Safety CaUsation, Benefits and Efficiency (SafetyCube) is a European Commission supported Horizon 2020 project. The project’s main objective is the development of an innovative road safety Decision Support System (DSS) that will enable policy-makers and stakeholders to select and implement the most appropriate and cost-effective strategies, measures and approaches to reduce casualties of all road user types and of all severities.
Work Package 7 is dedicated to serious traffic injuries, their health impacts and their costs. This deliverable (D7.1) focuses on the determination of the number of serious traffic injuries, defined as casualties with an injury level of MAIS ≥ 3 (MAIS ≥ 3 casualties).
BACKGROUND AND OBJECTIVES
Crashes also cause numerous serious traffic injuries, resulting in considerable economic and human costs. Given the burden of injury produced by traffic, using only fatalities as an indicator to monitor road safety gives a very small picture of the health impact of traffic crashes, just the tip of the iceberg. Moreover, in several countries during the last years the number of serious traffic injuries has not been decreasing as fast as the number of fatalities. In other countries the number of serious traffic injuries has even been increasing (Berecki-Gisolf et al., 2013; IRTAD Working Group on Serious Road Traffic Casualties, 2010; Weijermars et al., 2015).Therefore, serious traffic injuries are more commonly being adopted by policy makers as an additional indicator of road safety. Reducing the number of serious traffic injuries is one of the key priorities in the road safety programme 2011-2020 of the European Commission (EC, 2010).
To be able to compare performance and monitor developments in serious traffic injuries across Europe, a common definition of a serious road injury was necessary. In January 2013, the High Level Group on Road Safety, representing all EU Member States, established the definition of serious traffic injuries as road casualties with an injury level of MAIS ≥ 3. The Maximum AIS represents the most severe injury obtained by a casualty according to the Abbreviated Injury Scale (AIS).
Traditionally the main source of information on traffic accidents and injuries has been the police registration. This provides the official data for statistics at national and European level (CARE Database). Data reported by police usually is very detailed about the circumstances of the crash particularly if there are people injured or killed. But on the other hand police cannot assess the severity of injuries in a reliable way, due, obviously to their training. Therefore, police based data use to classify people involved in a crash as fatality, severe injured if hospitalised more than 24 hours and
slight injured if not hospitalised. Moreover, it is known that even a so clear definition as a fatality is not always well reported and produces underreporting. This is due to several factors such as lack of coverage of police at the scene or people dying at hospital not followed by police (Amoros et al., 2006; Broughton et al., 2007; Pérez et al., 2006).
Hospital records of patients with road traffic injuries usually include very little information on circumstances of the crash but it does contain data about the person, the hospitalisation (date of hospitalisation and discharge, medical diagnosis, mechanism or external cause of injury, and interventions). Hospital inpatient Discharge Register (HDR) offers an opportunity to complement police data on road traffic injuries. Medical diagnoses can be used to derive information about severity of injuries. Among others, one of the possible scales to measure injury severity is the
Abbreviated Injury Scale (AIS).
The High Level group identified three main ways Member States can collect data on serious traffic
injuries (MAIS ≥ 3):
- by applying a correction on police data,
- by using hospital data and
- by using linked police and hospital data.
Once one of these three ways is selected, several additional choices need to be made. In order to be able to compare injury data across different countries, it is important to understand the effects of methodological choices on the estimated numbers of serious traffic injuries. A number of questions arise: How to determine the correction factors that are to be applied to police data? How to select road traffic casualties in the hospital data and how to derive MAIS ≥ 3 casualties? How should police and hospital data be linked and how can the number of MAIS ≥ 3 casualties be determined on the basis of the linked data sources?
Currently, EU member states use different procedures to determine the number of MAIS ≥ 3 traffic injuries, dependent on the available data. Given the major differences in the procedures being applied, the quality of the data differs considerably and the numbers are not yet fully comparable between countries. In order to be able to compare injury data across different countries, it is important to understand the effects of methodological choices on the estimated numbers of serious traffic injuries.
Work Package 7 of SafetyCube project is dedicated to serious traffic injuries, their health impacts and their costs. One of the aims of work package 7 is to assess and improve the estimation of the number of serious traffic injuries.
The aim of this deliverable (D7.1) is to report practices in Europe concerning t he reporting of serious traffic injuries and to provide guidelines and recommendations applied to each of the three main ways to estimate the number of road traffic serious injuries.
