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.

185 13 Proposed Implementation of Results in MASH Introduction The purpose of this chapter is to synthesize the project results for the purpose of proposing potential modifications to existing MASH injury criteria and/or associated threshold values. Approach to Developing Proposed MASH Modification Options The sections below provide details on how the analyses conducted as part of the research effort were used to determine appropriate MASH modifications and/or MASH modification options. In general, the following five aspects were considered: 1. The statistically significant predictors of real-world occupant injury as determined through the model development process using the IAD training dataset. 2. The real-world occupant injury risk associated with the current MASH vehicle-based occupant injury risk thresholds, including risk ranges based on confounding factors. 3. Comparison of the ability of current MASH vehicle-based metrics and other alternative vehicle-based metrics to predict real-world crash injury as well as crash test dummy-based injury risk. 4. How well the current region-specific MASH vehicle occupant compartment intrusion limits correlate with real-world occupant injury. 5. Incorporation of associated injury risk curves into future MASH updates to provide user- agencies with additional means of interpreting numerical occupant risk values. As necessary, the implications of any developed modification options were assessed using a sample of previously conducted MASH tests with various roadside hardware devices. For all modifications, suggested language for inclusion in a future update to MASH was developed and included. 13.2.1 Developed Models and Statistically Significant Predictors The statistical model development process was used to identify the candidate injury metrics and other confounding crash and occupant factors that were statistically significant predictors of real- world occupant injury. Initial models were developed with all available covariates prior to the development of final models with only the statistically significant predictors. An examination of these statistically significant covariates provides insight into which metrics and associated factors are important predictors of real-world occupant injury risk. This was done for each crash mode investigated (i.e., frontal, side, and oblique crashes) and for overall injury as well as body region injury (if sufficient body region injury cases were available to permit model development). 13.2.2 Determine Occupant Injury Risk Bounds for Current MASH Thresholds The statistical models developed with the training dataset were used to compute the upper and lower injury risk bounds for the current MASH FSM preferred and maximum thresholds. Note that these injury risk bounds were only computed if the metric was found to have a statistically

186 significant correlation to real-world occupant injury. Similarly, the upper and lower injury risk bounds were computed for the ASI thresholds present in the analogous international roadside hardware crash test procedures. As the developed injury risk models include statistically significant factors other than the candidate metrics, these other factors were varied to generate the minimum and maximum injury risk estimates for each crash type modeled (i.e., frontal, side, and oblique). The associated injury risk bounds were examined to identify any large injury risk level disparities in the current thresholds relative to crash type and to provide context for the importance of accounting for factors other than the kinematics-based injury metric value. 13.2.3 Evaluate Candidate Injury Metric Performance The project compared the candidate injury metrics in several ways, as summarized in Table 13-1. Table 13-1. High-level summary of comparison of candidate roadside crash injury metrics. Data Type Analysis/Comparison Description Dataset Crash Type Injury Type(s) and Metric Real- World Crashes Compare statistical metrics, e.g., F2 score, accuracy, ROC AUC, across candidate injury metrics for predicting MAIS Training (NASS/CDS) Frontal, Side, and Oblique Overall MAIS2+F injury, body regions as available Compare statistical metrics, e.g., F2 score, accuracy, ROC AUC, across candidate injury metrics for predicting MAIS Test (CISS) Frontal, Side, and Oblique Overall MAIS2+F injury, body regions as available Compare statistical metrics, e.g., RMSE, across candidate injury metrics for predicting Harm Training (NASS/CDS) and Test (CISS) Frontal, Side, and Oblique Harm Compare performance of augmented metrics to candidate metrics Subset of Training Frontal only Overall MAIS2+F injury Crash Tests Compare statistical metrics, e.g., R2, across candidate injury metrics NHTSA vehicle crash database subset Frontal, Side, and Side Pole Body regions: head, chest, neck, lower extremity; Various ATD metrics The available real-world crash data were used to rank order the ability of the candidate metrics to predict overall and body region-specific injury for frontal, side, and oblique crashes. These rankings were completed for both the training and, more importantly, the test datasets. This included the ability of the models to predict injury as measured by the AIS as well as an alternative metric (i.e., the Harm metric). Two methods for including vehicle-specific restraint performance were examined on a subset of crashes and compared to the performance of the primary candidate metrics to determine if there was any injury prediction advantage to quantifying vehicle-specific restraint performance. Finally, late model full-scale vehicle crash test data were used to compare the ability of the candidate metrics to predict occupant injury risk as measured by an ATD (i.e., crash test dummy).

187 13.2.4 Review Correlation of MASH Intrusion Criteria with Real-World Crash Injury The findings from the investigation of the current vehicle region-specific MASH intrusion thresholds were reviewed in light of the current intrusion-related text in MASH. The primary intent of the intrusion study portion of the current project was to provide additional data on how current MASH intrusion limits relate to real-world occupant injury. At present, the only real-world study used as a basis for the current MASH intrusion thresholds did not examine intrusion for all the vehicle regions specified in the current criteria. At a minimum, the MASH language should be updated to mention the more recent study conducted as part of this project. 13.2.5 Injury Risk Curves for MASH Metric(s) For any candidate injury metric(s) recommended for inclusion in an updated version of MASH, the corresponding injury risk models by crash type will be suggested for inclusion in MASH. Provision of the analogous injury risk models is consistent with full-scale vehicle crash testing procedures where the ATD-based injury metrics have corresponding injury risk curves available that can be used to convert an injury metric value to a corresponding probability of injury. A specific benefit of including these injury risk curves is that they would allow MASH users to better quantify differences in occupant risk values for different roadside hardware devices, and thus allow agencies to make better informed decisions about specific hardware to install. 13.2.6 Develop Proposed Options for Updating MASH and Evaluate Potential Implications The results of the five previously described summary analyses/factors were used to develop potential options for updating the MASH occupant risk criteria. A sample of already-conducted MASH crash tests selected from the three primary U.S. crash test facilities (i.e., MwRSF, TTI, and FHWA FOIL) was used to assess the implications of any potential changes to the existing MASH criteria. The following aspects were identified with respect to any proposed changes to the existing MASH occupant injury risk procedures: • Change in the number of passing/failing tests based on the occupant risk criteria • Any difficulties or ambiguities encountered when computing any alternative metrics Both of these aspects were considered when developing and evaluating potential modifications to the existing MASH occupant risk procedures. Based on all the available information from the analyses conducted for the present study, the research team recommended one of the developed options for updating the MASH occupant risk criteria. 13.2.7 Develop Suggested Modifications to Existing MASH Text A copy of the existing MASH (AASHTO 2016) text related to occupant risk has been provided to AASHTO. This includes a relatively large section of MASH Chapter 5 – Evaluation Criteria. Specifically, the excerpt includes Table 5-1B, section 5.2.2 Occupant Risk, as well as the associated commentary (section A5.2.2). For the recommended MASH occupant risk modification scenario, suggested modifications have been included directly to the MASH text with proposed new text shown in red.

188 Results and Discussion The following sections detail the results and associated discussion related to the development of the proposed modifications to the MASH occupant risk evaluation procedures. 13.3.1 Developed Models and Statistically Significant Predictors A summary of developed frontal, side, and oblique MAIS2+F models is provided in Table 13-2, Table 13-3, and Table 13-4, respectively. This includes the total number of unweighted and weighted training cases used as well as an indication of the statistically significant predictor variables. Table 13-2. Case counts and statistically significant variables for the final frontal impact MAIS2+F models. Frontal Model Raw Cases (Weighted) Metric Statistically Significant Predictors for MAIS2+F Injury Metric Belt Status Age Vehicle Type PDOF Binary RA Full Body 494 (167,330) MDV OIV x x x OLC x x x x OIV + Binary RA x x x x ASI VPI x x x Head/Face 489 (166,777) MDV, OIV, OLC, ASI, VPI x x x Neck C-Spine MDV, OIV, OLC, ASI, VPI x x Thorax Abdomen L- & T-Spine MDV, OIV, OLC, ASI, VPI x x Table 13-3. Case counts and statistically significant variables for the final oblique impact MAIS2+F models. Oblique Model Raw Cases (Weighted) Metric Statistically Significant Predictors for MAIS2+F Injury Metric Belt Status GAD Full Body 176 (67,107) MDV OIV OLC VPI x x x ASI x *Note: Not sufficient cases to develop body region models for the oblique crash mode.

