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.

99 6 Analyze IAD for Side 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 side crashes. The side impact 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. The injury models were evaluated with an equivalent test dataset composed of side impact 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 183 sampled occupants representing 66,124 occupants involved in real- world side impact crashes from NASS/CDS and 101 sampled occupants representing 72,372 occupants involved in real-world side impact crashes from CISS. The training dataset contained occupants in real-world side impacts sampled in NASS/CDS with associated EDR data in the VT EDR database. Similarly, the test dataset contained occupants in real-world side impacts sampled in CISS with associated EDR data in the VT EDR database. 6.2.1 Metric Computation and Validation The methods used to validate the computation of each metric are described in the frontal crash IAD analysis chapter (Section 5.2.1). When building the side crash IAD, the occupant had to cross the lateral FSM boundary prior to crossing the longitudinal FSM boundary, or the occupant had to cross only the lateral FSM boundary. Since this method was used to select laterally dominant crashes, only the lateral delta-v crash pulse was used to calculate OIV. Additionally, only the lateral delta-v crash pulse was used to compute the remaining metrics: MDV, OLC, ASI, VPI. 6.2.2 Injury Risk Modeling The injury models were developed using 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 rating was considered injured. Occupants with an unknown injury severity were excluded unless their treatment status was for that of a fatal injury. 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 enough cases to build reliable predictive models (Table 4-8). The following predictor variables were used within the model to predict injury: • Crash Severity Metrics. Five injury models were built for side 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 lateral delta-v. The delta-v versus time series data were obtained from the EDR.

100 • Belt Use. Belt use was a binary variable, where a value of 1 indicates an occupant 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. • Impact Type. Impact type was a binary variable, where 1 indicates a nearside impact and 0 indicates a far-side impact. Occupant seating location and GAD were used to determine whether the impact was a far-side impact or a nearside impact. Drivers with GAD = left and right front passengers with GAD = right are nearside impacts. Drivers with GAD = right and right front passengers with GAD = left are far-side impacts. • Sex. Sex was a binary variable, where 1 indicates male and 0 indicates female. • BMI. Body mass index (kg/m2) was a binary 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 binary 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 our dataset fell into one of these categories. Vehicle type was a binary variable, where a value of 1 indicates a PC and 0 indicates an LTV. Impact type was included because this is a variable that describes the seating location of the occupant relative to what side of the vehicle was struck. The NASS/CDS airbag table was used to determine presence of a side airbag; however, very little data were available, so it was not possible to include it as a predictor variable. First, logistic regression models were developed using the full set of 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. 6.2.3 Body 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. The NASS/CDS regions were combined into HF, N, and TALT regions (Table

101 6-1). There were too few MAIS2+F injuries in the N and TALT regions to construct side body region injury models. Table 6-1. NASS/CDS body regions used to form the side body regions models. NASS/CDS Region NASS/CDS Specific Anatomic Structure Model Region Head All HF Face All Neck All N – not modeled Thorax All TALT – not modeled Abdomen All Spine C-Spine N – not modeled T-Spine TALT – not modeled L-Spine Upper extremity All Injury risk not modeled Lower extremity All 6.2.4 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. 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. 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. A higher F2 score indicates better predictive capability of the model and was the primary comparison metric. When calculating precision and recall, a decision threshold in the form of a percent injury risk must be chosen. This threshold was found 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. This is because a lower decision threshold corresponds to a higher recall value. The NASS/CDS and CISS weighted values were used to calculate each of these comparison metrics. Overall Injury Model Results 6.3.1 Initial Injury Risk Models The initial models used belt status, impact type, age, sex, obesity, and seating location, and vehicle type as covariates. Each model additionally used one of the five crash severity metrics as a covariate. Table D-1 through Table D-5 show the regression coefficients for each of the injury risk models. 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.