Specific objectives for this deliverable are to:
- Describe the current state of collection of data on serious traffic injuries across Europe
- Provide practical guidelines for the estimation of the number of serious traffic injuries for each of the three ways identified by the High Level Group
- Examine how the estimated number of serious traffic injuries is affected by differences in methodology.
METHODS USED
The practical guidelines for the determination of the number of serious traffic injuries were developed using:
- A survey carried out among experts in EU Member States
- Current practices and experiences from a number of countries
- Specific analysis in which different procedures were applied to the same data.
A survey was carried out among experts in EU Member States in order to provide an overview of the data and procedures that are applied for estimating the number of MAIS ≥ 3 casualties across Europe. The questionnaire was inspired by the a survey that had been conducted by FERSI (Auerbach and Schmucker, 2016). The 72 questions of the survey were clustered in six groups: (1) Prime contact in the country; (2) General information on collection practices and responsibilities; (3) MAIS ≥ 3 methodology and planned changes; (4) Detailed information on hospital data; (5) Detailed information on applied method; (6) Concrete figures: fatalities and serious injuries police / MAIS ≥ 3.
Current practices and experiences from some countries allowed to explore the following topics:
- Methods to apply correction factors have been explored using data from Belgium, France and Austria
- Inclusion and exclusion criteria have been defined using Hospital Discharge Data based from Spain and the Netherlands. It includes a sensitivity analysis of the impact of using different inclusion/exclusion criteria.
- A sensitivity analysis has been carried out to assess the impact of obtaining MAIS ≥ 3 using different methods, either coding AIS directly or recoding from ICD diagnosis with a conversion tool. We used data from Spain, Belgium, the Netherlands and Germany.
- How to derive the number of serious traffic injuries using police and hospital record linkage has been explored with data from France, the Netherlands and Slovenia
Finally, a comparison of three methods proposed by the High Level Group to estimate the number of seriously injured (factors, hospital and record linkage) was carried out using data from the Netherlands.
MAIN RESULTS
State of data collection on serious traffic injuries across Europe
As of June 2016, 17 of the 26 countries that responded to the survey had either delivered MAIS ≥ 3 estimates to DG-MOVE – or had reported that they would be in the position to do so shortly. In the remaining 14 countries, the process for estimating the number of MAIS ≥ 3 traffic injuries appeared to be only in a very early stage or had not even started yet. One of the central problems in these countries, due to privacy regulations, was to get access to hospital discharge data.
The methods for estimating the number of MAIS ≥ 3 casualties differ between the countries. Most of the countries (9) use only hospital data, whereas two countries apply corrections to police data and four countries use a linkage of police and hospital data. France and Germany apply a combination of methods: in France a generalization based on the Rhone Trauma Register and Germany a generalisation based on GIDAS in depth data and data from the German Trauma Register DGU. Several countries plan to modify their method in the future, the majority of them towards linking police and hospital data.
As of June 2016, 13 of the 26 countries that responded to the survey had MAIS ≥ 3 estimates for 2014 readily available. The ratio of MAIS ≥ 3 casualties and fatalities differs considerably between these countries, from o.6 MAIS ≥ 3 casualties per fatality in Poland to 13.2 MAIS ≥ 3 casualties per fatality in the Netherlands. This difference illustrates the considerable differences between the methodologies used and also indicates that extreme care should be taken in comparing national estimates on MAIS ≥ 3 at this stage.
Application of correction factors to police data
Basically, the first method proposed by the High Level Group estimates the actual or the registered number of MAIS ≥ 3 casualties on the basis of the number of casualties that is registered by the police.
Both previous research and current practices show that correction factors vary substantially between countries. This variability is due to the variation between police registrations, hospital registrations and the distribution of traffic injuries across countries. Thus, correction factors are country specific and it is strongly recommended not to apply the correction factors used in one country to another country.
In order to determine the correction factors on police data, it is imperative to have access to at least a sample of hospital data. Such a sample could be from part of the country routine data (e.g. as is currently the case in France and Germany) and/or for a limited time period (like in Belgium). Using such correction factors starts from the assumption that there is relative stability in both police and hospital registrations of casualties over time. However, as shown by the comparison of the three methods using Dutch data, the accuracy of police and hospital registration may change over time. Therefore, correction factors need to be validated and updated on a regular basis.