189 Table 13-4. Case counts and statistically significant variables for the final side impact MAIS2+F models. Side Model Raw Cases (Weighted) Metric Statistically Significant Predictors for MAIS2+F Injury Metric Belt Status Impact Type Full Body 183 (66,124) MDV OIV OLC ASI VPI x x x Head/Face 183 (66,124) MDV OIV OLC ASI VPI x *Note: Not sufficient cases to develop N and TALT body region models for the side crash mode. The vast majority of candidate injury metrics were found to be statistically significant predictors of real-world occupant overall and body region-specific injury, independent of crash modality. A notable exception was the RA metric; the RA metric alone was not found to be a statistically significant predictor of overall real-world MAIS2+F occupant injury for frontal crashes (and hence not included in Table 13-2). Given this finding and the smaller number of cases available for the other two crash modes, corresponding RA metric only models were not developed for the frontal body region, side impacts, or oblique impacts. As RA is used in tandem with OIV to assess occupant injury risk in MASH, frontal impact statistical models were also developed for two additional scenarios: • Both the OIV and RA value present in a model, and • Both OIV and a binary RA variable (i.e., above or below a specific threshold RA). Similar to the RA metric only model, the first scenario did not produce any statistical significance for the RA metric. Inclusion of both the RA and OIV vehicle-based models in a single model would not be expected to create any multicollinearity issues, as these metrics are based on different portions of the crash pulse (i.e., OIV prior to occupant impact with the idealized vehicle interior and RA after occupant impact with the interior). For the second scenario, all the RA values for the training data were below the current MASH maximum threshold value of 20.49 G, so this RA threshold could not be used to meaningfully populate a binary RA variable. There were two unweighted (446 weighted; 0.27% of the entire frontal dataset) occupants with RA values above the MASH preferred threshold of 15 G, so this threshold was used to generate the binary RA indicator. Using this binary RA variable with 15 G threshold did result in binary RA being a statistically significant predictor of MAIS2+F injury in frontal crashes. 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: • 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

190 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%. • 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. Crashes with multiple events relatively close together in time (within a few seconds) would be more likely to have a high RA value. Other than the vehicle-based injury metrics, belt use (three-point belt or no belt) was found to be the largest contributing factor to overall occupant injury. Belt use was statistically significant in four of the six developed full body frontal models, four of the five developed full body oblique models, and all five of the developed full body side models. For specific body region injury, however, this effect was much less pronounced with belt status only found to be statistically significant in predicting HF injury in frontal impacts. Occupant age (age ≥ 65 years old or 13 ≤ age < 65) was also found to be another substantial contributing factor for occupant injury in the frontal crash mode. Age was statistically significant in all of the frontal full-body models developed and 10 of the 15 frontal body region models developed. To a lesser extent, the PDOF for the impact was also important for predicting overall injury in frontal impacts; PDOF was found to be statistically significant in half of the developed frontal overall injury models. For overall injury in side impacts, impact type (nearside or far-side) was found to be statistically significant in all of the developed models. For overall injury in oblique impacts, the general area of vehicle damage (front or side) was found to be statistically significant in four of the five developed models. Based on the available data, other potentially contributing factors found to be less important for predicting occupant injury (as measured by AIS) included: occupant sex (male or female), occupant BMI (BMI ≥ 30 kg/m2 or BMI < 30 kg/m2), occupant seating location (driver or right front passenger), and vehicle type (passenger car or light truck/vans). Occupant sex, BMI, and seating location were not found to be statistically significant predictors of overall or body region- specific injury in any of the developed MAIS2+F models for any of the three crash modes. Vehicle type was only found to be statistically significant when predicting HF injury in frontal impacts. Given the larger data sample available for frontal impacts, these factors are less likely important factors for predicting injury in the frontal crash mode. The results of the injury risk models developed using the alternate method to quantify injury (i.e., Harm) are summarized in Table 13-5. Similar to the MAIS2+F models developed, the vehicle-based injury metrics were found to be statistically significant in all cases (corresponding RA only models were not developed). In general, more crash and occupant factors were found to be statistically significant predictors of Harm compared to AIS-based injury. Similar to the AIS- based models, occupant belt use is found to be the primary crash factor other than the value of the metric for predicting Harm. Belt use was found to be statistically significant in all but one of the

191 developed Harm models. Age was also found to be an important factor in predicting Harm as it was statistically significant in all of the frontal and side impact Harm models. For frontal impacts, occupant BMI, sex, and seat location were also found to be important factors; BMI and seat location were statistically significant in all frontal Harm models while occupant sex was statistically significant in four of the five models. Type of object struck (tree/pole or other) was also found to be an important factor in frontal and oblique impacts, as it was found to be statistically significant in three of the five developed frontal Harm models and all of the oblique Harm models. For oblique crashes, the GAD appeared less important to predicting Harm compared to predicting MAIS2+F injury, as this predictor was only statistically significant in one of the five oblique Harm models developed. Similar to the AIS-based injury models, impact type (nearside or far-side) was found to be an important predictor of Harm in side impacts, as it was statistically significant in all of the developed side impact Harm models. Table 13-5. Case counts and statistically significant variables for the final Harm models. Impact Type Raw Cases (Weighted) Metric Statistically Significant Predictors for Harm Metric Belt Use Sex Age BMI Seat Location Object Struck GAD Impact Type Frontal 490 (166,860) MDV OIV x x x x x x x OLC x x x x x ASI VPI x x x x x x Oblique 175 (66,990) MDV x x x x OIV OLC ASI VPI x x x Side 183 (66,124) MDV OIV OLC ASI VPI x x x x For the investigation of the two methods to incorporate vehicle-specific restraint performance into the injury prediction process (i.e., Pjoint and vehicle-specific VPI), neither metric was found to be a statistically significant predictor of overall occupant injury in the frontal crash mode. As a result, final models were not constructed, and these metrics were not considered further at this time. Note that the developed models were on a subset of the available data due to availability of corresponding full-scale vehicle crash test data; not all the vehicle makes and models present in the IAD have full-scale crash test data present in the NHTSA vehicle crash test database. 13.3.2 Determine Occupant Injury Risk Bounds for Current MASH Thresholds The statistical models developed with the training dataset were used to compute the upper and lower bounds of real-world occupant injury risk for the OIV and ASI roadside hardware injury risk metrics for each crash mode (i.e., frontal, side, and oblique). Note that the OIV limits, 9.1 m/s preferred and 12.2 m/s maximum, are prescribed in MASH while the ASI thresholds are only used in international roadside hardware crash test procedures. RA was not included as the RA metric alone was not found to be a statistically significant predictor of real-world occupant injury risk, as previously discussed.

192 The overall occupant injury risk values for frontal, side, and oblique impacts are summarized in Table 13-6, Table 13-8, and Table 13-10, respectively. Occupant injury risk by body region for frontal and side impacts are summarized in Table 13-7 and Table 13-9, respectively. There were not sufficient oblique body region injuries present in the training dataset to develop the analogous body region models for the oblique impacts. Note that all the developed injury risk models predicted MAIS2+F occupant injury, but the original intent of the FSM thresholds was to distinguish serious from severe injury (e.g., MAIS3+F). Sufficient severe injury cases were not available in the training dataset to develop the analogous MAIS3+F statistical models. Nevertheless, the MAIS2+F can be used to identify any relative differences between occupant injury risk for different crash modes. The injury costs from the models developed to predict Harm in frontal, side, and oblique impacts are summarized in Table 13-11, Table 13-12, and Table 13-13, respectively. Table 13-6. Summary of real-world occupant injury risk associated with current roadside hardware injury metric thresholds: Overall MAIS2+F injury in frontal crashes. 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 13-7. Summary of real-world occupant injury risk associated with current roadside hardware injury metric thresholds: Body region MAIS2+F injury in frontal crashes. Body Region Crash and Occupant Conditions Injury Risk Scenario Belt Status Age (years) Vehicle Type OIV Threshold (m/s) ASI Threshold (--) 9.1 12.2 1.0 1.4 1.9 TALT 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% HF 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% N 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% Table 13-8. Summary of real-world occupant injury risk associated with current roadside hardware injury metric thresholds: Overall MAIS2+F injury in side crashes. 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%

