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.

221 15 Conclusions Summary of Current Practices 15.1.1 Existing MASH Roadside Hardware Crash Testing and Occupant Risk Procedures MASH prescribes the current procedures for crash testing and evaluation of roadside safety hardware devices such as guardrail and impact attenuators. MASH procedures evaluate test results based on the structural adequacy of the device, the post-impact trajectory of the vehicle, and the potential for injury to vehicle occupants. As a surrogate for instrumented ATDs, the FSM is currently used to assess vehicle occupant injury risk. The FSM assumes occupant injury risk occurs via two possible mechanisms: the OIV and the RA. OIV is the occupant’s velocity relative to the vehicle occupant compartment at the instant the occupant first crosses a simplified occupant compartment boundary. RA is the maximum 10-ms average vehicle acceleration subsequent to occupant impact with the simplified occupant compartment boundary. Larger OIV and RA values correspond to a higher risk of occupant injury, and computed values are compared to established threshold values to limit occupant injury risk potential. As a base assumption of the FSM is that the vehicle occupant compartment remains intact, MASH procedures prescribe thresholds for static post-test vehicle occupant compartment deformations. These intrusion thresholds are prescribed for nine different vehicle areas (windshield, side panel, toe pan, etc.), with the values varying by vehicle area. 15.1.2 Gaps and Research Needs • Since the inception of the FSM, passive safety has been greatly improved and become far more prevalent in the U.S. fleet. Seat belt usage rates have exceeded 80%, airbags have been mandated for all new vehicles, and vehicle structures include collapsible steering columns and crumple zones to reduce RA. There is a need to investigate the current FSM thresholds in a modern vehicle fleet with improved passive safety. • The influence of crash and occupant factors other than the value of the FSM parameters, such as belt use, occupant sex, and occupant age, on resulting occupant injury risk are not well known. Unlike vehicle crash testing, which has available injury risk curves to translate injury criteria values to a corresponding probability of injury, the current roadside hardware crash injury metrics lack these injury risk curves. • European roadside hardware crash testing standards also implement the ASI metric as an instrumented ATD surrogate. This metric assumes the vehicle occupant is belted. Besides OIV, RA, and ASI, other vehicle-based crash severity metrics exist, including the MDV, OLC, and the VPI. There is a need to compare the ability of the FSM to predict vehicle occupant injury with the capabilities of other vehicle-based metrics. • There has not been a detailed investigation of how the current MASH area-specific intrusion thresholds correlate with real-world crash occupant injury. While the study used as a partial basis for the MASH intrusion limits did use real-world crashes, the study did not investigate all the vehicle areas with prescribed intrusion limits.

222 Methods of Evaluation 15.2.1 Building the IAD NASS/CDS and CISS were used to build the IAD. NASS/CDS and CISS are two in-depth, nationally representative databases comprising real-world crashes in the United States. Both databases provide detailed occupant injury data and can be paired with the VT EDR database to access EDR crash pulse data to compute vehicle-based metrics such as the FSM. NASS/CDS cases were used to train the models, while CISS cases were used to test the models. From the IAD, frontal, side, and oblique crash IAD subsets were constructed. Each of these subsets was filtered through an additional set of inclusion criteria, unique to the crash type. A total of 494 training cases and 351 test cases comprised the frontal IAD, representing 167,330 and 215,913 weighted cases, respectively. A total of 183 training cases and 101 test cases comprised the side IAD, representing 66,124 and 72,372 weighted cases, respectively. A total of 176 training cases and 179 test cases comprised the oblique IAD, representing 67,107 and 103,839 weighted cases, respectively. 15.2.2 Analysis of Candidate Injury Metrics in Frontal, Side, and Oblique Crashes For each of the frontal, side, and oblique IAD subsets, the available NASS/CDS cases were used to train a set of binary logistic regression models to predict MAIS2+F injury. There were not sufficient MAIS3+ injury cases to develop models using this more severe injury severity threshold. Whole body injury models were constructed for the frontal, side, and oblique crash IAD subsets. Three body region models (HF, N, and TALT) were constructed for the frontal crash IAD subset. A head and face body region model was constructed for the side crash IAD subset. There were not enough MAIS2+F injured occupants in the oblique crash IAD subset to construct body region- specific models for this crash type. The candidate metrics investigated in this study were MDV, OIV, OLC, ASI, and VPI. An additional component to the FSM is the occupant RA. This metric was investigated within the frontal crash IAD only and was also modeled as a binary covariate in a model alongside OIV. For each crash type, one model was constructed for each candidate crash severity metric. An initial set of models was constructed using several available covariates that were specific to each crash type. A second and final set of models was then constructed using only the covariates significant (p < 0.05) in the initial models. There were six final whole-body models for the frontal crash IAD and five final body region models. There were five final whole-body and five final body region models for the side crash IAD. There were five final whole-body models for the oblique crash IAD. The final models were then run on the available CISS cases in each IAD crash type subset. The F2 score was used to compare the predictive capability of the models. The F2 score is a metric that tests a model’s ability to correctly classify as many positive cases as possible. Harm is a measurement of the societal cost of a traffic crash and accounts for both medical and indirect costs. A dollar amount is assigned to each injury an occupant experiences, based on the region of injury and the injury’s AIS score. For each occupant in the frontal, side, and oblique crash IADs, the sum of the costs of the injuries experienced was computed to obtain that occupant’s Harm value. Much like the injury models, the NASS/CDS cases in each IAD subset were used to

