STAIRS - Final Summary Report
Standardisation of Accident and Injury Registration Systems
Project Co-ordinator :
(FR) CETE (SO)
-Jean Louis MARTIN
Reference period :
||1 October1996 to 31 December 1998
1.1 The Objectives of STAIRS
The aim of STAIRS was the standardisation of in-depth road accident data collection and methodologies. This objective focuses mainly on secondary safety or injury prevention and is divided into five Work Packages:
1.1.1 Work Package 1 - In-depth Data Collection
This Work Package involved developing a protocol for the collection and processing of injury and vehicle data from road crashes. It specifies the criteria required for data to be included in a harmonised European injury database. The Work Package concentrates on the systems and data required for the development of injury prevention methods, or for the evaluation of any new safety measures that may be introduced in the future. It includes the following topics:
1. A review of existing studies,
2. Values and variables,
3. Data collection,
4. Data accuracy,
6. Confidentiality and ethics.
The following are the conclusions or recommendations given from each topic.
220.127.116.11 1 - A review of existing studies
There are many areas of comparability throughout all three studies but only in the area of secondary safety. At a conceptual level there are many areas of similarity, but at the detailed level there are many differences.
Agreement on the objectives of collecting the data should point to which aspects of the crash are to be used and at what level. Consensus will then be required as to the methodology of investigation. The common focus concerns car occupant injuries and there are many areas of comparability. The main area of lower comparability concern case selection procedures. The CCIS study results in a more cost effective and statistically efficient model; whereas the random sampling procedures of INRETS and MUH allows for primary safety studies. As all partners do post-crash investigations on at least some of their cases, this would seem to be the best method to employ. Some on-site, post-crash recordings may be integrated into the system to include primary safety.
Agreement as to the mathematical model or software used for calculating crash severity would define at least some of the variables that are required for collection (unless the programs are non-contradictory). As STAIRS is fundamentally a secondary safety study, analysis of the performance of the vehicle and it's safety mechanisms is obviously an important factor. These two points go some way to providing a lead as to the type of methodology to be employed. Each study collects data on the above, but in a slightly different manner, and it was not difficult to focus on the key measurements required.
Differences in methods of collection are likely to remain; they are a consequence of local practicalities as well as technical precision. The aims and objectives of the STAIRS project are achievable, if at a cost to each partner, and a European-wide database will be beneficial to all.
18.104.22.168 2 - Values and variables
The objective of this task was to define a set of variables along with detailed data fields and a glossary of terms that will form the basis of a harmonised crash injury database. In order to achieve this goal, it was decided to determine a nucleus of variables, which were seen as essential for the realisation of a secondary safety orientated, data collection system. The group tried to specify the different values required for these variables, but of course, one must almost systematically add values such as "unknown" or "inapplicable". Although only a small selection of road users are used as examples, the basics principles are transferable with little or no modification. This document is seen as a starting point and it will evolve over time.
The work has tried to capture within its remit other major contributions to safety, such as the CREST program on child restraints, the COST 327 project on research into helmets, and Euro NCAP. The requirements for these studies have been assessed and variables included that are compatible with them.
As much as possible the variable list has tried to use commonly accepted systems for data collection, such as AIS 90, but it has been found necessary to expand on this in order to obtain some of the in-depth details required. Wherever possible, backward compatibility has been retained.
In some cases a new system had to be devised due to the incompatibility of those already in place. Wherever possible these systems have been kept very simple with the hope that they will be easily understood and still collect the necessary details. They will also allow for the 'mapping' of other systems onto a common framework.
The document as a whole is designed in a modular fashion, in order to allow the relevant sections to be extracted for whatever aim or objectives a study may have. It is expected that other variables will be required to be added. The list within this document is seen as the minimum requirement and has limitations.
22.214.171.124 3 - Data collection
The objective of this task was to present different ways of collecting data that we will qualify in this report as in-depth investigations. To choose a system you have to make an assessment according to the principal criteria of:
- the objective of the study and therefore the data that you wish to gather,
- the means that you can bring to bear, on a human level as well as on a financial one,
the local organisation of the emergency services.