193 Table 13-9. Summary of real-world occupant injury risk associated with current roadside hardware injury metric thresholds: Body region MAIS2+F injury in side crashes. Body Region Occupant Scenario Injury Risk OIV Threshold (m/s) ASI Threshold (--) 9.1 12.2 1.0 1.4 1.9 TALT Insufficient Injury Cases Available to Build Injury Models HF Best and Worst Case 41.6% 78.3% 8.5% 19.3% 43.8% N Insufficient Injury Cases Available to Build Injury Models Table 13-10. Summary of real-world occupant injury risk associated with current roadside hardware injury metric thresholds: Overall MAIS2+F injury in oblique crashes. Crash and Occupant Conditions Injury Risk Scenario Belt Status GAD OIV Threshold (m/s) ASI Threshold (--) 9.1 12.2 1.0 1.4 1.9 Best Case Belted Frontal damage 5.0% 14.3% 8.7% 19.9% 44.8% Worst Case Unbelted Side damage 69.0% 87.5% 8.7% 19.9% 44.8% Table 13-11. Summary of real-world occupant Harm cost associated with current roadside hardware injury metric thresholds: Frontal crashes. Crash and Occupant Conditions Harm (US$100) Scenario Belt Status Age (years) Sex BMI (kg/m2) Seating Location Object Struck OIV Threshold (m/s) ASI Threshold (--) 9.1 12.2 1.0 1.4 1.9 Best Case Belted < 65 Male < 30 RF Passenger Not narrow 17.3 107.8 13.0 92.5 293.1 Worst Case Unbelted ≥ 65 Female ≥ 30 Driver Narrow object 685.0 1049.3 419.9 702.1 1156.1 Table 13-12. Summary of real-world occupant Harm cost associated with current roadside hardware injury metric thresholds: Side crashes. Crash and Occupant Conditions Harm (US$100) Scenario Belt Status Age (years) Impact Type OIV Threshold (m/s) ASI Threshold (--) 9.1 12.2 1.0 1.4 1.9 Best Case Belted < 65 years Far-Side 30.4 66.9 3.8 13.0 32.5 Worst Case Unbelted ≥ 65 Nearside 587.6 723.9 435.3 507.9 606.5 Table 13-13. Summary of real-world occupant Harm cost associated with current roadside hardware injury metric thresholds: Oblique crashes. Crash and Occupant Conditions Harm (US$100) Scenario Belt Status Object Struck OIV Threshold (m/s) ASI Threshold (--) 9.1 12.2 1.0 1.4 1.9 Best Case Belted Not narrow 43.9 86.0 23.2 39.7 66.7 Worst Case Unbelted Narrow object 342.0 446.9 264.9 315.6 385.2 Based on the occupant injury risk values present in Table 13-6 through Table 13-13, the following observations were made: • The current MASH lateral OIV thresholds appear to be less conservative than the current longitudinal thresholds. For the best-case occupant scenario, MAIS2+F injury risk is approximately twice as high for occupants exposed to a preferred or maximum lateral OIV value compared to the analogous longitudinal OIV in a frontal crash or an analogous resultant OIV value in an oblique crash. The worst-case occupant differences, however, are less pronounced, especially for the maximum threshold where the difference is 10% or less.

194 For the preferred threshold, a worst-case side impact occupant has a 10% to 30% higher risk of MAIS2+F injury, compared to an analogous longitudinal OIV in a frontal crash or resultant OIV in an oblique crash, respectively. • For a given crash mode, the difference between the best and worst case MAIS2+F injury risk is large, suggesting that factors beyond the injury metric value do have a significant effect on occupant injury risk. In terms of overall injury at the current maximum OIV threshold, the worst-case occupants for frontal and oblique have roughly 6 times the injury risk of the best-case occupants in those crash modes. For the side impacts at the current maximum OIV threshold, the difference between the best- and worst-case occupant injury risk is roughly a factor of 3. • For overall occupant injury, irrespective of crash mode, increasing OIV from the preferred to maximum threshold value generally increases MAIS2+F injury risk by approximately 10% to 20%. • When comparing the current OIV maximum threshold to the ASI thresholds for frontal crashes, the OIV maximum threshold generally has an occupant injury risk between the ASI thresholds of 1.4 and 1.9. For side crashes, the maximum OIV threshold has an occupant injury risk in excess of the ASI 1.9 threshold. • For the frontal crash mode, injury risk was found to vary by body region. The TALT region had the least variation in injury risk, and all probability of injury values for this region were much less than those for the corresponding overall occupant injury. For the HF and N regions, however, there were very large injury risk variations between the best- and worst- case occupants. • With the exception of the best-case occupant at the preferred OIV threshold, frontal crashes with OIV values at the MASH thresholds were observed to have higher Harm costs than side and oblique impacts with OIV values at the MASH thresholds. • The difference in Harm cost between the preferred and maximum OIV thresholds was also highest for the frontal impacts, between $10,000 and $37,000 for the best-case and worst- case occupant, respectively. This difference for side and oblique impacts was comparable at approximately $4,000 for best-case occupants and $10,000 to $14,000 for worst-case occupants. • Similar to the AIS-based models, the current OIV maximum threshold Harm costs for frontal impact were between the costs for the ASI thresholds of 1.4 and 1.9. For both side and oblique impacts, the OIV maximum threshold Harm predicted cost exceeds the cost predicted for the ASI maximum threshold of 1.9. 13.3.3 Evaluate Candidate Injury Metric Performance To compare the ability of the candidate injury metrics to predict real-world crash occupant injury, several statistical metrics were tabulated for the models developed with the training (NASS/CDS) dataset and the same models applied to the test (CISS) dataset. The results of these computations for the overall MAIS2+F injury in frontal, side, and oblique impacts are summarized in Table 13-14, Table 13-16, and Table 13-18, respectively. The body region results for the frontal and side impacts are summarized in Table 13-15 and Table 13-17, respectively. Note that there

195 were not sufficient body region-specific injury cases in the oblique crash mode to develop any of the oblique crash body region models. Table 13-14. Summary of final candidate metric model performance for overall MAIS2+F injury: Frontal crash mode. 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 32% 0.82 0.78 0.94 0.54 0.45 0.47 ASI 24% 0.81 0.80 0.94 0.58 0.57 0.57 VPI 24% 0.81 0.79 0.95 0.61 0.58 0.58 Table 13-15. Summary of final candidate metric model performance for body region MAIS2+F injury: Frontal crash mode. Body Region Candidate Metric Decision Threshold Training Test F2 Scores (HF) ROC AUC Accuracy Precision Recall F2 Scores (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 HF 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 Table 13-16. Summary of final candidate metric model performance for overall MAIS2+F injury: Side crash mode. Candidate Metric 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

196 Table 13-17. Summary of final candidate metric model performance for body region MAIS2+F injury: Side crash mode. Body Region Candidate Metric Decision Threshold Training Test F2 Scores (HF) ROC AUC Accuracy Precision Recall F2 Scores (HF) 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 HF 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 N Insufficient Injury Cases Available to Build Injury Models TALT Table 13-18. Summary of final candidate metric model performance for overall MAIS2+F injury: Oblique crash mode. Candidate Metric Decision Threshold Training Test F2 Scores ROC AUC Accuracy Precision Recall F2 Score MDV 14% 0.69 0.90 84% 0.23 0.87 0.56 OIV 14% 0.69 0.86 85% 0.22 0.73 0.51 OLC 20% 0.64 0.88 92% 0.37 0.73 0.61 ASI 6% 0.60 0.88 77% 0.17 0.84 0.47 VPI 7% 0.66 0.88 75% 0.16 0.88 0.47 Based on the values present in Table 13-14 through Table 13-18, the following observations were made: • Based on the F2 scores for the overall injury models developed with the training IAD data, there is very little difference between the candidate injury metrics. In the frontal and side impact modes, the difference between the best and worst performing metric F2 score was no more than 3%. For the oblique impact mode, the observed differences were larger but still within 15% of one another. • A similar observation was made relative to the F2 scores for the training IAD body region models. For the side impact HF models and the frontal N and TALT models, the F2 scores were within 5% of one another for all the investigated vehicle-based metrics. The most variation between the candidate injury metrics was observed in the frontal HF body region models with approximately a 15% difference between the F2 score of the best and worst performing metric. • Based on the F2 scores from the frontal impact test dataset, the OIV + Binary RA metric performed the best but was followed closely by the VPI, OIV, ASI, and MDV metrics. The performances of these five models were within approximately 7% of one another. Note that the binary RA metric had only two unweighted cases in the training dataset above the 15 G RA cutoff threshold, and there were no cases with RA > 15 G in the test dataset. The OLC metric was found to be the worst performer. • Based on the F2 scores from the side impact test dataset, the VPI and ASI were the best performing metrics. The other three metrics had very low or zero F2 scores. For the zero F2 scores, all MAIS2+F injury occupants in the test data were below the corresponding decision threshold for these metrics. Precision and recall both rely on the presence of true positives and false negatives. Since there were none of these, it was not possible to calculate

