the models perform fairly well. However, similarly to accuracy, the AUC for ROCs are not a great diagnostic tool for determining how well a model performs when it is exposed to highly imbalanced data. This is because ROCs consider the false positive rate, which is calculated using the total number of true negatives. When the number of true negatives is very large, the false positive rate does not drop drastically. On the contrary, precision is a metric very sensitive to the number of false positives.

94 Table 5-26. F2 scores for the six metrics’ frontal models. These values come from the final model. Metric Model Decision Threshold Training Test F2 Scores ROC AUC Accuracy Precision Recall F2 Score MDV 26% 0.82 0.82 0.93 0.48 0.58 0.56 OIV 24% 0.82 0.82 0.94 0.52 0.58 0.57 OIV+RA Bin 31% 0.82 0.82 0.96 0.70 0.58 0.60 OLC 24% 0.82 0.78 0.93 0.49 0.58 0.56 ASI 24% 0.82 0.78 0.94 0.54 0.45 0.47 VPI 24% 0.81 0.80 0.94 0.58 0.57 0.57 The identical performance evaluation method was performed on the region-specific models (Table 5-27). Of the 344 sampled occupants used for testing, 21, 21, and 24 were injured for the HF, N, and TALT body regions, respectively. As a result, the F2 score is largely influenced by a small population of the dataset. Table 5-27. F2 scores for the region-specific frontal models. Region Metric Model Decision Threshold Training Test F2 Score ROC AUC Accuracy Precision Recall F2 Score HF MDV 10% 0.61 0.94 0.94 0.21 0.02 0.02 OIV 15% 0.59 0.93 0.94 0.36 0.01 0.01 OLC 19% 0.55 0.95 0.94 0.60 0.02 0.02 ASI 14% 0.54 0.94 0.94 0.24 0.02 0.02 VPI 17% 0.55 0.94 0.94 0.60 0.02 0.02 N MDV 4% 0.67 0.76 0.93 0.42 0.54 0.51 OIV 40% 0.70 0.76 0.94 0.78 0.02 0.02 OLC 28% 0.70 0.74 0.96 0.91 0.27 0.31 ASI 6% 0.69 0.73 0.95 0.58 0.47 0.49 VPI 7% 0.70 0.75 0.95 0.63 0.44 0.47 TALT MDV 43% 0.92 0.88 0.97 0.53 0.56 0.56 OIV 53% 0.90 0.87 0.96 0.36 0.27 0.29 OLC 45% 0.90 0.84 0.96 0.39 0.31 0.32 ASI 45% 0.90 0.81 0.96 0.39 0.31 0.32 VPI 29% 0.90 0.83 0.96 0.38 0.31 0.32 Discussion Occupant age and the crash severity metric were significant predictors in all six final regression models. Belt status was significant in the MDV, OIV, OIV + binary RA, and OLC models, and PDOF was significant in the OLC, ASI, and VPI models. The metrics of interest and age were significant in every N and TALT model. The candidate injury metric, belt status, and the vehicle type were significant in every HF model. For MDV, OIV, OIV + binary RA, and OLC, holding all covariates constant, the risk of injury was substantially lower for a belted occupant than for an unbelted occupant. Using the F2 scores from the test dataset as a metric of prediction capability, the OIV model with a binary RA covariate is most equipped to accurately predict occupant injury outcomes in frontal crashes. However, there were only two cases in the test data where RA was greater than 15 G, and both of those occupants experienced an MAIS2+F injury. Therefore, this model may be overfit. For each pair of injury risk curves, the confidence intervals grow wider as the risk of injury approaches 50%. This is due to the variability in injury outcomes for medium-severity crashes. The MDV model is most equipped to accurately predict occupant N and TALT injury outcomes in frontal crashes. For the HF models,