In order to carry out the study, you have to clearly establish the data list that you intend to collect. Keeping in mind that the further you go back in time away from the moment of reference, (the instant where all movement of vehicles or victims in a trajectory sense have come to an end), the more difficult will become an on-the-spot- study. Obviously the setting up of on-the-spot studies is more expensive in human terms and could therefore limit the choices available.
Finally, and this is not necessarily the easiest problem to resolve, the local organisation of the emergency services may influence the choice of collection methodology. It is impossible to set up a system of accident notification for an on-the-spot study if there is a categorical refusal to collaborate by the emergency services. In such a case either change the system or change the study site, if that is possible. The following table sums up the different possibilities:
ON THE SPOT
||Very little data on vehicles
||Need specific studies
||Need specific studies
|Cost by case
||Very low cost/case
|Specific team (gathering)
||"Universal specialist" (except for medical data)
||Except for specific studies
|Special agreement (except for involved people)
Police (access to the scene)
Hospital (public and private)
||More difficult, more often fixed
126.96.36.199 4 - Data accuracy
The objective of this task was to define a minimum set of procedures that could be used to ensure that data is collected and recorded to a sufficient level of accuracy. The recommendations from this report are:
1. A balanced team should be selected, with the appropriate specialists in place.
2. Training and a constant updating of the skills necessary to ensure high quality collection of the data should be a main priority.
3. A similar process for the coding of the information should also occur.
4. A glossary of terms, updated as necessary, should be in place with a clear, precise understanding of the terminology and conventions used.
5. An objective method of recording data, either as the primary or secondary tool for investigation should be used wherever possible.
6. Assembling the case should have at least two stages:
- The initial methodology to bring together all the separate parts which should include a manual logic check,
and a second, more dispassionate check.
7. Checks should be used to ensure conversion of the data into an electronic format is correct.
8. Validation of the coding and plausibility of the electronic data should occur to check for self-consistency.
9. Management check.
10. Feedback loops should be established throughout the system to allow for errors to be corrected and new conventions or training identified at the earliest possible stage with National Accident Databases.
188.8.131.52 5 - Validation
With the variables and values, collection processes and data accuracy procedures decided upon, the next step was to validate these procedures and check if the information could be collected in the manner described.
In order to do this it was decided that each partner would collect three cases covering three different road user types. The French partner covered pedestrian crashes, the German partner would investigate two wheeled motor vehicles and finally, the UK would look at car crashes. This covered the three main modules described in the value and variables list and was designed to highlight any problems inherent within the scope of the document. It also allows the collection methodologies to be tested to the accuracy levels given in the latter two documents.
All three collection systems experienced similar problems, but only at the detailed level, especially concerning coding developments. This would indicate that further work is required in the development of work sheets for the coding process. This could be facilitated by the different groups exchanging staff and allowing the free flow of information between the groups. Regular workshops could be scheduled with a different focus for each one e.g. coding of a particular variable in a specific way. Case review conferences could also be arranged for any particularly interesting or difficult crashes.
This may not be as difficult as previously thought. With the advent of computer technology and video conferencing; the exchange of information via e-mail and other means, allows for faster and more accurate feedback which can greatly aid this process. Every effort should be made to allow this interchange and further enhance the development of the protocol.
184.108.40.206 6 - Confidentiality and Ethics
Within the framework of in-depth investigation of traffic accidents, personal information is required. The type and volume of this information depends upon the relation to the research plan and approach. Most of the information concerns injury pattern and human individual data such as age, driving circumstances and others. There are some regulations for the receipt and use of such personal data. It can, however, not be ignored that the legal regulations for the protection of the private sphere of patients are still relatively new and that the interpretation of these regulations is still partly unknown or difficult or even diverging. Often it will have to be decided between individual rights of the patient and the importance of the pursued research objective for common benefit. There still exists the principle "All has to be done to protect personal individual information".
Legislation within the EU and much of the national legislation seeks to protect the rights of the individual from unreasonable intrusion and it is likely to be very effective. However, one consequence of this is that the opportunities for health and safety research have become more restricted. The difficulty of access to individual records means that the numbers of cases available for an analysis may be insufficient for a decisive result. Including a requirement to seek permission, when casualties may be seriously injured or dead, can be insensitive and can mean that a longer period of research is needed before a clear result is achieved. While further data is being collected other road users are being injured and are unable to take advantage of new countermeasures.
