Article Text


Translating road safety into health outcomes using a quantitative impact assessment model
  1. Stijn Dhondt1,
  2. Ali Pirdavani2,
  3. Cathy Macharis3,
  4. Tom Bellemans2,
  5. Koen Putman1,4
  1. 1Department of Medical Sociology and Health Sciences, Vrije Universiteit Brussel, Brussels, Belgium
  2. 2Transportation Research Institute (IMOB), Faculty of Applied Economics, Hasselt University, Diepenbeek, Belgium
  3. 3Department MOSI-Transport and Logistics, Vrije Universiteit Brussel, Brussels, Belgium
  4. 4Interuniversity Centre for Health Economics Research (I-CHER), Vrije Universiteit Brussel, Brussels, Belgium
  1. Correspondence to Stijn Dhondt, Department of Medical Sociology and Health Sciences, Vrije Universiteit Brussels, Laarbeeklaan 103, Brussels B-1090, Belgium; stijn.dhondt{at}


Introduction The majority of traffic safety policies are limited to preventing mortality. However, non-fatal injuries also impose a significant risk of adverse health. Therefore, both mortality and morbidity outcomes should be included in the evaluation of traffic safety policies. The authors propose a method to evaluate different policy options taking into account both fatalities and serious injuries.

Methods A health impact model was developed and aligned with a transport and road safety model, calculating the health impact of fatalities and seriously injured traffic victims for two transport scenarios in Flanders and Brussels (Belgium): a base scenario and a fuel price increase of 20% as an alternative. Victim counts were expressed as disability adjusted life years, using a combination of police and medical data. Seriously injured victims were assigned an injury, using injury distributions derived from hospital data, to estimate the resulting health impact from each crash. Health impact of fatalities was taken as the remaining life expectancy at the moment of the fatal crash.

Results The fuel price scenario resulted in a decrease of health impact due to fatalities of 5.53%–5.85% and 3.37%–3.88% for severe injuries. This decrease was however not equal among all road users.

Conclusions With this method, the impact of traffic polices can be evaluated on both mortality and morbidity, while taking into account the variability of different injuries following a road crash. This model however still underestimates the impact due to non-fatal injuries.

  • Health impact assessment
  • road traffic injury
  • disability adjusted life years
  • traffic safety
  • policy
  • policy analysis
  • process/impact evaluation
  • disability
  • mortality
  • engineering
  • motor vehicle occupant
  • rehabilitation
  • public health
  • economic analysis
  • costs
View Full Text

Statistics from


Injuries from road traffic crashes are the most direct impact of transport on health. Public authorities at all levels draft traffic safety policies to reduce the number of victims from road traffic crashes, through European level directives (COD 2008/0062 and 2011/82/EU), the national ‘GOFORZERO’ campaign or regional targets (Aim 16 of the ‘Flanders in Action Plan 2020’).1 The majority of these policies are focused on preventing fatalities. However, involvement in non-fatal but serious injury-producing crashes may result in long-term adverse health.2 ,3 Thus, traffic safety indicators relying on mortality are insufficient;3 a measure of morbidity should be included in policy evaluations. Traffic crashes result in a wide range of injuries, with varying severities, hindering an easy comparison both among injured victims and between injured victims and fatalities. A summary health measure of population health,2 ,4 such as disability adjusted life years (DALYs), may help to overcome this methodological issue. DALYs have a proven usefulness and relevance as a population health indicator for cost-effectiveness analysis in public health and traffic safety5–7 and have been widely applied in health impact assessment, which is commonly used to evaluate transport policy measures.4 ,8

Policy evaluations need to be supported by evidence, such as local data and scientific understanding.9 Police records are the most accessible and most common data source for the evaluation of transport policy. But, it is not always possible to translate the numbers of road traffic victims into DALYs. Medical data could complement police data in the evaluation of traffic safety.10 Police data provide detailed information on traffic circumstances, which is valuable for making future projections,11 while medical data provide information on the health burden following a crash.

