Quality Criteria for the Safety Assessment of Cars Based on Real-World Crashes

Use of In-Depth Data in Comparing EuroNCAP and Real-World Crash Results

Report of Sub-Task 2.3

CEA/EC SARAC II QUALITY CRITERIA FOR THE SAFETY ASSESSME

OF CARS BASED ON REAL-WORLD CRASHES

Funded by the European Commission,

Directorate General TREN

SARAC II

Quality Criteria for the Safety Assessment of Cars based on Real-World Crashes

Project Number: SUB/B27020B-E3-S07.17321-2002

Report of Sub-Task 2.3

Use of In-Depth Data in Comparing EuroNCAP and Real-World Crash Results

Stuart Newstead, Amanda Delaney and Max Cameron Monash University Accident Research Centre

March 2006 CEA/EC SARAC II QUALITY CRITERIA FOR THE SAFETY ASSESSME

OF CARS BASED ON REAL-WORLD CRASHES

Funded by the European Commission,

Directorate General TREN

International Project Management Comité Européen des Assurances (CEA) Prof. Dr. Klaus Langwieder

SARAC Members European Commission (EC) Comité Européen des Assurances (CEA) DG TREN 26 Boulevard Haussmann 28 Rue Demot FR-75009 Paris B-1040 Brussels

Monash University Helsinki University of Technology Accident Research Centre (MUARC) Laboratory of Transportation Engineering Building 70, P.O. Box 2100 Clayton, 3800 Victoria, Australia FIN-02015 HUT, Finland

BMW Group Bundesanstalt für Straßenwesen Centro Zaragoza Vehicle Safety (BASt) Instituto de Investigación Sobre D-80788 München Brüderstraße 53 Reparación de Vehiculos, S.A. D-51427 Bergisch Gladbach Carretera Nacional 232, km 273 E-50690 Pedrola (Zaragoza) DaimlerChrysler AG Department for Transport FIA Foundation for the Automobile Zone 1/29a Great Minister House and Society D-71059 Sindelfingen 76 Marsham Street 8 Place de la Concorde London, SW1P 4DR United Kingdom Paris 75008 France

Ministry of Transport and Finnish Motor Insurers’ Centre FOLKSAM Insurance Group Communications of Finland (VALT) Research/Traffic Safety P.O. Box 31 Bulevardi 28, S-106 60 Stockholm FIN 0023 Helsinki FIN-00120 Helsinki

Ford Motor Company German Insurance Association Honda Motor Europe Safety Data Analysis (SDA) (GDV) Wijngaardveld 1 Office (ASO) German Insurance Institute for Traffic 9300 Aalst Belgium Köln-Merkenich / Spessartstraße Engineering D-50725 Köln Friedrichstrasse 191, D-10117 Berlin Insurance Institute for Highway ITARDA IVT Heilbronn Safety (IIHS) & Institute for Traffic Accident Institut für Verkehrs- und Highway Loss Data Institute (HLDI) Research and Data Analysis Tourismusforschung e. V. 1005 N. Glebe Road Kojimachi Tokyu Bldg. 6-6 Kojimachi, Kreuzäckerstr. 15 Arlington, VA 22201 USA Chiyoda-ku Tokyo 102-0083 Japan D-74081 Heilbronn Japanese Automobile Research Laboratory of Accidentology, Loughborough University Institute (JARI) Biomechanics and Human Vehicle Safety Research Centre 2530 Karima, Tsukuba Behaviour PSA Peugeot- Holywell Building Loughborough Ibaraki 305-0822, Japan Citroën/RENAULT (LAB) Leicestershire LE 11 3 UZ UK 132 Rue des Suisses 92000 Nanterre (France) National Organisation for Swedish Road Administration Technische Universität Automotive Safety and Victims Aid (SRA) Braunschweig (NASVA) Röda Vägen Institut für Mathematische Stochastik 6-1-25, Kojimachi Chiyoda-Ku, S-78187 Borlange Pockelsstr. 14 Tokyo, 102-0083, Japan D-38106 Braunschweig Verband der Automobilindustie Volkswagen AG (VDA) Westendstr. 61 1777 Unfallforschung D-60325 Frankfurt/Main D-38436 Wolfsburg CEA/EC SARAC II QUALITY CRITERIA FOR THE SAFETY ASSESSME

OF CARS BASED ON REAL-WORLD CRASHES

Funded by the European Commission,

Directorate General TREN

Document Retrieval Information

Report No. Date Pages 251 March 2006 12

Title and Subtitle Use of in-depth data in comparing EuroNCAP and real-world crash results.

