COMBINING CRASH RECORDER AND PAIRED COMPARISON TECHNIQUE: INJURY RISK FUNCTIONS IN FRONTAL AND REAR IMPACTS WITH SPECIAL REFERENCE TO NECK INJURIES

Transcription

1 COMBINING CRASH RECORDER AND AIRED COMARISON TECHNIQUE: INJURY RISK FUNCTIONS IN FRONTAL AND REAR IMACTS WITH SECIAL REFERENCE TO NECK INJURIES Anders Kullgren, Maria Krafft Folksa Research, 66 Stockhol, Sweden Claes Tingvall Swedish National Road Adinistration, 7887 Borlänge, Sweden, Monash University Accident Research Centre, Clayton, VIC, 368, Australia Anders Lie Swedish National Road Adinistration, 7887 Borlänge, Sweden aper No. 44 ABSTRACT Knowledge fro real-world crashes is iportant in the design of a crashworthy road transportation syste. Such design ust be based on the huan injury tolerance liits. Links between ipact severity and injury outcoe are iportant and could be used in order to achieve such tolerance liits. Traditionally ipact severity has been calculated with retrospective reconstruction technique, although recently, injury risk functions have been presented where ipact severity has been easured with crash pulse recorders. The ais of this paper were to present injury risk functions, with special reference to neck injuries, calculated with crash recorder and paired coparison technique, and to propose a way of cobining the two ethods. By cobining coprehensive statistical aterial with in depth crash recorder inforation, injury risk functions for injuries to different body regions were established. functions for AIS neck injuries both in frontal and rear-end ipacts have also been established. It was found that the data fro the crash pulse recorder generated risk functions could be used to validate and calibrate risk functions based on the atched-paired technique. Moreover, it was found that the shape of the injury risk curves differed significantly for injuries to different body regions. It was also found that the neck injury risk differed significantly for frontal and rear-end ipacts. It is concluded, that adding new techniques to the existing techniques based on reconstruction can further refine generating risk functions. The injury risks found are iportant for the understanding of injury tolerance liits for injuries to different body regions, but also for the understanding of injury echaniss for different injury types. INTRODUCTION In the construction of a crashworthy road transportation syste, knowledge fro real-world collisions describing tolerance liits for occupants as well as knowledge of how well a vehicle can protect its occupants are fundaental. An essential issue fro that perspective is injury probability functions versus crash severity or injury risk functions as they often ay be called. functions can be used to find threshold values for axiu echanical force with injury occurring, or for finding any threshold level. functions can also be used to validate injury criteria, especially looking at the elasticity of the criteria. This is done by estiating the rate at which an injury (or an injury criterion) will increase with increased ipact severity (or echanical force). An experiental test should be at least as sensitive to echanical force as real-life analyses show. If the risk of injury increases with say %, with a 5% increase in echanical force, this should be reflected in the experient. functions can be calculated directly by studying the ratio in nuber of injured and uninjured at different crash severity levels. Several studies of injury risk functions have been presented by for exaple Norin (995) and Evans (994). In those studies the crash severity, ost often change of velocity, was estiated by using crash reconstruction techniques. More recent studies, where crash severity have been easured with onboard crash pulse recorders, have been presented by for exaple Kullgren (998), Kullgren et al (999), and Krafft et al. (22). Crash recorders ight have the possibility to increase the quality of the estiates of ipact severity, which has been shown to have an iportant effect on the estiates of risk functions (Kullgren 998). Another way of calculating risk functions fro real-life crashes has been proposed by Krafft et al (2). In that study, the injury risk functions were calculated with a statistical ethod based on the paired coparison technique (Hägg et al 99). By

