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.

5 2 Synthesis of Existing and Potential Roadside Crash Injury Metrics and Identification of Relevant Data Sources Scope and Approach The purpose of this section is to summarize the available literature, both U.S. and international, related to existing and potential roadside hardware crash injury metrics. There was a specific focus on the biomechanical basis, associated limitations, and practicality of computing each metric as well as an assessment of the applicability of currently available data to assess the existing and potential roadside hardware crash injury metrics. This review focused primarily on the available literature related to the following five areas: 1. Existing roadside hardware crash injury metrics, including biomechanical basis, associated limitations, and computational practicality of each metric. 2. Potential vehicle-based crash injury metrics not currently used in roadside hardware testing but that could potentially be adapted for use, including their biomechanical basis, any associated limitations, and computational practicality of each potential metric. 3. Studies examining occupant compartment intrusion and associated occupant injury, including studies that serve as the basis for the MASH acceptable intrusion limits. 4. Injury characterization metrics to supplement the proposed use of the Abbreviated Injury Scale (AIS) in the current project. 5. Available data and datasets that could be used to assess both existing and potential roadside hardware crash injury metrics. Existing Roadside Crash Injury Metrics The following sections summarize the existing U.S. crash injury metrics that have been used to evaluate occupant risk in full-scale vehicle-to-roadside hardware crash tests. For historical context, Table 2-1 summarizes the U.S. roadside hardware crash testing procedural documents and associated occupant risk metrics prescribed in each document. Note that since the NCHRP Report 153 procedures were published, there has been a three-tiered approach to evaluating roadside hardware crash tests that includes structural adequacy and post-impact vehicle trajectory in addition to the focus of this literature review, the occupant risk evaluation. A high-level summary of all the published studies found relating extant roadside crash injury metrics (FSM and newer) to injury can be found in Appendix A (Table A-1).

6 Table 2-1. Historical summary of roadside hardware crash test procedures and associated occupant risk metrics. Roadside Crash Test Standard Year Published Occupant Risk Metric(s) HRB Circular 482 1962 Limit vehicle average, maximum decelerations and/or momentum change NCHRP Report 153 1974 TRC 191 1978 NCHRP Report 230 1981 FSM NCHRP Report 350 1993 MASH 2009 FSM + Location/component specific vehicle deformation limits 2016 2.2.1 FSM Predecessors Prior to the FSM, the evaluation of occupant risk consisted of comparing vehicular decelerations or momentum change to established limits. These criteria appeared in NCHRP Report 153 (Bronstad and Michie 1974) guidelines and in the subsequent TRC 191 (TRB 1978) guidelines under the guise of “impact severity,” although the intent was to estimate occupant risk. The underlying assumption is a relationship between vehicle dynamics and occupant injury potential; presumably, the lower the maximum vehicle decelerations and change in momentum, the lower the potential for severe occupant injury. The authors of these guidelines, however, cautioned that this relationship is “tenuous,” affected by numerous other factors including occupant differences and vehicle interior padding differences, and the evaluation procedures “are not directly applicable to the complex highway collision” (TRB 1978). For oblique (<15° impact angle) collisions with longitudinal barriers, NCHRP Report 153 (Bronstad and Michie 1974) specified limits on the longitudinal, lateral, and total 50-ms average accelerations of the vehicle during the impact (summarized in Table 2-2). There was no guidance on whether the “total” category is found by adding the lateral and longitudinal accelerations or by evaluating the resultant accelerations. For crash cushion impacts and guardrail end terminals where the vehicle is decelerated to a stop, the authors recommended a maximum average deceleration of 12 G, with the desirable average acceleration between 6 G and 8 G, in addition to the limits shown in Table 2-2. For breakaway or yielding sign supports, the authors prescribed a maximum momentum change of 1,100 lbf-s, with a preferred limit of 750 lbf-s. Table 2-2. NCHRP Report 153 redirection impact severity thresholds (Bronstad and Michie 1974). Peak 50-ms Average Vehicle Accelerations (G) Category Longitudinal Lateral Total Preferred 5 3 6 Acceptable 10 5 12 A majority of the prescribed occupant risk criteria was retained in the subsequent TRC 191 (TRB 1978), including the limiting values shown in Table 2-2 for redirectional impacts and the deceleration and momentum change limits noted above. The most significant change was the provision of additional details for applying the momentum change limits in the two prescribed breakaway/yielding support tests. Criticisms of this approach to occupant risk included the piecewise nature (e.g., using peak 50- ms acceleration for barriers but average deceleration and momentum change for other devices) and

7 that the criteria were overly conservative in some respects (e.g., widely used concrete barriers typically did not satisfy the preferred thresholds; Chi 1976; Michie 1981a). These criticisms ultimately led to the development of the FSM. While the pre-FSM injury criteria present in NCHRP Report 153 and TRC 191 limit peak vehicle decelerations, there was evidence in the available literature of at least two other approaches to determining injury risk: (1) relating observed vehicle damage to average vehicle decelerations and occupant injury, and (2) developing a “severity index” similar to the present-day ASI metric. In tandem with developing guidelines for highway bridge railings, Olson et al. (1970) developed parabolic curve fit equations (see Table 2-3) relating vehicle damage (rated on a scale of 1 to 7) to average vehicle deceleration and probability of occupant injury for both frontal and oblique collisions. These equations were an extension of work by Michalski (1968) that examined approximately 950 traffic crashes in Oregon resulting in 184 injuries and seven fatalities. Note the distinction between average vehicle deceleration shown in Table 2-3 compared to the peak 50-ms accelerations noted in NCHRP Report 153 and TRC 191 shown in Table 2-2. As the second method (i.e., the severity index) is very similar to the current ASI metric, the early work relating to the severity index will be further discussed in tandem with the ASI metric. Table 2-3. NCHRP Report 86 equations relating vehicle damage, average vehicle deceleration, and occupant injury potential (Olson et al. 1970). Impact Type Developed Mathematical Relationship Front Glongitudinal = 0.280*R2 = 13.7*P Angle Glateral = 0.204*R2 = 10.0*P Where: G = average vehicle deceleration (component) R = vehicle damage rating (integer value, 1 to 7) P = proportion of vehicles in which injury occurred 2.2.2 FSM Description, Assumptions, and Evolution: Michie (1981a) introduced the FSM as part of the NCHRP Report 230 (Michie 1981b) crash test procedures. The FSM was retained in the subsequent NCHRP Report 350 (Ross et al. 1993a) procedures and in the current MASH procedures. The FSM provides a single measure of occupant risk regardless of the evaluated device and is more indicative of the involved physical phenomena than prior injury risk metrics. The FSM represents the occupant-vehicle interaction during a collision as a simplified two- phase event (Michie 1981a): 1. At vehicle impact, the occupant remains at the pre-impact vehicle velocity while the vehicle compartment surfaces begin to accelerate toward the occupant. The occupant impacts an interior surface of the vehicle (e.g., dashboard, windshield, steering wheel). 2. After occupant contact with the vehicle interior, the occupant remains in contact with the interior surface and is subjected to the same acceleration forces as the vehicle. Injury can occur in either phase, resulting in two distinct metrics: the occupant impact velocity (OIV) and the occupant ridedown acceleration (RA). The OIV is the velocity difference between the occupant and vehicle occupant compartment at the end of the first phase, while the RA is the maximum 10-ms vehicle acceleration subsequent to the occupant-interior impact during the second

8 phase. Computed OIV and RA values are compared to threshold values to ensure that occupant injury risk is not excessive. The original FSM uses several simplifying assumptions (Michie 1981a) as depicted in part in Figure 2-1: • The occupant is assumed to be an unrestrained point mass positioned at the vehicle center of mass. • Distances that the occupant is allowed to “flail” prior to impacting the vehicle interior are assumed to be 0.6 meters (2 feet) longitudinally (i.e., parallel with the typical vehicle travel direction) and 0.3 meters (1 foot) in the lateral direction. • Occupant impact with the vehicle interior is completely inelastic (i.e., the occupant subsequently remains in contact with the vehicle interior and does not rebound). • Vertical accelerations of the vehicle, as well as vehicle roll, pitch, and yaw motions, are ignored. This implies the longitudinal/lateral motions of the occupant are independent. • The vehicle remains upright during the collision (i.e., no rollover), the occupant compartment remains intact (i.e., no inward penetrations that would affect occupant trajectory), and the occupant is not partially or fully ejected (i.e., the vehicle windows and doors remain closed). Figure 2-1. FSM assumptions and simplifications (Adapted from Gabauer and Gabler 2008a). Generally, these assumptions either (a) serve to simplify the computations associated with the OIV or RA metrics, to exclude scenarios where occupant injury could be caused by other means,

9 or (b) were intended to represent a “practical worst-case” at the time of model inception. Sled and vehicle crash tests indicate that an unrestrained occupant moves as a free body inside the occupant compartment within experimental accuracy (Michie 1981a). Vehicle motion during the crash test is typically measured at or near the vehicle center of mass; a coincident occupant initial location simplifies the OIV computations. Michie (1981a) also noted that front seat occupants are positioned near and just behind the center of mass for small sedans, so the vehicle pitching and rolling will not significantly affect OIV values. Michie noted belt usage rates of less than 20% at the time of development of the model as support for the unrestrained occupant assumption. This assumption represents a worst case for the OIV metric because belted occupants would have a lower OIV, and it also simplifies the OIV computation. For the no occupant rebound assumption, Michie observed ATD kinematics from available roadside hardware crash test video (number of tests not reported) and noted that the ATD remains in contact with the vehicle surface at least through the period of high vehicle accelerations. The “flail” dimensions, 2 feet longitudinally and 1 foot laterally, were initially based on a 50th percentile male in a normal upright and seated position inside a vehicle (number and specific vehicle types not reported; Michie 1981a). Michie noted that smaller occupant compartments or occupants seated closer to the vehicle interior will generally result in lower OIV values. The stipulation of the crash test procedures for the vehicle to remain upright (i.e., no rollover) minimizes the vertical vehicle accelerations. Also, occupant injury mechanisms in a rollover would be different than as described by the FSM approach. Similarly, occupant injury mechanisms would likely be different if the occupant compartment is penetrated or the occupant is partially or fully ejected. The FSM was retained, essentially unchanged, in the subsequent NCHRP Report 350 (Ross et al. 1993a) crash test procedures. The only change was a minor computation clarification for the case where the occupant does not reach either edge of the idealized vehicle interior (i.e., 0.6 meters longitudinally or 0.3 meters laterally). For this case, NCHRP Report 350 specified that the OIV be set equal to the vehicle velocity change occurring during vehicle contact with the test article. If the device remains in contact with the vehicle after impact, the velocity change is computed from the time when the vehicle clears the footing of the test article. Note that this is generally applicable to tests with shorter duration (e.g., impacts with breakaway poles or supports). As part of the development of the NCHRP Report 350 procedures, Ray et al. (1987b) investigated the FSM occupant compartment dimensions using measurements from approximately 160 New Car Assessment Program (NCAP) frontal crash test vehicles. All vehicles were passenger vehicles, model years 1978 through 1984, with 50th percentile male ATD occupants in the driver and right front passenger positions and the seats in the mid-track position. As part of the NCAP at the time, the distance between the ATD and occupant compartment interior was measured at nine locations (see Figure A-1). The researchers examined the variation in these nine measurements for the available vehicles and compared the values to the assumed FSM “flail” dimensions noted above. The analysis also included the measurement ranges with vehicles stratified into four different vehicle weight range categories. For a 5th percentile female in the right front position and seat adjusted rearmost, the 75th percentile value of the head-to-windshield longitudinal distance (HW in Figure A-1) was estimated to be 22 inches, suggesting that the corresponding FSM value of 2 feet is an “appropriately conservative yet realistic value” (Ray et al. 1987b). For the lateral direction, the head-to-side window distance (HS in Figure A-1) ranged from 7 to 13 inches, suggesting the corresponding FSM value of 1 foot was “appropriate” (Ray et al. 1987b). The final result of the analysis was verification of the appropriateness of the original FSM occupant

10 compartment dimensions. The same results were also presented in Ray et al. (1986). Note that the authors indicated that a 50th percentile male ATD from the available NCAP test data was used but presented occupant compartment dimensions assuming a “worst case” of a 5th percentile female seated in the rearmost position; no details are provided on how the authors transformed the available 50th percentile dimensions to the presented 5th percentile dimensions. Ray et al. (1986, 1987b) also conducted seven sled tests with a 1979 Honda Civic buck to further investigate the FSM (see Figure A-2 for a more detailed summary of these tests). These included three frontal tests using a 5th percentile female ATD (Part 572, 105 lb) and four side impact tests using a 50th percentile male side impact ATD (165 lb). The impact speeds were between 20 to 45 fps (6.1 to 13.7 m/s). The presence of the instrumented ATD allowed for a comparison of the FSM-estimated impact velocity (OIV calculated) and the actual impact velocity of the unrestrained ATD (OIV measured), as summarized in Table 2-4. Note that the FSM- estimated OIV used the actual distance between the ATD and the occupant interior, 22.5 inches longitudinally and 6.5 inches laterally. There was reasonable agreement between the estimated and actual OIV values, especially in the longitudinal direction and at higher impact velocities. At lower impact velocities, the FSM procedures generally overestimated the actual impact velocity. This study provided additional validation that the computed FSM OIV is a reasonable representation of the impact velocity of an actual unrestrained occupant. Table 2-4. Summary of sled tests to assess validity of the FSM (Ray et al. 1986, 1987b). Measurement [fps] Frontal Impact Sled Test # Side Impact Sled Test # 2538 2537 2539 2534 2533 2535 2540 Velocity Change [fps] 25 35 45 20 30 35 40 OIV measured [fps] 21.1 33.2 45.6 7.7 14.8 21.8 24.9 OIV calculated [fps] 23.5 34.4 45.8 9.5 18.1 22.2 25.3 Error [%] 11.37 3.61 0.44 23.38 22.30 1.83 1.61 A second study by Hinch et al. (1988), which described results from 20 full-scale impact attenuator crash tests at approximately 60 mph, reported the OIV using a typical 2-foot flail distance, the actual flail distance for the ATD present in each test, and the corresponding OIV associated with the actual flail distance (refer to Figure A-6). Comparing these values resulted in conclusions similar to Ray et al. (1986, 1987b). In all but two cases, the actual available flail space was less than or equal to the typically assumed 2-foot value, resulting in the traditional OIV overestimating the OIV (based on the actual flail distance) by approximately 9%. In the two cases where the flail distance exceeded 2 feet, the excess distance was an inch or less. Also, Hinch et al. (1988) noted that the ATD response data suggest injury (with the exception of femur) typically occurs when the unrestrained occupant first impacts the vehicle interior, providing additional support for the initial FSM injury mechanism assumptions. The following modifications to the FSM were considered for incorporation into the NCHRP Report 350 occupant risk procedures (Ross et al. 1993b): a) Position occupant in the driver or right front passenger seat, not the vehicle center of mass; b) Account for vehicle yaw motions in the FSM computations; and c) Alter the dimensions of the occupant compartment to reflect actual distances (e.g., the 1-foot lateral distance approximates the distance from the driver to the occupant compartment left of the driver, but not to the right of the driver).