Author(s) Newstead S., Delaney A, Cameron M.

Performing Organisation Accident Research Centre Building 70 Monash University, 3800, Victoria Australia

Sub-Task Participants Pilot: Monash University Accident Research Centre

Sub-contractors: Monash University Accident Research Centre BASt Advisors: LAB Observers: NASVA GDV JARI UK Department for Transport Ford

Abstract In previous SARAC work comparing the relationship between injury outcomes in real crashes and the results of EuroNCAP testing, injury outcomes in real crashes have been assessed at the broadest level with the real crash outcome measure being an average overall risk of death or serious injury to the vehicle driver across all body regions. Being able to make comparisons between injury outcomes on a finer injury severity scale and by body region with results from EuroNCAP testing by body region on the would allow much finer assessment of the ability of EuroNCAP to reflect real world outcomes in crashes. The aim of this sub task was to assess the suitability of available European in-depth crash data sources for comparing real world crash outcomes by body region and on a finer injury severity scale with results of EuroNCAP testing by body region. This study examines the suitability of two existing European databases for in-depth analysis and summarises the data requirements for future analysis.

Keywords NEW CAR ASSESSMENT PROGRAM (NCAP), REAL-WORLD DATA, , CORRELATION

The views expressed are those of the authors and do not necessarily represent those of CEA or any of the participants of the SARAC committee.

CEA/EC SARAC II Table of Contents

Table of Contents

EXECUTIVE SUMMARY...... 1 1 BACKGROUND AND AIMS ...... 3 2 AUSTRALIAN RESEARCH EXPERIENCE...... 3 3 EUROPEAN IN-DEPTH DATA SOURCES ...... 6 3.1 CCIS ...... 6 3.2 PENDANT...... 7 4 SUITABILITY OF IN-DEPTH DATA SOURCES ...... 8

4.1 CCIS DATA...... 8

4.2 PENDANT DATA ...... 9 4.3 SUMMARY ...... 9 5 SUMMARY OF DATA REQUIREMENTS FOR FUTURE ANALYSIS ...... 10

5.1 OTHER POSSIBLE DATA SOURCES ...... 11 6 REFERENCES ...... 12

CEA/EC SARAC II EXECUTIVE SUMMARY

EXECUTIVE SUMMARY

In previous SARAC work comparing the relationship between injury outcomes in real crashes and the results of EuroNCAP testing, injury outcomes in real crashes have been assessed at the broadest level with the real crash outcome measure being an average overall risk of death or serious injury to the vehicle driver across all body regions. Being able to make comparisons between injury outcomes on a finer injury severity scale and by body region with results from EuroNCAP testing by body region on the crash test dummy would allow much finer assessment of the ability of EuroNCAP to reflect real world outcomes in crashes. The aim of this sub task was to assess the suitability of available European in-depth crash data sources for comparing real world crash outcomes by body region and on a finer injury severity scale with results of EuroNCAP testing by body region.

Earlier Australian analysis shows the potential for comparing injury severity outcomes in real crashes and NCAP testing score components by body region and using the finer AIS injury severity scoring system. It was able to identify significant correlations between score components and real crash outcomes in specific body regions in each of the test configurations considered. The analysis was limited, however, in that it did not include adjustment for differences in driver characteristics and crash circumstances between vehicle models that may have confounded the results presented. Ideally, analysis should have included adjustment of such differences. However, the limited data available for analysis meant the adjusted analysis was not feasible.