2 directly coparing the injury outcoe in two-car collisions, where the cars were categorised in ass intervals, a easure of relative injury risk versus a relative easure of change of velocity could be calculated (Krafft et al 2). The huan tolerance to echanical force ay be estiated by cadaver crash tests. This could be done especially to establish the huan tolerance to fractures. To establish tolerance levels for soft tissue injuries as for exaple AIS neck injuries, volunteer tests could be done. However, for ethical reasons it is only possible to run tests below injury tolerance levels. Results fro cadaver tests and volunteer tests could be used to design and validate duies and coputer siulation odels. The ai of this paper is to present injury risk functions calculated with two different ethods and to propose how these ethods can be cobined. The ethods used are a direct easure of injury risk using on-board crash pulse recorder data and a relative easure of injury risk using a statistical ethod based on the paired coparison technique. An additional ai is to present injury risk functions for injuries to different body regions, and especially AIS neck injury risk functions in both frontal and rear ipacts. METHODS Crash ulse Recorders Ipact severity was easured with a Crash ulse Recorder, CR, which easured the acceleration tie history in the ipact phase in one direction. The CR is based on a spring ass syste where the oveents of the ass in an ipact are easured. The displaceent of the ass is registered on a photographic fil. The circuit has its own power cell and does not need an external power unit. The CR has a trigger level of approxiately 3g. When the characteristic paraeters for each CR have been easured, such as spring coefficient and frictional drag, and with knowledge of the displaceent tie history, the acceleration tie history were calculated. The change of velocity was then calculated fro the acceleration tie history. The crash pulses were filtered at approxiately Hz. The CR and the analysis of the recordings fro the CR have been described by Aldan et al. (99) and Kullgren (998). The standard deviation of the easureent of the CR has been evaluated and estiated to be approxiately 5% (Kullgren 998). The ipact severity easureents were divided into intervals, and the injury risk was calculated in each interval. Sooth curve fits were used in the plots of injury risks. aired Coparisons The basis for this statistical ethod is the paired coparison technique, where two car accidents are used to create relative risks. The ethod was initially developed by Evans (986), but has been developed further for car to car collisions by Hägg et. al. (992). The assuption for the ethod is that the risk of injury is a continuous function of change of velocity. This assuption ight conflict with safety features such as airbags that ight generate a step-function. Another assuption is that injuries in one car are independent fro the injuries in the other car, given a certain crash severity. For a given change of velocity the risk of an injury is p and p 2 in the two cars, respectively. Suing over all change of velocities, the outcoe will be as presented in Table. Basically, the change of velocity can be calculated fro the law of the conservation of oentu, where: Delta v V rel (M 2 /(M +M 2 )), where V rel is the relative velocity and M and M 2 the asses of the two vehicles colliding. This relation is true even if the two vehicles involved do not have a coon velocity after the ipact. If the asses are equal, both vehicles will undergo the sae change of velocity. This ethod uses this fact, and that any deviation in ass can be transferred to differences in change of velocity, as long as the individual asses are known, see Figure. The ethod cannot generate absolute figures, only risks relative to each other. Instead of generating new risk functions, the ethod uses the change on the exposure distributions and the resulting change in risk. nuber of ipacts equal ass f(s)f(s) 2 unequal ass f(s) unequal ass f(s) 2 ipact severity Figure. Ipact severity (delta-v) for cars in atching crashes for equal ass: f (s) f 2 (s) and unequal ass: f (s) f 2 (s) where car is of less ass than car 2.