223 construct linear models to predict Harm. Since Harm can span several orders of magnitude, the linear models were constructed to predict the square root of Harm (√Harm). For each crash type, one model was constructed for each of these five metrics: MDV, OIV, OLC, ASI, and VPI. A set of initial models was constructed, and the final models included only the covariates that were significant in the initial pass of models. The final models were run on the test dataset, comprising CISS cases, and the RMSE was used to assess the amount of error associated with each model. The R2 value was used to measure how much of the models’ variance was accounted for by the included covariates. Two related investigations were also conducted: (1) to determine if considering vehicle-specific restraint performance significantly improved the real-world injury prediction capabilities of vehicle-based metrics, and (2) to compare how well the candidate injury risk metrics predicted ATD-based injury metrics used in full-scale vehicle crash tests. The results from all of these analyses were synthesized and used to propose any new or modified injury risk procedures that may be adopted in a future version of MASH. A sample of previously conducted MASH crash tests was used to assess the possible implications of any proposed new or modified occupant injury risk criteria. 15.2.3 Investigation of MASH Occupant Compartment Intrusion Limits MASH guidelines on acceptable intrusion limits were investigated using real-world crashes available from NASS/CDS. The analysis had three primary components: (1) an update to the study originally serving as a basis for the MASH intrusion guidelines, (2) an evaluation of the current MASH occupant compartment intrusion limits, and (3) estimation of the frequency of vehicle damage patterns in real-world roadside hardware crashes. Updating the previous intrusion study used the most recent years of NASS/CDS to determine correlations between occupant injuries and toe pan, floor pan, and side panel intrusions. The evaluation of the current MASH intrusion thresholds involved classifying real-world NASS/CDS crashes using the MASH occupant compartment intrusion criteria and then examining corresponding maximum occupant injury. Statistical models were developed to relate binary intrusion presence indicators (i.e., above/below the associated intrusion threshold) and other potentially confounding factors to observed occupant injury. To estimate the frequency of damage patterns in crashes with roadside safety hardware, vehicle damage photographs available in NASS/CDS were reviewed for cases involving an impact with specific roadside hardware devices to provide an estimate of the proportion of crashes where each damage mode is present. Research Findings 15.3.1 Analysis of the Candidate Injury Metrics in Frontal, Side, and Oblique Crashes Although there were some differences noted between the candidate metrics as described in more detail in the paragraphs below, no one metric was found to consistently outperform the OIV metric in terms of predicting real-world crash occupant injury at the MAIS2+F level in the investigated crash modes (i.e., frontal, side, or oblique). Age was a significant predictor in each of the final whole-body frontal crash models. Additionally, belt use was significant in the MDV, OIV, OIV+RA, and OLC models. PDOF was

224 significant in the OLC, ASI, and VPI models. Within the final frontal HF body region models, belt status and vehicle type were significant. In the N and TALT models, age was the only significant covariate. Using the F2 score as a metric of comparison, the OIV+RA was the best performing whole body frontal model with an F2 score of 0.60. MDV was the best performing model for the N and TALT body regions. The F2 scores for the HF region were all either 0.02 or 0.01. Belt status and impact type were the significant predictors in each of the final whole-body side crash models. No additional covariates were included in the final HF body region models. VPI was the best performing side crash whole body model with an F2 score of 0.42; however, ASI was close behind with an F2 score of 0.41. F2 scores could not be computed for the side crash body region models because there were too few injured occupants in the test dataset. Belt status and GAD were significant predictors in the final MDV, OIV, OLC, and VPI whole- body oblique crash models. No additional covariates were included in the final ASI whole-body model. The OLC model performed the best on the test dataset, with an F2 score of 0.61. For each model, the predicted MAIS2+F injury risks associated with current OIV and ASI thresholds were computed for a best-case and worst-case scenario occupant. This allowed comparison of acceptable risk values across multiple metrics and multiple crash types. In frontal crashes, the risks associated with the minimum thresholds for the best-case scenario occupant are close in range. Similarly, the risks associated with the maximum thresholds for the worst-case scenario occupant are close in range. This pattern does not hold true for side crashes. The risks associated with the minimum OIV threshold for a best- and worst-case scenario occupant are similar to those associated with the maximum ASI threshold for a best- and worst-case scenario occupant. For the best-case scenario occupant in an oblique crash, the risk values associated with the OIV thresholds are similar to the risks associated with the minimum and middle ASI thresholds. The worst-case scenario occupant risks associated with OIV thresholds differ vastly from the ASI model predictions. In every final frontal crash Harm model, age, BMI, and seating location were significant predictors. Sex was significant in every model besides OLC. Object contacted was significant in the MDV, OIV, and OLC models. The model with the lowest RMSE value was the OLC model, with an RMSE of 64.40. There was very little variation in the R2 values of these models, as they ranged from 46% to 52%. Belt status, age, and side impact type were significant predictors in every final side crash Harm model. The model with the lowest RMSE value was the OLC model, with an RMSE of 64.40. The model with the lowest RMSE value was the OIV model, with an RMSE of 62.40. However, there was very little variation among the RMSE values for the final side crash models. Additionally, there was very little variation in the R2 values of these models, as they ranged from 31% to 34%. Belt status and object contacted were significant predicters in every final oblique crash Harm model. The GAD was an additional significant predictor in the final MDV model. The model with the lowest RMSE value was the ASI model, with an RMSE of 69.30. There was very little variation in the R2 values of these models, as they ranged from 36% to 42%.