Two European in-depth crash inspection data sources were made available for assessing the general suitability of such data sources for examining the relationship between real crash outcomes and EuroNCAP test results by body region of injury. Suitability of both the CCIS and PENDANT databases for assessing the relationship between EuroNCAP tests and real crash outcomes by body region and more specific injury severity outcome are related to the content and coverage of the data. Both databases appear in theory to be suitable for comparing to EuroNCAP results by body region and more specific levels of injury severity. Both have the information on injury severity by body region and sufficient details on vehicle model and year of manufacture to match with EuroNCAP test data. They also both appear to contain the required range of variables to undertake analysis controlled by driver and crash characteristics between vehicle models. This was less certain for the CCIS data from the sample supplied.

Coverage of both datasets was inadequate to measure injury risk but both could be used to measure relative injury severity which would suit the purpose of the planned comparisons. The PENDANT data was most suitable in this respect as it included representative case

1 CEA/EC SARAC II EXECUTIVE SUMMARY weights whereas CCIS did not appear to contain weights. Further, both databases were of insufficient size with the number of cases of EuroNCAP tested vehicle models in the data not great enough to undertake meaningful analysis, particularly when disaggregated by body region of injury. Furthermore, the CCIS data would need to be provided in unit record format to be entirely useful for analysis.

The limitations of both the sample CCIS and PENDANT in-depth crash data meant meaningful analysis of the relationship between EuroNCAP test outcome and real world injury levels by body region could no be meaningfully undertaken. The key output of the sub- task review then was a summary of ideal requirements of in-depth data. These are as follows. • Data should be provided in unit record format for analysis where every case in the database represents a driver or front left passenger involved in a crash in a vehicle model tested by EuroNCAP or a pedestrian struck by a EuroNCAP tested vehicle model.

• Vehicle make and model information and year of manufacture should be supplied directly or be able to be derived to allow precise selection of vehicle models for comparison with the corresponding EuroNCAP test.

• Factors independent from the vehicle that ideally should be recorded in the individual case data for analysis

• The data sampling frame should be well articulated, particularly with respect to injury and crash severity levels. Ideally weightings to make the data representative of the broader crash population should be provided to allow comparative injury severity outcomes between vehicles to be accurately represented.

• The data should contain a reasonable number of cases on each EuroNCAP tested vehicle model to allow meaningful analysis to be undertaken. Although the exact number required is difficult to quantify without estimates of the effect sizes to be measured, a rough guide of a minimum of 30 cases per vehicle would be reasonable although significantly more cases could be required for more complex comparisons, including those adjusted for confounding factors.

Australian analysis correlating injury outcome by body region with A-NCAP results demonstrates that other data sources such as injury compensation claims can be successfully used in this type of analysis. Other European data that may be useful are hospital admissions databases such as that being assembled in Europe under Working Package 3 of the PENDANT project. The availability and potential use of insurance claims

2 CEA/EC SARAC II BACKGROUND AND AIMS

and hospital admission databases in Europe should also be investigated as part of future research. 1 BACKGROUND AND AIMS

SARAC 2 sub-tasks 2.1 and 2.1 have compared the relationship between injury outcomes in real crashes and the results of EuroNCAP testing. In these comparisons, injury outcomes in real crashes have been assessed at the broadest level with the real crash outcome measure being an average overall risk of death or serious injury to the vehicle driver across all body regions. Similarly, only the aggregate EuroNCAP score has been considered, either for the overall vehicle assessment of from each individual test configuration.

In this previous work, no attempt was made to compare the specific EuroNCAP test component outcome scores by body region with injury outcomes to specific body regions of the driver in real-world crashes. This was primarily because police crash data only indicate the overall injury severity to a driver using a coarse four or five point scale of injury outcome. It does not generally code specific injuries and does not indicate the specific regions of the body injured. Nor does it code injury severity to specific body regions or use an overall severity scale with a finer gradient, such as the Abbreviated Injury Scale (AIS) or Injury Severity Score (ISS). This is because coding such detail in the data generally requires a level of information beyond what it is possible for police to collect.

Being able to make comparisons between injury outcomes on a finer injury severity scale and by body region with results from EuroNCAP testing by body region on the crash test dummy would allow much finer assessment of the ability of EuroNCAP to reflect real world outcomes in crashes. It would allow assessment of the ability of each EuroNCAP score component to reflect its corresponding specific real world injury outcome. For example, it would allow assessment of the EuroNCAP offset frontal or side impact test head assessment component in representing the risk of serious head injury in real world crashes of the same configuration. Comparison on this specific basis would allow targeted evaluation of the relevance of each EuroNCAP score component.