In this study, we apply DALYs to the evaluation of traffic safety policies. Using both police and medical data in Flanders and Brussels (Belgium), we translate the number of fatalities and serious injuries into DALYs. Policy scenarios can then be evaluated on a wider range of effects, while also providing disaggregated analysis of the impact on different subpopulations. We demonstrate this by evaluating the health impact from the current situation and an alternative policy in which fuel prices are increased 20%.


Model description

A health impact model was developed as part of a full modelling chain of exposure (distance travelled), risk (victims per distance travelled) and health impact. A transport model, predicting the amount of distance travelled for different policy scenarios, was previously aligned with a road safety model. We provide a brief overview of both models and then describe the health impact model.

Transportation and traffic safety modelling

For the transport modelling, the activity-based model Forecasting Evolutionary Activity-Travel of Households and their Environmental Repercussions (FEATHERS) was used.12 In general, FEATHERS models the pattern of activity and travel episodes for each individual within the population by a scheduling model.13 Scenarios corresponding to particular policy measures consist of changes in this scheduling model. The impact of a fuel price increase of 20% was assessed as a possible motivation to reduce distances travelled by individual motorised transport.14 Changes in travel demand, vehicle choice and reallocation of activities were estimated.15 This alternative scenario, the ‘fuel’ scenario, was compared with a base scenario, both for the year 2007. The transport model provided the distance travelled (kilometres) for both scenarios for each age stratum (18–34, 35–54, 55–64 and >64) and gender category as well as four road user categories: drivers (car driver/motorcyclist), car passengers, bus and slow modes (including pedestrians, cyclists and mopeds).

The number of traffic victims for the base scenario is derived from the police register in 2007. In Belgium, traffic crashes with injuries must be reported to the police, providing detailed information on the circumstances of crash and the victims.16 For the fuel scenario, victim rates (victims per distance travelled) were derived from both fatal (deceased within 30 days after the crash) and hospitalised traffic victim data and the distance travelled estimated in the base scenario. We omitted slightly injured victims (those not hospitalised according to the police), as these numbers are highly under-reported.

Victim rates were derived for the four road user categories, conflict type (single or multiple vehicle crash), and the age and gender categories corresponding to the transport model. Disaggregation of victim rates permits different risk behaviour to be taken into account as well as changing mobility patterns of the different subgroups.17 ,18 Next, the estimated victim rates were multiplied by the predicted distance travelled by subgroup for the fuel scenario. This provided the number of fatalities and hospitalised victims stratified by road user, conflict type, age and gender.

Health burden modelling

Knowing the number of fatal and hospitalised traffic victims for both scenarios, DALYs were estimated (figure 1). We used injury data from hospital registries to assign a plausible distribution of injuries to the population of non-fatally injured traffic victims. We used a Monte Carlo simulation (with 10 000 iterations) performed with Matlab R2011b (Mathworks, Natick, Massachusetts, USA) to incorporate the variability and uncertainty around the parameters in estimating the final DALY.

Figure 1

Health impact model: horizontal bars represent the different steps in the health impact modelling; italic descriptions represent model input.


DALYs consider the impact of mortality (years life lost; YLL) along with morbidity (years lived with disability; YLD) at the same time,5 ,19 simplifying comparisons between different health outcomes and/or subgroups in a population. YLL represents the time lost due to premature mortality and was calculated using the formula as endorsed by WHO.5 YLD represents the healthy time lost while living with a disability. To calculate the YLD, the numbers of patients with injuries are multiplied with their respective disability weight, and duration to recovery or death.

Injury sampling

Injury data were obtained from nationwide hospital discharge data (Minimal Clinical Data), governed by the Federal Public Service of Health, Food Chain Safety and Environment. Selection was restricted to people from Flanders and Brussels, ICD-9-Clinical Modification transport-related external causes (E-codes: E810-816, E818-819 and E826) and the years 2003–2007 (n >48 000). To minimise double counting, we excluded transfers, day-patients and readmissions.20 ,21 Additionally, admissions lacking information on conflict type were excluded. Excluded admissions did not differ in their injury distribution.

We included 33 000 admissions in the analysis. Injury information was gathered from the principal diagnosis (reason of admission) and was aggregated to the EUROCOST classification.22 ,23 The EUROCOST classification comprises of 39 injury groups and offers a link with disability. Both the medical and police registries use the same definition of seriously injured traffic victims (hospitalised for more than 24 h).