225 15.3.2 Investigation of MASH Occupant Compartment Intrusion Limits Based on the developed statistical models, MASH intrusion limits were found to be strong predictors of occupant injury at the MAIS1+, MAIS2+, and MAIS3+ levels. The odds of occupant injury were found to range between 10 and 30 times higher for nearside occupants where one or more of the MASH intrusion thresholds were exceeded compared to nearside occupants where none of the MASH intrusion thresholds were exceeded. With respect to the confounding factors, older, obese, and unbelted occupants were found to have a statistically significant increased risk of injury, regardless of injury level threshold. Statistical models developed using the individual vehicle region intrusion indicators suggest similar results with regard to the specific MASH intrusion limits. Nearly all the odds ratios exceeded 1.0 and were statistically significant, suggesting an increased occupant injury risk if the corresponding threshold is exceeded. The available data suggest that exceeding the MASH toe pan intrusion limit has the largest influence on occupant injury risk. The odds ratio estimates from the developed models also suggested that the windshield, A/B-pillar, and floor pan areas are influential to injury risk prediction and that the side door area intrusion becomes more influential as the injury threshold level is increased. With the exception of breakaway poles, vehicle occupant compartment intrusion for single vehicle real-world impacts with roadside hardware devices was estimated to occur in less than 10% of crashes. A manual review of vehicles damaged in single vehicle real-world impacts with cable barriers suggested that cables interact with the vehicle A-pillar in approximately half of the crashes but that the interaction generally results in minor damage and no severing of the A-pillar. No evidence of glued seam separation of unibody type vehicles was observed. Recommendations This study investigated the ability of existing MASH occupant risk metrics, as well as several vehicle-based alternative methods, to predict real-world occupant crash injury in the current vehicle fleet. The analyses conducted build on previous studies using real-world frontal crashes and expand knowledge in other less explored areas (i.e., side and oblique crashes). The developed statistical models include the influence of other relevant crash and occupant factors that influence injury risk and can be employed by MASH users to compare devices more comprehensively with differing occupant risk values based on full-scale crash testing results. The research effort also provided a first evaluation of the MASH area-specific intrusion thresholds using real-world crash data. Based on these findings, following are the recommendations of this study: • Consider retaining FSM for an updated MASH. Although the FSM assumptions do not necessarily align with the occupants of the current vehicle fleet, the model, especially the OIV, was not consistently outperformed by any other alternative metric. As a result, the FSM is recommended for inclusion for an update to MASH. To better align with the vehicle safety community and aid with end users’ interpretation of computed occupant risk values from crash tests, inclusion of the developed injury risk curves is also

226 recommended in an update to MASH. These injury risk curves also allow users to incorporate other crash and occupant factors that influence injury risk potential. • Consider retaining occupant compartment intrusion limits for an updated MASH. In general, the current region-specific MASH intrusion thresholds were found to have a strong correlation to occupant injury risk both in aggregate and individually. These findings serve as support for the current MASH intrusion threshold values, and additional MASH text has been proposed. • Revisit vehicle-based metrics periodically using more recent real-world crash data. The vehicle fleet and associated technologies are continually evolving. Similar to the ongoing efforts to update the MASH test vehicles, real-world crash data should be examined periodically to (1) update the injury risk curves developed as part of this research effort, and (2) evaluate the need to update the vehicle-based injury metrics and/or associated threshold values used to evaluate occupant injury risk. • Considerations for future studies examining vehicle-based occupant risk metrics. Findings of interest from this study include that the delta-v and VPI metrics would likely introduce additional subjectivity into the crash test procedures, as these metrics are sensitive to the selected analysis window and thus are not likely good candidates for future occupant risk procedures. Although not found to consistently outperform OIV in predicting real-world crash injury, OLC continues to be a metric of interest, as it specifically models a belted occupant and was found to generally outperform the other candidate metrics in predicting ATD-based injury metrics. Additional study of RA is needed given the current EDR data limitations. While the RA has been omitted in analogous international crash test procedures in favor of the ASI metric, making this switch in MASH would result in relatively large changes to the currently accepted MASH hardware.