The aim of this sub task was to assess the suitability of available European in-depth crash data sources for comparing real world crash outcomes by body region and on a finer injury severity scale with results of EuroNCAP testing by body region.

2 AUSTRALIAN RESEARCH EXPERIENCE

To illustrate the potential for assessment of EuroNCAP score components by body region, research completed in Australia some years ago was reviewed (Newstead et al, 1997). The Australian research was not able to utilise in-depth crash data since the Australian collection

3 CEA/EC SARAC II AUSTRALIAN RESEARCH EXPERIENCE of such data was extremely limited. Instead, the project was based on the analysis of data from third party insurance claims for injuries suffered in motor vehicle crashes from the state of Victoria in Australia. The insurance data contained detailed coding of injury outcomes to vehicle drivers that could be converted into AIS injury severity scores by body region of injury and subsequently ISS scores for the driver. Analysis focused on comparing the average maximum AIS scores by body region for injured drivers in each vehicle make and model with results of Australian NCAP testing. It should be noted that the A-NCAP testing protocol and scoring system at the time of the study was different to the current EuroNCAP practices with a full frontal and offset frontal impact test being carried out and injury assessment based only on Head injury Criteria (HIC), Chest Acceleration (Chest G) and Femur Loading (FL).

Although Victoria has over 15,000 claims for injury compensation from motor vehicle crashes each year, the Victorian claims data contained information on only 2982 observations of driver injury across 28 ANCAP tested vehicles. After excluding vehicle models with very small numbers of cases (less than 30), analysis was limited to 19 vehicle models with full frontal A-NCAP scores and 12 vehicle models with offset A-NCAP scores. For each vehicle model, the average maximum AIS score for each of the head, chest and leg regions was calculated and compared to the corresponding A-NCAP score.

Tables 1 and 2 summarise the maximum AIS scores against the corresponding ANCAP readings (HIC, Chest Loading and Femur Loading) for each of the three body regions for each vehicle model for full frontal and offset ANCAP results respectively presented in the Australian study. Also included at the bottom of these tables is the correlation between each ANCAP measure and the corresponding average maximum AIS score as a measure of the association between the two variables.

Table 1: Full Frontal ANCAP Test Results and Average Maximum AIS Scores by Body Region for 19 ANCAP Tested Models.

BODY REGION HEAD CHEST FEMUR

MODEL No. ANCAP Av. ANCAP Av. ANCAP Av. Cases HIC Max. Chest G Max. Femur L Max. AIS AIS AIS

CIVIC (92-95) 30 1456 0.00 63 0.13 3.20 0.23

NISSAN PATROL (88-90) 32 1750 0.19 67 0.13 2.80 0.06

HYUNDAI EXCEL (95-96) 36 1411 0.22 60 0.11 2.60 0.08

SUBARU LIBERTY (89-94) 43 1360 0.09 58 0.16 3.90 0.21

MAZDA 626/MX6 (92-94) 57 1160 0.14 60 0.35 2.60 0.26

TOYOTA CAMRY (93-96) 65 1040 0.09 61 0.20 1.90 0.08

4 CEA/EC SARAC II AUSTRALIAN RESEARCH EXPERIENCE

NISSAN PULSAR (92-95) 66 1464 0.15 50 0.29 4.80 0.11

MAZDA 121 (91-96) 72 1525 0.11 61 0.21 4.70 0.22

FORD FALCON EF (94-96) 104 910 0.27 74 0.18 7.40 0.23

NISSAN PINTARA (89-92) 107 1750 0.11 64 0.26 2.40 0.20

FORD LASER (91-94) 129 1903 0.20 68 0.25 8.60 0.26

HYDAI EXCEL (90-94) 139 1318 0.20 54 0.14 3.60 0.19

BARINA (89-93) 146 1005 0.14 59 0.25 3.90 0.16

FORD FALCON EB SERIES II (92-94) 156 1340 0.11 74 0.16 6.00 0.15

MITSUBITSHI MAGNA TR/TS (91-95) 163 1140 0.08 60 0.15 3.80 0.15

HOLDEN COMMODORE VR/VS (93-96) 194 1170 0.08 51 0.15 3.20 0.15

TOYOTA COROLLA (90-94) 294 1499 0.13 60 0.21 9.40 0.19

TOYOTA CAMRY (88-92) 353 1090 0.08 63 0.28 3.90 0.15

HOLDEN COMMODORE VN/VP (87-93) 640 1690 0.19 82 0.21 1.20 0.16 CORRELATION HIC with Av. CG with Av. FL with Av. ANALYSES Max. AIS to Max. AIS to Max. AIS to