102 For a binary covariate, the non-baseline condition reduces the injury risk. For example, the metrics all have positive coefficients, because an increase in any of the metrics will heighten the occupant’s risk of injury. Additionally, impact type always has a positive coefficient because occupants in nearside crashes are more likely to suffer an MAIS2+F injury. Equation 8 is the final form of the model and Equation 11 is the logit expanded. 𝑃𝑃[𝑀𝑀𝐴𝐴𝑀𝑀𝑀𝑀2+ F] = 1 1 + 𝑃𝑃−𝑅𝑅𝑃𝑃𝑙𝑙𝑃𝑃𝑙𝑙 (8) 𝑅𝑅𝑃𝑃𝑙𝑙𝑃𝑃𝑙𝑙 = 𝛽𝛽0 + 𝛽𝛽1 ⋅ (𝑃𝑃𝑃𝑃𝑖𝑖𝐴𝐴𝑃𝑃𝐴𝐴 𝑚𝑚𝑃𝑃𝑙𝑙𝑃𝑃𝑃𝑃𝑃𝑃) + 𝛽𝛽2 ⋅ 𝑏𝑏𝑃𝑃𝑅𝑅𝑙𝑙_𝑃𝑃𝑙𝑙𝑅𝑅𝑙𝑙𝐴𝐴𝑃𝑃 + 𝛽𝛽3 ⋅ 𝑃𝑃𝑚𝑚𝑡𝑡𝑅𝑅𝑃𝑃𝑙𝑙_𝑙𝑙𝐴𝐴𝑡𝑡𝑃𝑃 + 𝛽𝛽4 ⋅ 𝑅𝑅𝑙𝑙𝑃𝑃 + 𝛽𝛽5 ⋅ 𝑃𝑃𝑃𝑃𝑠𝑠 + 𝛽𝛽6 ⋅ 𝑃𝑃𝑏𝑏𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃 + 𝛽𝛽7 ⋅ 𝑃𝑃𝑃𝑃𝑅𝑅𝑙𝑙𝑃𝑃𝑃𝑃𝑙𝑙𝑅𝑅𝑃𝑃𝑃𝑃𝑅𝑅𝑙𝑙𝑃𝑃𝑃𝑃𝑃𝑃 + 𝛽𝛽8 ⋅ 𝑣𝑣𝑃𝑃ℎ𝑃𝑃𝑃𝑃𝑅𝑅𝑃𝑃𝑙𝑙𝐴𝐴𝑡𝑡𝑃𝑃 (10) 6.3.2 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. The metric, belt status, and impact type were significant in every initial model. Table 6-2 through Table 6-6 show the regression coefficients for each of the final models. Table 6-2. Parameters for the MDV final side 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 -2.946 0.912 0.002** Lateral Delta-v β1, Delta-v (m/s) 0.386 0.130 < 0.004** Belt Status β2, Belted -2.595 0.878 0.004** Impact Type β3, Nearside 1.632 0.645 0.014**

103 Table 6-3. Parameters for the OIV final side 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 -2.858 0.888 0.002** Lateral OIV β1, OIV (m/s) 0.383 0.130 0.005** Belt Status β2, Belted -2.573 0.873 0.005** Impact Type β3, Nearside 1.566 0.646 0.018** Table 6-4. Parameters for the OLC final side 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 -2.044 0.782 0.011** Lateral OLC β1, OLC (G) 0.158 0.049 0.002** Belt Status β2, Belted -2.401 0.917 0.011** Impact Type β3, Nearside 1.856 0.648 0.006** Table 6-5. Parameters for the ASI final side 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 -3.070 0.967 0.002** Lateral ASI β1, ASI 1.947 0.581 0.001** Belt Status β2, Belted -2.429 0.925 0.011** Impact Type β3, Nearside 1.831 0.690 0.010** Table 6-6. Parameters for the VPI final side 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 -3.273 0.987 0.002** Lateral VPI β1, VPI (m/s2) 0.010 0.003 0.002** Belt Status β2, Belted -2.418 0.902 0.009** Impact Type β3, Nearside 1.848 0.672 0.008** 6.3.3 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 belted and unbelted 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. This process was performed once for a belted occupant and again for an unbelted occupant for both near- and far-side crashes. Figure 6-1 through Figure 6-5 plot the risk curves for a belted and unbelted occupant in near- and far-side crashes for all five models. Because the original values for these metrics, besides VPI, were either zero or negative, they were multiplied by -1 prior to plotting to maintain a positive x-axis. The shaded regions represent 95% confidence intervals.