197 an F2 score for MDV and OIV. Additional data would be required to evaluate these metrics more fully in this crash mode. • Based on the F2 scores from the oblique impact test dataset, the OLC metric performed the best but was followed closely by the MDV and OIV metrics. The performances of these three models were within approximately 10% of one another. The ASI and VPI metrics were the worst performing metrics. • The MDV metric was the best performer for predicting injury to the N region and the TALT body region in frontal impacts. The ASI and VPI metrics also performed well in these scenarios. None of the investigated metrics were able to predict injury to the HF region in frontal or side crashes in the test dataset. • While there were some differences observed between the candidate injury metrics, most differences were relatively small and no one metric was found to consistently predict real- world occupant injury risk better than the other metrics in the three crash modes. To further compare the ability of the candidate injury metrics to predict real-world crash injury, linear regression models were developed with the training dataset to relate the candidate injury metrics to an alternate method of quantifying occupant injury (i.e., Harm). The results of this comparison for the frontal, side, and oblique impacts are summarized in Table 13-19. Note that the RMSE values in the table compare the predicted to observed Harm values from the test dataset. Table 13-19. Summary of final candidate metric model performance for Harm predictions: Frontal, oblique, and side crash modes. Crash Type Metric Model RMSE Frontal MDV 65.91 OIV 67.49 OLC 64.40 ASI 65.71 VPI 66.40 Oblique MDV 75.75 OIV 70.83 OLC 71.41 ASI 69.30 VPI 71.48 Side MDV 62.89 OIV 62.40 OLC 62.86 ASI 62.90 VPI 63.26 Similar to the MAIS2+F models developed, the Harm models suggest relatively little difference in the predictive capability of the investigated vehicle-based metrics. For frontal and side impacts, the largest difference in RMSE between the best and worst performing metrics was 5% or less. The differences in the oblique mode were more pronounced, but all the metrics were within 10% of one another. Even though the differences were relatively small, there was also no single metric found to consistently have the lowest RMSE value; OLC, ASI, and OIV had the lowest RMSE values for the frontal, oblique, and side impacts, respectively.

198 The original intent was to complete similar comparisons between the candidate injury metrics and two additional vehicle-specific restraint metrics investigated (i.e., Pjoint and vehicle-specific VPI). Since the models developed with these two additional metrics indicated that these metrics were not statistically significant predictors of overall frontal occupant injury, no further models were developed. Finally, late model full-scale vehicle crash test data were used to compare the ability of the candidate metrics to predict occupant injury risk as measured by an ATD. Linear regression models were developed to determine the relationship between the candidate injury metric values and corresponding ATD-based injury metrics. The results of the analysis are summarized in Table 13-20. Note that the numerical values represent the R2 values for the corresponding relationship (e.g., OIV predicted 3-ms clip for drivers in frontal crashes with an R2 value of 0.238) and that the highlighted cells indicate statistical significance. Table 13-20. Summary of final candidate metric model R2 values for ATD-based injury: Frontal, side, and side pole crash tests. Crash Mode Seat Location ATD Injury Metric OIV ASI MDV OLC VPI Frontal Driver HIC15 0.0011 0.0040 0.0010 0.0253 0.0077 HIC36 0.0320 0.0428 0.0051 0.0903 0.0457 3ms Clip 0.2380 0.2609 0.0351 0.2693 0.2445 Chest Compression 0.0260 0.1354 0.0108 0.0689 0.0907 Femur Load 0.0080 0.0258 0.0342 0.0441 0.0357 Tibia Load 0.0522 0.0979 0.0047 0.1021 0.0813 Nij 0.0013 0.1068 0.1474 0.1104 0.0961 Right Front Passenger HIC15 0.0026 0.0000 0.0201 0.0001 0.0006 HIC36 0.0188 0.0004 0.0500 0.0003 0.0004 3ms Clip 0.2589 0.3175 0.0908 0.3003 0.3233 Chest Compression 0.0324 0.0508 0.0002 0.0164 0.0109 Femur Load 0.0031 0.0398 0.0040 0.0459 0.0287 Tibia Load 0.0016 0.0005 0.0003 0.0008 0.0006 Nij 0.0030 0.0477 0.0048 0.0989 0.0548 Side Drivers HIC15 0.1328 0.1702 0.0729 0.1391 0.0051 HIC36 0.1121 0.1596 0.0497 0.1291 0.0275 Rib Deflection 0.0967 0.0884 0.0728 0.0982 0.0063 Lower Spine Resultant 0.0436 0.0615 0.0377 0.0608 0.0008 Rib Deflection Rate 0.0637 0.0802 0.0762 0.0854 0.0233 Left Rear Passengers HIC15 0.0461 0.0889 0.0199 0.0550 0.0244 HIC36 0.0486 0.0938 0.0186 0.0628 0.0373 Rib Deflection 0.1184 0.1851 0.0821 0.1415 0.0023 Lower Spine Resultant 0.1263 0.1459 0.0671 0.1716 0.0446 Rib Deflection Rate 0.1647 0.2047 0.1625 0.2178 0.0106 Side Pole Drivers HIC15 0.1588 0.1002 0.0847 0.1554 0.1247 HIC36 0.1104 0.0867 0.0328 0.1105 0.1081 Rib Deflection 0.1594 0.0508 0.0835 0.1866 0.1126 Lower Spine Resultant 0.0005 0.0049 0.0154 0.0005 0.0132 Rib Deflection Rate 0.0720 0.0155 0.0058 0.0690 0.0158 Statistically Significant Predictor Count 11 14 4 17 9 *Note: Highlighted cell indicates statistical significance at the 0.05 level Based on the data shown in Table 13-20, the following observations were made:

199 • The vehicle-based metrics were found to have the strongest relationship to the chest-based ATD metrics (i.e., maximum chest acceleration, rib deflection, and rib deflection rate), depending on the crash mode. • Approximately 40% (55 of 145) of the developed models found a statistically significant fit for the candidate injury metric compared to an intercept-only (i.e., single value) model. From a practical standpoint, however, none of the vehicle-based metrics accounted for more than one third of the variation observed in the corresponding ATD-based metrics. In the frontal crash mode, the metrics were found to account for 30% or less of the ATD-based metric variation. For the side impact and side pole crashes, the metrics accounted for 20% or less of the ATD-based metric variation. • The OLC was found to have the highest number of statistically significant correlations to the ATD-based injury metrics (17 of 29), followed by ASI (14 of 29), OIV (11 of 29), VPI (nine of 29), and MDV (four of 29). In terms of the different crash modes, OLC had the highest number of statistically significant correlations in the frontal (seven) and side pole (four) tests and trailed only ASI in the side impact tests (seven for OLC, eight for ASI). 13.3.4 Review Correlation of MASH Intrusion Criteria with Real-World Crash Injury Updating the previous Eigen and Glassbrenner (2003) study with the most recent NASS/CDS data (2000 through 2015) echoed the original study findings that a strong link exists between occupant compartment intrusion and linked occupant injury level. The numerical results are summarized in Table 13-21. For the four investigated categories (any relevant intrusion with vehicle contact, toe pan intrusion with vehicle contact, any relevant intrusion with non-vehicle contact, and toe pan intrusion with non-vehicle contact), most of the intrusion and injury level combinations were found to be statistically significant. Note that “relevant intrusion” included non-zero intrusion in one or more of the following areas: toe pan, floor pan, and/or forward of the A-pillar. Table 13-21. Summary of statistically significant chi-square test results comparing maximum relevant intrusion level to nearside occupant maximum linked injury level. Contact with another vehicle and toe pan, floor pan, and/or forward of A-pillar intrusion present Contact with another vehicle and toe pan intrusion present Contact with non-vehicle and toe pan, floor pan, and/or forward of A-pillar intrusion present Contact with non- vehicle and toe pan intrusion present Intrusion Injury p Intrusion Injury p Intrusion Injury p Intrusion Injury p ≥ 8 cm AIS ≥ 1 <0.001 ≥ 8 cm AIS ≥ 1 <0.001 ≥ 8 cm AIS ≥ 1 <0.001 ≥ 8 cm AIS ≥ 1 <0.001 ≥ 8 cm AIS ≥ 2 <0.001 ≥ 8 cm AIS ≥ 2 <0.001 ≥ 8 cm AIS ≥ 2 <0.001 ≥ 8 cm AIS ≥ 2 <0.001 ≥ 8 cm AIS ≥ 3 <0.001 ≥ 8 cm AIS ≥ 3 <0.001 ≥ 8 cm AIS ≥ 3 <0.001 ≥ 8 cm AIS ≥ 3 <0.001 ≥ 15 cm AIS ≥ 1 <0.001 ≥ 15 cm AIS ≥ 1 <0.001 ≥ 15 cm AIS ≥ 2 <0.001 ≥ 15 cm AIS ≥ 3 0.0014 ≥ 15 cm AIS ≥ 2 <0.001 ≥ 15 cm AIS ≥ 2 <0.001 ≥ 15 cm AIS ≥ 3 0.0003 ≥ 46 cm AIS ≥ 1 <0.001 ≥ 30 cm AIS ≥ 1 0.0454 ≥ 15 cm AIS ≥ 3 <0.001 ≥ 30 cm AIS ≥ 2 <0.001 ≥ 46 cm AIS ≥ 2 0.0018 ≥ 30 cm AIS ≥ 2 0.0047 ≥ 30 cm AIS ≥ 1 0.0409 ≥ 46 cm AIS ≥ 1 <0.001 ≥ 46 cm AIS ≥ 3 0.0003 ≥ 46 cm AIS ≥ 2 <0.001 ≥ 30 cm AIS ≥ 2 0.0057 ≥ 46 cm AIS ≥ 2 0.0010 ≥ 61 cm AIS ≥ 1 <0.001 ≥ 46 cm AIS ≥ 3 0.0079 ≥ 30 cm AIS ≥ 3 0.0026 ≥ 46 cm AIS ≥ 3 0.0008 ≥ 61 cm AIS ≥ 2 0.0008 ≥ 61 cm AIS ≥ 2 0.0065 ≥ 46 cm AIS ≥ 2 <0.001 ≥ 61 cm AIS ≥ 1 <0.001 ≥ 46 cm AIS ≥ 3 <0.001 ≥ 61 cm AIS ≥ 2 0.0002 ≥ 61 cm AIS ≥ 2 0.0034

