Methods

The simulations

We selected two policy options, which have actually been under consideration by the Dutch government, to reduce exceeding air pollution limits near highways and to improve mobility. Possible effects of these interventions on air pollution, noise and traffic intensities had been investigated by the Department of Waterways and Public Works.13 14 Our HIAs were based on these previous investigations but were no part of the decision-making process. Supplementary fig A shows where the involved highways are located.

The first simulation concerned speed limit reduction by 20 km/h at nine highway sections in the Netherlands where air pollution levels are frequently exceeding limits, representing about 0.1% of all highways in the Netherlands. The speed limit changed from 100 to 80 km/h at seven of those sections and 120 to 100 km/h at the other two sections. This intervention has been introduced at five of these nine highway sections in the meantime.

The second simulation was the construction of a new highway section (dotted line in supplementary fig A), to improve mobility in the corridor Schiphol-Amsterdam-Almere. Traffic intensities in this corridor in 2020 had been estimated with and without this intervention with a traffic model.13

Health impact assessment

We basically followed the steps of environmental impact assessment, as described by Hertz-Picciotto15 (see supplementary fig B). We identified six steps in the HIA process, which are described below. Steps 3–6 were performed twice in each simulation, to estimate the disease burden before and after the introduction of the transport policy. The net health impact of the simulated policies was estimated by comparing the estimated pre- and post-policy disease burden, as proposed by Joffe and Mindell.12

Step 1: Selection of exposures

As the risk factor under study — transport — is related to a range of exposures, the first step of this HIA was to decide which needed to be included. First, noticeable effects of our simulated policies would probably include local changes in the emission of air pollution and noise, which is why we have focused on those factors. In the speed limit reduction simulation, we also included traffic accidents, because it has been shown previously that speed limit reduction reduces the number of traffic accidents, and thereby injuries and fatalities.16 17 As it is not allowed to cycle or walk on the involved highway sections, it is not likely that physical activity, an important determinant of health, is directly affected by these interventions. Therefore, psychical activity was not included, although more indirect effects, for example, by changing choices of transport modes because of more or less traffic jams cannot be excluded.

We used NO₂ levels as an indicator of all traffic-related air pollutants, which is widely accepted.18 Particulate matter with a diameter ⩽10 μm has also been used as an indicator in epidemiological studies, but is less specific for traffic-related air pollution.18

Step 2: Selection of health endpoints

We selected health endpoints of which WHO18 19 20 and the Dutch Health Council21 22 concluded that there is evidence of a causal relationship with traffic-related air pollution and road traffic noise and for which it is likely that they occur at typical levels near roads. Generally, only diseases which impair people’s daily functioning were included. Elevated blood pressure by noise exposure, for example, was excluded because this does not necessarily impair daily functioning. On the other hand, we did include annoyance and sleep disturbance, because they affect wellbeing, which is included in WHO’s definition of health.23 Also ischaemic heart disease due to noise was included. Health effects of short-time increases in air pollution, such as myocardial infarction, were not included because no daily air pollution data were available.

Step 3: Population exposure

We assumed that all people living within 500 m of the involved highway sections were “at risk” of adverse traffic-related health effects. Air pollution levels drop to background levels within 300 m from main roads.18 Noise levels also drop to background levels within a few hundred metres from roads, but the dispersion heavily depends on local characteristics, such as the location of buildings. As noise effects may extend beyond 300 m from the highways, an impact zone of 500 m was applied in the simulations.

Demographic data, that is, how many people live in the involved areas of the simulations, were combined with exposure distribution data to estimate population exposure, using a geographical information system (GIS). Demographic data on postal code level were obtained from Statistics Netherlands. Background exposure distribution data were provided by the Netherlands Environmental Assessment Agency (MNP), which yearly produces noise and air pollution maps of the Netherlands, with a resolution of 25×25 m and 1000×1000 m, respectively. Models underlying these maps combine information of the location of different sources of air pollution and noise (road traffic, aircraft, railroad traffic, and industry) with characteristics of those sources (emissions, traffic intensity, velocity, type of vehicles), and characteristics of its environment (buildings, noise barriers). Corrections are made for meteorological conditions.24 25 In simulation 2, local changes in air pollution and noise caused by the intervention were calculated using the same emission dispersion models of MNP as those underlying the maps, though restricted to road traffic. For air pollution, a local component was added to the large-scale model, adding local pollution by busy and main roads. For simulation 1, reduction percentages of noise and air pollution levels had been previously published13 and these were applied to each distance from the highway sections after expert consultation (W F Blom, personal communication, 2006), with a maximum of 300 m for air pollution and 500 m for noise. Last but not least, we assumed that traffic flow was not affected (eg, no larger number of cars or different car fleet).