Victims in both scenarios were assigned with an injury from the hospital data through stratified random sampling with replacement. One injury diagnosis per victim was sampled. Stratifications were based on age, gender, road user and conflict type, as these variables were found to have a significant effect on injury pattern.24 ,25 No information on conflict type was available for the ‘bus’ category. Variable definitions were identical between the police and medical dataset, except for road user. The hospital dataset has a separate category for motorised two-wheeled vehicles, while the transport model partly incorporated these in the drivers group for heavy motorcycles and in the slow modes for mopeds. Based on the description of road user (type of motorcycle) in the original police registry, it was possible to derive the proportions for drivers and slow modes. A proportion of injuries from these road user categories were then sampled from the motorised two-wheeled vehicles category in the hospital data. A binomial distribution was used to sample these proportions. The injury distribution of mopeds and heavy motorcycles was assumed to be similar.26

DALY calculation

The health consequences following an injury were calculated using the burden of injury method.27 Some injuries lead to temporary disability only, while others result in lifelong sequelae. Consequently, a distinction was made between injuries with temporary or lifelong injury consequences. The proportions of injuries with lifelong disability per EUROCOST diagnose followed a binomial distribution.

Let Hijkl represent the number of hospitalised victims for a particular injury and Fjkl the number of fatal cases, where i indexes one of the EUROCOST injury types, j age, k gender and l road user. DWi is the corresponding disability weight for injury i, either temporary (equation 1) or lifelong (equation 2). The duration of temporary disability is 1 year. Ljk is the 2007 Belgian life expectancy for a particular age and gender class.

YLDijkltemporary=Hijkltemporary×DWitemporary (1) YLDijkllifelong=Hijkllifelong×DWilifelong×Ljk (2) YLLjkl=Fjkl×Ljk (3)

Disability weights for both temporary and lifelong disabilities were based on Haagsma et al.27 We assumed a normal distribution around the mean. The disability weight from the Global Burden of Disease was used for spinal cord injuries,28 as they were not provided in Haagsma et al. 27

The transport model only provides broad age classes and therefore age was randomly sampled with replacement from the original police registries within the matching age, gender and road user categories. This sampled age was used to calculate the life expectancy needed for both lifelong YLD and YLL. The age distribution of hospitalised people was used for lifelong YLD, and for YLL (equation 3) we used the age distribution of fatalities. As there were very few victims among bus passengers, the age distribution of car passengers was applied. No age weighting or time discounting was applied.

Results are presented as median values from the 10 000 iterations with their 2.5–97.5 percentile range representing the uncertainty with assigning injuries to traffic victims, disability weights and duration estimates. Difference in health impact between scenarios was tested with the dependent samples t test (comparison of the 10 000 iterations between the two scenarios).


The fuel scenario resulted in less distance travelled compared with the base scenario. This was mainly due to a reduction in distance travelled as a driver, while other road user categories show a slight increase (table 1). This resulted in an overall decrease in number of fatalities and serious injuries.

Table 1

Output of modelling chain for both scenarios: amount of distance travelled, number of victims and health burden expressed in years lived with disability (YLD) and years life lost (YLL) for both fatal and hospitalised victims (absolute and rates per 100 000 inhabitants)

Injury distributions for seriously injured victims were comparable between the base and fuel scenario, with persistent differences in injury site based on road user and conflict type (table 2). Compared with other road users, slow modes had a higher proportion of injuries to the lower extremities. Both drivers and passengers had more injuries to the spine compared with others. The high proportion of injuries to the lower extremities in the drivers group, compared with passengers, was due to the inclusion of motorcyclists. The type of crash (multiple or single vehicle) influenced the injury pattern for all road users (χ2 test, p value for all road users <0.001). For example, within the car passengers’ category there were more head injuries in single vehicle crashes, while in multiple vehicle crashes there was a higher proportion of injuries to the spine and thorax/upper extremities. Similarly, there was a higher proportion of injuries to the thorax/upper extremities for slow modes in single vehicle crashes, while injuries to the head and lower extremities were more frequent in multiple vehicle crashes.

