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.

140 9 Analyze the Effect of Pjoint and Two Modified VPI Metrics on the Frontal Crash Injury Prediction Models Introduction The purpose of this chapter is to present frontal crash full body injury models (1) using Pjoint as an additional model covariate and (2) using a vehicle-specific version of the VPI metric. These models will use a subset of the cases from the Chapter 5 frontal crash dataset. The IAD was composed of NASS/CDS crashes with available EDR data. This frontal IAD was used to train MAIS2+F whole body and region-specific logistic models based on five candidate injury metrics, MDV, OIV, OLC, VPI, and ASI. Methods 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. The occupants in the training and test datasets were from the IAD and subsets of the same dataset analyzed in Chapter 5. Information necessary to calculate the modified VPI metrics and the Pjoint covariate were obtained by analyzing the NHTSA crash test database. After obtaining the crash test data, a vehicle matching process was performed so the VPI metrics and Pjoint covariates could be applied to the maximum number of cases within the IAD. 9.2.1 Pjoint Definition In the NHTSA frontal barrier test, the test vehicle is towed into a fixed rigid barrier at 56 km/h (35 mph). Two ATDs, sometimes referred to as crash test dummies, are placed in the vehicle. A Hybrid III 50th percentile male ATD is seated in the driver seat, and a Hybrid III 5th percentile female ATD is seated in the front outboard passenger seat. Both ATDs are belted. For each ATD, the probability of injury at the AIS 3 level or greater (AIS3+) is computed for the head, neck, and chest, and the probability of AIS 2+ injury is computed for the femur. NHTSA uses this joint measure of risk experienced by the front row ATDs to assign a star rating for both front row occupants. Pjoint values of 5%, 10%, and 15% are the thresholds for five-star, four-star, and three- star vehicles, respectively. A joint measure of risk is computed for each ATD (Equation 17): 𝑃𝑃𝑁𝑁𝑓𝑓𝑖𝑖𝑁𝑁𝑓𝑓 = 1 − (1 − 𝑃𝑃ℎ𝑁𝑁𝑙𝑙𝑠𝑠)(1− 𝑃𝑃𝑙𝑙ℎ𝑁𝑁𝑁𝑁𝑓𝑓)(1 − 𝑃𝑃𝑙𝑙𝑜𝑜𝑠𝑠𝑓𝑓𝑁𝑁𝑁𝑁𝑁𝑁)�1 − 𝑃𝑃𝑝𝑝𝑁𝑁𝑙𝑙𝑝𝑝𝑖𝑖𝑁𝑁� (16) For each crash test in the dataset, the Pjoint value was computed for both the driver ATD and, if applicable, the RF passenger ATD. If data were missing for any of the four body regions used to compute Pjoint, that ATD was excluded from the dataset. After the crash test Pjoint dataset was finalized, vehicle matching was performed. The vehicle matching process used IIHS data to determine which combinations of vehicle make, model, model year, body type, and number of doors correlated to which vehicles in the frontal IAD. For example, a four-door Chevrolet Cruze sedan remained essentially unchanged for model years 2011 to 2015.