It is recommended that legal exemptions should be available for research improving public health and safety, Publicly approved studies, which handle data according to an agreed code of practise, should be able to use this data for studies without seeking permission from the casualties or their relatives. These legal exemptions would be incorporated within both EU and national legislation
1.1.2 Work Package 2 - Linkage with National Accident Databases
Workpackage 2 is about the linking of in-depth accident databases to Regional, National accident databases. For the purposes of this document the methodological approach has only been applied to the 1st step of the linkage. Linkage between In-depth and Regional accident databases has been demonstrated for France, Germany and the UK. In all three countries it has been necessary to check the adequacy of the linking and weighting process, this has been illustrated by means of examples. For National levels, statistical requirements and sampling considerations have been proposed without examples.
220.127.116.11 Methodological approach
The sampling strategy affects what can be estimated from the In-depth database. Different strategies result in different constraints and assumptions on the estimation procedures.
The population of accidents for In-depth databases for at-the-scene investigations are not known a priori and hence must be selected opportunistically, however the sampling schemes applied ensure that there is no built in bias. Even though the strict statistical requirements are not being met, some linking to Regional or National databases may be possible. However, it is essential that the constraints of the data and the underlying assumptions are stated if Regional or National estimates are produced.
Methodologically, the estimation approach should first be to determine if the sample of in-depth accidents are representative of the Regional population from which they are sampled. This will check on the adequacy of the sampling method and the ability of any linking variables to provide adequate estimates at a Regional level. If the Regional level estimation was satisfactory then National estimates might be made. If the National level estimates also check out satisfactorily then the data may be suitable for pan-European purposes.
The reduction of bias by use of appropriate weighting variables has been discussed. Weighting is required because the in-depth database is probably a biased sub-set of the Regional database and the Regional database probably a biased sub-sample from the National database. The selection of appropriate weighting variables will then reduce the bias. It cannot however, eliminate the bias caused because some specific accident data were never included. The following points arise:
- the need to identify key variables for weighting purposes
- to try and identify types of variables which would use the same set(s) of weighting variables
- to use proxy weighting variables when necessary
- to accept that there may not be suitable weighting variables and/or data in the In-depth database to provide the required estimate
- that it is essential to estimate the confidence interval on any estimate
to realise that weighting from a small In-depth database to a Regional or National estimate may be imprecise and so have a large associated confidence interval
It is important to remember when applying the suggested methodology that the estimation process must make sense, in other words the population of interest must be available within the database and the data must be interpreted in an appropriate way. In order to achieve this it is vital to have a good working knowledge of the data, the data collection process, the data limitations and the statistical methodology.
It is also important to realise that some analyses will be interested in relationships between sets of variables, and that this relationship may not be affected by weighting and so the results are applicable at an in-depth, a Regional, a National or a pan-European level. This is in contrast to analyses where the required answer is a number or a proportion, these estimates should be weighted to properly reflect the underlying population of interest.
On the basis of these analyses it should be possible to develop an appropriate linkage for each Country. Such analyses will also be required if a linkage to a pan-European level is to be achieved. This is further complicated by the fact that different countries use different coding schemes. The CARE+ project is providing a mechanism to facilitate the comparison of accident statistics between member states using data from the member states disaggregated data. It would be helpful for STAIRS to have a mechanism for data exchange across Europe using a common coding scheme.
18.104.22.168 Guidelines for Pan European In depth Accident and Injury data base
An In-depth database may not be the most efficient way of providing a specific answer, but if an answer is needed now and not in 'n' years time, then an In-depth database may contain the required data. It must therefore, have sufficient data in depth and breadth to be able to answer most likely questions.