Since the accuracy of police registrations differs between road user groups (age, gender, transport mode) and accident types (single vs. multi-vehicle, place of occurrence, etc.), it is necessary to derive and apply different correction factors for different groups of road users. A first useful step to determine such correction factors is to model the effects of a series of variables (such as year, type of road user, age, gender…) on the ratios of police/hospital registrations. This step allows to identify which variables significantly affect these ratios and consequently it is possible to determine a series of correction factors on police data in order to predict the number of hospital registrations
Using hospital data
The availability of hospital data is essential for the determination of the number of serious traffic injuries. When such data is available all over a country and can be accessed easily and timely, it can be used to determine the number of MAIS ≥ 3 traffic injuries. The main source for hospital data is the Hospital Discharge Register (HDR) that includes all hospitalisations for diseases and injuries from all or some public and/or private hospitals of the country. Hospital data is not always accessable for institutions that are responsible for the determination of the number of serious traffic injuries. Such data is indeed often highly protected by privacy legislation because it includes very sensitive information such as individual health information. However, practice from different countries shows that it is possible to anonymize the data in such a way that it is not possible to identify a particular person, and hence such data can be made available and accessible for research or statistical purposes. At national level it is advised to establish inter-sectorial collaboration between the health and the transport or interior ministries in order to facilitate the access to HDR data in view of calculating the MAIS ≥ 3 numbers. At European level, it is recommended to reinforce institutional collaboration between the European Commisison (DG MOVE), Eurostat, OECD-IRTAD and WHO to facilitate and improve reporting serious road traffic injuries in Europe.
Recording and handling systems of the HDR differ by country so the data should be compared with caution. Moreover, MAIS ≥ 3 road traffic casualties should be selected from the hospital data. This can be done in several ways and also this process influences the estimated number of serious traffic injuries. In this report we analysed (1) the effects of applying different in- and exclusion criteria to select road casualties from hospital data, and (2) the difference between direct AIS coding and the use of various recoding tools for the determination of MAIS.
In- and exclusion criteria to select road casualties from hospital data
All methods used for estimating the number of serious traffic injuries (MAIS ≥ 3) are in one way or another based on a selection of hospital records. So it is important to have clear criteria for inclusion or exclusion of hospital records in order to establish the population of people injured in traffic.
Hospital discharge registers use the International Statistical Classification of Diseases (ICD) to codify the main diagnosis, or reason for the hospital admission. Currently, hospital data in Europe are coded with either ICD-9 or ICD-10. It is on the basis of these codes that traumatic injuries can be identified. According to the ICD9-CM (clinical modification) specification codes 800 to 959 refer to injuries. When using ICD10 the range S00-T88 relates to injuries.
Since not all injury patients admitted to the hospital are road traffic casualties, one also needs to know the injury mechanism in order to properly identify traffic injuries. This can be done on the basis of the codes for external causes (the E-codes) that are part of the ICD nomenclature - provided that such a code has been allocated (which is not always the case). When identifying traffic injuries, it is recommended to include records with the following E-codes: E810-E819, E826, E827, E829 and E988.5 and excluding E828.
For non-motorized vehicles, the ICD9 coding scheme does not make a distinction between “Traffic accident” (any vehicle accident on a public road) and “Non-traffic accident” (any vehicle accident occurring entirely somewhere other than on a public road). There is a specific code to designate the place of occurrence of the event (E849) but usually it is not reported. Thus, on the basis of the Ecodes alone, the number of traffic casualties may be somewhat overestimated. In order to avoid this, one may use other codes for casualties (if available) and/or weight or correct for non-public traffic accidents.
It should be noted that several countries suffer from incomplete specification of external causes in their hospital injury records. In Belgium for example, despite the compulsory registration of E-codes in hospitals, E-codes are missing for almost 20% of all casualties (although the percentage of missing causes for traffic injuries is probably lower). This leads to an underestimation of the number of traffic injuries. Some countries look for other variables to identify traffic injury cases like the insurance company that pays the hospitalisation (vehicle insurance).