141 Therefore, if Pjoint was calculated for ATDs in a 2013 four-door Chevrolet Cruze sedan crash test, the Pjoint values were applied to model years 2011 to 2015 for this vehicle type in the frontal crash IAD. The resulting dataset was a subset of the frontal crash dataset presented in Chapter 5, comprising 121 sampled cases representing 48,101 weighted cases. 9.2.2 Modified VPI Metric The ISO VPI standard recommends values of 2,500 N/m, 0.03 m, and 1 kg for vehicle stiffness (k), seat belt slack (s), and occupant mass (m). These values were used to calculate the VPI metric for each of the models presented so far. Since vehicles can vary in vehicle stiffness and seat belt slack, two modified versions of VPI were tested on the frontal crash injury prediction model: a vehicle-specific VPI metric and an occupant-specific VPI metric. For each case in our crash test list, NHTSA’s SignalBrowser software was used to access and download accelerometer data from the rear of the vehicle, as well as the ATD pelvis accelerometer data for both front seated ATDs, all in UDS file format. Acceleration data from the rear of the vehicle was necessary to compute the vehicle stiffness. The accelerometers must have been located toward the rear of the vehicle because there is little to no deformation at the rear. Pelvis data from the ATDs were necessary to compute the seat belt slack. When acceleration data were available from the left rear and right rear of the vehicle, the acceleration data were averaged together. If only the left rear or right rear data were available, that acceleration was used. When acceleration data were available from the driver and RF passenger pelvises, the acceleration data were averaged together. If only the driver or RF passenger data were available, that acceleration was used. If a crash test was missing either all the rear vehicle acceleration data or all the pelvis acceleration data, the case was excluded. All the data were filtered at channel filter class (CFC) 180 based on the procedures set forth by SAE J211 (SAE 2007). NHTSA’s SISAME software was then used to extract the vehicle stiffness and seat belt slack values for every crash test vehicle with the correct available acceleration data. Following the acquisition of these parameters, the same vehicle matching process was again performed. The resulting dataset was a subset of the frontal crash dataset presented in Chapter 5, comprising 118 sampled cases representing 45,182 weighted cases. To compute the vehicle-specific VPI metric, the acquired stiffness and slack values for each case were implemented in the VPI calculation process, rather than the standard 0.03 m and 2,500 N/m values. To compute the occupant-specific VPI metric, occupant mass data were extracted from the occupant table in the NASS/CDS database. The VPI computation process was repeated using the occupant mass in each case instead of a standard 1 kg. 9.2.3 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

142 be built, as this injury category did not contain enough cases to build reliable predictive models (Table 4-7). The following predictor variables were used specifically to analyze the predictive capabilities of models using Pjoint and modified VPI metrics as predictors for injury: • Crash Severity Metrics. To analyze the effect of adding in Pjoint as a covariate to the models, five injury models were built for frontal crashes. Each model used one of the five 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. • Pjoint. Pjoint was modeled as a continuous predictor variable. Pjoint was calculated using ATD accelerometer data from a series of vehicle crash tests. • Vehicle- and Occupant-Specific VPI. To analyze the effect of the VPI metric being tailored to specific vehicles, two injury models were built for frontal crashes. One model used vehicle-specific VPI as an independent predictor variable, while the other used occupant-specific VPI. These modified VPI metrics are a function of the longitudinal delta- v and consider the vehicle stiffness, seat belt slack, and occupant mass. The remaining covariates are listed below and were used in the initial models for both the Pjoint and the modified VPI analyses: • Belt Use. Belt use was 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 was a categorical variable, where 1 indicates ≥ 65 years old and 0 indicates ≥ 13 years old and < 65 years old. • Sex. Sex was a categorical variable, where 1 indicates male and 0 indicates female. • BMI. Body mass index (kg/m2) was 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 was 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 of the vehicles in the dataset fell into one of these categories. Vehicle type was a categorical variable, where a value of 1 indicates a PC and 0 indicates an LTV. • PDOF. For the analysis, PDOF was redefined to go 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.