11 These potential modifications, however, were ultimately not incorporated into the NCHRP Report 350 occupant risk procedures. For typical redirectional crashes, the authors noted that incorporating options (a) and (b) would not have “significant effects on controlling factors” (Ross et al. 1993b). The authors noted that inclusion of option (c) would result in almost all longitudinal barriers not meeting the limiting occupant risk threshold values. Several possible explanations provided by the authors included that NCHRP Report 230 devices are indeed unsafe, crash test impact conditions are more severe than most real-world barrier crashes, the OIV and RA thresholds are too low, or occupants do not flail about the occupant compartment as suggested by option (c). Since the NCHRP Report 230 devices appeared to be performing satisfactorily and the FSM was intended to be an index of risk, rather than an absolute measure, the authors concluded that inclusion of option (c) did not appear warranted (Ross et al. 1993b). The authors also noted that inclusion of any of these options would require a “rather complex” standardized computer program to compute the occupant risk values. The FSM was also retained in the subsequent MASH (AASHTO 2009; 2016) crash test procedures. The MASH commentary (2009) noted the following with regard to retaining the FSM: The flail space model has served its intended purpose well, and there are no indications that features designed and assessed thereby have performed adversely in service. Thus, it was decided to retain the flail space model for the present document. The commentary also noted that roadside hardware testing agencies are now using a standardized computer program to compute MASH-specified occupant risk values, which promotes consistency and accuracy of the computed values. The commentary provided no detail on any potential changes to the FSM considered for inclusion in MASH. With regard to the FSM, the commentary in the second edition of MASH (2016) is identical to that in the first edition of MASH (2009). Threshold Values and Biomechanical Basis: Using the biomechanical research available at the time, the intent of the original OIV and RA threshold values was to indicate severe but not life- threatening injury, distinguishing between an AIS value of 3 or lower and AIS values of 4 and higher (Michie 1981a). The AIS numerically describes injury severity in terms of threat to life and categorizes any injury using an integer value from 1 to 6, where 1 represents minor injury and 6 represents maximum/not survivable injury (Association for the Advancement of Automotive Medicine [AAAM] 2008). Note that many researchers unofficially use an AIS score of 0 as an indicator of no occupant injury. The same overall philosophy of severe but not life-threatening injury was used in the Federal Motor Vehicle Safety Standards (FMVSS) for occupant protection in frontal impacts (Michie 1981a). In addition to these upper thresholds, the original FSM (and future FSM versions) provided design/preferred limits for both the OIV and RA metrics to encourage the development of safer roadside devices. The design/preferred limits were computed by dividing the upper thresholds by a corresponding factor of safety. Table 2-5 summarizes the prescribed FSM threshold and preferred values from FSM inception to the present MASH crash test procedures. Note that, in some cases, the FSM limits vary not only by direction (e.g., longitudinal vs. lateral), but also by type of roadside hardware impacted in the full-scale crash test. The values presented in Table 2-5 are those presented in each corresponding crash test procedural document and reflect differing unit systems (i.e., NCHRP Report 230 indicated thresholds only in U.S. Customary units, NCRHP Report 350 indicated thresholds only in SI units, and MASH indicated thresholds in both unit systems).

12 The RA threshold has remained essentially unchanged since the FSM’s inception. The slight RA increase in MASH was to match current practice, as some roadside hardware devices had been considered acceptable if the RA rounded down to the original integer limit of 20 G, that is, 20.49 G or less (AASHTO 2009; 2016). More significant, but still relatively minor, changes have been applied to the OIV limits. The original lateral OIV threshold was found to be overly conservative by Ray et al. (1986, 1987b), so this threshold was increased from 30 ft/s in NCHRP Report 230 to essentially 40 ft/s (12 m/s = 39.4 ft/s) in NCHRP Report 350. This higher lateral OIV limit was maintained in the subsequent MASH publications. The slight difference in the NCHRP Report 350 and MASH OIV limits reflects issues converting between SI and U.S. Customary units; since several safety devices have been approved with a limit of 40 ft/s (12.2 m/s), this slightly higher limit was adopted for MASH (2009; 2016). For support structures and work zone devices, the current OIV limits are lower than those initially proposed for breakaway signs and luminaires in NCHRP Report 230. This change in part reflected “state of the possible” in terms of support structure design and harmonization with AASHTO (1985) support structure design standards. Table 2-5. Summary of FSM threshold and preferred values. Crash Test Procedure Direction Device Type(s) OIV Limit (preferred) RA Limit (preferred) NCHRP Report 230 (Michie 1981b) Longitudinal Breakaway signs and luminaires 40 ft/s (15 ft/s) 20 G (15 G) All Others 40 ft/s (30 ft/s) Lateral Redirectional Barriers 30 ft/s (20 ft/s) NCHRP Report 350 (Ross et al. 1993a) Longitudinal Support structures and work zone devices 5 m/s (3 m/s) 20 G (15 G) All Others 12 m/s (9 m/s) Lateral All (except support structures and work zone devices) 12 m/s (9 m/s) MASH (AASHTO 2009; 2016) Longitudinal Support structures and work zone devices 16 ft/s (10 ft/s) 4.9 m/s (3.0 m/s) 20.49 G (15.0 G) All Others 40 ft/s (30 ft/s) 12.2 m/s (9.1 m/s) Lateral All (N/A for support structures and work zone devices) 40 ft/s (30 ft/s) 12.2 m/s (9.1 m/s) The FSM threshold values shown in Table 2-5 are based on a limited number of experimental data. The NCHRP Report 230 longitudinal OIV thresholds are based primarily on 38 (Kay et al. 1973) and 99 (Begeman et al. 1978) frontal sled test sets performed at impact speeds up to 55 km/h (15 m/s) and 60 km/h (17 m/s), respectively. Refer to Figure A-4 and Figure A-5, respectively, for a more detailed summary of the Begeman et al. (1978) and Kay et al. (1973) sled tests. For the NCHRP Report 350 longitudinal OIV, an additional three (Ray et al. 1986, 1987b) frontal sled tests were performed at impact speeds up to 50 km/h (13.7 m/s). NCHRP Report 350 commentary also cites a paper that presented results of 20 full-scale crash tests—16 head-on and four oblique impacts—into impact attenuators at approximately 60 mph (Hinch et al. 1988). Refer to Figure A-6 for a more detailed summary of the Hinch et al. (1988) impact attenuator crash tests. All four studies utilized an ATD to infer potential occupant injury potential. The head injury criterion (HIC), a method developed by NHTSA to assess head injury risk for occupants involved in frontal collisions, was the primary injury metric used in the Kay et al. (1973) and Begeman et al. (1978) studies. In addition to HIC, the Ray et al. (1986, 1987b) study included peak chest acceleration.

13 The Hinch et al. (1988) study included peak chest acceleration and maximum femur loads in addition to HIC. NCHRP Report 350 commentary also noted consultation with biomechanics experts and a General Motors (GM) research study (Viano and Lau 1989) as support for retaining the longitudinal OIV thresholds. The first edition MASH (2009) commentary related to the FSM threshold values focuses primarily on the unit conversion/rounding issues mentioned previously with no additional biomechanical data or references that support the longitudinal OIV thresholds. With regard to the longitudinal OIV thresholds, the commentary in the second edition of MASH (2016) is identical to that present in the first edition of MASH (2009). Human injury tolerance data are also limited in the lateral direction; the original lateral OIV threshold of 30 ft/s (9 m/s) was based principally on a total of 296 lateral impacts obtained from 6 years of French accident data starting in 1970 (Hartman et al. 1976). For the NCHRP Report 350 procedures, the increase of this value to 40 ft/s (12 m/s) was prompted by an analysis of four lateral sled tests with an ATD (see Figure A-2) and a total of 17 reconstructed oblique collisions into longitudinal barriers (Ray et al. 1986, 1987b). The measured ATD responses, summarized in Table 2-6, were used to determine head (HIC) and thoracic (Thoracic Trauma Index) injury risk in each of the four sled tests. Note that even in the most severe test (#2540), the HIC value was relatively low, 316 compared to the typical 1,000 threshold at the time, and the computed Thoracic Trauma Index was associated with a 16% risk of severe chest injury. The OIV in this test, however, exceeded the preferred lateral limit of 20 ft/s. Note, however, that HIC was developed to predict head injury in frontal collisions and not side impact crashes. The cases reconstructed by Ray et al. (1986, 1987b) were selected from a total of 679 in-depth barrier crashes from two different sources: (1) the National Automotive Sampling System (NASS) Longitudinal Barrier Special Studies (Erinle et al. 1994) and (2) the FHWA Narrow Bridge Study (Mak and Calcote 1983). Cases were first narrowed to those where the barrier length-of-need section (with end treatments excluded) was the first object struck and the vehicle was tracking, smoothly redirected, and did not roll over as a result of the initial impact. Of the 165 suitable cases, only 17 had serious occupant injury (i.e., a maximum AIS score [MAIS] of 3 or 4) and were subsequently reconstructed. The reconstructions suggested that none of the observed severe injuries could be attributed to the first impact (i.e., with the barrier). Occupant injury severity (first impact only) was plotted as a function of lateral OIV, estimated from the reconstruction (see Figure A-3). Based on these results, the authors suggested severe occupant injuries (≥MAIS 4) occur at lateral OIV values exceeding 40 ft/s. This higher NCHRP Report 350 lateral OIV threshold was maintained in MASH. Neither edition of MASH provides any additional biomechanical data or references to support the current lateral OIV threshold.

14 Table 2-6. Summary of lateral sled tests and ATD-based injury risk (Ray et al. 1986, 1987b). Parameter Side Impact Sled Test # 2534 2533 2535 2540 Velocity Change [fps] 20 30 35 40 OIV measured [fps] 7.7 14.8 21.8 24.9 OIV calculated [fps] 9.5 18.1 22.2 25.3 Head Injury Criterion (HIC) 37 121 193 316 Thoracic Trauma Index 69 91 97 113 P(AIS≥3) based on Thoracic Trauma Index [%] 0 3 6 16 Due to the extensive amount of research in the aeronautics field on tolerable limits of acceleration, the RA limits are based on more extensive research. The original NCHRP Report 230 lateral and longitudinal RA threshold of 20 G was chosen based on an extensive literature review by Snyder (1970) and a critical review (Chi 1976) of the pre-FSM occupant risk procedures. The review by Snyder (1970) summarized the findings of 35 different human-subject and 40 animal- subject forward-facing deceleration studies. The referenced forward-facing studies included more than 400 human subject tests and more than 500 animal subject tests. For the lateral direction, the data are more limited. Snyder summarized the findings of eight different human-subject studies (124 total human subject tests) and 12 different animal subject studies (at least 95 animal subject tests). NCHRP Report 350 commentary also notes consultation with biomechanics experts and a GM research study (Viano and Lau 1989) as support for retaining the RA thresholds. The RA threshold was maintained in MASH, but neither edition of MASH provided any additional biomechanical data or references in support of this threshold value. Evaluation of Link to Injury: Early efforts to link the FSM to injury in real-world roadside hardware crashes used crash reconstruction techniques to estimate FSM parameters in real-world crashes or attempted to match full-scale crash tests to real-world crashes. Ray et al. (1986, 1987b), discussed above, estimated lateral OIV in 17 real-world barrier crashes using dynamics relationships and several simplifying assumptions, including a constant vehicle deceleration during the time interval in which the vehicle becomes parallel to the barrier. Ray et al. (1987a) compared this estimated lateral OIV to the observed lateral OIV in 16 full-scale crash tests. The error associated with the estimation procedure was approximately 10% on average, but the available data suggested the OIV could be overestimated or underestimated by nearly 25%. Council and Stewart (1993) utilized accident data in an attempt to link occupant risk (as calculated in crash tests) to actual injury attained in collisions. The procedure matched instrumented full- scale crash tests with similar vehicle characteristics (make, model, and year), crash characteristics (object struck, impact location on vehicle, etc.), and crash severity (as measured by vehicle deformation) in real-world roadside hardware crashes. The authors used North Carolina crash data from 1973 to 1986 and included guardrail, median barrier, bridge rail, and sign support impacts. For the investigation of pre-FSM occupant risk measures, 232 crashes (53 guardrail, five median barrier, 122 bridge rail, and 52 sign support impacts) were matched with analogous full-scale crash tests. For the FSM investigation, 62 crashes (34 guardrail, 24 median barrier, and four bridge rail) were matched with analogous full-scale crash tests. Contingency table analysis was used to determine relationships between occupant injury and crash test occupant risk measures. Results of the study indicated the lack of a strong relationship between injury severity and vehicle momentum change and 50-ms peak acceleration values. With regard to the FSM, the limited amount of data in the study prevented any conclusions. An approximate comparison done by Michie (1981a),

15 though, suggested that there is not a significant disparity between the previous 50-ms criterion and the OIV portion of the FSM. The accuracy of the occupant risk parameters for the real-world crashes, as determined by the Council and Stewart (1993) matching methodology, is not known. Aside from the relatively small sample sizes, the primary limitation of these early efforts was the uncertainty associated with estimating FSM risk values without the vehicle crash pulse information (i.e., typically used to compute the OIV and RA in full-scale crash tests). The crash reconstruction methods and efforts to match real-world crashes with a laboratory crash test introduce potentially large uncertainties in the FSM risk value determination. To overcome this limitation, more recent attempts to link FSM to injury in real-world crashes leveraged the vehicle crash pulse data provided by event data recorders (EDRs). EDRs are devices that record vehicle crash characteristics (e.g., vehicle delta-v). Initially installed in GM vehicles for the 1994 model year (Chidester et al. 1999), NHTSA estimated that 96% of model year 2013 light vehicles were equipped (NHTSA 2013). Several studies have evaluated the accuracy of EDRs (Chidester et al. 1999; Niehoff et al. 2005; Tsoi et al. 2013, 2014) with a primary focus on longitudinal maximum delta-v (MDV) and found them to be accurate within 7%. Table 2-7 provides a high-level summary of these EDR-based studies, and a brief discussion of each appears in the following paragraphs. Several of these studies addressed the accuracy of determining the FSM metrics using available EDR data; this is noted when applicable. Table 2-7. Summary of EDR-based studies relating the FSM to real-world occupant injury. Study (Year) Available Cases (weighted) Injury Criteria and Associated Occupant Injury Metrics Gabauer and Gabler (2004a) 58 - 66 Longitudinal OIV and RA. Overall (MAIS) injury (58 cases) and injury by body region (66 cases, OIV only). Body regions included head, chest, upper/lower extremity, and neck. Gabauer and Gabler (2004b) 69 Longitudinal OIV and ASI. Overall (MAIS) injury and injury by body region. Body regions included head and thoracic/abdomen. Gabauer and Gabler (2006) 191 Longitudinal OIV and Delta-v. Overall (MAIS 3+) injury. Gabauer and Gabler (2008a) 180 Longitudinal OIV, ASI, and Delta-v. Overall (MAIS 2+ and 3+) injury Tsoi and Gabler (2015) 334 (102,744) Longitudinal OIV, ASI, Delta-v, VPI. MAIS 3+ injury based on thorax, abdomen, and spine body regions. Gabauer and Gabler (2004a) developed a methodology to evaluate the FSM using EDR data to compute OIV and ORA. The authors presented a preliminary analysis of OIV and RA based on 58 available real-world crash cases (see Figure 2-2) and a body region–specific OIV analysis based on 66 cases. The dataset was limited to single impact frontal collisions with airbag deployment, no rollover, complete EDR velocity versus time data, and known occupant injury data. No restriction was placed on the impacted object; approximately three fourths of the cases were vehicle-to-vehicle crashes, and the remaining one fourth were vehicle-to-fixed object impacts. Also, approximately 77% of the occupants were restrained by a seat belt and airbag in the cases used for analysis. Based on six full-scale frontal rigid barrier crash tests where vehicle EDR data were also available, the EDR-computed longitudinal OIV was noted to have an average error of 4% (6% maximum), while EDR-computed longitudinal RA consistently overestimated the observed RA by 40% on average (Gabauer and Gabler 2004a). The primary finding of the study was that OIV was a substantial predictor of overall occupant injury, but the predictive ability varied by body region. OIV was found to be a good predictor of chest injury, and to a lesser extent, lower