200 Since the previous study did not address all the vehicle regions specified in the current MASH intrusion limit guidance, the available NASS/CDS cases were used to investigate the relationship between the MASH intrusion guidelines and real-world occupant injury while accounting for confounding factors. Binary logistic regression models were developed to relate vehicle occupant compartment intrusion levels based on current MASH intrusion thresholds to three different occupant injury thresholds, MAIS1+F, MAIS2+F, and MAIS3+F, while accounting for the same confounding factors used in the models developed to compare the candidate injury metrics. Two different measures were used to indicate whether a particular occupant in a vehicle had intrusion above the MASH threshold: 1. A single “overall” binary variable indicating one or more areas are in excess of the corresponding MASH threshold. 2. Vehicle region-specific binary indicators (seven total) that indicate whether the corresponding MASH intrusion threshold was exceeded. The estimated odds ratio values and associated confidence limits for the binary MASH intrusion models are summarized in Table 13-22. The odds ratios and associated confidence limits for the models with vehicle region-specific MASH binary intrusion indicators are summarized in Table 13-23.

201 Table 13-22. Summary of odds ratio results for the MAIS1+F, MAIS2+F, and MAIS3+F intrusion-injury models. Model Predictor Variable Value Comparison Group Odds Ratio 95% CI MAIS1+F Exceed 1+ MASH Limit Yes No Intrusion > MASH Limits 9.69 5.6 – 16.8 Belt Use Belted Unbelted 0.49 0.41 – 0.59 Sex Male Female 0.54 0.46 – 0.62 Age ≥ 65 years < 65 years 1.09 0.92 – 1.29 BMI ≥ 30 kg/m2 < 30 kg/m2 1.66 1.32 – 2.09 Vehicle Type Passenger Car LTV 1.15 1.01 – 1.32 MAIS2+F Exceed 1+ MASH Limit Yes No Intrusion > MASH Limits 16.68 12.2 – 22.8 Belt Use Belted Unbelted 0.34 0.27 – 0.44 Sex Male Female 0.68 0.58 – 0.79 Age ≥ 65 years < 65 years 1.82 1.40 – 2.37 BMI ≥ 30 kg/m2 < 30 kg/m2 1.45 1.12 – 1.88 Vehicle Type Passenger Car LTV 0.98 0.85 – 1.13 MAIS3+F Exceed 1+ MASH Limit Yes No Intrusion > MASH Limits 27.9 20.6 – 37.7 Belt Use Belted Unbelted 0.22 0.18 – 0.26 Sex Male Female 0.95 0.81 – 1.11 Age ≥ 65 years < 65 years 3.02 2.34 – 3.90 BMI ≥ 30 kg/m2 < 30 kg/m2 1.31 1.07 – 1.61 Vehicle Type Passenger Car LTV 1.00 0.83 – 1.22 Table 13-23. Abbreviated summary of odds ratio results for the MAIS1+F, MAIS2+F, and MAIS3+F intrusion-injury models with area specific intrusion variables. Model Predictor Variable Value Comparison Group Odds Ratio 95% CI MAIS1+F Exceed Windshield Limit Yes No 6.23 3.25 - 12.0 Exceed Roof Limit Yes No 6.52 3.13 – 13.6 Exceed A/B Pillar Limit Yes No 6.60 3.90 – 11.2 Exceed Toe Pan Limit Yes No 74.4 19.6 – 282 Exceed Side Door Limit Yes No 0.98 0.20 – 4.82 Exceed Side Panel Limit Yes No 17.7 2.52 – 124.8 Exceed Floor Pan Limit Yes No 9.94 2.42 – 40.8 MAIS2+F Exceed Windshield Limit Yes No 6.31 3.84 – 10.4 Exceed Roof Limit Yes No 3.52 2.23 – 5.55 Exceed A/B Pillar Limit Yes No 6.10 4.46 – 8.34 Exceed Toe Pan Limit Yes No 61.9 22.8 – 168 Exceed Side Door Limit Yes No 3.80 1.95 – 7.40 Exceed Side Panel Limit Yes No 20.1 2.03 – 200 Exceed Floor Pan Limit Yes No 5.40 1.91 – 15.1 MAIS3+F Exceed Windshield Limit Yes No 7.99 4.38 – 14.6 Exceed Roof Limit Yes No 3.64 2.24 – 5.90 Exceed A/B Pillar Limit Yes No 8.86 6.46 – 12.2 Exceed Toe Pan Limit Yes No 22.7 10.0 – 51.2 Exceed Side Door Limit Yes No 4.36 2.26 – 8.41 Exceed Side Panel Limit Yes No 1.87 0.53 – 6.56 Exceed Floor Pan Limit Yes No 5.07 2.43 – 10.6

202 Based on the data shown in Table 13-22 and Table 13-23, the following observations were made: • MASH intrusion limits are 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. In each model, this variable had the largest magnitude coefficient compared to the other included predictors, suggesting it has the largest effect on occupant injury risk. • 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. In general, males were found to be less likely to be injured, but this was only statistically significant at the MAIS1+ and MAIS2+ levels. For the MAIS1+ level, passenger car occupants had a statistically significant increase in injury risk, but vehicle type was not a statistically significant effect for the higher injury threshold models. • The models developed using the individual vehicle region intrusion indicators suggest similar results regarding the specific MASH intrusion limits. Except for the side door limit at the MAIS1+ level (which was not statistically significant), all the odds ratios exceeded 1.0 and were statistically significant, suggesting an increased occupant injury risk if the corresponding threshold is exceeded. • Based on the odds ratio values and associated lower 95% confidence bounds, exceeding the MASH toe pan intrusion limit appears to have the largest influence on occupant injury risk. At the lower injury levels (MAIS1+ and MAIS2+), exceeding the MASH side panel intrusion limit appears to have a large influence on injury risk, but this effect was not found to be statistically significant at the MAIS3+ level. Also, the lower 95% confidence bound of the side panel indicator was roughly the same as many of the other vehicle region indicators; this coupled with the large range on the confidence bounds suggest more cases with side panel intrusion are needed to better understand this relationship. The odds ratio estimates also suggest 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. 13.3.5 Injury Risk Curves for MASH Metric(s) For any candidate injury metric(s) recommended for inclusion in an updated version of MASH, the corresponding injury risk curves are suggested for inclusion to better align with full-scale vehicle crash test procedures and provide users of MASH test data with additional information to inform roadside hardware related decisions. The final AIS-based injury models developed as part of this project are summarized below. Table 13-24 provides the base equation for each model along with the impact type specific logit equation. Note that the corresponding logit includes only the statistically significant predictor variables based on the models developed with the IAD training dataset.