HEAD CHEST LEGS All Models 0.12 -0.06 0.40 (p=0.3148) (p=0.5949) (p=0.0451)

Table 1 shows a strong statistically significant association between full frontal ANCAP femur loading readings and average maximum AIS to the leg region in real crashes for the 19 models included in the analysis. Whilst table 1 also shows indication of a weak association between HIC and real crash head injury severity for this ANCAP test configuration, the result is not statistically significant. No association between full frontal ANCAP chest loading and maximum AIS to the chest region in real crashes was observed. Table 2 shows a strong statistically significant association between the offset ANCAP chest loading and average maximum AIS to the chest in real crashes for the 12 car models for which offset scores are available. No association was found between ANCAP and real crash measures for the head or leg regions.

Table 2: Offset ANCAP Test Results and Average Maximum AIS Scores by Body Region for 12 ANCAP Tested Models.

BODY REGION HEAD CHEST FEMUR

MODEL No. ANCAP Av. ANCAP Av. ANCAP Av. Max. Cases HIC Max. Chest G Max. Femur L AIS AIS AIS

CIVIC (92-95) 30 623 .03 40 .13 1.30 .23

NISSAN PATROL (88-90) 32 897 .00 37 .13 4.60 .06

5 CEA/EC SARAC II EUROPEAN IN-DEPTH DATA SOURCES

HYUNDAI EXCEL (95-96) 36 1270 .06 49 .11 4.70 .08

TOYOTA CAMRY (93-96) 65 640 .03 42 .20 3.50 .08

NISSAN PULSAR (92-95) 66 2161 .03 78 .29 18.00 .11

MAZDA 121 (91-96) 72 1566 .01 69 .21 7.40 .22

FORD FALCON EF (94-96) 104 596 .11 53 .18 3.70 .23

FORD LASER (91-94) 129 3234 .07 84 .25 11.20 .26

HYDAI EXCEL (90-94) 139 1195 .06 58 .14 4.90 .19

BARINA (89-93) 146 1213 .03 56 .25 8.30 .16

HOLDEN COMMODORE VR/VS (93-96) 194 730 .09 37 .15 2.60 .15

TOYOTA COROLLA (90-94) 294 1024 .06 52 .21 6.20 .19 CORRELATION HIC with Av. CG with Av. FL with Av. ANALYSES Max. AIS to Max. AIS to Max. AIS to

HEAD CHEST LEGS Models with offset ANCAP scores -0.04 0.74 -0.01 (p=0.5478) (p=0.0022) (p=0.5120)

The Australian analysis shows the potential for comparing injury severity outcomes in real crashes and NCAP testing score components by body region and using the finer AIS injury severity scoring system. It was able to identify significant correlations between score components and real crash outcomes in specific body regions in each of the test configurations considered. The analysis was limited, however, in that it did not include adjustment for differences in driver characteristics and crash circumstances between vehicle models that may have confounded the results presented. Ideally, analysis should have included adjustment of such differences. However, the limited data available for analysis meant the adjusted analysis was not feasible.

3 EUROPEAN IN-DEPTH DATA SOURCES

Two European in-depth crash inspection data sources were made available for assessing the general suitability of such data sources for examining the relationship between real crash outcomes and EuroNCAP test results by body region of injury.