16 extremity injury but a weak predictor of head and upper extremity injury for single event frontal collisions. With regard to RA, no apparent correlation was found between RA and occupant injury in frontal collisions. A follow-on study (Gabauer and Gabler 2004b) used essentially the same frontal crash dataset to compare the OIV and ASI metrics. The longitudinal OIV was found to be a stronger predictor of occupant injury than the longitudinal ASI based, but the findings relative to OIV were nearly identical to the initial Gabauer and Gabler (2004a) study. Gabauer and Gabler (2006) used logistic regression to develop injury risk curves relating OIV to serious (MAIS 3 and greater, or MAIS 3+) injury and compared OIV to the simpler change in vehicle velocity (delta-v) metric. Compared to the initial study (2004a), this study (2006) used an expanded dataset, 191 frontal crash cases compared to the 58 in the initial study, but there was approximately the same proportion of belted and airbag-restrained occupants (158 of 191 were belted; 83%) as the previous study. Accuracy of the EDR-estimated OIV reported as within 6% based on a larger number (37) of full-scale crash tests with EDR data available. OIV was found to be a better predictor of injury for unbelted occupants and found to offer no statistically significant advantage, in terms of injury prediction, than the simpler delta-v metric. Figure 2-2. Maximum occupant injury severity in 58 frontal collisions as a function of FSM risk values (Figure 2, Gabauer and Gabler 2004). Gabauer and Gabler (2008a) compared the injury predictive capability of OIV, delta-v, and ASI and also developed MAIS 2+ and 3+ injury risk curves for each metric (see Figure A-8 for the developed risk curves). A slightly smaller frontal crash dataset was used compared to the 2006 study, as only drivers were included (180 cases compared to 191 cases), but belt usage rates (81% of occupants belted) and objects struck (12% fixed object and 88% other vehicles) were similar to the previous studies. Logistic regression was used to develop the injury risk curves for two occupant subsets: (1) belted and airbag restrained and (2) airbag restrained only. Models were compared using the available fit statistics as well as the area under the receiver operating

17 characteristic (ROC) curves. Longitudinal OIV was found to be a better predictor of injury for unbelted occupants (compared to belted occupants) and to offer no statistically significant serious injury prediction advantage over the simpler delta-v metric. Tsoi and Gabler (2015) compared the injury predictive ability of OIV, ASI, VPI, and delta-v metrics using frontal crashes with available EDR data. A total of 334 raw cases were used for the analysis with 24 occupants sustaining MAIS 3+ injury. Using the available NASS/Crashworthiness Data System (CDS) weights and excluding cases with weights exceeding 5,000, approximately 102,000 weighted cases were available for analysis. The methods were similar to Gabauer and Gabler (2008) with the exception of the consideration of the NASS/CDS weights and the determination of MAIS that included only the thorax, abdomen, and spine body regions (as opposed to all body regions in the Gabauer and Gabler [2008a] study). This more recent study had a similar percentage of belted occupants (71% raw, 72% weighted); the distribution of object struck was not reported. Contrary to Gabauer and Gabler (2008a), OIV was found to offer a statistically significant improvement over the delta-v metric for the prediction of serious injury to belted occupants. For unbelted drivers, OIV had the highest area under the ROC curve, but no statistically significant difference in predictive capability was found between any of the analyzed metrics. Based on the available cases, Tsoi and Gabler determined that the best longitudinal OIV thresholds for distinguishing serious occupant injury were 9.7 m/s and 13.2 m/s for belted and unbelted occupants, respectively. Refer to Figure A-9 for the injury risk curves developed by Tsoi and Gabler (2015). As an alternative to using real-world crash data, several researchers have attempted to correlate the FSM to injury risk as measured by an instrumented ATD. Generally, these efforts either conducted additional crash tests or sled tests or used existing publicly available vehicle crashworthiness test data. A high-level summary of these previously published efforts is provided in Table 2-8, and a brief discussion of each appears in the following paragraphs.

18 Table 2-8. Summary of crash/sled test studies relating the FSM to ATD-based occupant injury metrics. Study (Year) # Tests (Occupants) Injury Criteria Test/Restraint Configuration/ Notes Ray et al. (1986, 1987b) 7* (7) Longitudinal OIV to HIC, chest acceleration; lateral OIV to HIC, Thoracic Trauma Index No restraint: frontal, 25-45 fps (3 tests) No restraint: side, 20-40 fps (4 tests) *Sled tests only, excludes 4 tests done with Lexan windshields. See Figure A-2 for a more detailed summary. Hinch et al. (1988) 20 (26) Longitudinal OIV/RA to HIC, chest acceleration, femur force All tests at approximately 60 mph; 24 of 26 occupants were unrestrained. See Figure A-6 for a more detailed summary. Gabauer and Thomson (2005) 24 (44) Longitudinal OIV, RA, ASI to HIC, chest acceleration, chest deflection, femur force Frontal – 25 to 40 mph (21 tests) Frontal offset – 40 mph (3 tests) Airbag only (16 occupants) and airbag + belted occupants (28 occupants) Gabauer and Gabler (2008d) 39 (71) Longitudinal OIV, RA to HIC, chest acceleration and combined head/chest No restraint: 97 km/h frontal (9 tests) Airbag only: 40 km/h frontal (10 tests) Belt only: 48 km/h frontal (10 tests) Airbag + belt: 56 km/h frontal (10 tests) Silvestri- Dobrovolny et al. (2016b; 2016c) 1 (1) OIV, RA to HIC15, chest deflection, neck tension, femur force Airbag + belt: 56 km/h frontal The Ray et al. studies (1986, 1987b), previously described in reference to the FSM assumptions and modification of the lateral OIV threshold, conducted frontal and side impact sled tests to correlate OIV to ATD-based occupant risk. With regard to the longitudinal OIV, a linear fit between OIV and HIC for the three available sled tests suggest that an OIV of 41 ft/s corresponds to an HIC of 1,000, the threshold for serious head injury at the time of the study. Similarly, a longitudinal OIV of 34.7 ft/s corresponds to a peak chest acceleration of 60 G, the threshold for serious chest injury at the time of the study. With regard to the lateral OIV, a linear fit between OIV and HIC for the four available sled tests suggests that an OIV of 70 ft/s would correspond to an HIC of 1,000. Similarly, a lateral OIV of 41 ft/s corresponds to a Thoracic Trauma Index value of 150, which correlates to approximately a 50% probability of AIS 3+ chest injury (Kuppa 2004). The Hinch et al. study (1988) was primarily aimed at investigating the performance of impact attenuators (two inertial barrier systems and a single energy absorbing system) with small and large passenger car test vehicles as well as differing inertial barrier conditions (e.g., pea gravel vs. sand fill, frozen vs. non-frozen sand, and bagged vs. loose sand fill). All of the 20 full-scale crash tests were performed with at least one instrumented ATD present and the HIC, maximum chest acceleration, and maximum femur force were documented (see Figure A-6 for a more detailed summary of numerical values). Plots of the available ATD response as functions of OIV and RA were generated using the available data and can be seen in Figure A-7. Note that only the unrestrained ATDs were included (24 ATD responses total). Although not explicitly noted in the published study, it is presumed that the computed RA values are longitudinal only as the tests were head-on or at small oblique angles and the OIV is noted as the “delta-V at a 2-foot flail,” the typical longitudinal limiting distance. Visual inspection of the plots suggests no strong correlation between OIV or RA and any of the available ATD response values. Linear fits between all six combinations (OIV to HIC, OIV to peak chest G, OIV to peak femur, RA to HIC, RA to peak chest G, and RA to peak femur) result in R2 values of 0.19 or less, suggesting weak correlations at

19 best. Hinch et al. noted that, except for one case, all the ATD occupant injury values were within prescribed limits while one or more of the vehicle-based injury indicators (OIV or RA) exceeded the NCHRP Report 230 “preferred” threshold in 13 of the 20 tests. Also, the RA limiting value was exceeded in four of the tests. Based on this, the authors suggested that the “preferred” NCHRP Report 230 OIV and RA limits may be overly conservative. Gabauer and Thomson (2005) compared existing roadside metrics (OIV, RA, and ASI) to ATD metrics (HIC, peak chest acceleration, maximum chest deflection, and maximum femur force) in 24 previously conducted full-scale, primarily full-width frontal barrier, vehicle crash tests. The examined vehicles included an approximately equal number of passenger cars and light trucks (e.g., pickup trucks, sport utility vehicles, full size vans, minivans). The analysis was primarily qualitative but indicated a wide variation in HIC and chest deflection within a relatively small corresponding range of OIV values. Based on the available data, RA was found to have the strongest correlation to HIC, and the OIV was found to be conservative for the frontal collision mode. Gabauer and Gabler (2008d) examined roadside metric and ATD-based injury risk variation in 39 previously conducted full-scale crash tests. Tests spanning four different restraint conditions were included: (1) no belt, no airbag, (2) airbag only, (3) 3-pt belt only, and (4) 3-pt belt and airbag. Impact speeds ranged from 40 km/h to nearly 100 km/h but were similar within the same restraint condition category. ATDs included either the Hybrid II or Hybrid III models depending on the restraint condition. For each restraint condition, the variation in OIV was compared to the normalized probability of injury based on HIC and peak chest acceleration. Similar to Gabauer and Thomson (2005), there was a relatively small variation in OIV but a wide variation in the corresponding ATD-based injury measures. Linear regression was used to examine the correlation between OIV and RA and the ATD risk measures (HIC, peak chest acceleration, and combined probability of head/chest injury). The strongest correlations were found in the no belt, no airbag restraint condition with OIV explaining roughly 30% of the variation in the ATD risk measures. Most of the correlations, however, were not statistically significant (even for the no belt, no airbag condition). The RA correlations were surprisingly weak; the highest R2 value was 0.122 with RA as a predictor. Silvestri-Dobrovolny et al. (2016b, 2016c) conducted a single full-scale crash test with a MASH 1100C test vehicle. The 2010 Toyota Yaris impacted a rigid wall at 90° and 35 mph (i.e., same impact conditions as specified by the U.S. NCAP tests). A belted Hybrid III ATD was seated in the driver seat with the passive safety (i.e., airbag) systems active. The measured ATD response was used to compute associated body region injury risk metrics. Longitudinal OIV was computed to be 54 ft/s, well above the 40 ft/s MASH threshold. The ATD-derived occupant risk parameters, however, were well within the NHTSA-prescribed limits for head, chest, neck, and femur body regions (HIC15 of 264, chest deflection of 50.9 mm, neck tension of 1930 N, and maximum femur force of 3764 N). The primary finding was that the MASH occupant risk is conservative. The researchers suggested that in certain cases when the OIV is in excess of the 40 ft/s limit, other recorded data from the crash test could be used to infer injury risk. Since the maximum 50-ms average acceleration value was approximately 30 G in the conducted crash test, the authors suggested this threshold as one possibility.

20 Another similar approach to linking the FSM to ATD-measures of occupant injury risk involves computer simulation of crash tests in lieu of conducting physical tests. Generally, the simulations are validated against one or more conducted or available physical crash tests, and then the validated model is used to determine both vehicle-based and ATD-based occupant injury risk measures. Typical simulation types are either finite element simulations (FE; e.g., using LS-DYNA or other similar programs) or multibody simulations (MBS; e.g., using MADYMO or other similar programs). A high-level summary of these previously published efforts is provided in Table 2-9, and a brief discussion of each appears in the following paragraph. Table 2-9. Summary of computer simulation studies relating the FSM injury metrics to ATD-based occupant injury metrics. Study (Year) # Tests (Simulations) Injury Criteria Test/Restraint Configuration/ Notes Li et al. (2015) (28) Lateral and longitudinal OIV and RA to HIC, maximum chest deflection Large pickup to concrete barrier (16) and w-beam barrier (12). Speeds from 50 to 120 km/h and impact angles varied from 15-30 for concrete barrier and 20-30 for w-beam. Li et al. (2015) developed LS-DYNA FE models of a large pickup truck (NCAC 2006 Ford F250) impacting concrete and w-beam barriers at various impact speeds and angles. The FE models were validated using a sled test (to validate ATD kinematics) and a full-width frontal rigid barrier test. No details, however, are provided of any FE model validation conducted using available vehicle-to-concrete or vehicle-to-w-beam barrier crash tests. A total of 16 simulations were conducted with the pickup impacting a concrete safety shape barrier at four different impact speeds (50, 70, 100, and 120 km/h) and four different impact angles (15°, 20°, 25°, and 30°). A total of 12 simulations were conducted with the pickup impacting a G4(1S) w-beam barrier at the same four impact speeds and three different angles (20°, 25°, and 30°). For each simulation, the lateral and longitudinal OIV and RA values are computed as well as the 15-ms HIC and maximum chest deflection. Regression equations, typically either exponential, power-law, or quadratic polynomial functions, were developed to predict the ATD-based metrics using the FSM metrics. For concrete barrier impacts, longitudinal OIV correlated well with HIC (R2 = 0.96), but none of the FSM metrics were found to correlate well with maximum chest deflection (R2 values of 0.60 or less). For w-beam barrier impacts, longitudinal OIV had the highest correlation with HIC (R2 = 0.58) and maximum chest deflection (R2 = 0.87) of the FSM metrics. All computed OIV and RA values were below the MASH threshold values, but two of the concrete barrier simulations and one of the w-beam simulations had HIC values in excess of 700. Based on the available plots, all the maximum chest deflection values were approximately 30 mm or less (typical limit is 76 mm). One other alternative to link the FSM to injury is to use available post-mortem human subject (PMHS) tests; a single study was found using this method. Tan et al. (2016) examined the current lateral OIV threshold using previously conducted PMHS tests involving lateral thorax impact. A total of 131 tests were included from a total of nine studies published between 1976 and 2004. The majority of the tests were sled tests (109 of 131) with a smaller number of pendulum impact tests (22). The thoracic impact velocity ranged from 3.6 to 11.9 m/s with a mean value of 8 m/s. The PMHS age ranged from 17 to 86 years with a mean age of 52 years. The included thoracic impacts included both padded (53 tests) and rigid (78 tests) surfaces. Three different impact side arm configurations were present in the included tests: (1) directly alongside the thorax, (2) slightly