It is not difficult to design an in-depth accident investigation database once the objectives of having the data have been fully defined. However, it is not possible to define the best database for all purposes so a compromise must be made, and the guidelines must reflect that from a pan-European perspective. The following questions should be asked:
- what is the objective of having the in-depth accident database
- what level of question are likely to be asked - National/Regional/Pan-European
- what depth and types of data are likely to be required
- what sampling strategy should be used - opportunistic/stratified/randomisation
- what mode of investigation is likely to be best - at scene/follow-up
- what comparisons with National/Regional databases can be made for linking
- should unique accident reference numbers be included for direct linking purposes
- which variables are more reliable
- how can a common pan-European coding of variables be achieved
- what constraints will be built-in, in terms of the questions that can be answered
what Regions (across Europe ?) could be used
The decision whether to have an at-the-scene investigation or a follow-up investigation, or perhaps a combination of both, is first a question of what will be the aim of the in-depth investigation. The main goal of the implementation of an in-depth investigation is the gathering of detailed information about road crashes: vehicle information, medical data, environmental details, psychological aspects and so on. The evaluation of injury prevention measures as well as accident prevention measures is dependent on the data collection form. The following steps should be considered when designing a within-country or a pan-European in-depth accident database:
- can we identify the types of questions/estimates required
- what is the population of interest
- what is the required precision on any estimate
- are resources limited - what are the realistic limits
- what is the optimal sampling approach
- what is the best data collection approach
- what will be the constraints on the answers that can be provided
what is the acceptable delay in obtaining an answer
If estimates are to be made on a Pan-European basis then certain conditions will have to be met. The first is the acceptability of the National estimates, which in turn require the acceptability of Regional estimates. Only once these pre-requisites have been made can a Pan-European estimate be considered. Any constraints which apply at a Regional or National level will also apply at a Pan-European level. Further, all the constraints from different Countries must be taken into account, hence if one Country only has data on new cars then a Pan-European estimate including that Country can only be based on new cars. This does not necessarily mean that different sampling schemes used in different Countries cannot be combined at a Pan-European level, but it does mean that some care should be exercised before combining such data. However, it does mean that the population of interest must be defined and should be compared to the population actually sampled in order to establish if the estimation required is possible.
The broader the available data the broader the range of questions that can be addressed. This suggests that, as a policy, it may be better to broaden the data and scope of in-depth databases - but not at the expense of sample sizes. The broadening of in-depth databases implies not just more accident data but data from more accidents.
1.1.3 Work Package 3 - External Consultation
The processes of external consultation that members of the consortia developed are the main topic of this Work Package. These range from: conference papers at prestigious events throughout the world; presentations to interested groups and organisations involved in, or associated with, in-depth crash investigation; a workshop attended by a selection of delegates from the areas of research, government and manufacturing; and the development of a World Wide Web site
in order to disseminate the group's work and findings to as large an audience as possible.
1.1.4 Work Package 4 Workshop - Presentation of Results
This is a full report on the workshop mentioned above. The idea of the workshop was to begin a two-way discussion between the interested parties and the STAIRS partners. It was hoped that from these discussions the protocols developed could better reflect the wants and needs of a wider target audience. Presentations were given on all of the areas in Work Packages 1 and 2 to a selection of delegates from the differing areas of research, government and vehicle manufacturing. Discussion and feedback was encouraged at key points throughout the day. The workshop was discussed by the group and it was agreed that the outcome was extremely positive. The feedback was helpful, and areas identified for further work or clarification.
1.2 Personnel involved in the STAIRS project
Transport Safety Epidemiology Laboratory,
Direction Régionale Rhône-Alpes,
Centre de Lyon-Bron,
25, avenue François Mitterrand, Case 24,
F-69675 BRON CEDEX
Vehicle Safety Research Centre,
Leicestershire. LE11 3UZ
Dr. Ingo Koßmann
D-51401 Bergisch Gladbach,
Dr. Bernard LAUMON
Transport Safety Epidemiology Laboratory,
Direction Régionale Rhône-Alpes,
Centre de Lyon-Bron,
25, avenue François Mitterrand, Case 24,
F-69675 BRON CEDEX
CETE, Sud-Ouest-Ministere des Transports,
Rue Pierre Ramond,
33165 St Médard en Jalles,
Accident Research Unit,
Medical University Hanover,
OE 6290, D-30623
Old Wokingham Road,
Berkshire. RG45 6AU