3 Table. Sus of probability of injury to driver in car and 2 for all segents of ipact severity. driver injured Driver of Car 2 driver not injured Total Driver of Car driver injured driver not injured i i n i i 2i x n i (- i ) 2i x 3 i i n i i (- 2i )x 2 n i (- i )(- 2i )x 4 i n i i 2i +n i i (- 2i )n Total i n i i 2i +n i (- i ) 2i n 2 The relative risk of an injury, for vehicle to 2, given a certain change of velocity distribution is therefore: R (x + x 2 ) / (x + x 3 nii ) ni2i ni ni i i 2i 2i + + n i i ni (- i) (- 2i ) 2i The ethod is unbiased for any cobination where the vehicles are of the sae weight; i.e. the ass ratio is. If the vehicles are of different weights, the two vehicles will undergo different changes of velocity, which will have to be copensated for. Generally, we can introduce any coponent, K, that will affect the risk of injury in either, or both of the vehicles. If we let K denote this factor in vehicle, and K 2 in vehicle 2, this will lead to: (Eq. ) n i i 2i K K 2 /n i 2i K n i i 2i K K 2 /n i 2i K 2 K i n i i 2i / i i n i i 2i K / n i 2i i n i 2i To solve the equation, cars of different weights will be used, where the weights are known. K will therefore denote the role of change of velocity, and could be a constant, or a function of, say, change of velocity. (Eq. ) is estiated by K (X /(X +X 3 )) (2) and, ( X ) /( X + X 3) b K ( X /( X + X 3) ) a (3) where, a and b are ass relations in the atched pairs. These ass relations are transfored to relative change of velocity by b a / 2 2. ( + 2 ) b ( + 2) a The analytical functions chosen to describe the risk functions have been applied siply using either a linear function or a power function. This issue would have to be further investigated using ore advanced aterial. Cobining crash pulse recorder and paired coparison technique While the iportance of a arginal change of velocity as well as parts of the risk function will be calculated using paired coparison technique, absolute values cannot be given with this ethod. Since the actual change of velocity in each crash not is known, only a relative change of velocity for each segent can be calculated. If absolute values are to be given, a key value ust be brought into the equation. Such key values can be estiated by coparing the relative risk functions, derived fro statistical data, with the absolute easures of injury risk calculated with data fro crash pulse recorders. Boththerelativeriskandtherelativechangeof velocity ust then be related to the absolute values. By coparing the average change of velocity of the crashes using the crash recorder data, with a relative change of velocity of for the ass data, a key factor for crash severity can be established. By coparing the shape of relative injury risks with absolute injury risks for the sae injury type, a key factor for injury risk can also be established. The relative risk functions can by that be transfored fro relative to absolute risks.

4 MATERIAL Crash pulse recorder data Since 992, CRs have been installed in approxiately 7, vehicles aied at easuring frontal ipacts and approxiately 5, vehicles aied at easuring rear-end ipacts. Regarding frontal ipacts crash pulse recorders have been installed in 4 different car akes and 22 odels and in rear-end ipacts in 7 car odels of the sae ake. The car fleet has been onitored since 992, and regarding frontal ipacts, accidents with a repair cost exceeding 5 US$ have been reported via a daage warranty insurance. Rear-end ipacts were reported irrespectively of repair cost. The accident data collection syste has previously been described by Karén et al. (99). This study includes ipact severity and driver injury data in 286 frontal ipacts with an overlap of ore than 25%. Eighty-three rear ipacts were analysed with known ipact severity and with injury data fro front seat occupants. In frontal ipacts only restrained drivers were included, as the neck injury risks ay differ between front seat passengers and drivers. There was not enough data to calculate the neck injury risk for the front seat passengers separately. Regarding rear-end ipacts both drivers and front seat passengers were included as the neck injury risk in rear ipacts could be regarded as siilar for both positions. In the frontal ipacts, belt use was verified fro inspections of the seat belt syste and in the rear ipacts, belt use was verified fro questionnaires to the involved occupants. Approxiately 4% of the frontal ipacts and none of the rear ipacts were rejected due to lack of belt use. In the frontal ipacts, 72% of the drivers were ale and 28% were feale. Regarding the rear ipacts, 47% of the front seat occupants were ales and 53% were feales. Data for the paired coparison risk functions The aterial used was two car crashes, front to front and front to side, fro