203 Table 13-24. Summary of binary logistic regression MAIS2+F model equations based on candidate vehicle- based metrics. Overarching Base Equation 𝑃𝑃[𝑀𝑀𝐴𝐴𝑀𝑀𝑀𝑀2+ F] = 1 1 + 𝑃𝑃−𝑙𝑙𝑓𝑓𝑙𝑙𝑖𝑖𝑓𝑓 Frontal Impacts 𝑅𝑅𝑃𝑃𝑙𝑙𝑃𝑃𝑙𝑙 = 𝛽𝛽0 + 𝛽𝛽1 ⋅ (𝑃𝑃𝑃𝑃𝑖𝑖𝐴𝐴𝑃𝑃𝐴𝐴 𝑚𝑚𝑃𝑃𝑙𝑙𝑃𝑃𝑃𝑃𝑃𝑃) + 𝛽𝛽2 ⋅ 𝑏𝑏𝑃𝑃𝑅𝑅𝑙𝑙𝑁𝑁𝑓𝑓𝑙𝑙𝑓𝑓𝑁𝑁𝑁𝑁 + 𝛽𝛽3 ⋅ 𝑅𝑅𝑙𝑙𝑃𝑃 + 𝛽𝛽4 ⋅ 𝑃𝑃𝑃𝑃𝑃𝑃𝐹𝐹 + 𝛽𝛽5 ⋅ 𝑅𝑅𝐴𝐴𝑜𝑜𝑖𝑖𝑁𝑁𝑙𝑙𝑁𝑁𝑏𝑏 Side Impacts 𝑅𝑅𝑃𝑃𝑙𝑙𝑃𝑃𝑙𝑙 = 𝛽𝛽0 + 𝛽𝛽1 ⋅ (𝑃𝑃𝑃𝑃𝑖𝑖𝐴𝐴𝑃𝑃𝐴𝐴 𝑚𝑚𝑃𝑃𝑙𝑙𝑃𝑃𝑃𝑃𝑃𝑃) + 𝛽𝛽2 ⋅ 𝑏𝑏𝑃𝑃𝑅𝑅𝑙𝑙𝑁𝑁𝑓𝑓𝑙𝑙𝑓𝑓𝑁𝑁𝑁𝑁 + 𝛽𝛽3 ⋅ 𝑃𝑃𝑚𝑚𝑡𝑡𝑅𝑅𝑃𝑃𝑙𝑙_𝑙𝑙𝐴𝐴𝑡𝑡𝑃𝑃 Oblique Impacts 𝑅𝑅𝑃𝑃𝑙𝑙𝑃𝑃𝑙𝑙 = 𝛽𝛽0 + 𝛽𝛽1 ⋅ (𝑃𝑃𝑃𝑃𝑃𝑃𝐴𝐴𝑅𝑅𝑙𝑙𝑅𝑅𝑃𝑃𝑙𝑙 𝑃𝑃𝑃𝑃𝑖𝑖𝐴𝐴𝑃𝑃𝐴𝐴 𝑚𝑚𝑃𝑃𝑙𝑙𝑃𝑃𝑃𝑃𝑃𝑃) + 𝛽𝛽2 ⋅ 𝑏𝑏𝑃𝑃𝑅𝑅𝑙𝑙𝑁𝑁𝑓𝑓𝑙𝑙𝑓𝑓𝑁𝑁𝑁𝑁 + 𝛽𝛽3 ⋅ 𝐺𝐺𝐴𝐴𝑃𝑃 Table 13-25, Table 13-26, and Table 13-27 summarize the parameters to be used with the logit equations for each metric for frontal, side, and oblique impacts, respectively. Note that not all models use all the available predictor variables. All continuous variables should be entered in the noted units, and binary variables are either 1 or 0 with the value listed in the parameter column corresponding to a value of 1 (e.g., for all belted occupants, beltstatus = 1 and all unbelted occupants, beltstatus = 0). Note that the OIV + Binary RA frontal model was excluded from Table 13-25. Although the binary RA variable was found to be a statistically significant predictor of MAIS2+F injury, this was only based on two unweighted cases with RA > 15 G in the training dataset and no cases in the test dataset with RA > 15 G. Given the lack of high RA cases and the previously articulated issues regarding the EDR-based RA accuracy, the OIV + Binary RA model has been included in the final project report for documentation and future research purposes. At this time, however, the developed OIV + Binary RA model is not recommended for inclusion in any updated version of MASH.

204 Table 13-25. Summary of MAIS2+F frontal impact logistic regression model parameters by candidate injury metric. Metric Predictor Variable Parameter Coefficient OIV --- β0, Intercept -8.663 Longitudinal OIV β1, OIV (m/s) 0.726 Belt Use β2, Belted -1.864 Age β3, Age ≥ 65 3.400 ASI --- β0, Intercept -7.839 Longitudinal ASI β1, ASI (g) 4.529 Age β2, Age ≥ 65 3.247 PDOF β4, PDOF -0.041 OLC --- β0, Intercept -5.531 Longitudinal OLC β1, OLC (g) 0.326 Belt Use β2, Belted -1.622 Age β3, Age ≥ 65 3.252 PDOF β4, PDOF -0.041 MDV --- β0, Intercept -8.445 Longitudinal Delta-v β1, Delta-v (m/s) 0.674 Belt Use β2, Belted -1.812 Age β3, Age ≥ 65 3.469 VPI --- β0, Intercept -8.362 Longitudinal VPI β1, VPI (m/s2) 0.019 Age β2, Age ≥ 65 3.146 PDOF β4, PDOF -0.046 Table 13-26. Summary of MAIS2+F side impact logistic regression model parameters by candidate injury metric. Metric Predictor Variable Parameter Coefficient OIV --- β0, Intercept -2.858 Lateral OIV β1, OIV (m/s) 0.383 Belt Status β2, Belted -2.573 Impact Type β3, Nearside 1.566 ASI --- β0, Intercept -3.070 Lateral ASI β1, ASI 1.947 Belt Status β2, Belted -2.429 Impact Type β3, Nearside 1.831 OLC --- β0, Intercept -2.044 Lateral OLC β1, OLC (g) 0.158 Belt Status β2, Belted -2.401 Impact Type β3, Nearside 1.856 MDV --- β0, Intercept -2.946 Lateral Delta-v β1, Delta-v (m/s) 0.386 Belt Status β2, Belted -2.595 Impact Type β3, Nearside 1.632 VPI --- β0, Intercept -3.273 Lateral VPI β1, VPI (m/s2) 0.010 Belt Status β2, Belted -2.418 Impact Type β3, Nearside 1.848 Table 13-27. Summary of MAIS2+F oblique impact logistic regression model parameters by candidate injury metric.