104 Figure 6-1. Side impact MDV injury risk curves for drivers and right front passengers, at least 13 years old, in passenger vehicles. Figure 6-2. Side impact OIV injury risk curves for drivers and right passengers, at least 13 years old, in passenger vehicles.

105 Figure 6-3. Side impact OLC injury risk curves for drivers and right front passengers, at least 13 years old, in passenger vehicles. Figure 6-4. Side impact ASI injury risk curves for drivers and right front passengers, at least 13 years old, in passenger vehicles.

106 Figure 6-5. Side impact VPI injury risk curves for drivers and right front passengers, at least 13 years old, in passenger vehicles. Body Region Model Results 6.4.1 Initial Injury Risk Models The initial models used belt status, impact type, age, sex, obesity, seating location, and vehicle type as covariates, in addition to one of the severity metrics. Seating location and the vehicle type were excluded as covariates because all nine injured occupants were passenger car drivers. A p- value < 0.05 was considered significant. For the HF models, the metric, belt status, and side impact type were significant. Equation 8 is the final form of the model and Equation 11 is the logit expanded for the initial models. The full models are not shown in table form. 𝑃𝑃[𝑀𝑀𝐴𝐴𝑀𝑀𝑀𝑀2+ F] = 1 1 + 𝑃𝑃−𝑅𝑅𝑃𝑃𝑙𝑙𝑃𝑃𝑙𝑙 (8) 𝑅𝑅𝑃𝑃𝑙𝑙𝑃𝑃𝑙𝑙 = 𝛽𝛽0 + 𝛽𝛽1 ⋅ (𝑃𝑃𝑃𝑃𝑖𝑖𝐴𝐴𝑃𝑃𝐴𝐴 𝑚𝑚𝑃𝑃𝑙𝑙𝑃𝑃𝑃𝑃𝑃𝑃) + 𝛽𝛽2 ⋅ 𝑏𝑏𝑃𝑃𝑅𝑅𝑙𝑙_𝑃𝑃𝑙𝑙𝑅𝑅𝑙𝑙𝐴𝐴𝑃𝑃 + 𝛽𝛽3 ⋅ 𝑃𝑃𝑚𝑚𝑡𝑡𝑅𝑅𝑃𝑃𝑙𝑙_𝑙𝑙𝐴𝐴𝑡𝑡𝑃𝑃 + 𝛽𝛽4 ⋅ 𝑅𝑅𝑙𝑙𝑃𝑃 + 𝛽𝛽5 ⋅ 𝑃𝑃𝑃𝑃𝑠𝑠 + 𝛽𝛽6 ⋅ 𝑃𝑃𝑏𝑏𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃 + 𝛽𝛽7 ⋅ 𝑃𝑃𝑃𝑃𝑅𝑅𝑙𝑙𝑃𝑃𝑃𝑃𝑙𝑙𝑅𝑅𝑃𝑃𝑃𝑃𝑅𝑅𝑙𝑙𝑃𝑃𝑃𝑃𝑃𝑃 + 𝛽𝛽8 ⋅ 𝑣𝑣𝑃𝑃ℎ𝑃𝑃𝑃𝑃𝑅𝑅𝑃𝑃𝑙𝑙𝐴𝐴𝑡𝑡𝑃𝑃 (10) 6.4.2 Final Injury Risk Models Based on the initial regression models, a second set of injury models was developed. We attempted to develop models that used the significant covariates from the initial models. Upon doing this, the covariate coefficients and p-values were highly indicative of overfitting the models to the training data. To avoid this overfitting, we constructed the final models using only the metrics, with no addition covariates.