21 anterior of the thorax with the lower arm placed on the lap, and (3) raised overhead. Although the exact number of tests with each arm position is not reported, the sled tests are noted to be configuration (1) or (2) while the pendulum tests use configuration (3). Multiple logistic regression analysis was used to model severe thoracic injury (AIS 3+ and AIS 4+) as a function of impact velocity, PMHS age and mass, test method (sled or pendulum), and impact interface (padding or no padding). PMHS sex and height were omitted due to missing data. The model results suggested that impact velocity, PMHS age, and impact interface were statistically significant, while PMHS mass and test method were not statistically significant. Variable significance was the same whether the pendulum tests were included or excluded. Single variable logistic regression models were used to develop AIS 3+ and AIS 4+ injury risk curves based on impact velocity. The developed curves suggest a high risk of severe thoracic injury at impact velocities in excess of 10 m/s and 10.8 m/s for AIS 3+ and AIS 4+, respectively. The current “preferred” lateral OIV threshold of 9 m/s corresponds to an 84% and 67% likelihood of sustaining an AIS 3+ or AIS 4+ thoracic injury, respectively. The current maximum lateral OIV threshold of 12 m/s corresponds to a 97% likelihood of AIS 3+ or AIS 4+ thoracic injury. In light of the developed risk curves, the authors suggested a lower limit of 6.4 m/s as this corresponds to approximately a 90% sensitivity and approximately 50% likelihood of serious (AIS 3+) thoracic injury. There are, however, several important limitations to the Tan et al. (2016) study. Since the data were combined from several previous studies, differences in instrumentation, test methods, and methods of injury assessment would likely influence the overall results. One such difference is the arm position differences noted above; previous research has found that arm position has an effect on thoracic injury outcome. Also, the average PMHS age is greater than the average age of the current population. Previous biomechanics research has indicated a higher likelihood of injury for older PMHSs subjected to the same impact force. In addition, these tests were generally localized impacts directly to the thorax and may not represent a whole-body lateral impact, as modeled in the FSM. Practicality of Computation: The assumptions that underpin the FSM provide for a reasonable ability to compute this metric from data typically obtained from full-scale roadside hardware crash tests. Additionally, the roadside safety community has used the FSM as a means to assess occupant risk for more than 30 years. While the FSM computations can be performed relatively easily with typical computer applications (e.g., Microsoft Excel, MATLAB), a standard computer program, the Test Risk Assessment Program (TRAP), does exist to compute the FSM risk values, the OIV, and RA (Bligh et al. 2000). The TRAP computer program is presumably the “standardized computer program” referred to in the MASH commentary (2009). Due to the extensive use of the FSM, this metric will serve as the reference for computation practicality; the computation practicality of other potential injury risk metrics will be judged relative to the computations required for the FSM. FSM Variants: Although the FSM provides a more physically correct approach to assessing occupant injury risk, researchers have long questioned the validity of the FSM simplifying assumptions. The numerous variations in the computation of the FSM that have emerged are a testament to this fact. Most notable variations are as follows:

22 • Ray et al. (1987a) • Ross et al. (1988) • Ray and Carney (1989) All the variations incorporate vehicle yaw motion and utilize the coupled equations of motion; the lateral and longitudinal motions are not assumed independent. The Ray et al. (1987a) algorithm also tracks the position of the occupant with respect to its original position and the relative velocity with respect to the vehicle interior at each time interval. Additional features of the Ross et al. (1988) version are a more exact “flail” space that accounts for different occupant seating locations (e.g., the driver can “flail” in excess of 1 foot laterally to the right) and a more exact tracking of the occupant movement. For instance, if the occupant impacts the side of the vehicle first, the lateral component of the velocity is set to zero, but the algorithm continues to track the occupant’s longitudinal motion. The Ray and Carney (1989) version also tracks the occupant position beyond the initial impact but accounts for rebound. At each impact with the vehicle interior, the algorithm determines the velocity normal to the contacted boundary and subtracts it from the occupant velocity in that direction to determine the rebound velocity. No previously published studies have been found relating any of these specific FSM variants to real-world occupant injury. Most of the noted studies, however, did compare the computed FSM variant values to the traditional FSM values. In general, the differences between the FSM and associated variants were small. Based on an analysis of data from two crash tests, Ray and Carney (1989) reported that using the uncoupled equations of motion can result in an OIV error of up to 6%. With regard to other more sophisticated techniques for tracking the occupant and vehicle interior positions, Ray and Carney note that these result in “small changes” to the impact velocity value for the first occupant-to-interior impact. Similarly, based on an analysis of two typical redirectional crash tests, Ray et al. (1987a) found an error in the computed OIV of 6% or less when neglecting the effects of vehicle yaw rate. 2.2.3 Theoretical Head Impact Velocity (THIV) and Post-Impact Head Deceleration (PHD) Description, Assumptions, and Evolution: THIV and PHD are analogous to the FSM OIV and RA, respectively. Both metrics have been adopted by CEN to assess occupant risk in roadside hardware crash tests (CEN 2010). Both metrics differ slightly in computation from its FSM counterpart since each uses the coupled equations of motion and the resultant velocity/acceleration values, as opposed to separating lateral and longitudinal velocity/acceleration components as in the FSM. Other than these noted differences, the assumptions that underpin the THIV and PHD metrics are the same as those that underpin the FSM (discussed previously). Up until 2010, CEN has prescribed both the THIV and PHD metrics for use in assessing occupant risk. From 2010 forward, only the THIV is required, with the PHD being removed based on empirical evidence and expert opinion that it is an unreliable metric (Anghileri 2013). Anghileri (2013) also noted that the RA/PHD concept is physically correct, but the practicality of its measurement is “too sensitive to oscillations in the acceleration trace.”

23 Threshold Values and Biomechanical Basis: A summary of the threshold values for the THIV and PHD metrics are summarized in Table 2-10. When used, the PHD metric limiting value was 20 G, identical to the RA limiting value in both the lateral and longitudinal direction. Since the PHD considers the resultant acceleration, rather than the individual lateral and longitudinal components separately, the 20 G limit is actually slightly more stringent than the RA threshold. The THIV threshold has remained at 33 km/h and is approximately equivalent to the “preferred” OIV threshold of 9.1 m/s. Again, since the THIV considers the resultant impact velocity and not the lateral and longitudinal components separately, the THIV threshold is slightly more conservative than the “preferred” OIV threshold. Note that the EN 1317 procedures use impact severity class to group devices into three categories based on the potential risk to vehicle occupants; a Class A barrier has a lower risk than a Class B barrier, and a Class B barrier has a lower risk than a Class C barrier. This classification, however, depends only on the ASI metric value. The THIV and (pre-2010) PHD values are the same for all three impact severity classes, as evident in Table 2-10. Table 2-10. Summary of THIV and PHD metric limiting values. Crash Test Procedure Impact Severity Class THIV Limit PHD Limit EN 1317 (1998) A and B 33 km/h 20 G EN 1317 (2007) A, B, and C EN 1317 (2010) A, B, and C N/A The available EN 1317 documents provide no details on how the prescribed THIV and PHD limits were determined, including any associated biomechanics study information. At least one other study found in the literature noted this issue as well (Roque and Cardoso 2013). Since the THIV/PHD are FSM variants and the CEN procedures have generally been developed in tandem with the U.S. roadside hardware crash test procedures, it is presumed that many of the studies that serve as the FSM basis (discussed previously) were also considered when developing the associated THIV/PHD metrics. Unfortunately, the lack of detail available in the CEN documents in this regard does not allow any further investigation. Evaluation of Link to Injury: In general, studies relating roadside occupant risk measures to occupant injury use either real-world crash data, full-scale crash data augmented with ATD responses, or computer simulations of full-scale crashes with ATDs. Compared to the FSM, there was much less published literature found relating THIV/PHD to injury. There were no previously published studies found specifically relating THIV and/or PHD to occupant injury using real-world crashes. A total of three studies were found using crash tests and/or simulations to relate THIV/PHD to ATD-based injury metrics or other roadside injury metrics. Table 2-11 provides a high-level summary of these previously published efforts, and a brief discussion of each appears in the following paragraphs.

24 Table 2-11. Summary of crash test and/or simulation studies relating THIV/PHD to other injury metrics. Study (Year) # Tests (Simulations) Injury Criteria Test/Restraint Configuration/ Notes Kammel (2007) (~30, exact number not reported) THIV to ASI (no analogous ATD metrics) Small car to concrete barrier MBS (no further details provided), full-scale crash data collected from 3 test locations (total number of tests collected not reported) Sturt and Fell (2009) 3 (50) THIV/ASI to HIC, chest deflection, neck forces/moments, Euro NCAP front/side scoring protocol Small car to concrete barrier Tests: ~110 km/h and 15°–20° angle, HIII ATD Simulations: 90 – 150 km/h and 10°– 25° angle, Restraint status varied Li et al. (2015) (28) THIV/PHD/ASI to HIC, maximum chest deflection Large pickup to concrete barrier (16) and w-beam barrier (12). Speeds from 50 to 120 km/h and impact angles varied from 15-30 for concrete barrier and 20-30 for w-beam. Kammel (2007) used multi-body dynamics simulation (MEPHISTO program) to model a small passenger car (900 kg) impacting a concrete barrier at 100 km/h and a 20° angle. Details are provided on varying the concrete barrier angle (e.g., vertical wall to slope of 20° from vertical) as well as the friction coefficient between the car and the barrier. The author, however, did not provide additional details on any other parameters varied in the simulations and did not clearly indicate the total number of simulations conducted as part of the study. The output from the simulations included THIV and ASI, and the simulation-generated THIV values were plotted as a function of the simulation-generated ASI values. THIV and ASI values from physical crash tests were also included on the plot. Based on the available plot, there were approximately 40 full-scale crash test results (exact number not provided) gathered from three different testing locations and approximately 30 simulations conducted. These data were used to develop a relationship between THIV and ASI. There was no direct correlation, however, to any ATD-based occupant injury metrics. Sturt and Fell (2009) examined the relationship between crash severity, as measured by the THIV and ASI, and associated occupant injury risk, measured primarily through the response of a real (Hybrid III) or a computer-simulated ATD (Hybrid III or Euro-SID). The authors conducted three physical crash tests of an ATD-equipped small car impacting a concrete barrier at approximately 110 km/h and angles between 15° and 20°. A total of 50 LS-DYNA computer simulations were conducted, including simulations of the three physical crash tests to provide validation. The 47 additional simulations were used to examine the effects of varying vehicle impact speed, impact angle, vehicle crush properties, occupant position relative to the vehicle interior, ATD type (e.g., front vs. side), belt status (e.g., belted vs. unbelted), seat type, and friction between ATD and seat. Based on the crash test and simulation results, the authors compared the roadside metrics (THIV and ASI) to available injury metrics and assessment reference values (as measured by the real or computer-simulated ATD). The ATD-based metrics included HIC, neck forces/moments, chest deflection, and European New Car Assessment Programme (Euro NCAP) frontal and side scoring protocols. For all simulations, THIV ranged from 15 to 42 km/h. For head and chest injury, the simulated injury measurements were found to be within acceptable limits for THIV values below 35 km/h. No unacceptable injury measurements were observed for the

25 abdomen and pelvis body regions. For neck injury, unacceptable injury measurements were observed for THIV values above 38 km/h. Based on these findings, Sturt and Fell concluded that THIV values below 33 km/h are unlikely to produce significant occupant injury provided there are no other risk factors present, such as the concrete barrier being high enough to be struck by the occupant’s head. As previously described, Li et al. (2015) developed LS-DYNA FE models of a large pickup truck (NCAC 2006 Ford F250) impacting concrete (16 simulations) and w-beam barriers (12 simulations) at various impact speeds and angles. For each simulation, the THIV and PHD values are computed as well as the 15-ms HIC and maximum chest deflection. Regression equations, typically either exponential, power-law, or quadratic polynomial functions, were developed to predict the ATD-based metrics using the THIV/PHD metrics. For concrete barrier impacts, THIV correlated well with HIC (R2 = 0.95), but the PHD metric was found to have a weaker correlation (R2 = 0.69). Neither metric was found to correlate well with maximum chest deflection (R2 = 0.60 for THIV and 0.28 for PHD). For w-beam barrier impacts, THIV and PHD were found to have relatively weak correlations to HIC (R2 = 0.54 for THIV and 0.67 for PHD) and stronger correlations to maximum chest deflection (R2 = 0.92 for THIV and 0.68 for PHD). The THIV and PHD values were not reported in tabular form, but based on the provided plots it appears the THIV values ranged from 3 to 14 m/s for the concrete barrier simulations and 5 to 9 m/s for the w-beam barrier simulations. The highest PHD value observed for simulations of either barrier system was 14 G. Based on the conducted simulations, the THIV values (in combination with ASI) were found to be more conservative than the FSM counterparts; that is, none of the simulations had FSM values in excess of the thresholds, while six of the concrete barrier simulations and one of the w- beam simulations exceeded one or more of the CEN metrics. The similarities between the FSM and CEN metrics suggest that the previously discussed FSM- research may apply, at least in some capacity, to the relationship between the THIV/PHD and occupant injury. As an example, in the purely frontal collision mode, the THIV and OIV should be equivalent. Any developed relationships between OIV and injury for frontal crashes would be expected to be similar to the relationship between THIV and injury for the same subset of purely frontal crashes. Also, the results from the FSM variants suggest that the inclusion or exclusion of the coupled equations of motion should only produce relatively small OIV differences (e.g., 6% or less). Practicality of Computation: The THIV and PHD include the coupled equations of motion and the resultant accelerations/velocities, which results in a slightly more complicated computation compared to the analogous FSM risk parameters. Despite this increase in complexity, however, the THIV and PHD are still reasonable to compute using standard computer-based methods, and both the THIV and PHD computational procedures are currently included in the TRAP program (Bligh et al. 2000). 2.2.4 ASI Description, Assumptions, and Evolution: The ASI metric is a single dimensionless quantity representing the severity of vehicle motion during an impact. The ASI is based only on the measured three-dimensional acceleration of the vehicle center of gravity and is computed using the following relationship (CEN 2010):

26 2 1 222 ˆˆˆ )(               +        +      = z z y y x x a a a a a a tASI where xa , ya , and za are the 50-ms average component vehicle accelerations and xâ , yâ , and zâ are corresponding threshold accelerations for each component direction. The threshold accelerations are 12 G, 9 G, and 10 G for the longitudinal (x), lateral (y), and vertical (z) directions, respectively. Since it utilizes only vehicle accelerations, the ASI inherently assumes that the occupant is continuously contacting the vehicle, which typically is achieved through the use of a seat belt. The maximum ASI value over the duration of the vehicle acceleration pulse provides a single measure of collision severity that is assumed to be proportional to occupant risk. To provide an assessment of occupant risk potential, the ASI value for a given collision acceleration pulse is compared to established threshold values. Essentially no background information on the ASI metric is provided in the EN 1317 documents; however, the ASI metric is nearly identical to the “severity index” (SI) used by Weaver and Marquis (1974) to assess occupant injury risk in slope-traversing events. The only difference is that the Weaver and Marquis SI specifies acceleration in each direction (presumably instantaneous) and does not specify the 50-ms moving average acceleration in each direction as the ASI does. Other previous researchers (Hirsch et al. 1972) did note a specific time duration of 50 ms, identical to that of the ASI. Threshold Values and Biomechanical Basis: Although a maximum ASI value of 1.0 is recommended, a maximum ASI value of up to 1.9 is considered acceptable (CEN 2010). As noted previously, the ASI is the sole metric that defines the impact severity class in the CEN standards. In contrast, the THIV/PHD limits are the same for all three levels. The ASI ranges for each impact severity class are summarized in Table 2-12. Although impact severity Class C was added in the 2007 EN 1317 version, the limits for Class A and B remained the same. Table 2-12. Summary of ASI metric limiting values. Crash Test Procedure Impact Severity Class ASI Range EN 1317 (1998) A ≤ 1.0 B 1.0 < ASI ≤ 1.4 EN 1317 (2007, 2010) A ≤ 1.0 B 1.0 < ASI ≤ 1.4 C 1.4 < ASI ≤ 1.9 Note that if two of the three vehicular acceleration components are zero, the ASI will reach the recommended threshold of unity only when the third component reaches the corresponding limit acceleration. If more than one component is non-zero, however, the unity threshold can be attained when the components are less than their corresponding limits. According to EN 1317 (CEN 2010), the ASI value of 1.0 corresponds to “light injuries, if any” but a corresponding AIS injury level is not specified. Also, no corresponding injury level description or AIS level is provided for the higher ASI threshold values (1.4 or 1.9).