Queensland, Australia, as well as rear-end crashes fro Sweden for the analysis of neck injuries. In both aterials, injuries were classified as to bodily localisation and severity in three classes, naely inor, serious and fatal injuries. The reason for using data fro two sources is that very few data sets would allow the kind of analysis ade here. Both aterial sources consisted of police reports, known to have soe probles with quality. While using only a few variables fro the police records, the ain quality issue lies with the under-reporting, and weak injury classification due to the lack of in-depth edical data. RESULTS Fro the first two figures it can be seen that the shape of the upper part of the injury risk for all injuries in Figure is equal to the relative injury risk presented in Figure 2. The average change of velocity for the collisions in Figure was 2 k/h. The average change of velocity should be siilar as the relative change of velocity of in Figure 2. This eans that the scaling for the x-axis in Figure 2 could be estiated and changed fro a relative scale to a fix scale. The slope of the injury risk for all injuries in Figure should then be siilar as the risk of any injury in Figure 2.,4 All injuries (n72/286),2 MAIS2+ injuries (n42/286) Change of velocity (k/h) Figure. Injury risk, all injuries and MAIS2+ injuries, versus change of velocity fro crash pulse recorder data. Relative injury risk,4,2 Frontal ipacts, all injuries,4 5,9,95,5,,5,2 Relative change of velocity Figure 2. Relative injury risk versus relative change of velocity fro police reported crashes in Australia. A clear difference in risk of an AIS neck injury in rear ipacts and of any injury in frontal ipacts was found in the crash recorder data, see Figure 3. Also in the police reported crashes a clear difference was found, see Figure 4. The risk of a neck injury in the struck car increased rapidly with increased crash severity while the risk of any injury in the striking car was always lower and had a lower slope. In Figure 5, risk functions with both ethods have been cobined. The relative injury risk was adjusted so that the relative injury risk for the striking car, at the relative change of velocity of, was equal to the absolute injury risk for all injuries at the average change of velocity (.9

5 k/h) in the crash recorder crashes. The slopes of the risk curves were siilar for both ethods.,4,2 Rear ipacts, neck injuries (n43/) Frontal ipacts, all injuries (n72/286) Change of velocity (k/h) Figure 3. of neck injury in rear ipacts and any injury in frontal ipacts versus delta-v, fro crash pulse recorder data. Relative injury risk,6,4,2 Neck injury-rear ipacts, struck car All injuries-rear ipacts, striking car,7,9,,2,3 Relative change of velocity Figure 4. Relative risk of neck injury in the struck car and any injury in the striking car versus relative change of velocity, fro police reported crashes in Sweden.,4 All injuries, frontal ipacts Neck injury, rear ipacts Crash recorder data Rear-end ipacts, target car,2 Rear-end ipacts, bullet car olice reported crashes Change of velocity (k/h) Figure 5. of neck injury in rear ipacts and any injury in frontal ipacts versus delta-v, fro crash pulse recorder and police reported data. In Figure 6 it can be seen that the AIS neck injury risk in rear ipacts differed significantly copared to the neck injury risk in frontal ipacts. In rear ipacts, the risk increased to % at approxiately 25 k/h, while in frontal ipacts the risk was only approxiately 3% at the sae change of velocity. The neck injury risk in frontal ipacts never exceeded 45% and was lowered above 35 k/h.,4,2 Frontal ipacts (n86/286) Rear ipacts (n43/) change of velocity (k/h) Figure 6. of neck injury in frontal and rear ipacts versus change of velocity, fro crash recorder data. Figures 7 and 8 present two attepts to differentiate injury probability in frontal ipacts for injuries to different body regions. Siilar differences in shapes of the risk functions were found for the two alternative ethods. However, the head injury risk differed between Figures 7 and 8. In Figure 7 it can also be seen that the neck injury risk shows the highest increase in risk at low severity, while the neck injury risk at high severity decreased. The shape of the injury risk for lower spine has a siilar decrease at high ipact severity as the neck injury risk.,4,2 Head Neck Chest Lubar spine Change of velocity (k/h) Figure 7. Injury risk for head, chest, neck and lower spine injuries versus change of velocity fro crash pulse recorder data. Relative injury probability,4,2,,8,6,4,2 Chest injuries Head injuries Neck injuries 5,9,95,5,,5,2 Relative change of velocity Figure 8. Relative risk of head, chest and neck injury versus relative change of velocity fro police reported crashes in Australia.