Persons who die within 30 days after the accident should be excluded from the hospital records, as they are counted as a fatality. Another group to exclude from the numbers are the readmissions, in order to avoid duplicates. On the basis of data from Spain and the Netherlands, we estimate that the inclusion of fatalities within 30 days results in an overestimation of the number of serious injuries of about 5% and that inclusion of readmissions results in an overestimation of about 3%. To account for these differences, weighting factors can be applied.
It should finally be noted that not all MAIS ≥ 3 traffic casualties end up being hospitalized. Based on data from France, it appears that the exclusion of non-hospitalized MAIS ≥ 3 casualties results in an underestimation of the number of serious traffic injuries of roughly 5%.
The impact of different coding mechanisms on the number of MAIS ≥ 3 casualties
The AIS level of injuries can be determined in several ways. AIS coding can be direct, i.e. when traffic victims are registered, an AIS code is given for each of the injuries (or diseases) of the casualty. In Europe, such direct AIS coding is not very common however. In most countrues, AIS codes can be derived from other injury coding systems, like ICD. Currently the following conversion tools are available to derive AIS from ICD codes: ICDmap90, ICDpic, DGT, ECIP, AGU or AAAM. The use of any of these conversions tools leads to the so-called ICD-derived AIS values. Some of these tools recode the ICD codes into the latest AIS© 2005/update 2008 codes, but other older tools recode ICD data into AIS codes that are based on previous versions of the AIS coding (AIS2005, AIS1998 or AIS1990). Recoding always has the disadvantage compared to direct coding, that some information gets lost or is not available so that a best match must be selected (in the recoding tool). This may have an effect on the severity that is assigned to a casualty and therefore also on the estimated number of MAIS ≥ 3 casualties.
Application of AIS1990/AIS1998 results in an overestimation of the number of MAIS ≥ 3 casualties by 12%. So in order to make data from different countries more comparable, the number of MAIS ≥ 3 casualties should be multiplied by a factor 0.89 when injuries are coded in AIS1990 or AIS1998 instead of AIS2005 or AIS2008.
In some cases, only a limited number of diagnoses is coded or available for analysis. The analyses conducted show that in case only 1 diagnosis is available, this leads to an underestimation of the number of MAIS ≥ 3 casualties of 22%; when only 3 diagnoses are available, the underestimation is around 5%. The principal diagnosis is not always the most severe, as also other criteria, like financial issues determine the principal diagnosis.
The estimated number of MAIS ≥ 3 casualties was compared for direct AIS coding and the ECIP conversion tool, using a small sample of German data. This analysis suggests that the ECIP recoding tool seems to result in reliable numbers of serious traffic injuries if codes are complete. The difference between the numbers generated by the different ICD9cm recoding tools is at most 7%. We were not able to investigate the difference between ICD9 tools and direct coding.
The majority of European countries now use a tool from AAAM (“Association for the Advancement of Automotive Medicine”) that has been provided by DG-MOVE. It became obvious, however, that the US-based AAAM10 table does not yet provide satisfying transformation rates for the ICD versions currently being used in Europe. The conversion algorithm actually uses ICD10CM. As most European countries use an older version of ICD10 without clinical modification, this does not fit with European practice.
Due to hospital practice or privacy regulations, some countries use 4-digits injury codes instead of 5-digits codes. The effects of this truncation depend on the recoding tool that is applied. AAAM10 and ICDpic do not seem to be able to deal well with truncated codes. Countries that use AAAM10 in combination with truncated injury codes report fail rates of the ICD to AIS transformation of about 20%. In case other tools then ICDpic and AAAM10 are used, underestimation is between 3% and 10%.
Using linked/matched police and hospital data
A third method for estimating the actual number of MAIS ≥ 3 injuries is linking police data and hospital data. The main benefit of such a data linkage is that it leads to a maximal use of the available data sources. The process also provides insight in the completeness of police and hospital data and also allows to identify, and possibly to reduce, selection biases and underreporting. In this way, one could for example correct for missing or misspecified external causes (E-codes) in hospital data.
The linking process is based on one or more variables that are included in the records of both databases. Ideally this variable is a unique personal identification number, which allows to identify 1-to-1 linkages and apply a relatively easy and straightforward deterministic linking. However, this variable is often unavailable in one or both databases for privacy reasons. In the absence of a such unique identifier, it is possible to apply a so-called probabilistic or distance-based linking process based on several variables at once. Linking variables that are commonly used are date and time of the crash (and/or date and time of hospital admission), location of the crash, gender and date of birth (or age) of the casualty, mode of transport.