107 6.4.2.1 Body Region 1: HF Table 6-7 through Table 6-11 show the regression coefficients for each of the final side crash HF models. In every final model, the metric was significant. A p-value < 0.05 was considered significant and is denoted by ** in the parameter tables. Table 6-7. Parameters for the MDV logistic regression side 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 -5.100 1.820 0.007** Lateral Delta-v β1, Delta-v (m/s) 0.512 0.212 0.019** Table 6-8. Parameters for the OIV logistic regression side 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 -5.103 1.801 0.006** Lateral OIV β1, OIV (m/s) 0.524 0.212 0.016** Table 6-9. Parameters for the OLC logistic regression side 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 -3.380 1.087 0.003** Lateral OLC β1, OLC (G) 0.188 0.078 0.019** Table 6-10. Parameters for the ASI logistic regression side 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 -4.748 1.587 0.004** Lateral ASI β1, ASI 2.368 0.912 0.012** Table 6-11. Parameters for the VPI logistic regression side 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 -5.078 1.654 0.003** Lateral VPI β1, VPI (m/s2) 0.012 0.004 0.008** 6.4.3 Injury Risk Curves The predict function in R was used to obtain injury probability values for each of the five final models. This function requires a new data frame as input. The new data frame must have constant values for every covariate except the crash severity metric. The severity metric was stepped from zero to the value where injury risk was 100% in increments of 0.01. 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. Sex was held constant as male when relevant. This process was performed for belted drivers and unbelted drivers. Figure 6-6 through Figure 6-10 plot the risk curves for a belted and unbelted occupant for all five models. Because the original values for these metrics, besides VPI, were either zero or negative, they were multiplied by -1 prior to plotting to maintain a positive x-axis. The shaded regions represent 95% confidence intervals.

108 Figure 6-6. Side impact MDV injury risk curves for the HF region for front-seated occupants at least 13 years old. Figure 6-7. Side impact OIV injury risk curves for the HF region for front-seated occupants at least 13 years old.

109 Figure 6-8. Side impact OLC injury risk curves for the HF region for front-seated occupants at least 13 years old. Figure 6-9. Side impact ASI injury risk curves for the HF region for front-seated occupants at least 13 years old.

110 Figure 6-10. Side impact VPI injury risk curves for the HF region for front-seated occupants at least 13 years old. 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 6-12 and Table 6-13). The accuracy of each model on the test dataset was also calculated. MDV and OIV yield similar decision thresholds, as do OLC, ASI, and VPI (Table 6-12). Additionally, the training F2 scores for each model are very similar. ASI and VPI performed better on the test dataset than the rest of the models with F2 scores of 0.41 and 0.42, respectively. For MDV and OIV, the decision threshold fell above every probability prediction in the test dataset, making it impossible to calculate precision and recall and, therefore, an F2 score. For every HF model, the decision threshold fell above every probability prediction in the test dataset, making it impossible to calculate precision and recall and, therefore, an F2 score. This is because the test dataset only had two injury cases. All of the 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 0.80, indicating the models perform fairly well. However, similar 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