27 Similar to the THIV and PHD, the EN 1317 documents provide no details on how the ASI limits were determined, including any associated biomechanics study information. Since ASI has not been included as an evaluative metric in the U.S. crash test procedures (i.e., MASH only requires that ASI be computed and reported), there is no detailed description on the development of the metric and its threshold values, as there is for the FSM. Due to the similarity of ASI to the Weaver and Marquis (1974) SI, however, additional insight into the ASI development may be gleaned from further investigation of the SI metric. The maximum threshold values proposed by Weaver and Marquis are shown in Table 2-13 based on the type of occupant restraint. Note that the “lap belt only” limits correspond to those utilized in the current ASI. According to Chi (1976), these limits are based principally on a military specification for upward ejection seats (U.S. Military 1967) and a study done by Hyde (1968). Chi (1976) also noted that neither study provided any “supporting documentation or references” for the presented information. Hirsch et al. (1972) applied the same SI approach to evaluate collisions with median barriers instead of the slope traversal application used by Weaver and Marquis (1974). The primary difference in the Hirsch et al. (1972) use of SI was the use of the “unrestrained” limits shown in Table 2-13 and the establishment of a threshold limit of 1.5 as an indicator of “major” injuries. Table 2-13. Summary of occupant acceleration limits by restraint type for ditch traversal (Weaver and Marquis 1974). Restraint Maximum Acceleration (G) Longitudinal Lateral Vertical Unrestrained 7 5 6 Lap Belt Only 12 9 10 Lap and Shoulder Belt 20 15 17 Evaluation of Link to Injury: The previously discussed early effort by Council and Stewart (1993) to link the FSM to injury provides some insight relative to ASI effectiveness. In an analysis based on 232 matched cases, the authors indicated a lack of strong relationship between the peak 50-ms lateral and longitudinal average accelerations and driver injury. Although the peak 50-ms criterion does not transform tri-axial accelerations into a single quantity, it does utilize the 50-ms average acceleration concept as in the ASI. Several EDR-based studies (a number discussed previously) have used real-world crash data to link ASI (and other roadside injury metrics) to occupant injury. A high-level summary of these studies is provided in Table 2-14, and a brief discussion of the ASI-related results of each appears in the following paragraphs.

28 Table 2-14. Summary of EDR-based studies relating the ASI to real-world occupant injury. Study (Year) Available Cases (weighted) Injury Criteria and Associated Occupant Injury Metrics Gabauer and Gabler (2004b) 69 Longitudinal ASI and OIV. Overall (MAIS) injury and injury by body region. Body regions included head and thoracic/abdomen. Gabauer and Gabler (2005) 120 Longitudinal ASI. Overall (MAIS) injury and injury by body region. Body regions included chest, head, lower/upper extremities. Gabauer and Gabler (2008a) 180 Longitudinal OIV, ASI, and Delta-v. Overall (MAIS 2+ and 3+) injury Tsoi and Gabler (2015) 334 (102,744) Longitudinal OIV, ASI, Delta-v, VPI. MAIS 3+ injury based on thorax, abdomen, and spine body regions. Gabauer and Gabler (2004b) presented a methodology to compute ASI using available EDR data and examined data from 69 frontal crash cases to compare the OIV and ASI metrics. The dataset was limited to single impact frontal collisions with airbag deployment, no rollover, complete EDR velocity versus time data, and known occupant injury data. No restriction was placed on impacted object or occupant belt use. Approximately three fourths of the cases were vehicle-to-vehicle crashes, and the remaining one fourth were vehicle-to-fixed object impacts. Similarly, approximately three fourths of the occupants were belted, and remaining one fourth were airbag restrained only. Based on six full-scale frontal rigid barrier crash tests with vehicle EDR data also available, the EDR-computed longitudinal ASI was noted to underestimate ASI by approximately 5% on average and within 10%. In terms of injury severity, Gabauer and Gabler interpreted the preferred ASI threshold of 1.0 (i.e., “light injury”) as AIS scores between 1 and 2. Based on primarily qualitative analysis, longitudinal ASI was found to be a weaker predictor of occupant injury compared to OIV. The ASI threshold value of 1.4 was found to correspond to approximately the same injury risk potential as the maximum OIV threshold (12 m/s). Gabauer and Gabler (2005) used an expanded dataset (120 cases compared to 69) to examine the correlation of ASI to occupant injury in single-event frontal collisions. Cases were selected in a similar fashion to the previous Gabauer and Gabler studies but with the addition of a belted-only occupant restriction, as ASI inherently assumes the occupant is in continuous contact with the vehicle. The distribution of crash type was similar to the previous Gabauer and Gabler studies (e.g., 17% vehicle-to-fixed object and 83% vehicle-to-vehicle crashes). The error in EDR- computed ASI and the interpretation of injury level (e.g., values at or below the 1.0 threshold correspond to AIS ≤ 1 injuries) was also the same as reported in Gabauer and Gabler (2004b). Gabauer and Gabler (2005) also included contingency table analysis and logistic regression models to further investigate the link between ASI and occupant injury. Primary findings were that the ASI is a good indicator (at least with respect to the 1.0 threshold) of occupant injury for belted occupants involved in frontal collisions, and the available data support the assertion that an ASI of 1.0 corresponds to “light” (AIS 0 or 1) injury if any. Gabauer and Gabler (2008a) compared the injury predictive capability of OIV, delta-v, and ASI and also developed MAIS 2+ and 3+ injury risk curves for each metric using 180 available cases (refer to Figure A-8 for the developed risk curves). Belt usage rates (81% of occupants belted) and objects struck (12% fixed object and 88% other vehicles) were similar to the previous studies. There was no stipulation limiting the cases to belted occupants as in Gabauer and Gabler (2005), but the analysis was repeated for both the belted and unbelted occupant subsets of the data. Despite the ASI metric modeling a belted occupant, longitudinal ASI was found to offer no advantage over

29 OIV or delta-v in predicting injury in belted occupants. Considering both belted and unbelted occupants, longitudinal ASI was found to offer no statistically significant serious injury prediction advantage over the simpler delta-v metric. As previously described, Tsoi and Gabler (2015) compared the injury predictive ability of OIV, ASI, VPI, and delta-v metrics using 334 frontal crashes (102,000 weighted cases) with available EDR data. The methods were similar to Gabauer and Gabler (2008), with the exception of the use of the NASS/CDS weights and the determination of MAIS that included only the thorax, abdomen, and spine body regions as opposed to all body regions. ASI was found to be the best predictor of serious (MAIS 3+) occupant injury for belted occupants based on the area under the ROC curve but not quite statistically better than the delta-v metric (p = 0.073). There was also no statistically significant difference found between both metrics intended to predict injury to belted drivers, ASI and VPI. With respect to unbelted occupants, ASI was not found to have any statistically significant difference from the other metrics for predicting severe injury. Based on the available cases, Tsoi and Gabler determined that the best longitudinal ASI thresholds for distinguishing serious occupant injury were 1.18 and 1.57 for belted and unbelted occupants, respectively. Several researchers (some previously discussed) have attempted to correlate the ASI to injury risk as measured by an instrumented ATD. A high-level summary of these previously published efforts is provided in Table 2-15, and a brief discussion of each appears in the following paragraphs. For studies previously described, only the findings relative to ASI are summarized. Table 2-15. Summary of crash test studies relating ASI to ATD-based occupant injury metrics. Study (Year) # Tests (Occupants) Injury Criteria Test/Restraint Configuration/ Notes Hinch et al. (1988) 20 (26) Longitudinal 50-ms peak acceleration to HIC, chest acceleration, femur force All tests at approximately 60 mph; 24 of 26 occupants were unrestrained. See Figure A-6 for a more detailed summary. Shojaati (2003) 9 (9) Lateral ASI to HIC Lateral tests only with a Hybrid III ATD (no further test details provided) Gabauer and Thomson (2005) 24 (44) Longitudinal ASI to HIC, chest acceleration, chest deflection, femur force Frontal – 25 to 40 mph (21 tests) Frontal offset – 40 mph (3 tests) Airbag only (16 occupants) and airbag + belted occupants (28 occupants) Wusk and Gabler (2017) 140 (131 - 140 depending on ATD metric) Longitudinal ASI to HIC, peak chest acceleration, chest deflection, and pelvic acceleration Frontal, full-width rigid barrier tests at approximately 56 km/h. Drivers only; belted and airbag restrained occupants Hinch et al. (1988) reported on 20 full-scale crash tests involving impact attenuators. Although ASI was not specifically investigated, the authors reported the maximum 50-ms average acceleration such that the longitudinal ASI can be computed for the head-on tests (i.e., 15 of the 20 tests). Of the head-on tests, there were 21 ATD occupants, 19 unrestrained and two restrained. The computed longitudinal ASI was plotted against the reported ATD metrics (HIC, maximum chest acceleration, and maximum femur force; refer to Figure A-10 in Appendix A). Visual inspection of the plots suggests no strong correlation between ASI and any of the available ATD response values. Linear fits between ASI and the three ATD metrics result in R2 values of 0.02 or less, suggesting weak correlations at best.

30 Shojaati (2003) attempted to correlate the ASI to risk of occupant injury via HIC. For nine lateral sled tests, the HIC determined from a Hybrid III dummy was plotted against the ASI as determined from the measured vehicle acceleration. No additional details for the conducted tests, such as impact speed range or exact test configuration, were provided. From the available plot, ASI varied from approximately 0.5 to 2.25, while HIC varied from less than 100 to more than 2,000. The available data suggested an exponential relation between HIC and ASI. The results also suggest that ASI values of 1.0 or less correspond to HIC values of 100 or lower, ASI values between 1.0 and 1.5 correspond to HIC values up to approximately 350, and ASI values of 2.0 correspond to HIC values of approximately 1,000. Note that HIC and its associated limiting value were developed specifically for frontal head impacts, not side impacts. There is currently no equivalent head injury metric for the lateral direction. Gabauer and Thomson (2005) compared existing roadside metrics (including ASI) to ATD metrics in 24 previously conducted full-scale, primarily full-width frontal barrier, vehicle crash tests. The analysis was primarily qualitative but indicated a wide variation in HIC and chest deflection within a relatively small corresponding range of ASI values. Based on the available data, ASI was found to have the strongest correlation to maximum chest acceleration and was found to be conservative for the frontal collision mode. More recently, Wusk and Gabler (2017) used approximately 140 frontal NCAP tests to investigate how longitudinal ASI correlates to ATD-based injury metrics. The ATD-based metrics were computed only for ATDs in the driver position and included HIC, peak chest acceleration, peak chest displacement, and peak pelvic acceleration. Based on a linear regression analysis, longitudinal ASI was not found to have a strong correlation to any of the investigated ATD-based metrics (i.e., R2 values of 0.039 or less). Note that all the crash tests had approximately the same impact speed, 56 km/h, and that ASI varied from 1.62 to 2.76. As each NCAP test also had EDR data available, the authors were able to investigate how incomplete EDR crash pulses affected computed ASI values. The available data suggest that complete crash pulses do not necessarily result in more accurate ASI values. Several researchers (some previously discussed) used computer simulation to link ASI to ATD- based injury metrics. Table 2-16 provides a high-level summary of these previously published efforts, and a brief discussion of each appears in the following paragraphs.

31 Table 2-16. Summary of crash test/simulation studies relating ASI to ATD-based injury metrics. Study (Year) # Tests (Simulations) Injury Criteria Test/Restraint Configuration/ Notes Thomson et al. (2006), Naing et al. (2008) (22 vehicle to barrier, 20 vehicle to pole) ASI to HIC (Authors note THIV/PHD as well as viscous criterion and chest deflection but no detailed results presented related to these metrics) MBS simulations using MADYMO. Impact speeds of 10 to 28 m/s Barrier Simulations: Impact angles of 5°–35°, Vehicle orientation from 0°– 45°. Pole Simulations: Vehicle orientation from -57°–57°. Pole radius, offset and stiffness also varied. Sturt and Fell (2009) 3 (50) THIV/ASI to HIC, chest deflection, neck forces/moments, Euro NCAP front/side scoring protocol Small car to concrete barrier Tests: ~110 km/h and 15°–20° angle, HIII ATD Simulations: 90 – 150 km/h and 10°– 25° angle, Restraint status varied Li et al. (2015) (28) THIV/PHD/ASI to HIC, maximum chest deflection Large pickup to concrete barrier (16) and w-beam barrier (12). Speeds from 50 to 120 km/h and impact angles varied from 15-30 for concrete barrier and 20-30 for w-beam. As part of the Roadside Infrastructure for Safer European Roads (RISER) project, researchers used MBS to investigate the correlation between ASI and HIC (Naing et al. 2008; Thomson et al. 2006) in w-beam barrier and pole crashes. A stochastic process was used to vary the simulation parameters, which included vehicle impact speed, angle and vehicle orientation for the barrier simulations and impact speed, vehicle orientation, pole diameter, and pole stiffness for the pole simulations. For the barrier simulations, the intent was to vary the parameters around the typical barrier crash test values (i.e., 100 km/h impact speed and 20° impact/orientation angle, or vehicle tracking at 20° prior to impact). For the 22 vehicle-to-barrier simulations, impact speed varied from approximately 36 to 100 km/h, the impact angle varied from 5° to 35°, while the vehicle orientation angle varied from 0° to 45°. For the pole simulations, the impact speed varied from 36 to 100 km/h, the vehicle orientation varied from -57° to 57°, and the offset varied from ±0.4 m from the vehicle centerline. Linear regression was used to correlate ASI to HIC (R2 = 0.82) using the w-beam barrier simulations (Naing et al. 2008). Note that the exact same ASI/HIC plot is present in Thomson et al. (2006) in reference to the conducted pole simulations; it is not clear to which simulations the plot corresponds. Also, Thomson et al. (2006) mentioned several other roadside (THIV and PHD) and ATD-based metrics (viscous criterion and chest deflection), but no detailed results relative to these metrics were presented in either source. Sturt and Fell (2009) examined the relationship between ASI and associated occupant injury risk, measured primarily through the response of a real (Hybrid III) or a computer-simulated ATD (Hybrid III or Euro-SID). Three physical vehicle-to-concrete barrier crashes and 50 LS-DYNA computer simulations were conducted. For all simulations, ASI ranged from 1.1 to 2.3. For head and chest injury, the simulated injury measurements were found to be within acceptable limits for ASI values below 2.0. No unacceptable injury measurements were observed for the abdomen and pelvis body regions. For neck injury, unacceptable injury measurements were observed for ASI values above 2.0. The simulation results suggested that the boundary between impact severity Class B and Class C barriers does not correspond to any significant increase in injury risk.