Table 2

Numbers and percentages of injuries of hospitalised traffic victims across different body regions for both scenarios and by single vehicle or multiple vehicle crashes

On average, 85% of injuries resulted in temporary disability (table 3), while only accounting for 9% of all YLD. The health impact due to mortality and injuries was the highest in the group of drivers in both scenarios. However, the proportion of YLD in the total health impact of the drivers was smaller compared with, for example, slow modes: 50 YLD per 100 000 compared with 192 YLL per 100 000 for the base scenario, while for slow modes this was 35 YLD per 100 000 inhabitants compared with 52 YLL per 100 000 (table 1). The larger weight of YLD in the total health impact of slow modes is due to slow modes having a greater number of injuries with lifelong consequences compared with other road users.

Table 3

Distribution of temporary and lifelong injuries and resulting disability for both scenarios

The fuel scenario resulted in a reduction of 5.17% (4.91–5.41, p value <0.001) DALY compared with the base scenario: from 21 428 DALY (18 105–25 842) to 20 324 DALY (17 108–24 643). YLD decreased from 5644 YLD (3037–9358) to 5441 (2919–9071) (table 1) or 3.59% (3.37–3.88, p value <0.001) (figure 2). The large ranges in the estimates result from the many parameters involved in the YLD calculation. These parameters are however strongly correlated between the two scenarios. YLL decreased from 15 784 (15 068–16 484) YLL to 14 883 (14 189–15 572) or a decrease of 5.69% (5.53–5.85, p value <0.001). Based on victim counts the difference between scenarios was 3.72%, with 3.58% for hospitalised and 5.02% for fatal traffic victims.

Figure 2

Change in morbidity (top) and mortality (bottom) for each road user and age category between the base and fuel scenarios. Change is expressed as the relative (%) change in victim numbers and health impact (years lived with disability and years life lost). Error bars for health impact estimates represent the 95% range.

The change in health impact of the policy scenario was not equal among road users. The fuel scenario mainly had a positive effect on drivers (figure 2), while there was a moderate increase in victims for passengers and slow modes. Looking at age specific changes, health burden estimates showed larger differences between scenarios at a younger age, as compared with estimates relying on counts. For disability in slow modes, differences between scenarios at later age were still higher or equal to differences using victim numbers only. This illustrates the importance of disability following injuries when calculating the health impact of traffic safety measures.


We illustrated an alternative evaluation method for traffic safety by combining police-based victim counts with medical injury information. Medical data are a valuable addition because police data lack accurate information on the consequences following a crash. This modelling is particularly useful for policy evaluations, such as health impact assessments, as it meets both the need to use local data and apply the most recent scientific evidence from previous studies.9 The stochastic approach allowed incorporating the variability in injuries and disability outcomes following a road crash. The calculation of DALYs gives a common unit for comparison. The disaggregated results provide information on the effect of policy on certain groups. Slow modes have, for example, more injuries with lifelong consequences than other road users, such as injuries to the lower extremities which are associated with poor health outcomes.29 By incorporating the variability in injury outcomes, policy makers can more accurately decide on the impact of their policies on certain groups, both for mortality and morbidity. Moreover, by relating the health impact to transport and road safety models, the whole chain can be assessed: from a proposed policy measure to the expected health impact, that is, exposure (distance travelled), risk (victims per distance travelled) and impact (health outcome).

To compare our population rates with other studies, road user specific rates are most reliable. However, these are not widely reported and so we looked at the totals. A French study found between 215 and 331 YLL and 130 YLD per 100 000 inhabitants,30 and a Dutch study2 reported a total of 270 YLL and 120 YLD per 100 000 inhabitants. These studies support our YLL modelling. Our YLD estimates are slightly lower, but comparisons between studies should be interpreted with caution.31 Both studies relied on medical incidence data, which usually give higher rates for non-fatal injuries than do police data32–34 and used other disability weights, which influence the DALY calculations.

