Article Text

Download PDFPDF

Whiplash risk estimation based on linked hospital–police road crash data from France and Spain
  1. J-L Martin1,
  2. K Pérez2,3,
  3. M Marí-Dell’Olmo4,
  4. M Chiron1
  1. 1
    UMRESTTE, INRETS, Bron, France
  2. 2
    Agència de Salut Pública de Barcelona, Barcelona, Spain
  3. 3
    CIBER en Epidemiología y Salud Pública (CIBERESP), Spain
  4. 4
    Institut d’Investigació en Atenció Primèria Jordi Gol, Barcelona, Spain
  1. J-L Martin, UMRESTTE, INRETS, 25 avenue François Mitterrand, 69675 Bron cedex, France; Jean-louis.martin{at}


Objectives: To investigate potential risk factors for whiplash injury as a function of crash configuration and driver’s characteristics, and to provide information on over-reporting and under-reporting of whiplash.

Design: A case–control study of drivers involved in two-car injury collisions. Cases were drivers who had a diagnosis of whiplash injury, with or without another injury. Controls were drivers without diagnosed whiplash injury.

Setting: Hospital registries linked to police crash databases for Barcelona (Spain) and the “Département du Rhône” (France).

Main outcome measures: Relative risks of whiplash and 95% confidence intervals were estimated using a modified Poisson regression.

Results: Of the 8720 drivers involved in car-to-car crashes recorded in the French database, 12.2% were diagnosed with whiplash; the corresponding figure in the Spanish database was 12.0% of 7558 drivers. Female drivers and drivers in rear-impact collisions were most likely to have a whiplash diagnosis, although the absolute number of whiplash cases was greater in front and side impacts. Wearing a seatbelt, being in a heavier car, and age greater than 65 years were associated with a lower risk of whiplash injury. Drivers with other injuries were also more often diagnosed as having a whiplash injury, except the most severely injured.

Conclusions: Devices aimed at reducing the occurrence of whiplash injuries, such as dynamic headrest systems, should be adapted to the characteristics of at-risk occupants, especially women, and should address the mechanics of front and side impacts in addition to rear impacts.

Statistics from

Whiplash injury is a bone or soft tissue injury of the cervical spine which may lead to a variety of clinical manifestations. It is defined as an acceleration–deceleration mechanism of energy transfer to the neck.1 Although it is generally a mild injury, whiplash can have long-term consequences.

Whiplash is the most common injury sustained by car occupants involved in a crash. The incidence of whiplash disorders in South Australia in 2001 was greater than 300 per 100 000 inhabitants.2 Of all the personal injury claims in the UK, 85% are related to whiplash injury, and cost of litigation is about £3 billion a year.2 It has been speculated that whiplash might be overdiagnosed with the objective of obtaining insurance compensation. Conversely, underdiagnosis is also possible because of the absence of visible injury and the often delayed appearance of symptoms.

Two sources of surveillance data related to road traffic injuries are police databases, which lack detailed information on injury patterns, and hospital registries, which may not have information on the circumstances of crashes or the vehicles involved. A practical way to obtain comprehensive information is to link police and health databases, thereby augmenting the potential for studying road traffic injuries in greater detail. For the study of non-severe injuries such as whiplash, medical sources must record all crash casualties whether they are hospitalized or not. In the European PENDANT project,3 these two linked data sources, ie, police records and hospital registries, were available for both the French and Spanish partners.

The main objective of this study was to investigate potential risk factors for whiplash injury as a function of crash configuration and the characteristics of the drivers involved. A secondary objective was to provide additional information on possible over-diagnosis or under-diagnosis of whiplash, especially when no other injury has been sustained.



Data sources were hospital registries linked to police crash databases for the two areas Barcelona (Spain) and the “Département du Rhône” (France).

The Barcelona hospital data include crash victims treated in seven emergency departments in the area of Barcelona (DUHAT) (85% coverage of all road traffic emergencies in the city); 86% of these victims are outpatients. The French hospital data (Rhône registry) include records of patients cared for by emergency departments and by medical, surgical and rehabilitation departments, mobile emergency units and forensic departments, inside the geographical area of the “Département du Rhône”; 81% of these victims are outpatients.