143 Overall Injury Model Results 9.3.1 Initial Injury Risk Models The initial models used seat belt usage, sex, age, BMI, seating location, vehicle type, and PDOF as covariates. The Pjoint models additionally used Pjoint as a covariate. Each model additionally used one of the five severity metrics as a covariate. Table 9-1 through Table 9-7 show the regression coefficients for each of the injury risk models. 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. In every Pjoint model, Pjoint was not significant. In every modified VPI model, modified VPI was not significant. Since the covariates of interest were never significant, final models were not built. Equation 8 is the form of the initial model and Equation 18 is the logit expanded. Pjoint was not included as a covariate in the modified VPI models. 𝑃𝑃[𝑀𝑀𝐴𝐴𝑀𝑀𝑀𝑀2+ F] = 1 1 + 𝑃𝑃−𝑅𝑅𝑃𝑃𝑙𝑙𝑃𝑃𝑙𝑙 (8) 𝑅𝑅𝑃𝑃𝑙𝑙𝑃𝑃𝑙𝑙 = 𝛽𝛽0 + 𝛽𝛽1 ⋅ (𝑃𝑃𝑃𝑃𝑖𝑖𝐴𝐴𝑃𝑃𝐴𝐴 𝑚𝑚𝑃𝑃𝑙𝑙𝑃𝑃𝑃𝑃𝑃𝑃) + 𝛽𝛽2 ⋅ 𝑏𝑏𝑃𝑃𝑅𝑅𝑙𝑙_𝑃𝑃𝑙𝑙𝑅𝑅𝑙𝑙𝐴𝐴𝑃𝑃 + 𝛽𝛽3 ⋅ 𝑃𝑃𝑃𝑃𝑠𝑠 + 𝛽𝛽4 ⋅ 𝑅𝑅𝑙𝑙𝑃𝑃 + 𝛽𝛽5 ⋅ 𝑃𝑃𝑏𝑏𝑃𝑃𝑃𝑃𝑃𝑃 + 𝛽𝛽6 ⋅ 𝑃𝑃𝑃𝑃𝑅𝑅𝑙𝑙𝑃𝑃𝑃𝑃𝑙𝑙𝑅𝑅𝑃𝑃𝑃𝑃𝑅𝑅𝑙𝑙𝑃𝑃𝑃𝑃𝑃𝑃 + 𝛽𝛽7 ⋅ 𝑉𝑉𝑃𝑃ℎ𝑃𝑃𝑃𝑃𝑅𝑅𝑃𝑃𝑇𝑇𝐴𝐴𝑡𝑡𝑃𝑃 + 𝛽𝛽8 ⋅ 𝐺𝐺𝐴𝐴𝑃𝑃 + 𝛽𝛽9 ⋅ 𝑃𝑃𝑖𝑖𝑃𝑃𝑃𝑃𝑃𝑃𝑙𝑙 (17)

144 9.3.1.1 Initial Pjoint Models Table 9-1. Parameters for the MDV-Pjoint 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 -3.639 2.205 0.106 Longitudinal MDV β1, MDV (m/s) 0.319 0.217 0.149 Belt Use β2, Belted -2.395 0.806 0.005** Sex β3, Male -0.792 0.982 0.424 Age β4, Age ≥ 65 1.089 0.925 0.246 BMI β5, BMI ≥ 30 kg/m2 -1.327 0.876 0.137 Seating Location β6, Driver Seat -0.270 1.133 0.813 Vehicle Type β7, Passenger Car -1.336 1.336 0.323 PDOF β8, PDOF -0.010 0.038 0.799 Pjoint β9, Pjoint 13.814 9.235 0.142 Table 9-2. Parameters for the OIV-Pjoint 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 -3.449 2.083 0.105 Longitudinal OIV β1, OIV (m/s) 0.281 0.248 0.265 Belt Use β2, Belted -2.366 0.802 0.005** Sex β3, Male -0.707 1.008 0.487 Age β4, Age ≥ 65 1.124 1.045 0.288 BMI β5, BMI ≥ 30 kg/m2 -1.264 0.854 0.146 Seating Location β6, Driver Seat -0.292 1.178 0.805 Vehicle Type β7, Passenger Car -1.109 1.308 0.401 PDOF β8, PDOF -0.009 0.038 0.816 Pjoint β9, Pjoint 14.004 10.009 0.169 Table 9-3. Parameters for the OLC-Pjoint 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 -2.668 1.750 0.135 Longitudinal OLC β1, OLC (g) 0.139 0.122 0.261 Belt Use β2, Belted -2.401 0.854 0.007** Sex β3, Male -0.741 1.056 0.487 Age β4, Age ≥ 65 1.068 1.025 0.303 BMI β5, BMI ≥ 30 kg/m2 -1.375 0.947 0.154 Seating Location β6, Driver Seat -0.242 1.211 0.842 Vehicle Type β7, Passenger Car -1.087 1.319 0.414 PDOF β8, PDOF -0.006 0.039 0.870 Pjoint β9, Pjoint 14.945 10.082 0.146 Table 9-4. Parameters for the ASI-Pjoint frontal logistic regression model used to predict occupant MAIS2+F injuries. ** indicates statistical significance (p-value < 0.05).