95 there is very little variability among the F2 scores. None of the models yield F2 scores high enough to deem them competent in predicting HF injuries. Based on the F2 scores, the models overall seem to predict N and TALT injury better than HF injuries. Given a lack of higher value RA cases available, additional study of the RA metric is warranted. Note that there were also no RA values above 20.49 or even 15 G observed in the CISS test dataset. A similar issue was encountered in previous EDR-based work (Gabauer and Gabler 2004a) evaluating the RA metric with no RA values above the preferred 15 G threshold in 58 analyzed cases. Two other important issues related to RA are described below; both could contribute to a lack of cases with high RA values: 1. Computational accuracy. As EDRs typically provide vehicle velocity information (and not acceleration) in increments of 10 ms, the RA computation is much less accurate than the other velocity-based metrics such as OIV. Based on six full-scale frontal rigid barrier crash tests where vehicle EDR data were also available, Gabauer and Gabler (2004a) initially reported the EDR-based RA overestimated the actual RA by 40% on average. A more recent analysis by the research team used 23 full-scale frontal rigid barrier crash tests and found the EDR-based RA produced an RMSE of 93.7% with no trend toward overestimating or underestimating the actual RA value. While ASI is also acceleration- based, the larger time window used (50-ms vs. the 10-ms for RA) reduces the potential for error. An initial estimate by Gabauer and Gabler (2005) suggested that, while the EDR- based ASI underestimates the actual ASI, the values were within 10%. 2. EDR recording duration. EDRs typically record vehicle change in velocity information for 300 ms or less. Many crashes with the potential to have a high RA have either a secondary crash or event later in the crash that may not be captured by the EDR. In this study, only single event crashes were included to be sure that the computed metric value is matched with the injury-causing event. Crashes with multiple events relatively close together in time (within a few seconds) would be more likely to have a high RA value but were excluded from this study due to an inability to identify the injury-causing event in those cases. The majority of cases in the validation dataset, much like the training dataset, are occupants who were not injured. The models are good at predicting these incidences, so they yield a high number of true negative cases. The F2 score does not consider true negatives, which means the F2 score does rely heavily on the positive cases (i.e., injured occupants). It is important to note some of the differences between NASS/CDS, the database used for training, and CISS, the database used for validation. While largely similar, CISS uses AIS 2015, while the NASS/CDS training data implements AIS 1998. Additionally, the tow-away criterion between the two databases varies slightly. NASS/CDS requires at least one vehicle be towed away from the crash scene due to damage. CISS requires at least one vehicle be towed away from the crash scene due to any reason. Finally, CISS implements more primary sampling units than NASS/CDS. In a study conducted by Gabauer and Gabler (2008a), MAIS2+F injury risk curves were built for the MDV, OIV, and ASI metrics. This was done using an unweighted dataset of 180 real-world frontal crashes with EDR data available. Two curves were built for each metric: one for unbelted occupants and one for belted occupants. In this study, only one curve was built for ASI, because belt status was not significant in the final model. When looking at delta-v, Gabauer and Gabler’s

96 unbelted occupant curve crosses the 50% MAIS2+F injury mark at about 12.5 m/s, which is very similar to the results of this study. When looking at the delta-v belted occupant curves, Gabauer and Gabler’s is less steep, approaching 100% MAIS2+F injury approximately 5 m/s after the curve presented in this study. Gabauer and Gabler’s ASI curves for belted and unbelted occupants approach 100% at approximately ASI = 4 and ASI = 3, respectively. The curve presented in this study approaches 100% at approximately ASI = 3.5. Finally, the OIV unbelted occupant curve presented by Gabauer and Gabler is much steeper than the one shown in this study. Their OIV belted occupant curve is more similar to the one presented here, but approaches 100% at a higher value. Currently, the U.S. and other roadside safety hardware crash test procedures prescribe preferred and maximum thresholds for the OIV and ASI metrics. The models presented in this chapter can provide injury risk values associated with these thresholds for the best- and worst-case scenario occupant (Table 5-28 and Table 5-29). For these models, a best-case scenario occupant would be a belted occupant, younger than 65 years old, in a vehicle with a PDOF of 40°. An unbelted occupant, over the age of 65, in a vehicle with a PDOF of 0° would be the worst-case scenario within these models. For the HF models, a best-case scenario occupant would be a belted occupant in a passenger car. For the N and TALT models, the best-case scenario would be an occupant younger than 65 years old. While other populations could be explored as well, this chapter examines only the two extremes. Calculating the MAIS2+F injury risk for the OIV and ASI thresholds makes it possible to compare the acceptable level of injury risk across the two metrics. For example, for the best-case scenario occupant, the risk associated with the lowest OIV and ASI thresholds is fairly similar, while the risk associated with the highest thresholds has a larger margin. The opposite holds true for the worst-case scenario occupant.