3.1 CCIS

The UK Co-operative Crash Injury Study (CCIS) is funded by the UK Department for Transport (DfT), Ford, Toyota, Renault PSA and Autoliv with data being collected by specialist teams a Loughborough and Birmingham Universities and an agency of the DfT. The teams investigate crashes and compile vehicle examination evidence and injury details on about 1500 cases per year, the majority coming from the East and West Midlands region

6 CEA/EC SARAC II EUROPEAN IN-DEPTH DATA SOURCES

of England. The data represents a stratified sample of crashes by injury severity involving passenger cars less than 8 years old and towed from the scene. The data can be weighted to be representative of the population although the unweighted data is intentionally biased towards the most serious injuries occurring in newer cars with damage. The data sampling and collection protocol has been relatively unchanged since 1983 with the total database containing well over 10,000 records.

A sample of CCIS data was provided for the study. It comprised records on 3098 vehicles from model series that had been tested by EuroNCAP at some point. Data were only provided in tabular and not unit record format and covered overall counts of vehicles as well as maximum AIS by body region of injury.

3.2 PENDANT

Pendant is the acronym for the Pan-European Co-Ordinated Accident and Injury Databases, an EC funded project co-ordinated by Loughborough University in England and with partners from organisations across Europe. The PENDANT project aims to assemble a number of different databases of different size and severity from various European sources including hospital data and in-depth crash inspection data. Consideration in this report focuses on the in-dept data being assembled as part of PENDANT Work Package 2.

Work Package 2 of PENDANT brings together the resources and infrastructures of existing accident and injury investigation groups to build a demonstration European Crash Injury database. The project aimed to have 1100 cases in the demonstration database although it is anticipated the project would facilitate ongoing harmonized data collection. The principal aim of analysis from the database was to examine the injury prevention priorities for future action and to provide feedback to casualty reduction measures such as the EuroNCAP rating system. Content of the database is significantly more detailed than that found in police reported crash data.

Groups collecting the data cover northern, middle and southern Europe giving a representative range of accident conditions. A special feature of the data is the case selection methodology that targets coverage of newer vehicles to give data that has most value for regulation and safety countermeasures. Only crashes involving a vehicle registered on or after the 1st January 1998 are collected with the age of the partner vehicle in a multi vehicle crash not important. Furthermore, crashes are only included in the sample if they involve an injured car occupant, but not necessarily to the occupant of the 1998 or later vehicle. Single vehicle crashes are included if the car is registered from 1998 onwards and an occupant is injured. At least 20% of the cases in the PENDANT in-depth data are severe

7 CEA/EC SARAC II SUITABILITY OF IN-DEPTH DATA SOURCES crashes involving a fatality or AIS 3+ injury. Remaining crashes are selected at random although must still meet the previous criteria. Representative weighting is provided for each crash.

A sample of the PENDANT in-depth data was provided for assessment in the study on a unit record basis and covered 1322 crash involved vehicles. A wide range of relevant fields from the database were included.

4 SUITABILITY OF IN-DEPTH DATA SOURCES

Suitability of both the CCIS and PENDANT databases for assessing the relationship between EuroNCAP tests and real crash outcomes by body region and more specific injury severity outcome are related to the content and coverage of the data.

4.1 CCIS DATA

Assessment of the content of the CCIS data was problematic as the data was only supplied in summary tabular format and not on a case by case basis. There appeared to be sufficient data on about 20 different NCAP tested vehicle models that had more than 30 examples of crashes vehicles in the database for meaningful analysis on an aggregate basis. Make and model of vehicle by year of manufacture were given to allow specific matching of EuroNCAP tested vehicles. AIS injury outcomes were also available by body region although when disaggregated to this level the number of cases became relatively sparse meaning analysis results would likely be of limited accuracy. Despite this, it appeared possible to analyse the CCIS data on this basis.

It is known that the CCIS data in unit record format contain information on driver characteristics such as age and gender and details of the crash circumstances although tabulation of these variables was not provided in the sample data. Differences in these variables between vehicle models could be controlled for in an analysis of the CCIS data on a unit record basis. Not having access to the CCIS data on a unit record basis represented a key limitation of the data for the purpose being assessed.

Another concern of the CCIS data for the EuroNCAP comparison by body region was the absence of a representative sample of non-injury data in the sample meaning injury risk could not be assessed. This is not considered a fatal flaw since other studies have indicated that EuroNCAP is more representative of relative injury severity rather than injury risk. However, the CCIS database is known to be biased towards most serious injuries so, unless suitable weighting factors could be provided for each case, it may be difficult to use the data effectively for assessing relative injury severity. In addition, since CCIS is focused towards

8 CEA/EC SARAC II SUITABILITY OF IN-DEPTH DATA SOURCES

vehicle occupants, use of the database to assess relationships with the EuroNCAP pedestrian test would not be possible.