Although comparisons with other studies are promising, there are some limitations to our model. Our method relied heavily on police data. These are known to underestimate hospitalised traffic victims due to misclassification of severe injuries or under-reporting.34 Therefore, until police data on the number of traffic victims improve, this approach is likely to continue underestimating the actual health impact of seriously injured victims and the real disability burden will have to be sought within the upper limits of our uncertainty ranges. This is primarily applicable to victims from slow modes, the groups showing the highest levels of under-reporting.35 Moreover, we did not take into account slightly injured people, as reliable information on these victims was non-existent in current datasets. However, experience from the UK suggests that the cumulative YLD from people treated outside the hospital exceeds the cumulative YLD of people admitted to the hospital.36 We could therefore miss a large proportion of the disability following a traffic crash.

It could also be argued that it is less favourable to use a combination of hospital and police data (eg, applying age distributions found in police data to calculate lifelong disability from medical data), as the population could differ between the two datasets (under-reporting of certain groups). However, it has been shown that Belgian hospital data correspond to police data regarding age and gender distribution.37

The definition of road user categories also differs between the two datasets. The transport and road safety model based definitions on data availability and model assumptions. We acknowledge that a road user classification in line with the medical definition of road users would be more appropriate. Injury distributions from motorcycle crashes differ from those involving either motor vehicles or slow modes.38 The conversion between the two datasets adds another parameter of uncertainty to the calculations. Nevertheless, the method we proposed shows the flexibility of using medical and police data in combination to get the best possible estimate.

The use of nationwide datasets (police and hospitals) ensured full coverage of the study area. When using hospital discharge data it is important however to recognise that not all injury diagnoses have an external code. This is not a problem in our study as we were less interested in the actual number of admitted patients due to a traffic crash, but may pose difficulties when these under-reported cases have a different injury distribution. We also only gathered injury information from the principal diagnosis. As in other studies,23 we acknowledge that this can pose difficulties, in terms of selecting diagnoses based on resource use, rather than pure clinical severity.

Another source of uncertainty is the disability information. By using Dutch disability information we assume that the disability after an injury between the two neighbouring countries is comparable. There is also still no clear consensus on how to derive disability weights, leading to a potential for error in burden estimates.31 Furthermore, as in other studies we assumed that for long-term disabilities, individuals are affected for the remainder of their lives.39 However, a detailed understanding of the duration of disability is lacking. Moreover, many injuries, particularly those to older, frail individuals, result in elevated mortality over quite a long period.28 ,31 Therefore, the results of this study may still be provisional, until these data are verified or more accurate estimates become available.

Implications for prevention

Police data are the most common source for evaluating traffic safety. As more people survive traffic crashes, policies should also aim at reducing disability following a crash. Our method calculates DALYs from police data. We also captured the uncertainty that inevitably accompanies such calculations. By applying this method to the evaluation of alternative policies, we showed that this model could provide a more comprehensive evaluation of traffic safety outcomes.

What is already known on the subject

  • Mortality numbers have become insufficient to evaluate traffic safety.

  • Disability adjusted life years (DALYs) provides a common metric to represent diverse health outcomes.

  • Police reported injury counts are difficult to translate into summary health measures such as DALYs.

What this study adds

  • A stochastic method to translate police reported injury counts into disability adjusted life years is useful and can be applied for a concrete policy evaluation.

  • Using medical data, the wide range of different health consequences following a crash can be incorporated in the evaluation of traffic safety and the burden of injury of traffic victims can be estimated more accurately.


We acknowledge Yvette Van Norden and Frits Bijleveld (Dutch Institute for Road Safety Research) for their kind contribution in carrying out the traffic safety modelling. We are also grateful to Juanita Haagsma for providing additional information on the disability parameters and Nathalie Terryn for providing us with the injury information from the Minimal Clinical Data. Furthermore, we are grateful to the reviewers for useful comments and suggestions.


View Abstract


  • Funding This work was carried out with the financial support of the Flemish Agency for Innovation by Science and Technology (Grant number IWT412) and research funds of the Vrije Universiteit Brussel.

  • Competing interests None.

  • Provenance and peer review Not commissioned; externally peer reviewed.

  • Data sharing statement The hospital data from the injury population are available from the corresponding author.

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.