97 Table 5-28. Injury risk associated with the current OIV and ASI thresholds for the best- and worst-case frontal crash scenarios. Crash and Occupant Conditions Injury Risk Scenario Belt Status Age (years) PDOF (degrees) OIV Threshold (m/s) ASI Threshold (--) 9.1 12.2 1.0 1.4 1.9 Best Case Belted < 65 40 1.9% 15.8% 0.7% 4.2% 29.6% Worst Case Unbelted ≥ 65 0 79.3% 97.3% 48.4% 85.2% 98.2% Table 5-29. Injury risk by body region associated with the current OIV and ASI thresholds for the best- and worst-case frontal impact scenarios. Crash and Occupant Conditions Injury Risk Region Scenario Belt Status Age (years) Vehicle Type OIV Threshold (m/s) ASI Threshold (--) 9.1 12.2 1.0 1.4 1.9 HF Best Case Belted < 65 Passenger Car 0.5% 1.7% 0.5% 1.2% 3.9% Worst Case Unbelted ≥ 65 LTV 1.1% 3.4% 0.8% 2.0% 6.5% N Best Case Belted < 65 Passenger Car 0.3% 1.5% 0.2% 0.7% 3.9% Worst Case Unbelted ≥ 65 LTV 9.0% 32.0% 7.6% 24.5% 64.3% TALT Best Case Belted < 65 Passenger Car 0.04% 1.2% 0.02% 0.3% 10.0% Worst Case Unbelted ≥ 65 LTV 24.6% 89.3% 19.0% 79.0% 99% Limitations One of the limitations of this work is how few occupants suffered MAIS2+F injuries in both the training and test datasets. The value of the models lies in their ability to be able to predict occupants who are injured. The fewer injured occupants the models are exposed to in the training data, the weaker they will be when it comes to accurately predicting these cases in the test data. Additionally, because the F2 scores do not consider true negative cases, their calculations are fully dependent on the presence of injured occupants. The fewer injury cases there are in the test data, the less reliable the F2 score. Additionally, there were too few MAIS3+F injuries in the dataset to build MAIS3+F injury risk curves. This prevents extensive comparison between this work and previous MAIS3+F injury work. Another limitation to this study is that the inclusion criteria do not consider vehicle incompatibility. The size and weight differences between two colliding vehicles were not restricted. Despite this limitation, some of this incompatibility should be accounted for in the MDV values. The other metrics, however, may not compensate for this vehicle incompatibility as well. Conclusions For each of the crash severity metrics, whole body and region-specific MAIS2+F injury risk curves for frontal crashes were constructed with NASS/CDS cases in the IAD and tested with CISS cases in the IAD. In general, the higher the crash severity metric, the higher the risk of an occupant sustaining an MAIS2+F injury. Based on the F2 score, the OIV metric paired with a binary RA (>15 G) was the best at predicting MAIS2+F in the test dataset. The MDV, OIV, OLC, and VPI metrics, however, performed similarly well. Factors other than the crash severity metric are important predictors of occupant injury. Belted occupants and occupants under 65 years old were at a lower risk of MAIS2+F injury. These covariates can drastically affect the probability of injury even when the crash severity metric is held constant. At the maximum OIV, the best-case scenario

98 had a 15.8% probability of MAIS2+F injury, but the worst-case scenario had a 97.3% probability of MAIS2+F injury.