After the linking of hospital and police data has been completed, the number of traffic casualties recorded in hospital data but not identified as such can be estimated using the capture-recapture method. It is a method, which uses data linkage to estimate a total population. For example, by knowing numbers of casualties recorded by the police, by hospitals and by both (linked records), it is possible to estimate a lower bound of the number recorded by neither, and therefore to estimate the total number of casualties. To be valid, the capture-recapture approach must meet six conditions. Among them, three are particularly important: (i) the definition of the road casualty in the two data sources should be the same or included in one another; (ii) the two registrations are supposed to be independent; (iii) all subjects of interest should have the same probability of being registered by a given source. This third assumption is usually only valid within subgroups (e.g. mode of transport). These subgroups should therefore be taken into account by means of stratification or modelling.
Influence of the method on the estimated number of serious traffic injuries
Comparing the three methods proposed by the EC using data from the Netherlands illustrates that linking of police and hospital data has the potential to identify the highest number of serious traffic injuries, i.e. to have the lowest level of underreporting. Correction factors applied to police data identified the fewest MAIS ≥ 3 casualties in the Netherlands. In this case, this is due to a decrease in police registration level. Also in Austria, applying correction factors to police data resulted in lower number of serious traffic injuries than use of hospital data. This doesn’t mean that applying correction to police data always results in underestimation of the number of MAIS ≥ 3 casualties. Crucial requirements for correction factors to lead to reliable estimates of MAIS ≥ 3 casualties are the quality of the variables recorded and the stability and consistency of both police and hospital registrations.
OVERALL CONCLUSIONS AND RECOMMENDATIONS
The adoption of a common definition for serious injuries has certainly given an impetus for the collection of data on serious traffic injuries in the EU-Member states. However, in many countries the process for estimating the number of MAIS ≥ 3 traffic injuries is still in a very early stage or has not even started yet. One of the central problems in these countries is the restricted access to hospital discharge data due to privacy regulations.
Hospital data are essential for determining the number of serious traffic injuries, defined as MAIS ≥ 3 casualties. Even when applying correction to police data, it is necessary at some point to have hospital data to derive the correction factors. When hospital data are properly anonymized in a way in which it is not possible to identify a person, they should be available for research or statistical purposes. Thus, more efforts are needed in Europe to make hospital data available in such a way that an accurate estimate of the number of serious traffic injuries can be made. This implies at least the availability of 4 diagnoses of injuries, no truncation of ICD codes, registration of E-codes, and the use of the latest version of AIS (2008). To this end, there should be more inter-sectorial collaboration between the health and the transport actors at national and international level.
The methods for estimating MAIS ≥ 3 vary considerably across countries, the differences seem to be heavily determined by the data available. The methods used clearly affect the estimated numbers of serious traffic injuries. Factors like in- and exclusion criteria applied, missing E-codes, AIS version, ICD-AIS recoding tool applied and the number of injuries taken into account when determining MAIS, can have a large influence on the estimated number of MAIS ≥ 3 casualties. It is important to discuss, report and interpret the estimation results taking also into account the different specific methodologies they are derived from.
For policy purposes, it is important to be able to monitor changes over time. For this purpose, it may be sufficient to use a method, which is less accurate i.e., is known under- or overreport the number of seriously injured. As long as any under/overreporting remains consistent across years it will still be possible to observe important trends in serious traffic injuries.
It is recognized that all three methods for estimating the number of serious traffic injuries – (1) applying correction factors to police data; (2) use of hospital data; (3) linking police and hospital data – have both advantages and limitations. Which method(s) to choose will depend on the context and constraints of each individual country. When using correction factors on police data, it is important to be assured of the stability of the police registration practice, and to have regular access to at least a sample of high quality hospital data. Also, it may be necessary to apply correction factors to hospital data, if there is evidence that some MAIS ≥ 3 traffic injuries are not identifiable as traffic victims within the hospital data. Whenever possible, an attempt should be made to link hospital with police data. This allows identifying road traffic casualties that are not recognisable as such in the hospital data and therefore provides a more accurate estimate of the number of serious traffic injuries.
Further harmonisation of methods over the next years is desirable in order to ensure that the estimated numbers of MAIS ≥ 3 road traffic injuries are comparable across Europe.