The Spanish registry covers a mostly urban area with a population of about 1.5 million, and the French registry covers the “Département du Rhone”, which is also mostly (80%) urban, with a population of over 1.6 million. The annual number of casualties is 16 000–18 000 in the Spanish registry and 10 000–11 000 in the French registry.

At the time of the analysis, French police–hospital linkage was available for the years 1997–2003, and for the Spanish data, probabilistic linkage4 had been carried out for the years 2002–2004. There are no personal identifiers in any of the linked databases, and different linkage methods are used in each country. The key variables in the linkage process are the individual’s date of birth, sex, crash date, crash location, and type of vehicle. The Spanish system also used hospital name.

The Spanish process is fully computerised, but certain final linkage decisions are completed manually. The French process is mainly manual, although it makes great use of computer software. It allows free unformatted text data as a linking variable for crash location.5

Design and study population

The analysis was restricted to two-car collisions with information available for both drivers. Eligible subjects were identified from police data, and a hospital record was sought for each driver. When a link was found, the injury pattern was noted, in particular the presence or absence of whiplash. When no link could be established and the police considered the driver uninjured, the driver was included in the analysis. If the driver in an unlinked case was considered injured in police records, the case was excluded from the analysis (as nothing is known of the location and extent of the injuries).

We designed a case–control study of drivers involved in crashes. Cases were drivers with a diagnosis of whiplash injury. In the French database, whiplash injury is identified by the Abbreviated Injury Scale (AIS) code as 640278.1 (acute strain of the cervical spine without fracture or dislocation). This coding is only used after radiographic diagnosis of abnormal stiffness of the cervical spine. In the Spanish database, whiplash is identified by code 847.0 of the International Classification of Diseases 9th Revision, Clinical Modification (ICD-9-CM). Radiographs are not always available at the time of coding.

Controls were drivers without a diagnosis of whiplash injury. They might be uninjured or have a diagnosis of non-whiplash injury.


The outcome variable was whiplash injury. Independent variables included car impact area, driver age and sex, seatbelt use, crash location (main road or other, urban or not, crossroads or not), time period (night or day, weekend or weekday), traffic density (available for Spanish data), car mass (available for French data) and associated injury severity.

The Rhône registry uses the AIS to code injuries.6 DUHAT uses ICD-9-CM to code injuries. A comparable AIS score was derived from ICD-9 diagnoses using the method implemented in the ICDMAP90 software.7

Statistical analysis

Because the number of controls is not fixed but random, relative risks (RRs) can be estimated directly using model-based methods.89 A Poisson model was chosen for application to binomial data, and combined with a robust error variance estimation procedure10 to avoid overestimation of the RR estimate error. This was performed using SAS software (Genmod procedure).11

The results are shown as RR (95% CI). The global effects of risk factors are evaluated by the likelihood ratio test (comparison of two nested models). The results show adjusted RRs estimated by multivariate regression, including only significant factors (to the threshold of 5%).


Of the 8720 drivers involved in two-car collisions identified in the French database, 12.2% had a diagnosis of whiplash; the corresponding figure in the Spanish database was 12.0% of 7558 drivers (table 1). These percentages were higher on stratification by associated injury: 19.0% of French drivers and 35.1% of Spanish drivers with another injury (whatever the severity) sustained whiplash, whereas only 7.8% and 9.1%, respectively, of those otherwise uninjured sustained whiplash.

Table 1 Distribution of whiplash and other injuries in drivers involved in two-car collisions

Table 2 shows the proportions for whiplash and the corresponding crude RRs associated with certain crash and driver characteristics. Female drivers were more than twice as likely as male drivers to have a diagnosis of whiplash in both countries. There was a lower rate of diagnosed whiplash in the French database for younger drivers (⩽25 years old) and for older drivers (⩾50 years old). The Spanish data showed an increased risk of a diagnosis of whiplash for drivers up to 34 years old compared with the 35–49 age group.

Table 2 Crude relative risks of whiplash among drivers involved in two-car collisions: Poisson regression models

Compared with otherwise uninjured drivers, the proportion of drivers sustaining a whiplash injury was highest when they also suffered from another AIS 1 injury. This proportion was also increased when they suffered an AIS 2 injury. Spanish observations did not contain many casualties with maximum abbreviated injury score (MAIS) 3 or worse. For the French observations, the proportion of casualties with a diagnosis of whiplash was lower when they sustained an AIS 3 injury or worse.