4.2 PENDANT DATA

A fuller assessment of the PENDANT in-depth data was possible since essentially the complete database was supplied for assessment on a unit record basis. The content of the data was highly detailed in its coverage of information. Key fields required for the EuroNCAP comparison by body region were all present including AIS injury outcome by body region for each case. Make and model of vehicle and year of manufacture were also given in the data to allow accurate matching with EuroNCAP tested vehicle models.

For the same reasons as CCIS, PENDANT data could not be used to assess injury risk. However, use of the PENDANT data to assess relative injury severity would be feasible considering representative weighting factors for cases in the database are available. The data also covers pedestrian injury outcome meaning comparisons with the EuroNCAP pedestrian test outcomes would be possible. Information about crash circumstances and occupant characteristics are also available to allow a controlled analysis to be undertaken.

Reflecting the fact that PENDANT is an example harmonised European in-depth database, its key problem for use in comparing with EuroNCAP outcomes is the coverage of the database. Despite crash data on over 1300 vehicle being available, when analysed by vehicle make and model, only 1 vehicle model had more than 30 cases in the database before restriction to specific years of manufacture for matching with the EuroNCAP test outcomes. Clearly this is insufficient to allow analysis of injury outcome at an overall level let alone by specific body region and injury severity level. Clearly ongoing collection of the PENDANT in-depth data would have to happen before it was ultimately suitable for the analysis purpose being assessed.

4.3 SUMMARY

Both the CCIS and PENDANT databases appear in theory to be suitable for comparing to EuroNCAP results by body region and more specific levels of injury severity. Both have the information on injury severity by body region and sufficient details on vehicle model and year of manufacture to match with EuroNCAP test data. They also both appear to contain the required range of variables to undertake analysis controlled by driver and crash characteristics between vehicle models. This was less certain for the CCIS data from the sample supplied.

9 SUMMARY OF DATA REQUIREMENTS FOR CEA/EC SARAC II FUTURE ANALYSIS

Coverage of both datasets was inadequate to measure injury risk but both could be used to measure relative injury severity which would suit the purpose of the planned comparisons. The PENDANT data was most suitable in this respect as it included representative case weights whereas CCIS did not appear to contain weights.

Both databases were of insufficient size with the number of cases of EuroNCAP tested vehicle models in the data not great enough to undertake meaningful analysis, particularly when disaggregated by body region of injury. Furthermore, the CCIS data would need to be provided in unit record format to be entirely useful for analysis.

5 SUMMARY OF DATA REQUIREMENTS FOR FUTURE ANALYSIS

The limitations of both the sample CCIS and PENDANT in-depth crash data meant meaningful analysis of the relationship between EuroNCAP test outcome and real world injury levels by body region could no be meaningfully undertaken. The key output of the sub- task review then was a summary of ideal requirements of in-depth data. These are as follows.

• Data should be provided in unit record format for analysis where every case in the database represents a driver or front left passenger involved in a crash in a vehicle model tested by EuroNCAP or a pedestrian struck by a EuroNCAP tested vehicle model. Analysis of individual case data allows maximum opportunity to control for factors external to the vehicle in determining injury outcome hence making the comparison with EuroNCAP results more pure. The external factors may relate to either the injured occupants physical characteristics, their mode within the vehicle, the crash impact severity and crash configuration.

• Vehicle make an model information and year of manufacture should be supplied directly or be able to be derived from, for example, Vehicle Identification Number, to allow precise selection of vehicle models for comparison with the corresponding EuroNCAP test. Alternatively, each record should have a code linking it to the appropriate EuroNCAP test information if vehicle model details are not to be provided explicitly.