145 Predictor Variable Parameter Coefficient Std. Error p-Value --- β0, Intercept -3.075 1.871 0.108 Longitudinal ASI β1, ASI 1.777 1.792 0.327 Belt Use β2, Belted -2.319 0.810 0.006** Sex β3, Male -0.647 1.000 0.521 Age β4, Age ≥ 65 1.067 1.028 0.305 BMI β5, BMI ≥ 30 kg/m2 -1.394 0.937 0.144 Seating Location β6, Driver Seat -0.280 1.181 0.814 Vehicle Type β7, Passenger Car -1.091 1.316 0.412 PDOF β8, PDOF -0.006 0.039 0.879 Pjoint β9, Pjoint 15.351 10.145 0.138 Table 9-5. Parameters for the VPI-Pjoint 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 -3.714 2.065 0.079 Longitudinal VPI β1, VPI (m/s2) 0.009 0.007 0.243 Belt Use β2, Belted -2.333 0.798 0.006** Sex β3, Male -0.692 0.989 0.488 Age β4, Age ≥ 65 1.174 0.964 0.230 BMI β5, BMI ≥ 30 kg/m2 -1.545 0.981 0.123 Seating Location β6, Driver Seat -0.149 1.181 0.900 Vehicle Type β7, Passenger Car -1.186 1.301 0.367 PDOF β8, PDOF -0.012 0.038 0.757 Pjoint β9, Pjoint 16.622 9.957 0.102 9.3.1.2 Initial Modified VPI Models Table 9-6. Parameters for the vehicle-specific VPI 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 -1.537 2.011 0.449 Longitudinal Vehicle-Specific VPI β1, VPI (m/s2) 0.011 0.007 0.139 Belt Use β2, Belted -2.048 0.885 0.026** Sex β3, Male -0.569 1.008 0.576 Age β4, Age ≥ 65 1.174 0.941 0.219 BMI β5, BMI ≥ 30 kg/m2 -1.589 1.031 0.131 Seating Location β6, Driver Seat -0.854 1.164 0.467 Vehicle Type β7, Passenger Car -0.974 1.380 0.484 PDOF β8, PDOF -0.019 0.042 0.654 Table 9-7. Parameters for the occupant-specific VPI frontal logistic regression frontal 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 -3.408 2.998 0.262 Longitudinal Occupant-Specific VPI β1, VPI (m/s2) 0.110 0.067 0.108 Belt Use β2, Belted -2.124 1.071 0.054 Sex β3, Male -0.354 1.039 0.735 Age β4, Age ≥ 65 1.692 1.070 0.121 BMI β5, BMI ≥ 30 kg/m2 -1.597 1.065 0.141 Seating Location β6, Driver Seat -1.025 1.234 0.411 Vehicle Type β7, Passenger Car -0.798 1.192 0.507 PDOF β8, PDOF -s0.028 0.031 0.376

146 Conclusions For each of the crash severity metrics, a set of initial models were built for front crashes using all the same initial covariates, plus the Pjoint covariate. Two additional initial models were constructed using vehicle- and occupant-specific VPI crash severity metrics. These models were constructed using the NASS/CDS cases in the IAD. These models were not tested on a new dataset because none of the models yielded significance for the Pjoint covariate or the augmented VPI metrics. The only significant covariate in the Pjoint models was belt status. Belt status was also significant in the vehicle-specific VPI model, while no covariates were significant in the occupant- specific VPI model.