6 DISCUSSION Valid and reliable risk functions, describing the link between ipact severity and risk of injury, can be used for any purposes. One of the ost iportant areas is to validate injury criteria for experiental crash testing or siulations. The sensitivity and elasticity of an injury criterion would have to atch real-life experience in order to be accepted as a valuable candidate. This study deonstrates that risk functions are different for different types of injury, and it deonstrates that two independent ways to generate risk functions can be cobined. It is believed that calculations of injury risk functions based on data fro low-quality crash reconstruction can fundaentally influence the shape of the risk curves, possibly guiding developent of injury criteria in a wrong direction (Kullgren and Lie 998). This study shows that risk functions possibly can be estiated with sall errors. The aterial used was a ix of collisions fro Sweden and fro Australia. The crash pulse recorder data did coe fro Sweden, while the police reported crashes showing injury risk to different body regions were fro Australia. The results in Figure 6 were fro police reported crashes in Sweden. As different car fleets were used in the study and as the belt use differs between the countries, it is difficult to directly copare risk curves fro the different saples. Ideally, databases fro the sae country and with a siilar ix of car odels should be used. The nuber of car odels included in the crash recorder project was low, which eans that the results cannot be generalised to the whole accident population. The risk curves should because of that be handled with soe care. Also, in the calculations of injury risks for injuries to different body regions, the nuber of injuries, especially at high ipact severity, was relatively low. The differences between the risk curves for different body regions will still be valid, although the true shape of the risk curves could differ fro the ones presented. The advantage with large databases with police reported crashes is that risk functions for injuries to different body regions can be easily and accurately calculated, although only in a narrow interval in crash severity. This liitation ight be resolved by cobining risk functions for several types of injuries in a broader spectru of ipact severity. The advantages with crash pulse recorders are priarily that accurate easureents of crash severity are available, allowing risk curves to be calculated for a large variation in crash severity and also for different crash severity paraeters. However, the availability of data is often liited. Since the use of crash recorders in accident reconstruction is growing, ore can be done in the future. There is a need for valid ethodologies to get good value of this new opportunity. Injury risk curves are ost often regarded as continuously increasing functions versus ipact severity. The findings in this study show that there ight be large variations in the shape of the risk functions when studying injury risks for injuries to different body regions. Both ethods showed that especially the neck injury risk in frontal ipacts differed copared to head and chest injury risks. The results fro crash pulse recorders showed that both neck and lower spine injury risks decreased to an alost zero-level above certain changes of velocity. This effect is not due to asking of other injuries. Since there are large variations in risk functions for different body regions and since the risk functions not always are increasing at increased ipact severity, it will be iportant to take this into account in the design of crash tests. The chosen test speed will have a significant influence on the injury types that will be covered. The reason for the decrease in neck injury risk at high crash severity ight be due to a positive influence on neck injury risk of airbags (Kullgren at al. 2, Morris et al. 2). Another effect could be that other ore severe injuries are doinating at high severity crashes, and ay in these crashes lead to an under-reporting of AIS neck injuries. The crash pulse recorder data showed that the highest risk at high severity ipacts was for chest injuries, followed by head injuries, see Figure 7. In the police reported crashes, see Figure 8, head injuries showed the highest risk. The explanation for the discrepancy ight be the classification of head injuries as well as different proportions of airbags in Sweden and Australia. Sweden has a higher proportion of airbags, which reduces the head injury risk. Two injury risk functions showed continuously increasing risk values, naely the neck injury risk in rear ipacts and the risk of any injury in the frontal ipacts. Relative injury risks were copared with the absolute risk easures for these two risk functions. Both ethods showed that the neck injury in rear ipacts had the steepest slope in the risk function. Better links between real-life and experiental data is needed. It is iportant to better understand how crash test duies respond. In experiental tests, a change in test speed or acceleration level should be reflected in the easured duy readings corresponding to the increase in injury risk calculated fro real-life crashes. With the paired coparison technique it is possible to study injury risk functions for several different

7 injury types and ipact directions. However, future studies with ore hoogeneous data sources are necessary to be able to fully cobine the different ethods to be able to transfor the relative risk functions into absolute ones. The paired coparison technique akes it possible to in the future calculate risk curves for different vehicle categories and even for separate car odels. Even if the risk curves are relative ones, differences between car odels and vehicle categories could be analysed. The possibility to use atched-paired technique to generate relative injury risks and risk functions, stresses the need for high-quality injury classification or ass data, whereas estiates of exposure is not iportant in this type of analysis. A aterial consisting of ICD-codes would therefor be beneficial to use. CONCLUSIONS It was found that injuries to different body regions ay have very different shapes of the injury risk curves. Most of the have continuously increasing risk functions, while injuries to the spine in frontal ipacts showed increasing risks at low severity and decreasing risks at higher ipact severity. Different shapes of risk functions for injuries to different body regions should be considered in crash tests chosen at different test speeds. A correlation between the risk curves calculated with atched-paired and crash recorder techniques was found, allowing the two ethods to be cobined and cross validated. In the interval to 2 k/h, the neck injury risk in rear ipacts increased fro approxiately 45% to 8%, while the neck injury risk in frontal ipacts increased fro approxiately 27% to 33%. Such changes in risk for a certain change in crash severity should be reflected in duy readings fro crash tests and coputer siulations.