Still considering crude estimates, the proportion of whiplash was highest in cases of rear impact (usually the vehicle struck), and lowest in cases of frontal impact (usually the striking vehicle), especially when against the rear of another vehicle.

Car mass, only available in the French database for about half the observations, appeared to be significant. The higher the mass, the lower the associated risk of sustaining whiplash.

The proportion of diagnosed whiplash was also higher among drivers who did not wear a seatbelt, although this association was only significant in the French dataset.

The risk of whiplash was slightly higher during the daytime and on weekdays. From Spanish observations, the risk of whiplash appeared higher when traffic was dense, particularly in traffic jams. It was also slightly higher on ring and arterial roads.

Table 3 shows RR of whiplash adjusted for all previous significant factors. Once adjusted, the risk factors that remained significant in both datasets were injury severity, impact area (or the type of crash), seatbelt use, age, and sex. The results show that a diagnosis of whiplash was more likely in drivers of cars hit from the rear, in female drivers, and drivers with associated, but not severe, injuries, and was less likely in drivers wearing a seatbelt and drivers over 50.

Table 3 Adjusted risk of whiplash among drivers involved in two-car collisions including only significant factors: Poisson regression models

Car mass was not considered in this multivariate analysis, as there were many missing values and no Spanish data were available. However, results not shown here using French data and including car mass in the regression (4628 observations) yielded corresponding RRs very close to those shown in table 3.


This paper analyses the risk of whiplash among drivers involved in car-to-car crashes, using linked information from health and police data sources in two areas in different countries. Its design makes it possible to control confounding factors, including certain crash characteristics and injury severity.

Main findings

We found that female drivers were two to three times more likely to be diagnosed with whiplash than male drivers. This was observed even after taking into account a number of potentially confounding variables. This higher proportion has been observed in most research papers on the subject.1214 The explanations most usually put forward are anthropometric and physiological differences, leading to differences in tolerance to mechanical loading. This could be partly addressed by changing seat characteristics in terms of shape and stiffness.15

Also in keeping with previously observations,121617 older drivers (especially those above 65) were diagnosed with whiplash less often than others. Again, this was not because they were more severely injured, as adjustments were made for this aspect.

Whiplash injuries were most likely to be observed in rear-impact collisions. This was expected, as it corresponds to the main injury mechanism suspected and has been observed in most whiplash studies.1821 However, it is worth noting that, although the risk of sustaining whiplash was higher in the case of rear-end collisions, most whiplash injuries were observed in other crash configurations as the percentage of whiplash cases associated with rear impacts was 22% for French data and 48% for Spanish data (table 2). This implies that countermeasures should not focus exclusively on rear impacts.

Seatbelt use was associated with a lower risk of whiplash injury, which was not obvious considering the probable injury mechanism. This result is interesting but needs to be confirmed by more in-depth investigations,1 with details on different possible seatbelt technologies. Car mass too was associated with whiplash risk, highlighting, from a more general point of view, the issue of compatibility between vehicles.22

Lastly, although there may be reasons for the over-diagnosis of whiplash—for example, declared by an accident victim so as to reserve the right to make an insurance claim—we did not see evidence of this bias in our data. The proportion of whiplash was clearly higher among drivers suffering from at least one minor or moderate associated injury, suggesting that the probability of whiplash depends on crash severity. This is true for other AIS 1 or 2 injuries, but when a more severe injury is observed (AIS 3 and over), the probability of observing whiplash was lower. Whiplash injuries may be underestimated when there are more severe injuries. This underestimation could occur because only more severe injuries are recorded, or because, initially, victims do not identify a whiplash injury when they are more severely injured elsewhere.

Limitations of the study

Police data are known to be under-reported. However, this analysis focused on car drivers involved in collisions, a category of police data known to be among the most complete.23 The linking process also resulted in loss of subjects present in both sources and may have led to underestimation of the number of injured drivers. The number of victims suffering from whiplash is also probably underestimated, as police data underestimate mild injuries more often than serious injuries. If all car drivers with hospital data (linked or not linked to police data) are considered, the proportion of people diagnosed with a whiplash injury (34.2% for French data; 46.0% for Spanish data) is much higher than in our linked sample. This ascertainment bias on the basis of severity is partially taken into account in the multivariate model by adjusting for injury severity.