• Factors independent from the vehicle that ideally should be recorded in the individual case data for analysis, in rough order of priority by group are as follows:

Occupant Characteristics • Injury outcome by body region (AIS, ICD)

10 SUMMARY OF DATA REQUIREMENTS FOR CEA/EC SARAC II FUTURE ANALYSIS

• Seating position • Age • Gender • Restraint use • Height Crash Characteristics • Impact severity (Delta V or EBS) • Impact point on the vehicle • Impact angle • Impact partner vehicle details including body type, mass, point of impact (multi-vehicle crashes) • Object struck (single vehicle crashes or where relevant) Pedestrian Characteristics • Age • Gender • Height • The data sampling frame should be well articulated, particularly with respect to injury and crash severity levels. Ideally weightings to make the data representative of the broader crash population should be provided to allow comparative injury severity outcomes between vehicles to be accurately represented.

• The data should contain a reasonable number of cases on each EuroNCAP tested vehicle model to allow meaningful analysis to be undertaken. Although the exact number required is difficult to quantify without estimates of the effect sizes to be measured, a rough guide of a minimum of 30 cases per vehicle would be reasonable although significantly more cases could be required for more complex comparisons, including those adjusted for confounding factors.

Of the two sample in-depth data sets reviewed, the PENDANT data comes closest to meeting the above requirements apart from sample size. If data collection under the PENDANT framework is continued it could be useful for the proposed analysis by body region and specific injury severity outcome in the future.

5.1 Other Possible Data Sources

As demonstrated by the Australian analysis correlating injury outcome by body region with A- NCAP results, other sources of data exist to facilitate such comparisons. The Australian analysis demonstrates the successful use of an injury compensation claims database for the

11 CEA/EC SARAC II REFERENCES analysis. Other data that may be useful are hospital admissions databases such as that being assembled in Europe under Working Package 3 of the PENDANT project. To be useful for the purpose of the proposed analysis, the data would have to conform to the ideal requirements of the in-depth data listed above. In general, this would require the claims or hospital data to be linked to suitably detailed crash records and potentially vehicle registration records to provide the required crash and vehicle information. The key benefit of such data is that it provides many more cases for analysis than can typically be obtained from in-depth crash inspection databases but still has rich information in occupant and pedestrian injury outcomes that can be used to derive the required injury outcomes by body region. The availability and potential use of insurance claims and hospital admission databases in Europe should be investigated as part of future research.

6 REFERENCES

Morris, A.P. (2004) Harmonised Approaches to Collection of In-Depth Accident Data in the New European Union Workshop Report, PENDANT Project, EC Contact GMA2-2001-52066 S07.17215, University of Loughborough, UK

Bradford, M. Review of EU Accident Databases in “Review of the Co-operative Crash injury Study (CCIS) Accident Database Research Project No. SO115/VC” Report to UK DTLR by Mo Bradford, Private Consultant - Vehicle Safety

Newstead, S. and Cameron, M. (1999) Updated Correlation of Results from the Australian New Car Assessment Program with Real Crash Data from 1987 To 1996, Report No. 152, Monash University Accident Research Centre, Melbourne, Australia.

12 Comité Européen des Assurances (CEA) Commission of the European Communities (EC)

International Project Management Comité Européen des Assurances (CEA) Prof. Dr. Klaus Langwieder

Australia Japan Monash University Accident Research Centre National Organisation for Automotive Safety Belgium and Victims Aid (NASVA) Honda Motor Europe Institute for Traffic Accident Research and Data Analysis France (ITARDA) FIA Foundation for the Automobile and Society Japanese Automobile Research Institute (JARI) Laboratory of Accidentology, Biomechanics Spain and Human Behaviour, Centro Zaragoza, Instituto de Investigación Sobre PSA Peugeot, Citroen, Renault Reparación de Vehiculos Finland Sweden Helsinki University of Technology Swedish Road Administration (SRA) Ministry of Transport and Communication of Finland Folksam Insurance Group Finnish Motor Insurers’ Centre (VALT) United Kingdom Germany Department for Transport Bundesanstalt für Straßenwesen (BASt) Loughborough University (VSRC) German Insurance Association (GDV) USA Institute for Applied Transport and Insurance Institute for Highway Safety (IIHS) Tourisme Research (IVT) Technische Universität Braunschweig BMW Group DaimlerChrysler AG Ford Motor Company Volkswagen AG Verband der Automobil Industrie (VDA)