32 As previously described, Li et al. (2015) developed LS-DYNA FE models of a large pickup truck impacting concrete (16 simulations) and w-beam barriers (12 simulations) at various impact speeds and angles. For each simulation, ASI was computed along with 15-ms HIC and maximum chest deflection. Regression equations, typically either exponential, power-law, or quadratic polynomial functions, were developed to predict the ATD-based metrics using the ASI metrics. For concrete barrier impacts, ASI was found to have the best correlation to HIC (R2 = 0.98) but had a relatively weak correlation to maximum chest deflection (R2 = 0.46). For w-beam barrier impacts, ASI was also found to have a strong correlation to HIC (R2 = 0.91) but a slightly weaker correlation to maximum chest deflection (R2 = 0.81). The ASI values were not reported in tabular form, but based on the provided plots it appears the ASI values ranged from 0.6 to 2.5 for the concrete barrier simulations and 0.4 to 2.2 for the w-beam barrier simulations. Based on the conducted simulations, the CEN metrics (THIV, PHD, and ASI in combination) were found to be more conservative than the FSM counterparts (i.e., none of the simulations had FSM values in excess of the thresholds while six of the concrete barrier simulations and one of the w-beam simulations exceeded one or more of the CEN metrics). Practicality of Computation: Although the ASI is not included in the traditional FSM computations, the metric has long been used by CEN in its assessment of occupant risk for roadside hardware crash tests. The ASI is reasonable to compute using standard computer-based methods and is also present in the computational procedures currently included in the TRAP program (Bligh et al. 2000). Potential Roadside Crash Injury Metrics The following sections summarize published literature related to potential vehicle-based crash injury metrics not currently used, but that could potentially be adapted for use, in roadside hardware crash testing. These potential metrics are summarized in Table 2-17. Table 2-17. Summary of potential vehicle-based crash injury metrics. Injury Metric Type Injury Metrics Crash Pulse Only Delta-v Maximum moving average acceleration Crash Pulse + Assumed Occupant Response Vehicle Pulse Index (VPI) Occupant Load Criterion (OLC) Crash Pulse + Actual Occupant Response Ride-down Efficiency Restraint Quotient (RQ) Relative Kinetic Energy Factor A high-level summary of all the published studies found relating potential roadside crash injury metrics to injury can be found in Appendix A (

33 Table A-2). 2.3.1 Crash Pulse Only Metrics Description and Assumptions: Crash pulse only metrics are derived solely from the vehicle response during the crash event. These metrics primarily include vehicle change in velocity (delta- v) and maximum moving average acceleration. While other vehicle crash pulse only metrics do exist, such as maximum acceleration, time to zero velocity, and maximum dynamic displacement, they have had relatively limited application in the prediction of occupant injury, especially compared to delta-v and the maximum moving average acceleration. For delta-v and the maximum moving average acceleration metrics, the assumption is that higher values increase the likelihood of occupant injury. Note that the maximum moving average acceleration has served as a metric for assessing occupant injury risk in full-scale roadside hardware crash tests prior to the FSM (previously discussed). The same moving average acceleration concept is currently used in the RA computation (using a 10-ms time window) and in the ASI (using a 50-ms time window). Threshold Values and Biomechanical Basis: Existing threshold values and biomechanical basis for the maximum moving average acceleration metric have been previously discussed both relative to the FSM predecessors, the RA metric, and the ASI metric. See Table 2-2 for the limits previously established using a 50-ms time window. The RA limits (i.e., a 10-ms time window) are shown in Table 2-5. No delta-v threshold value was identified through the available literature. Evaluation of Link to Injury: Results of the Council and Stewart (1993) study indicated a lack of strong relationship between the peak 50-ms lateral and longitudinal average accelerations and driver injury. Also, a more recent study by Gabauer and Gabler (2008b) found that both the 10-ms and 50-ms maximum moving average acceleration were poor predictors of serious injury to belted occupants in frontal collisions. Although a specific delta-v threshold has not been established, there are several previous studies correlating real-world crash injury risk to delta-v (Augenstein et al. 2003; Bahouth et al. 2004; Kononen et al. 2011). Although an early EDR-based effort indicated no significant difference between delta-v and the more complicated ASI and OIV metrics (Gabauer and Gabler 2008a), the most recent effort suggests (at least for belted occupants) that more sophisticated metrics do offer a significant improvement in predicting occupant injury risk (Tsoi and Gabler 2015). Practicality of Computation: Both the delta-v and moving average acceleration metrics are simpler to compute than the FSM metric. 2.3.2 VPI Description and Assumptions: The VPI metric, recently proposed by the International Organization for Standardization (ISO, 2013), models vehicle occupant motion while explicitly considering occupant restraints (e.g., seat belts and airbags). As illustrated in Figure 2-3 and Equations 1–3, VPI is based upon a single degree of freedom (SDOF) model described by a lumped mass-spring system. The model is used to estimate the motion of a restrained vehicle occupant, denoted by y(t).

34 ( )ymaxVPI = (1) ( )tPkyyM =+ (2) ( ) ( )   − = sxk tP 0 sx sx ≥ < , , (3) Figure 2-3. Schematic of the VPI impact severity metric that explicitly models the occupant and restraints as a mass-spring system (Tsoi and Gabler 2015). The mass, M, represents the occupant, and the vehicle motion is shown as x(t). The vehicle and occupant motion are connected by the restraint systems, modeled using the function P(t), in terms of a spring of stiffness, k, and slack, s. ISO recommends a value of 1 kg for mass, 2,500 N/m for the restraint stiffness, and 0.03 m for the available slack. Using these recommended values, the only unknown is the vehicle motion as a function of time. With known vehicle motion as a function of time, either recorded by lab instrumentation in a full-scale crash test or an EDR in a real-world crash, the model is used to estimate the maximum occupant acceleration that occurs during the crash event. Threshold Values and Biomechanical Basis: The ISO standard describing VPI only provides the computation procedure. There are no threshold values provided, such as limits on the estimated maximum occupant acceleration, that are intended to assess injury risk potential. Evaluation of Link to Injury: A single study was found relating VPI to occupant injury in real- world crashes. Tsoi and Gabler (2015) compared VPI to other traditional vehicle-based crash metrics using 334 frontal crashes (102,000 weighted cases) with available EDR data. For the VPI computation, Tsoi and Gabler used available full-scale crash test data to compute vehicle-specific stiffness (k) and slack (s) values for each crash-involved vehicle as opposed to the ISO recommendation to use fixed values for these two parameters. Based on the 65 GM models included in the study, the computed vehicle slack averaged 5.88 mm while the stiffness averaged 1,964 N/m, both lower than the ISO recommended values (i.e., 30 mm slack and 2,500 N/m stiffness). Based on the generated plots, the computed VPI values ranged from less than 100 m/s2 (10 G) to approximately 700 m/s2 (71 G). VPI was found to be the best predictor of serious (MAIS 3+) occupant injury for belted occupants based on the best threshold contingency table, although there was no statistically significant difference found between VPI and ASI for belted drivers. Along with ASI and OIV, VPI was found to offer a statistically significant advantage over the simpler delta-v metric for seriously injured, belted drivers. With respect to unbelted occupants, VPI was not found to have any statistically significant difference from the other metrics for predicting serious injury. Based on the available cases, Tsoi and Gabler determined the best longitudinal VPI thresholds for distinguishing serious occupant injury was 309 m/s2 (31.5 G) and 334 m/s2 (31.4 G) for belted and unbelted occupants, respectively. This is the only known study correlating VPI to occupant injury risk in real-world crashes. There was a single crash test study found relating VPI to ATD-based injury metrics. Prasad and Weston (2011) conducted 16 sled tests to examine the effect of different rear seat cushion characteristics on the resulting occupant crash dynamics. Each of the sled tests used a 3-year-old

35 child ATD as well as a 5th percentile female ATD. The crash pulse used was representative of a high severity frontal impact with a mid-size passenger car (i.e., essentially a 35 mph NCAP test). Stepwise linear regression was used to build models to predict observed ATD-based injury metrics including HIC, peak chest acceleration, peak chest deflection, two neck injury criteria, and head excursion distance, using several metrics to characterize the crash pulse, including VPI. Computed VPI values ranged from approximately 43 to 54 G. VPI was found to be the best predictor of HIC and peak chest acceleration for both ATDs but was found to have a weak correlation to neck axial force and head excursion distance. Practicality of Computation: Compared to the FSM, the computations for VPI are more complex. While using the original ISO specified version with fixed slack and stiffness values would simplify the computations, this would also mute the ability of VPI to consider restraint differences across different vehicles. Computing vehicle-specific slack and restraint stiffness requires full-scale vehicle crash test data to be available and be used to extract the model parameters (i.e., slack and stiffness). To perform the model extraction, Tsoi and Gabler (2015) used readily available Structural Impact Simulation and Model Extraction (SISAME) software. 2.3.3 OLC Description and Assumptions: The OLC is a vehicle-based metric with two distinct phases: (1) occupant free flight and (2) ideal restraint of the occupant. The free flight phase begins at the initial vehicle impact and concludes when the relative distance between the occupant and vehicle is 65 mm (t1). The occupant is then assumed to be ideally restrained until the relative distance between the occupant and vehicle is 300 mm (t2). The OLC is defined as the slope between points t1 and t2. (Kübler et al. 2009). A graphical representation of the OLC is shown in Figure 2-4 (adapted from Wusk and Gabler 2017). Threshold Values and Biomechanical Basis: Similar to VPI, OLC provides a means of quantifying occupant motion but does not specify any limiting values intended to be used to assess occupant injury risk potential.

36 Figure 2-4. Graphical representation of the OLC (Adapted from Wusk and Gabler 2017, Figure 4). Evaluation of Link to Injury: There were no previously published studies relating OLC to occupant injury in real-world crashes. A total of three studies (one previously discussed) were found using crash tests and/or simulations to relate OLC to ATD-based injury metrics. Table 2-18 provides a high-level summary of these previously published efforts, and a brief discussion of each appears in the following paragraphs. Table 2-18. Summary of crash test/simulation studies relating OLC to ATD-based occupant injury metrics. Study (Year) # Tests (Simulations) Injury Criteria Test/Restraint Configuration/ Notes Kübler et al. (2009) (Exact number not reported) Longitudinal OLC to HIC, peak chest acceleration MBS (MADYMO) simulations of NCAP tests using 50th HIII Ellipsoid ATD Park and Kan (2015) 60 Longitudinal OLC to HIC, peak chest acceleration, chest deflection Frontal, full-width rigid barrier tests at approximately 56 km/h. Drivers and RF passengers; belted and airbag restrained occupants Wusk and Gabler (2017) 140 Longitudinal OLC to HIC, peak chest acceleration, chest deflection, and pelvic acceleration Frontal, full-width rigid barrier tests at approximately 56 km/h. Drivers only; belted and airbag restrained occupants Kübler et al. (2009) used multibody simulation (MADYMO) to investigate crash pulse characteristics that influence ATD-based injury metrics, specifically HIC and peak chest acceleration. The simulations were conducted using a large number of U.S. NCAP crash pulses (exact number not reported but appears to be at least 50 based on the available plots) imposed on the model. A second order polynomial regression was used to correlate various crash pulse -70 -60 -50 -40 -30 -20 -10 0 0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 Ve lo ci ty [k m /h r] Time [seconds] Vehicle Occupant Compartment Occupant (Based on OLC Model) The Occupant Load Criteria (OLC) value is the slope of this line, e.g. -23.8 g

37 characteristics, including OLC, to the ATD-based injury metrics. The OLC generally had the lowest root mean square error term (i.e., best fit to the ATD-based metrics). Park and Kan (2015) examined 60 U.S. NCAP tests from 2012 to determine relationships between vehicle-based metrics and resulting ATD-based injury metrics for drivers (H3 50th Male) and right front passengers (H3 5th Female). The ATD metrics included HIC15, peak chest acceleration, and peak chest deflection. Using linear regression analysis, none of the examined metrics had a strong correlation to any of the ATD-based metrics (R2 values of 0.34 or less). The OLC to peak chest acceleration was the strongest correlation identified (RF passengers only; R2 value of 0.34), but the vast majority of the correlation coefficients were 0.1 or lower. Wusk and Gabler (2017) used approximately 140 frontal NCAP tests to investigate how longitudinal OLC correlates to ATD-based injury metrics, including HIC, peak chest acceleration, peak chest displacement, and peak pelvic acceleration. Based on a linear regression analysis, longitudinal OLC was not found to have a strong correlation to any of the investigated ATD-based metrics (i.e., R2 values of 0.056 or less). Note that all the crash tests had approximately the same impact speed, 56 km/h, and that OLC varied from 21.2 G to 40.6 G. As each NCAP test also had EDR data available, the authors were able to investigate how incomplete EDR crash pulses affected computed OLC values. The available data suggest that the OLC metric accuracy was drastically affected by incomplete crash pulses; that is, the OLC was not able to be computed in 15 of the available cases. While the Wusk and Gabler study used only full-scale crash tests, the use of EDR data indicates that OLC can be computed for real-world crashes. Practicality of Computation: The computations for OLC are more complex than those needed for the FSM, but the work of Wusk and Gabler (2017) suggests these computations are possible and can be applied to real-world crashes with complete EDR data available. 2.3.4 Crash Pulse + Actual Occupant Response Description and Assumptions: The metrics in this category seek to quantify the performance of the restraints or the performance of the vehicle structure and occupant restraints combined. Several metrics have been proposed in the published literature, including those listed in the bottom portion of Table 2-17. A brief description of each is provided in the following paragraphs. The ridedown efficiency, µ, is defined as follows (Huang et al. 1995): 2 2 1 max| o rd V e =µ where Vo is the initial velocity of the vehicle and erd is the vehicle ridedown energy density, defined as follows: ∫= vord dxxe  where ox represents the acceleration of the occupant (i.e., ATD) and xv is the displacement of the occupant compartment. This metric reflects the percentage of total kinetic energy absorbed by the vehicle structure and has been found to be closely related to vehicle dynamic crush (Huang et