In addition, differential whiplash ascertainment by age is possible. Younger drivers are reported less often in the police database than older ones,23 which could produce an underestimation of whiplash in young drivers. It is also possible that ignoring crashes in which no one was injured led to an underestimation of the seatbelt’s effect.

People classified as not injured by the police were included in the “controls” if not found in the medical data. A proportion of these may have had whiplash unrecognized by this study. This can result in case–control misclassification and then possible differential bias. The direction could depend, for example, on whether any association exists between rates of ascertainment and sex or age group, which is largely unknown.

Cases could also be missed because of non-diagnosed whiplash. Clinical signs (pain, stiffness) may not be immediately apparent. The existence of these false negatives could lead to an underestimation of the associated risks.

Key points

  • Factors associated with whiplash include female gender (twice the risk of male drivers) and age. Older drivers (especially those aged 65 and over) were less often diagnosed as having whiplash.

  • The risk of being diagnosed with a whiplash injury was highest in the case of rear impacts, but most whiplash injuries were observed in other crash configurations; hence countermeasures must not focus exclusively on rear impacts.

  • Seatbelt use was associated with a lower risk of whiplash injury.

Only two-car collisions are considered here to allow assessment of impact characteristics. Whiplash injuries can occur in other crash configurations. The most common crash type is single-vehicle crashes, but with low whiplash rates, perhaps reflecting the low proportion of rear-end impacts (5% in French data).

This work has focused on drivers, and considers all the drivers involved whether injured or not. Other studies have estimated the risks of sustaining whiplash as being quite similar for front seat passengers and lower for rear seat passengers.12 Whiplash rates are also much lower for other categories of people involved in crashes (according to the Rhône registry, about 5% of pedestrians and motorcyclists injured in crashes were diagnosed with whiplash, and DUHAT figures show that about 5% of pedestrians and 7% of motorcyclists were affected).

There is a natural pairing of drivers from the two cars involved in the same collision. We could have analysed data as matched data, by using conditional logistic regression, but this would have excluded a large number of drivers (the French database has only 746 pairs), leading to a considerable loss of statistical power. Pairing is therefore only taken into account by adjustment as a function of crash circumstances.

Finally, this study has not been able to address many associated issues because of a lack of information in the datasets. These include the effect of headrests, possible vehicle seat response, and airbag effects. Nor is any direct information available on impact speed (type of road and location are used as proxy), but this is partly balanced by the use of comparisons between drivers involved in the same collisions.

Strengths of the study

To obtain these results, it was necessary to combine available hospital data providing precise injury descriptions, including whiplash, with linked police data to identify two-car collisions and to discover the main characteristics of the crash, but also to identify drivers who were involved in the crash but not injured. This made it possible to count four groups of drivers: those with a diagnosis of whiplash, with or without associated injury, and those with no diagnosis of whiplash with or without another injury. We were also able to include all casualties involved in crashes and not only those that had been hospitalized, as most people suffering from whiplash are not hospitalized.

Another strength of this study is the fact that, in spite of many differences between the French and Spanish sampling and injury coding methods used, the relative risks are shown to be quite coherent, thereby giving greater credence to our estimations.


For some years, car manufacturers have been working on specific devices, such as dynamic headrest systems, designed to reduce the occurrence of whiplash injuries. This study suggests that progress could be achieved by adapting these devices to the characteristics of at-risk occupants, especially women, and addressing the mechanics of front and side impacts in addition to rear impacts.


This study is part of the Pan-European Co-ordinated Accident and Injury Databases (PENDANT) and was partially funded by the general directorate of energy and transport (UE) in the framework of the competitive and sustainable growth program (contract number GMA2-2001-52066). We thank all the staff responsible for data collection and codification, especially Amina Ndiaye (ARVAC), Irène Vergnes, Blandine Gadegbeku (INRETS), Pere Arribas, Isabel Ricart and Monica Guxens (ASPB).



  • Competing interests: None.

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.