205 Metric Predictor Variable Parameter Coefficient OIV --- β0, Intercept -2.588 Resultant OIV β1, OIV (m/s) 0.372 Belt Status β2, Belted -1.773 GAD β3, Frontal Damage -1.966 ASI --- β0, Intercept -4.715 ASI β1, ASI 2.371 OLC --- β0, Intercept -0.874 Resultant OLC β1, OLC (g) 0.127 Belt Status β2, Belted -1.761 GAD β3, Frontal Damage -1.641 MDV --- β0, Intercept -2.073 Resultant MDV β1, MDV (m/s) 0.299 Belt Status β2, Belted -1.941 GAD β3, Frontal Damage -1.921 VPI --- β0, Intercept -- Resultant VPI β1, VPI (m/s2) 0.009 Belt Status β2, Belted -1.648 GAD β3, Frontal Damage -1.322 For the current MASH vehicle region intrusion limits, the analogous MAIS2+F model developed is also summarized below. Table 13-28 provides the base equation and the associated logit equation containing the statistically significant predictor variables. Table 13-29 summarizes the parameters to be used with the logit equation. Note that all variables in the model are binary variables with the value listed in the parameter column corresponding to a value of 1 (e.g., for all belted occupants, beltstatus = 1 and all unbelted occupants, beltstatus = 0). Table 13-28. Summary of binary logistic regression model equations for MAIS2+F intrusion-based model. Overarching Base Equation 𝑃𝑃[𝑀𝑀𝐴𝐴𝑀𝑀𝑀𝑀2+ F] = 1 1 + 𝑃𝑃−𝑙𝑙𝑓𝑓𝑙𝑙𝑖𝑖𝑓𝑓 All Fixed Object Hits 𝑅𝑅𝑃𝑃𝑙𝑙𝑃𝑃𝑙𝑙 = 𝛽𝛽0 + 𝛽𝛽1 ⋅ 𝑏𝑏𝑃𝑃𝑅𝑅𝑙𝑙𝑃𝑃𝑙𝑙𝑅𝑅𝑙𝑙𝐴𝐴𝑃𝑃 + 𝛽𝛽2 ⋅ 𝑃𝑃𝑃𝑃𝑠𝑠 + 𝛽𝛽3 ⋅ 𝑃𝑃𝑏𝑏𝑃𝑃𝑃𝑃𝑃𝑃 + 𝛽𝛽4 ⋅ 𝑅𝑅𝑙𝑙𝑃𝑃 + 𝛽𝛽5 ⋅ 𝑊𝑊𝑃𝑃𝑃𝑃𝑏𝑏𝑃𝑃ℎ𝑃𝑃𝑃𝑃𝑅𝑅𝑏𝑏 + 𝛽𝛽6 ⋅ 𝑅𝑅𝑃𝑃𝑃𝑃𝑅𝑅 + 𝛽𝛽7 ⋅ 𝐴𝐴𝐴𝐴 𝑃𝑃𝑃𝑃𝑅𝑅𝑅𝑅𝑅𝑅𝑃𝑃 + 𝛽𝛽8 ⋅ 𝑇𝑇𝑃𝑃𝑃𝑃 𝑃𝑃𝑅𝑅𝑃𝑃 + 𝛽𝛽9 ⋅ 𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃 + 𝛽𝛽10 ⋅ 𝑀𝑀𝑃𝑃𝑏𝑏𝑃𝑃 𝑃𝑃𝑅𝑅𝑃𝑃𝑃𝑃𝑅𝑅 + 𝛽𝛽11 ⋅ 𝐹𝐹𝑅𝑅𝑃𝑃𝑃𝑃𝑃𝑃 𝑃𝑃𝑅𝑅𝑃𝑃 Table 13-29. Summary of MAIS2+F frontal impact logistic regression model parameters for intrusion-based model. Predictor Variable Parameter Coefficient --- β0, Intercept 5.445 Belt status β1, Belted -0.519 Sex β2, Male -0.218 Obese β3, BMI ≥ 30 kg/m2 0.195 Age β4, Age ≥ 65 0.324 Windshield Intrusion β5, Exceed MASH 0.863 Roof Intrusion β6, Exceed MASH 0.583 A/B Pillar Intrusion β7, Exceed MASH 0.940 Toe Pan Intrusion β8, Exceed MASH 1.961 Door Intrusion β9, Exceed MASH 0.686 Side Panel Intrusion β10, Exceed MASH 1.401 Floor Pan Intrusion β11, Exceed MASH 0.819 Although the developed intrusion model is useful in capturing the effect of other crash and occupant related factors, the binary intrusion variables would not provide any discernment between

206 tests that meet all the MASH intrusion criteria but have differing levels of intrusion. The relatively crude ranges of occupant intrusion available in the NASS/CDS data, which were used to develop the models, did not permit development of more detailed models at this time. As MASH user agencies would be interested in comparing two or more devices with varying levels of intrusion below the current MASH thresholds, the developed intrusion models have been included in the final report for documentation purposes as well as future research but are not recommended for inclusion in an updated version of MASH. 13.3.6 Proposed Options for Updating MASH Based on the results of the analyses conducted as part of the project, four possible MASH occupant risk modifications have been developed, as summarized in Table 13-30, including a “no change” option. A brief rationale summary for each potential modification option is also provided. In all cases, the associated injury risk curves are suggested for inclusion to provide MASH users with an alternative means to quantify differences in occupant risk across hardware based on the vehicle-based metric. Table 13-30. Summary of MASH occupant risk modification options. Option Designation / Description Rationale 1 / Modify lateral OIV thresholds + Injury risk curves • MAIS2+F occupant injury risk as well as Harm models suggest lateral thresholds are less conservative than the associated longitudinal thresholds, especially for the best-case occupants. For side impacts, the current maximum lateral OIV threshold corresponds to an injury risk above the maximum ASI threshold of 1.9. For frontal impacts, however, the current maximum longitudinal OIV threshold corresponds to an injury risk between the ASI thresholds of 1.4 and 1.9. • The original NCHRP Report 230 lateral OIV threshold was lower than the longitudinal OIV threshold. This threshold was increased in subsequent crash testing procedures based only on a small number of sled tests (4 total) and reconstructed oblique real-world crashes (17 total). • The current project used real-world crash data from 494 frontal, 183 side, and 176 oblique impact involved occupants to develop the associated models. Applying the applicable statistical weighting factors, these cases represent approximately 167,000 frontal, 66,000 side, and 67,000 oblique impacts. Also, the models developed in this research effort were based on crashes occurring between 2006 and 2015 (vehicle model years 2005 to 2016). • Analogous international crash test procedures use the THIV, essentially a resultant OIV, with maximum threshold at approximately the current preferred MASH OIV threshold. 2 / Replace OIV with OLC + Injury risk curves • The current study found only relatively small differences between the real- world injury prediction capabilities of the candidate injury metrics with no candidate injury metric consistently performing best across the three different crash modes or datasets (training vs. test). While this may suggest the current OIV metric is sufficient, the OLC metric specifically models a belted occupant, which is better representative of the occupants in the current vehicle fleet. • The OLC consistently performed the best in predicting ATD-based injury criteria in full-scale vehicle crash tests compared to the other candidate metrics. In terms of real-world crash injury, OLC did perform the best on oblique impacts in the test dataset.

207 Option Designation / Description Rationale 3 / Eliminate RA and replace with ASI + Injury risk curves • International crash test procedures no longer use the PHD metric, the analog to the RA metric, citing issues with the reliability of measuring this metric. The international procedures, however, do specify an alternate acceleration-based metric, the ASI. • In this project, the RA alone was the only candidate injury metric not found to have a statistically significant relationship with real-world occupant injury risk. ASI was found to have a statistically significant relationship to real-world occupant injury risk. 4 / No changes + Injury risk curves • The current study found only relatively small differences between the real- world injury prediction capabilities of the candidate injury metrics with no candidate injury metric consistently performing best across the three different crash modes or datasets (training vs. test). This suggests the current OIV metric, while it does model unbelted occupants, is sufficient. • While the belt usage rate has substantially increased since the inception of the FSM (from approximately 15% to more than 80%), an unbelted occupant still represents the practical worst-case occupant scenario. • While there are injury risk differences observed between the lateral and longitudinal OIV thresholds, the differences are primarily observed for the best- case occupant scenario. The differences between the lateral and longitudinal injury risk at the current MASH OIV thresholds for the worst-case occupant are much less pronounced. 13.3.7 Evaluate Potential Implications of Proposed Options for Updating MASH Based on the available sample MASH crash tests, each of the potential MASH occupant risk modification options were evaluated. There were a total of 124 sampled MASH tests: 84 passing tests, one marginal test (due to a high lateral OIV value), and 39 tests not meeting one or more of the MASH evaluation requirements. The vast majority of the tests that failed to meet MASH requirements were a result of vehicle rollover and/or the test vehicle penetrating the system via rail rupture or override. Only four of the failing tests were a result of not meeting one or more of the occupant risk criteria threshold values (i.e., OIV or RA). MASH modification Option 1 described above suggests lowering the lateral OIV threshold to better align the lateral thresholds with the longitudinal thresholds based on real-world occupant injury risk. Table 13-31 summarizes the implications of lowering the lateral OIV threshold to various levels using the available passing/marginal passing sample MASH tests. The test designations and associated hardware type tested are also identified to provide additional context. Table 13-32 provides a summary of the associated MAIS2+F occupant injury risk for the various potential OIV lateral thresholds along with the corresponding MAIS2+F injury risk for the current longitudinal OIV threshold (i.e., 12.2 m/s).