38 al. 1995). A slight variant on ridedown efficiency has been proposed by Katoh and Nakahama (1982) where erd is computed over the interval from zero to the maximum vehicle deflection. The restraint quotient (RQ) was proposed by Viano and Arepally (1990) and is computed using the resultant acceleration (longitudinal and vertical directions only) of the occupant combined with the longitudinal deceleration of the vehicle occupant compartment. The RQ is typically computed for the thorax (RQc) using the following relation: max)( V c c x VRQ  = where Vc is the resultant velocity of the thorax with respect to the moving vehicle reference frame. This is computed by subtracting the respective velocity from that of the vehicle occupant compartment. max)( Vx is simply the maximum velocity change of the vehicle during the crash test. The restraint quotient varies from 0 to 1; where a value of 0 represents an occupant rigidly coupled to the vehicle interior and a value of 1 indicates an occupant attains the full velocity change of the vehicle prior to impacting the vehicle interior. Also suggested by Viano and Arepally (1990), the relative kinetic energy factor is simply a normalized measure of occupant kinetic energy (normalized by a velocity of 5 m/s). This metric (Ec) is computed based on the thorax accelerations using the following relation: 25 )max( 2c c VE = As in the restraint quotient, Vc, is the resultant velocities of the thorax with respect to the moving vehicle reference frame. Threshold Values and Biomechanical Basis: Similar to the VPI and OLC, all of the presented metrics simply provide a quantification of occupant restraint and/or vehicle structure performance but do not provide any threshold values. Evaluation of Link to Injury: There was a single study published relating these metrics to occupant injury in real-world crashes. Gabauer (2008) investigated three different models, each a different linear combination of delta-v and one or more of the restraint performance metrics just discussed, for the prediction of occupant injury in real-world crashes. The restraint performance metrics were computed using available full-width frontal crash test data. A total of 214 real-world crash cases were available with associated EDR data. The ability of the three different models to predict serious (MAIS 2+ and 3+) injury was compared to delta-v using ROC curve analysis. The available data suggested a small but not statistically significant improvement over the delta-v metric. The analysis, however, did not include the NASS weighting values. Two studies examined how these metrics relate to ATD-based occupant injury metrics. Park and Kan (2015) examined 60 U.S. NCAP tests from 2012 to determine relationships between vehicle-based metrics and resulting ATD-based injury metrics for drivers (H3 50th Male) and right front passengers (H3 5th Female). The ATD metrics included HIC15, peak chest acceleration, and peak chest deflection. Using linear regression analysis, none of the examined metrics had a strong correlation to any of the ATD-based metrics (R2 values of 0.34 or less). Among the three metrics

39 above, the best correlation was found to peak chest acceleration, primarily for the right front passenger (R2 values between 0.178 and 0.201). Gabauer and Gabler (2008c) examined 619 full-scale frontal crash tests to investigate whether delta-v can be modified with a measure of vehicle structure performance and occupant restraint performance to better predict occupant peak chest acceleration during a frontal crash. Multiple linear regression was used to correlate combinations of crash severity, vehicle structure performance, and occupant restraint performance descriptors to the maximum measured crash test dummy chest acceleration. The best combination of metrics was selected and then compared to a baseline model that used only delta-v to predict occupant chest kinematics. The combination of delta-v, ridedown efficiency, and the kinetic energy factor was found to provide the best prediction of the occupant chest acceleration. This combination accounted for approximately 4 times the variation in the maximum chest acceleration when compared to a model based solely on vehicle delta-v. Despite this improvement, the best model predicted only 42% of the variation in peak chest acceleration (compared to approximately 10% for delta-v alone). Practicality of Computation: The computations for all these metrics are more complex than those needed for the FSM and require matched full-scale crash test data. The work by Gabauer and Gabler (2008c) and Gabauer (2008) suggest, however, that large-scale computation of these metrics is possible. Occupant Compartment Intrusion and MASH Limits In the pre-MASH crash test criteria, the occupant compartment intrusion criterion was broad and relatively ambiguous with no specific numerical limits on occupant compartment deformation. NCHRP Report 230 procedures simply specify that the “integrity of the passenger compartment must be maintained with essentially no deformation or intrusion” (Michie 1981b). The subsequent NCHRP Report 350 (Ross et al. 1993) procedures relaxed this criterion to some extent by allowing occupant compartment deformations as long as deformation would not cause “serious injuries.” However, no definition of “serious injury” was provided, and the authors indicated that this factor “must be assessed in large part by the judgment of the test agency and the user agency, or both.” In the absence of specific numeric limits, the authors stressed the importance of documentation of the observed vehicle occupant compartment deformation, both using photographs and physical measurements (Ross et al. 1993). The Occupant Compartment Deformation Index (OCDI) was suggested as a means to uniformly quantify the deformation of the vehicle occupant compartment. The authors also indicated that injury risk due to deformation is dependent on the specific location, the extent of deformation, and the rate of deformation (Ross et al. 1993). While the deformation extent and specific location can readily be obtained by a post-test vehicle examination, the rate of deformation may not be possible to obtain depending on the instrumentation and/or camera placement locations available for the test. In contrast to the previous roadside crash testing protocols, MASH (AASHTO 2016) provides specific numeric static, post-test deformation limits for nine areas of the vehicle as summarized in Table 2-19. MASH commentary indicates that these were based on: (1) recommended guidelines developed by the Insurance Institute for Highway Safety (IIHS) to evaluate vehicle structural performance in offset frontal crash tests and (2) a Federal Highway Administration (FHWA) study that provided interim guidance on maximum acceptable occupant compartment intrusion limits.

40 The intrusion limits are nearly identical between the first (2009) and second (2016) editions of MASH, with the exception of the addition of the A- and B-pillar criteria in the second edition with no additional commentary provided in the appendix of the second edition of MASH related to the addition of this criterion. While the MASH commentary still references “serious injury,” no specific definition or corresponding AIS injury level is specified. Table 2-19. Summary of MASH occupant compartment deformation limits (AASHTO 2016). Vehicle Component or Area Static Post-Test Deformation Limit / Criteria Windshield ≤ 3 inches and no tear of plastic liner Roof ≤ 4 inches A and B pillars ≤ 5 inches of resultant deformation (≤ 3 inches laterally). No complete severing of support member. Wheel/foot well and toe pan ≤ 9 inches Front side door area (above seat) Side front panel (forward of A pillar) ≤ 12 inches Front side door area (below seat) Floor pan and transmission tunnel areas Window No shattering of side window resulting from direct contact with a structural member of the test article, except for tall, continuous barrier elements. For laminated side windows, windshield guidelines apply. Although not specifically cited, the FHWA study referenced by MASH is presumably the study conducted by Eigen and Glassbrenner (2003). The authors examined 10 years of NASS/CDS data, 1991 through 2001, to investigate the relationship between occupant compartment intrusion levels and corresponding occupant injury. The primary focus was to evaluate the current (at the time) FHWA intrusion limit guideline for roadside hardware crash tests (i.e., less than 15 cm occupant compartment intrusion). The study included non-rollover crashes with intrusion in one or more of the following vehicle areas: (1) the toe pan, (2) the floor pan, and (3) forward of the A-pillar. No restriction was placed on the object struck as sufficient cases of roadside hardware only impacts were not available; the cases were classified, however, as either striking a vehicle or striking a fixed object. Also, the study only included drivers, right front passengers (with intrusion present near the corresponding seat location), and occupants 13 years or older. Based on this selection criteria, there were approximately 1,625 relevant vehicles involved with relevant intrusions, representing approximately 350,000 vehicles after the applicable weights were applied. The primary method of analysis was chi-square tests to test for significance for various intrusion levels. Eigen and Glassbrenner examined both AIS 2+ and AIS 3+ injury levels as there were relatively few higher severity injuries present in the available data (i.e., AIS ≥ 4) and a significant number of lower limb injuries (i.e., these injuries generally have lower mortality/AIS scores but can produce long-term disability). A primary finding of the study was that moderate to maximum occupant injury does occur at less than 15 cm of occupant compartment intrusion. Even without controlling for other potentially confounding factors, such as seat belt use and age, a statistically significant relationship was found between non-minor occupant injuries and intrusions between 8 and 15 cm. Limitations of the study (primarily a function of small sample sizes in the data) include the inability to control for confounding factors, examining all fixed objects in a single category, and no control for specific vehicle types. Recommendations for future study included examining additional years of NASS/CDS, examining available anecdotal crash cases, and controlling for vehicle body types.

41 The literature review revealed several previously published studies that investigated how vehicle occupant compartment intrusion relates to occupant injury, as summarized in Table 2-20. Table 2-20. Summary of previously published studies related to occupant compartment intrusion. Source Data Source [Cases] Crash Type(s) Notes Kim et al. (2017) Hospital-based study [344 patients] Frontal crashes Logistic regression to predict severe injury using binary (yes/no) intrusion and Collision Deformation Code (CDC) - based deformation extent (DE). DE ≥ 4 / intrusion found to increase risk of injury by 2.4 / 5.2 times. Isenberg et al. (2011) Hospital-based study [608 patients] All crashes Primary purpose was to evaluate intrusion as a criterion to determine if admitted patients will use trauma center resources. Evans et al. (2009) Hospital-based study [808 patients] All crashes Children occupants only (0-15 years). Intrusion found to increase serious (AIS2+, 3+) injury by 2.9% and 4.0%, respectively. Conroy et al. (2008) CIREN [794 drivers] Head-on / wide or narrow Logistic regression model to predict severe injury using binary (yes/no) intrusion indicator only. Intrusion present approximately doubled injury risk. Stefanopoulos et al. (2003) Hospital-based study [48 vehicle crashes] Head-on, no ejection Only 11 cases with intrusion. Extent of intrusion and restraint use more important than involved vehicle component. None of the existing studies examined intrusion for the vehicle specific areas noted in MASH. Only a single study, Stefanopoulos et al. (2003), classified intrusion in different areas, which included only steering wheel, windshield, and control panel. The vast majority of the studies also treated intrusion as a binary variable (i.e., present or absent). Furthermore, the majority of the studies focused on frontal crashes and had relatively small sample sizes not necessarily selected in a statistically random method. Injury Characterization Metrics While the AIS has been the predominant scale used to quantify crash injury level, the AIS is based only on threat to life (AAAM 2008). Typically, the MAIS score has been used to quantify injury. The MAIS metric quantifies the highest severity injury to an occupant in a specific body region but will not distinguish between individuals with multiple injuries at a specific AIS (or lower) level and a single injury at the same AIS level. Other metrics exist, however, that consider multiple body regions and/or quantify other aspects of injury, such as post-injury functioning of the individual or post-injury disability level. The following section summarizes these available injury characterization metrics. 2.5.1 Injury Severity Score (ISS) Initially proposed by Baker et al. (1974), the ISS is the sum of the squared AIS scores from the three body regions with the most severe injury. ISS considers six distinct body regions: (1) head or neck, including cervical spine, (2) face, (3) chest, (4) abdomen/pelvic contents, (5) extremities or pelvic girdle, and (6) external. ISS scores range from 1 to 75 (i.e., AIS injury levels of 1 to 5 in one to three body regions). People with one or more AIS 6 level injuries are automatically assigned

42 the maximum ISS value (i.e., 75). For blunt injury, this metric has been found to be related to percent mortality (Baker and O’Neill 1976; Bull 1975). 2.5.2 Injury Impairment Scale (IIS) States and Viano (1990) proposed the IIS scale as a method to quantify loss of function following injury healing. Similar to AIS, the IIS rates impairment on a 0 to 6 scale, and the maximum IIS score is suggested as a measure of whole-body impairment. Criteria considered in the rating include mobility/dexterity, cognitive/physical, cosmetic/disfigurement, sensory, pain, and sexual/reproduction. Published studies evaluating the IIS suggested there is only limited association between disability and IIS (Nhac-Vu et al. 2012; Waller et al. 1995). 2.5.3 Functional Capacity Index (FCI) MacKenzie et al. (1994, 1996) developed the FCI, which maps AIS injury descriptions into scores that reflect the expected reduction in functional capacity of a person 1 year post-injury. The rating scale varies from 0 to 100, with 100 representing no deficit in a person’s everyday functioning. A three-step process was used for the development including: (1) identifying 10 relevant dimensions of function with corresponding capacity levels, (2) determining the relative severity of different levels of function in terms of daily living impact, and (3) using clinical experts to assign scores to injury descriptions based on the likely 1-year consequences. The original FCI was developed using AIS 90, and a revised version of FCI was developed using AIS 2005 (Gotschall 2005). A further revised FCI using AIS 2008 is available but has not been validated (Palmer et al. 2017). Although the initial investigation of this metric with lower extremity injuries suggested it is a modest predictor of function after 1 year (Gotschall 2005; MacKenzie et al. 1994), a review of the validation and use of the FCI (Palmer et al. 2017) suggested relatively poor performance of the metric and the need for additional validation. Palmer et al. (2017) further suggested that FCI “must be more rigorously evaluated before more comprehensive predictive tools can be developed from it.” 2.5.4 Measuring Societal Cost with Harm The Harm metric is a means of measuring the societal cost of traffic crashes. The Harm metric is frequently used in the evaluation of impact injury countermeasures. The Harm metric was first developed by Malliaris et al. (1982, 1985) as a means of balancing number of injuries with the severity or cost of an injury. Severity is measured using the AIS. Using the Malliaris Harm metric, each AIS level has a prescribed societal cost. This societal cost includes both medical costs and indirect costs, such as loss of wages. For each injured person, the Harm is the societal cost which corresponds to their MAIS injury level. This original Harm metric had two limitations. First, societal cost is not a function exclusively of AIS level. The societal cost of injury varies by body region as well as by injury severity. For example, an AIS 3 head injury has a higher societal cost than an AIS 3 leg injury. Second, the original Harm metric assigned a cost to only the injury of highest severity. This approach can underestimate the total societal cost of a person who suffers multiple injuries, as multiple injuries can aggravate the total threat to a crash victim’s life.

43 Fildes et al. (1992, 1994) developed an improved Harm metric that addressed these two issues. The improved method assigns a societal cost to each injury and sums these costs to estimate a total societal cost of injury. ∑ = = sNumInjurie i i AISbodyregionCostHarm 1 ),( In the above equation, Costi, the societal cost of an injury i as defined by Fildes et al. (1992), was used as a measure of societal cost. Costi is a function of the injury severity as measured by the AIS and the body region that has been injured. The cost components include not only treatment and rehabilitation costs but also all other costs to society such as loss of wages and productivity, medical and emergency service infrastructure costs, legal and insurance costs, family and associated losses, and allowances for pain and suffering. Gabler et al. (2005) developed a variation of the Fildes method for computation of Harm. In some cases, there may be multiple injuries to a single body region. In this methodology, the maximum injury to a single body region is used when assigning costs, as costs are typically assigned to treat a single body region, not individual injuries of that body region. The costs used for the Fildes Harm metric were normalized to cost of a fatality and are presented in Table 2-21. Table 2-21. Average cost per injury (normalized to the cost of a fatal injury). INJURY SEVERITY BODY Minor Moderate Serious Severe Critical Maximum Unknown REGION (AIS = 1) (AIS = 2) (AIS =3) (AIS = 4) (AIS = 5) (AIS = 6) External 0.0045 0.0250 0.0698 0.1135 0.1646 1.0000 0.0045 Head 0.0063 0.0295 0.1213 0.2796 0.9877 1.0000 0.0045 Face 0.0063 0.0295 0.1213 0.1601 0.3277 1.0000 0.0045 Neck 0.0063 0.0295 0.1213 0.1601 0.3277 1.0000 0.0045 Chest 0.0045 0.0250 0.0698 0.1135 0.1646 1.0000 0.0045 Abdomen 0.0045 0.0250 0.0698 0.1135 0.1646 1.0000 0.0045 Pelvis 0.0045 0.0250 0.0698 0.1135 0.1646 1.0000 0.0045 Spine 0.0045 0.0250 0.1631 1.4054 1.6804 1.0000 0.0045 Upper Extremity 0.0063 0.0433 0.1026 - - - 0.0045 Lower Extremity 0.0045 0.0433 0.1303 0.1926 0.3277 - 0.0045 2.5.5 Multi-Harm Approach Mallory et al. (2017) proposed a multi-Harm approach to analyzing crash and injury data. Essentially, the approach uses several of the existing metrics including the traditional MAIS metric, the FCI metric, injury cost (i.e., Harm), and attributable fatality. A framework was presented using an 11-year retrospective analysis of NASS/CDS data. The results suggested that while MAIS 3+ can be used as a single measure of Harm, an analysis based on multiple measures provide a more detailed account of risk for a particular injury or set of crash conditions.