111 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. Table 6-12. F2 scores for the five metrics’ side models. These values come from the final model. Metric Model Decision Threshold Training Test F2 Scores ROC AUC Accuracy Precision Recall F2 Score MDV 51% 0.71 0.85 98% 0.00 0.00 0.00 OIV 52% 0.71 0.86 98% 0.00 0.00 0.00 OLC 14% 0.69 0.80 89% 0.02 0.08 0.05 ASI 14% 0.70 0.83 89% 0.15 0.75 0.41 VPI 13% 0.71 0.84 88% 0.14 0.78 0.42 Table 6-13. F2 scores for the side HF models. Metric Model Decision Threshold Training Test F2 Scores ROC AUC Accuracy Precision Recall F2 Scores MDV 66% 0.62 0.94 99% 0.00 0.00 0.00 OIV 69% 0.62 0.93 99% 0.00 0.00 0.00 OLC 66% 0.62 0.95 99% 0.00 0.00 0.00 ASI 68% 0.62 0.94 99% 0.00 0.00 0.00 VPI 68% 0.62 0.94 99% 0.00 0.00 0.00 Discussion Belt status, impact type, and the crash severity metric were significant predictors in all five final regression models. For each severity metric holding all covariates constant, the risk of injury was substantially lower for a belted occupant than for an unbelted occupant. Additionally, risk of injury was greater for occupants in nearside crashes than for those in far-side crashes. Using the F2 score as a metric of prediction capability, the ASI and VPI models are most equipped to accurately predict occupant injury outcomes in side crashes; however, their F2 scores were still fairly low. The metric was a significant predictor in all five HF final regression models. F2 scores were calculated for the training data but could not be calculated for the test dataset, due to how few HF injury cases were available. The injury risk curves do not always display realistic probabilities. For unbelted occupants in nearside crashes, each metric predicts a non-zero chance of an MAIS2+F injury at the metric value of zero. This is likely because this dataset only includes two cases where an unbelted occupant suffered an MAIS2+F injury in a nearside crash. Additionally, the 95% confidence intervals for all of the curves cover a wide range of injury risk values, particularly for far-side crashes. 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 6-14 and Table 6-15). For these models, a best-case scenario occupant would be a belted occupant in a far-side crash. An unbelted occupant in a nearside crash would be the worst- case scenario within these models. 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 both the best- and worst-case scenario occupant, the risk associated with the preferred OIV threshold is similar

112 to the risk associated with the highest ASI threshold. This indicates there is a lack of consistency between the two metrics in terms of their allowable MAIS2+F percent injury risk. Table 6-14. Injury risk associated with the current OIV and ASI thresholds for the best- and worst-case side impact scenarios. Crash and Occupant Conditions Injury Risk Scenario Belt Status Impact Type OIV Threshold (m/s) ASI Threshold (--) 9.1 12.2 1.0 1.4 1.9 Best Case Belted Far-side 12.5% 32.0% 2.8% 5.9% 14.2% Worst Case Unbelted Nearside 90.0% 96.7% 67.0% 81.6% 92.1% Table 6-15. HF injury risk associated with the current OIV and ASI thresholds for the best- and worst-case side impact scenarios. Occupant Scenario Injury Risk OIV Threshold (m/s) ASI Threshold (--) 9.1 12.2 1.0 1.4 1.9 Best and Worst Case 41.6% 78.3% 8.5% 19.3% 43.8% 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. Similar to the frontal models developed, there were too few MAIS3+F injuries in the side impact dataset to build MAIS3+F injury risk curves. 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 incompatibility as well. Conclusions For each of the candidate injury metrics, whole body and HF body region MAIS2+F injury risk curves for side impact crashes were constructed with NASS/CDS cases in the IAD and tested with CISS cases in the IAD. Based on the findings of the literature review conducted as part of this research effort, this is the first study to use real-world crash data with EDR data to investigate the performance of vehicle-based metrics in side impact crashes. Similar to the frontal crash findings, the higher the crash severity metric, the higher the risk of an occupant sustaining an MAIS2+F injury. Based on the F2 score, the ASI and VPI metrics were the best at predicting MAIS2+F in the test dataset. Factors other than the crash severity metric were also found to be important predictors of occupant injury. Belted occupants and far-side occupants 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,

113 the best-case scenario had a 32% probability of MAIS2+F injury, but the worst-case scenario had a 96.7% probability of MAIS2+F injury.