208 Table 13-31. Summary of potential implications of modifying the lateral OIV threshold (Option 1). Revised Lateral OIV Threshold Number of Passing/Marginal Tests that would fail the revised lateral threshold (% of sampled tests) MASH Test Designation Hardware Type 9.1 m/s 7 (8.3%) 3-10 1 concrete barrier 2 bridge rails 3-11 3 bridge rails 3-20 1 transition 10 m/s 4 (4.8%) 3-10 1 concrete barrier 1 bridge rail 3-11 1 bridge rails 3-20 1 transition 10.5 m/s 2 (2.4%) 3-10 1 concrete barrier 3-11 1 bridge rail 11 m/s 11.5 m/s 1 (1.2%) 3-11 1 bridge rail Table 13-32. Summary of occupant MAIS2+F injury risk for various lateral OIV thresholds (Option 1). Revised Lateral OIV Threshold Best Case Occupant P(MAIS2+F) [%] Worst Case Occupant P(MAIS2+F) [%] Corresponding Longitudinal Occupant P(MAIS2+F) Range for Current Longitudinal Threshold [%] 9.1 m/s 12.5 90.0 15.8 – 97.3 10 m/s 16.8 92.7 10.5 m/s 19.6 93.9 11 m/s 22.8 94.9 11.5 m/s 26.4 95.7 Based on the data presented in Table 13-31, there is relatively little impact on the currently accepted roadside hardware if the lateral OIV threshold was lowered even to 10.5 m/s. Less than 3% of the sampled MASH test would be affected by this lowered threshold, primarily the stiffer barriers (i.e., bridge rails and concrete barriers). This proportion is even lower if the marginal pass test is classified as not meeting MASH occupant risk criteria (the lateral OIV for that test was 12.4 m/s, slightly higher than the 12.2 m/s maximum threshold). Based on the data presented in Table 13-32, lowering the lateral OIV threshold to 10.5 m/s would result in a closer match with the corresponding injury risk probabilities associated with the longitudinal OIV threshold value of 12.2 m/s, primarily for the best-case occupant. Note, however, that roadside hardware crash test procedures have traditionally adopted the “practical worst case” philosophy, so it is more appropriate to compare the worst-case occupant risk values. Comparing the worst-case occupant injury risk at the current maximum threshold value reveals only a small difference (96.7% compared to 97.3%) in injury risk by crash mode. The larger difference between the best-case occupants could reflect the relative efficacy of the occupant restraints in each direction (i.e., frontal-based restraints have a longer development history and have the advantage of more space available to slow the occupant in a crash scenario compared to side impact restraints). Given that the differences in injury risk appear limited to restrained occupants, reducing the lateral threshold value would only be prudent if there is a change in the underlying philosophy, such as to protect belted occupants, as these represent the majority rather than the unbelted worst-case occupant, or when occupant belt usage in the fleet reaches 100%. Potential MASH modification Option 2 described above suggests replacing the OIV metric with the OLC metric as this metric was developed for restrained occupants. Figure 13-1 shows how the

209 sample test OIV values compare to the current MASH preferred and maximum threshold values. The marker types distinguish whether the test was judged to meet, marginally meet, or fail to meet the MASH evaluation criteria (all applicable MASH criteria, not just relevant occupant risk criteria). As noted earlier, most of the MASH tests that failed to meet MASH criteria did so in areas other than the prescribed FSM/OIV thresholds. Figure 13-1. Longitudinal and lateral OIV values by test outcome for sample MASH tests compared to current threshold values. Figure 13-2 shows how the sample test OLC values compare to a potential preferred OLC threshold of 14 G and a maximum threshold of 19 G. Based on the available data, the potential thresholds would not result in any changes to the number of passing tests, other than the previously “marginal” test would be considered a failing test. Also, there would be approximately the same number of tests situated between the preferred and maximum thresholds for the OLC (16 tests) as the OIV (18 tests). This plot demonstrates that the OLC could be used in place of OIV if deemed appropriate. Similar to Option 1, however, using OLC in place of OIV would be prudent if there is a change in the underlying crash test philosophy or if belt usage of occupants in the fleet was 100%. For reference, Table 13-33 provides a summary of the occupant MAIS2+F injury risk values for both the current OIV thresholds as well as the identified potential OLC thresholds. Overall, the 0 2 4 6 8 10 12 14 16 0 2 4 6 8 10 12 14 16 La te ra l O IV [m /s ] Longitudinal OIV [m/s] Passing Tests Marginal Test Failing Tests OIV Maximum Threshold OIV Preferred Threshold

210 values are similar, especially for the worst-case occupant scenarios where the injury risk is within 3% for the maximum threshold values. In general, the best-case occupant with the potential OLC thresholds have a slightly lower risk of MAIS2+F injury. Figure 13-2. Longitudinal and lateral OLC values by test outcome for sample MASH tests compared to potential threshold values. Table 13-33. Summary of occupant MAIS2+F injury risk for OIV and potential OLC thresholds (Option 2). Impact Type Threshold Best Case Occupant P(MAIS2+F) [%] Worst Case Occupant P(MAIS2+F) [%] Frontal OIVx = 9.1 m/s 1.9 79.3 OLCx = 14 G 1.4 90.8 OIVx = 12.2 m/s 15.8 97.3 OLCx = 19 G 6.9 98.0 Side OIVy = 9.1 m/s 12.5 90.0 OLCy = 14 G 9.7 88.3 OIVy = 12.2 m/s 32.0 96.7 OLCy = 19 G 19.1 94.3 Potential MASH modification Option 3 suggests omitting the RA metric and instead using the ASI metric, similar to the approach used by the analogous international roadside hardware crash test procedures. To investigate this possibility, the ASI values for the sample tests were examined in isolation and relative to the RA threshold value (see Table 13-34). Despite the ASI and RA both 0 5 10 15 20 25 30 0 5 10 15 20 25 30 La te ra l O LC [G ] Longitudinal OLC [G] Passing Tests Marginal Test Failing Tests Potential OLC Maximum Threshold Potential OLC Preferred Threshold

211 being acceleration-based metrics, there is a significant number of passing tests (11 of 84; 13%) that would fail the current international crash test procedure ASI limit of 1.9. For the tests failing to meet the MASH RA thresholds, only half have an ASI greater than 1.4 and only one with ASI greater than 1.9. Based on these data, simply removing RA and using ASI based on the entire crash pulse would result in relatively significant changes to the number of passing MASH tests. Table 13-34. Summary of potential implications of replacing RA with ASI (Option 3). Test Type Tests with RA > 20 G Tests with ASI > 1.4 Tests with ASI > 1.9 Tests with ASI > 1.4 and RA > 20 G Tests with ASI > 1.9 and RA > 20 G Passing + Marginal N/A 24 11 N/A N/A Failing 4 6 3 2 1 Aside from assessing the implications of potential changes to MASH criteria, the sample MASH tests revealed computation issues with two of the candidate metrics. Both the MDV and VPI computations are sensitive to the analysis window selected. Many MASH tests record data for multiple seconds, many up to 5 or 10 seconds surrounding the crash event. These long durations mean that the sensors can include data when the test facilities are applying the brakes to stop the vehicle after the crash event. When using the full extent of electronic sensor data available for all sampled MASH tests, approximately half of the tests had very large MDV and VPI values. The research team has examined these cases and appropriately truncated the data to include only the data relevant to the crash event. While in many cases it is relatively clear when the crash event ends, adopting a metric that is sensitive to the analysis window selected could introduce additional subjectivity into the analysis procedures. In contrast, the other metrics (OIV, RA, ASI, OLC) were insensitive to the analysis window selected. The values of these metrics changed on average less than 0.01% when computed using the full extent of electronic sensor data or the truncated electronic sensor data. MASH Modification Recommendation Considering the four potential MASH modification options previously presented along with the potential implications identified based on the available sample MASH tests, the research team makes the following recommendations: • Option 4 is recommended for inclusion in an updated MASH. None of the investigated candidate injury metrics provided a consistently better prediction of real-world occupant injury risk than the current OIV metric. Although RA alone was not found to be a statistically significant predictor of real-world occupant injury risk, RA was found to be a significant predictor as a binary indicator along with the OIV metric, which is typically how this metric is employed in the MASH procedures. 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. Given that the vehicle-based occupant risk criteria are rarely the discerning factor in whether a test is judged as passing or failing, this switch is not recommended at this time. • Inclusion of the developed injury risk curves will provide agencies that rely on MASH tests results with additional context to interpret reported occupant risk values, including considering other external factors not related to vehicle kinematics. This additional context

212 would be expected to help agencies make better informed decisions related to roadside safety hardware installation. • Additional study of the RA metric is warranted. Should more detailed EDR data become available in the future, this may allow for a more accurate computation of the RA metric. Longer EDR recording times would also help, as the RA focuses on the later portion of the crash event and the EDRs to date typically only record for 150 to 300 ms. Consideration should also be given to the use of a modified version of ASI as this computation is more accurate using current EDR data and could be modified to compute an ASI value for the same portion of the crash pulse that the RA uses (i.e., after occupant contact with the idealized vehicle interior). • Future MASH occupant risk updates including the reduction of the lateral OIV threshold and the use of OLC in place of OIV would likely only be prudent if there was a shift in underlying crash test philosophy (i.e., protecting belted occupants as they represent the majority rather than the worst-case unbelted occupant) or if occupant belt usage in the vehicle fleet is at 100%. • Computation of the MDV and VPI metrics on sample MASH tests revealed that these two metrics are sensitive to the analysis window selected, which would introduce additional subjectivity. The other investigated candidate injury metrics, the OIV, RA, ASI, and OLC, were insensitive to the analysis window selected and thus better suited as metrics to be included in the MASH procedures.