8 REFERENCES Aldan B, Kullgren A, Lie A, Tingvall C. Crash ulse Recorder (CR) - Developent and Evaluation of a Low Cost Device for Measuring Crash ulse and Delta-V in Real-Life Accidents. roc. of the 3th ESV Conf., aris, 99. Boströ O, Svensson M Y, Aldan B, Hansson H A, Håland Y, Lövsund, Seean T, Sunesson A, Säljö A, Örtengren T. A New Neck Injury Criterion Candidate- Based on Injury Findings in the Cervical Spinal Ganglia After Experiental Neck Extension Traua. roc. IRCOBI Conf. On Bioechanics, Dublin, Ireland, 996. Boströ O, Bohan K, Håland Y. Kullgren A, Krafft M. A New Neck Injury Criterion Candidate for Frontal Collisions. roc. of the 2 IRCOBI Conf. on the Bioechanics of Ipacts, Montpellier, France, 2. DavidssonJ,SvenssonMY,FlogårdA,HålandY, Jacobsson L, Linder A, Lövsund, Wiklund K. BioRID - A New Biofidelic Rear Ipacr Duy. roc. IRCOBI Conf. on Bioechanics, Göteborg, Sweden, 998. Evans L. Double pair coparison a new ethod to deterine how occupant characteristics affect fatality risk in traffic crashes. Accid. Anal. and rev., Vol. 8, No. 3, 986. pp Evans L. Driver injury and fatality risk in two-car crashes versus ass ratio inferred using Newtonian echanics, Accident Analysis and revention, vol. 26, no 5, pp , 994. HäggA,KarénB,vKochM,KullgrenA,LieA, Malstedt B, Nygren Å, Tingvall C. Folksa Car Model Safety Rating 99-92, Folksa 6 6 Stockhol, Stockhol, 992. Karén B, Krafft M, Kullgren A, Lie A, Nygren Å, Tingvall C. Advanced Accident Data Collection - Description and otentials of a Coprehensive Data Collection Syste. roc. of the 3th ESV Conf., aris, 99. KrafftM,KullgrenA,LesM,LieA,TingvallC. Injury as a function of change of velocity, an alternative ethod to derive risk functions. roc. of the Crash 2 Conf., London, 2. Kullgren A, Lie A. Vehicle Collision Accident Data - Validity and Reliability. Journal of Traffic Medicine, Vol 26, No. 3-4, 998. Kullgren A. Validity and Reliability of Vehicle Collision Data: Crash ulse recorders for Ipact Severity and Injury Assessents in Real-Life Frontal Collisions. Thesis for the degree of Doctor in hilosophy, faculty of Medicine, Folksa Research 6 6 Stockhol, 998. Kullgren A, Tingvall C, Ydenius A. Experiences of Crash ulse Recorders in Analyses of Real- World Collisions, resented at the 999 Road Safety Research, olicing and Education Conf., Canberra, Australia, 999. Kullgren A, Krafft M, Tingvall C. Neck Injury in Rear-End and Frontal Ipacts: Results fro Real-Life Ipacts with Recorded Crash ulses. roc. of the Ipact Bioechanics Australia Conference: Neck Injury 2, Sydney, Australia, 2. Morris A, Kullgren A, Barnes J, Truedsson N, Olsson T, Fildes B. revention of neck injury in frontal ipacts. roc. of the Ipact Bioechanics Australia Conference: Neck Injury 2, Sydney, Australia, 2. Norin H. Evaluating the Crash Safety Level of Coponents in Cars. Thesis for the degree of Doctor in hilosophy, ISBN , Karolinska Institutet, Stockhol, 995