44 Potential Data Sources for Evaluating Roadside Crash Injury Metrics The following sections provide details on available datasets that could potentially be used to assess both existing and potential roadside hardware crash injury metrics. Table 2-22 provides a high-level summary of the available datasets. This includes an indication of whether the dataset is nationally representative and if it has been used in previously published studies to evaluate one or more vehicle-based crash injury metrics. Table 2-22. Summary of potential data sources to evaluate roadside crash injury metrics. Data Source Nationally Representative? Used in Previous Studies to Evaluate Vehicle-Based Metrics? Notes NASS/CDS Y Y 2015 is last year of NASS/CDS data collection. Replaced by CISS. CISS Y N 2017 onward. 2017 – 2019 is available (~7500 cases total) VT EDR Y Y Linked to NASS/CDS or CISS. 11,000+ cases available NCHRP 17-43 Y N Roadway departure crashes only. Linked to NASS/CDS. CIREN N N Detailed injury and injury causation data. ~300 cases/year SCI N N Special crash circumstances or vehicle technologies. ~150 cases/year NTDB Y N Detailed injury/hospital data but limited crash data. 2.6.1 NASS/CDS NASS/CDS provides details for 4,000 to 5,000 tow-away crashes investigated each year by NHTSA at 24 locations throughout the United States. NASS/CDS includes only passenger cars and light trucks involved in crashes of severity sufficient to require that at least one vehicle be towed from the scene. NASS/CDS is an appealing data source because of its extensive descriptions of crash severity (delta-v), occupant injury, vehicle deformation, and restraint usage. NASS/CDS also is a nationally representative sample of the U.S. crash population such that the available cases can be weighted to represent all U.S. crashes. The vast majority of the previously discussed real- world crash studies linking existing vehicle-based crash metrics to occupant injury have utilized the NASS/CDS dataset. 2.6.2 NHTSA Crash Investigation Sampling System (CISS) CISS is the in-depth follow-on crash investigation system to NASS/CDS (NHTSA 2020). The last year of NASS/CDS data collected by NHTSA was 2015. CISS has altered the locations of the primary sampling units so that the dataset better reflects U.S. population shifts over the last two decades. However, the database follows much the same structure as NASS/CDS. The first year of CISS data collection that was released in full to the public was CISS 2017. Currently there are approximately 7,500 cases available in total across CISS 2017, 2018, and 2019 (NHTSA 2021a).

45 2.6.3 Virginia Tech (VT) EDR Database EDRs are devices that record vehicle crash characteristics, such as vehicle delta-v. Virginia Tech has developed a unique dataset of over 11,000 EDR downloads that match with existing NASS/CDS, CISS, Special Crash Investigations (SCI), and Crash Injury Research and Engineering Network (CIREN) cases. An EDR in a vehicle can provide a comprehensive snapshot of the entire crash event: pre-crash, crash, and post-crash events including vehicle speed, brake status, and engine throttle position. As evidenced by the previously discussed studies of both existing and potential vehicle-based injury metrics, EDR data can feasibly be used to compute vehicle-based metrics for real-world crashes, allowing a linkage between these metrics and real- world crash occupant injury. 2.6.4 NCHRP 17-43 Road Departure Dataset As part of NCHRP Project 17-43, Virginia Tech is developing a database of roadway departure crashes. Currently, the NCHRP Project 17-43 database comprises 821 road departures extracted from NASS/CDS, with full reconstructions and trajectories (Riexinger and Gabler 2020). The dataset comprises NASS/CDS cases from 2011 to 2015. In addition to the detailed vehicle deformation, occupant injury, and restraint information from NASS/CDS, the NCHRP Project 17- 43 database includes more detailed roadway, roadside, and roadside hardware information. These supplementary collected data include specific roadside device type and the performance of the device. 2.6.5 CIREN CIREN is an in-depth crash investigation system sponsored by NHTSA and select automakers. CIREN investigates approximately 200 cases each year, which are collected by five Level 1 trauma centers. Multi-disciplinary CIREN teams combine the expertise of trauma physicians, crash investigators, biomechanics and automotive engineers, and emergency medical personnel to determine the detailed injury mechanisms associated with each crash. CIREN cases provide extraordinary detail on the injuries suffered by each crash victim. However, CIREN is a clinical sample of people who enter a Level 1 trauma center and, unlike NASS/CDS, is not nationally representative of the U.S. population. 2.6.6 NHTSA SCI The NHTSA SCI provides NHTSA with detailed crash data collected to study new and rapidly changing vehicle technologies or special crash circumstances. Approximately 100 SCI cases are collected annually. Case selection is based on U.S. DOT priorities and may include investigation of urgent crash safety issues and new emerging technologies. Of particular importance to this project are SCI that were performed as part of the “FHWA Pilot In-Service Performance Evaluation of Guardrail End Treatments” project (FHWA 2018). Data collected on SCI cases may include police reports, EDR downloads, insurance crash reports, and field investigations from professional crash investigation teams. SCI crash investigations provide incredible detail on these target cases, including comprehensive descriptions of vehicle, scene, occupants, safety systems, injuries, and injury mechanisms. Similar to CIREN cases, however, the SCI cases are not nationally representative.

46 2.6.7 National Trauma Data Bank (NTDB) The NTDB is the largest aggregation of trauma registry data ever assembled (American College of Surgeons 2007). It is supported by the American College of Surgeons and provides information about patients, injuries, and treatments. NTDB collects trauma registry data from participating trauma centers on an annual basis. Patient demographic, injury severity, and injury origin data are collected along with descriptive accounts of each traumatic incident. One limitation of the trauma registries that compose the NTDB is that few crash details are available aside from the International Classification of Diseases (ICD) external cause of injury code (e-code) indicating that a crash was the mechanism of injury. While this dataset will be investigated, it does not provide sufficient information for the evaluation of roadside crash injury metrics, especially the needed vehicle dynamics information required to compute the roadside crash injury metric values. Conclusions Based on a focused review of the available literature, both U.S. and international, related to existing and potential roadside hardware crash injury metrics, several conclusions can be drawn about each of the five focus areas of this literature review. These conclusions are summarized below. 2.7.1 Existing Roadside Hardware Crash Injury Metrics • There are four current metrics used to evaluate occupant injury risk in crash tests with roadside safety hardware: o OIV o Occupant RA o THIV o ASI The OIV and RA are used in MASH while the THIV and ASI metrics are used in international roadside hardware crash test procedures. The OIV and THIV computations differ primarily in the consideration of lateral and longitudinal directions; THIV considers the resultant value, while the OIV considers the lateral and longitudinal values separately. The RA is an acceleration-based metric used in MASH, while the ASI is an acceleration- based metric used in international roadside hardware crash test procedures. • The PHD metric, the international counterpart to the RA metric, is no longer used for international roadside hardware crash tests. The rationale for exclusion of this metric was cited as empirical evidence and expert opinion that the PHD is an unreliable metric. • For all the metrics, there have been very few changes to the established threshold values and little additional biomechanical information cited in support of the thresholds. The existing MASH occupant risk threshold values were originally intended to differentiate serious (AIS 3) from severe (AIS 4+) occupant injury.

47 • Of all the existing metrics, OIV and ASI have been studied most extensively. The majority of the existing studies either relate the metric values to real-world injury risk using EDRs or relate the metric values to ATD-based injury risk using crash test data. • The vast majority of the studies examining the existing metrics focus on injury risk in the longitudinal direction. Very little research relates the existing metrics to injury risk in other crash modes (i.e., side and oblique impacts). • The most recent and comprehensive EDR-based study in the literature found that OIV offers a statistically significant improvement over delta-v for predicting serious thorax/abdomen/spine injury to belted occupants in frontal crashes. For unbelted occupants, no statistically significant difference was found between OIV, ASI, and the simpler delta-v metric for predicting serious thorax/abdomen/spine injury. • Based on the studies in the literature using full-scale crash test data, the existing roadside hardware crash injury metrics appear to have a relatively weak relationship to ATD-based injury metrics, explaining 30% or less of the observed ATD-based metric variation. In general, OIV was found to have strongest correlation, followed by ASI and RA, respectively. • Although there are small differences in computational complexity between the existing metrics, all can be computed relatively easily with standard computer-based methods. Also, a standard computer program, called TRAP, does exist to compute these metrics. • All the existing metrics presume that the occupant compartment remains essentially intact with no significant reduction in front seat occupant space available. 2.7.2 Potential Vehicle-Based Crash Injury Metrics • Several vehicle-based metrics exist in addition to the existing roadside hardware crash injury metrics. These can be classified into three categories: o Crash pulse only o Crash pulse + assumed occupant response o Crash pulse + actual occupant response The crash pulse only metrics included delta-v as well as the maximum moving average acceleration, which serves as the basis for the RA and ASI metrics. The crash pulse with assumed occupant response metrics included the VPI and OLC metrics. The crash pulse with actual occupant response included the ridedown efficiency, RQ, and relative kinetic energy factor metrics. • All but the crash pulse only metrics account for the occupant being restrained via a seat belt and/or airbag. While the VPI and OLC metrics can be computed using generalized values, the ridedown efficiency, RQ, and relative kinetic energy factors all require corresponding full-scale vehicle crash test data to complete the computation.

48 • The vast majority of the identified potential metrics do not have existing threshold values. For the crash pulse only and crash pulse + assumed occupant response metric categories, however, the metrics are similar to the existing roadside hardware crash injury metrics in that higher values are intended to correspond with higher occupant injury risk. • Compared to the OIV and ASI, there is relatively little research on the majority of the identified potential vehicle-based metrics. Most of the studies involving the potential vehicle-based metrics used full-scale crash test data to relate the metric to occupant injury risk. Similar to the existing roadside hardware injury metrics, the majority of the studies focused on the longitudinal crash direction. • The most recent and comprehensive EDR-based study in the literature suggested that the vehicle-specific VPI metric performs at least as well as OIV and ASI in predicting serious thorax/abdomen/spine injury to belted occupants in frontal crashes. There were no existing studies available linking OLC to real-world occupant injury. • With the exception of the crash pulse only metrics, the potential vehicle-based injury metrics identified are slightly more complex to compute compared to the existing roadside hardware crash injury metrics. All the potential metrics, however, can be computed relatively easily with standard computer-based methods. 2.7.3 Occupant Compartment Intrusion and MASH Limits • Current MASH vehicle occupant compartment intrusion guidelines specify limiting values for nine different areas of the vehicle. The current guidelines were based on general guidance on intrusion from IIHS and a single FHWA study using real-world crash data. • The FHWA study used as a basis for the current MASH intrusion guidelines only examined a limited number of the MASH-specified vehicle regions. Although there are other previously published studies examining intrusion and subsequent occupant injury, none were sufficiently detailed to evaluate how the current MASH intrusion limits relate to occupant injury. 2.7.4 Supplemental Injury Characterization Metrics • At least four metrics exist as an alternative to the AIS to quantify crash injury level: o ISS o IIS o FCI o Harm ISS uses AIS but considers multiple injuries. The remaining metrics consider multiple injuries as well as the longer-term consequences of injury (i.e., disability). • Although the ISS considers multiple injuries, there are relatively small differences between MAIS and ISS. As an example, for a threshold of MAIS 3+, the minimum ISS score would be 9. All occupants with at least one AIS 3 injury would have an ISS of at least 9.

49 Considering ISS 9+, all the MAIS 3+ occupants would be included as well as any occupant with two AIS 2 injuries and one AIS 1 injury, or any occupant with three AIS 2 injuries. • The existing published studies suggested relatively poor performance of the IIS and FCI metrics. These two metrics represent two of the three identified metrics that also account for the longer-term consequences of crash injury. 2.7.5 Available Datasets to Assess Roadside Hardware Crash Injury Metrics • A total of seven data sources were identified that could potentially be used to assess both existing and potential roadside hardware crash injury metrics: o NASS/CDS o CISS o VT EDR Database o NCHRP 17-43 Road Departure Dataset o CIREN o NHTSA SCI o NTDB • Evaluation of the roadside crash injury metrics requires both a method to compute the metrics and detailed occupant injury information. EDR data are needed to compute the roadside crash injury metrics for real-world crashes, and in-depth crash data are required to provide AIS injury scores. Of the available data sources, NASS/CDS and CISS have this information available for a subset of cases and provide a nationally representative sample once the associated case weighting values are applied. Subsets of CIREN and SCI also have the needed EDR and injury information but are not nationally representative. • The VT EDR database contains EDR data for a subset of cases within the identified in- depth databases, such as NASS/CDS, CISS, CIREN, and SCI. The VT EDR database must be combined with the corresponding in-depth database injury information to evaluate the roadside crash injury metrics. • The NCHRP 17-43 Road Departure Dataset is a subset of NASS/CDS. By using NASS/CDS as a data source, any suitable road departure cases in the NCHRP 17-43 dataset would be included. • Although the NTDB contains detailed injury information for crash-involved people, there are no associated vehicle kinematics data available to compute any of the roadside crash injury metrics. This source cannot be used directly to evaluate the roadside crash injury metrics but could provide nationally representative injury information for people involved in single vehicle crashes. Gaps and Research Needs This literature review has identified a number of gaps in the literature and associated research needs related to roadside hardware crash injury metrics. The following is a summary listing:

50 • There is only limited information available relating vehicle-based metrics to real-world crash injury or ATD-based injury metrics for the side and, especially, the oblique crash mode. • The current MASH occupant injury risk criteria do not include any method for considering other crash and/or occupant factors that may influence injury risk, such as belt usage or occupant age. The relative magnitude of these potentially confounding factors is not well known relative to the current MASH occupant injury risk criteria/thresholds. • International roadside hardware crash test procedures no longer use a metric similar to the RA. The implications of excluding the RA from the MASH crash test procedures is not known. • There is currently no information related to how the current MASH occupant compartment intrusion limits relate to real-world occupant injury. • Nearly all of the ATD-based injury metrics have associated injury risk curves that transform injury metric values into an associated probability of occupant injury. With respect to the roadside hardware crash injury metrics, there is a general lack of these injury risk curves. • Roadside safety hardware crash tests are designed to represent worst case conditions and are assessed on a binary pass/fail scale. It remains unknown whether linking these criteria to AIS injury predictions to use in MASH test evaluations would translate directly to real- world crash scenarios.