Step 4: Exposure-effect relations

Consultation of experts (E E M M Van Kempen and P Fischer, personal communication, 2006) and a brief literature investigation, focusing on meta-analyses and reviews, were used to select exposure-effect relations for the selected health endpoints. For the relation between road traffic noise and cardiovascular disease the threshold of no-effect is still debatable. In line with our previous national study,11 we assumed 50 dB(A) was the threshold value. For traffic-related air pollution, no meta-analyses were available for the selected health endpoints, and we selected two often cited Dutch epidemiological studies which used NO₂ levels as an indicator of traffic-related air pollution.26 27 Results of these two studies were generally in line with results of other studies.28 29 30 We assumed that (inter)national exposure-effect relations were valid for the population under study.

Step 5: Estimation of the number of cases

In this step, the attributive burden, that is, the extra number of cases due to air pollution or noise from the involved highways, was calculated. The attributive burden is a function of the relative risks, baseline prevalence or incidence rates of the health endpoints under study and the number of people exposed.31 For more detail see Knol and Staatsen.11 National baseline prevalence rates of the health endpoints and mortality rates were obtained through the National Public Health Compass of RIVM32 and Statistics Netherlands. Unfortunately, local prevalence rates of diseases under study were not available. Therefore, we used national baseline prevalence rates, thereby assuming that national rates were similar to local rates. For ischaemic heart disease and cardiopulmonary mortality, national rates by sex and 5-year age groups were available, showing highest rates in elderly men. As a consequence, this group would be expected to benefit the most from reductions in exposure, if we assume that the exposure-effect relation for this group is similar to that for the rest of the population. For wheeze, no gender- and age-specific prevalence rates were available. Exposure-effect relations were not applied to age or sex groups other than for which they had been reported. We assumed that changes in the number of people severely annoyed or disturbed in their sleep were independent of changes in attitude towards noise exposure by the intervention, because — to our knowledge — limited quantitative data are available regarding this kind of “social effect” of traffic interventions.

The Dutch Institute for Road Safety Research provided estimated numbers of deaths and severe injuries due to accidents before and after the intervention, as a function of the measured changes in driving speed by the actual introduction of the new speed limits at some highway sections of simulation 1.33

Step 6: Disease burden

For each health endpoint, the disease burden was calculated by multiplying the number of attributable cases with a severity weight and estimate of the duration of the disease, or years of life lost for mortality11 (see supplementary table A for weight and duration factors applied). A Monte Carlo uncertainty analysis has been applied to all computations to estimate a 90% prediction interval around the disease burden. We included the uncertainty ranges (based on literature or expert judgements) for the exposure-effect relation, severity factor and duration (the latter only when incidence data were used). We used @Risk software version 4.5 (Palisade Decision Tools, Ithaca, New York, USA) to execute these simulations. The presented intervals only relate to the statistical uncertainty of the included input variables. Other types of uncertainty, like uncertainty related to assumptions, have not been included.

In order to facilitate (international) comparisons with other burden of disease estimates, and/or to adjust for changes in population size over time, it can be useful to express the disease burden per, for example, 1000 people. However, as the aim of our study was to compare the pre- and post-intervention burden of disease within each simulation, and not to compare the burden of disease across simulations or interventions or over time, we focus on the absolute burden of disease.

Results

Results of steps 1, 2 and 4, which merely define the context of this HIA and the choice of boundaries, have been summarised in table 1, together with the quantitative results described below.

Step 3: Population exposure

Approximately 93 000 and 60 000 people lived within 500 m of the involved highway sections in simulations 1 and 2, respectively. Table 1 shows for each health endpoint the size of the “population at risk”, which is the part of the population in the age category to which the corresponding exposure-effect relation referred.

Although the new highway section of simulation 2 was planned in a less densely populated area as compared with the areas in which the existing highways were located, approximately 3000 extra people came to live close to a highway, because the new section did not replace an existing highway section. However, as some part of traffic on the old highway section was simulated to move to the new highway section, reducing traffic-related emissions in the densely populated area, overall population exposure in the Schiphol-Almere area was estimated to decrease.

In both simulations, population exposure to NO₂ decreased about 2% by the interventions. Noise population exposure decreased 4% in simulation 1 and less than 1% in simulation 2. Figure 1 shows the estimated changes in population exposure graphically. In simulation 2, less people were exposed to noise levels above 60 dB(A) after the intervention, but on the other hand, the intervention resulted in more people living close to a highway and therefore, more people were exposed to somewhat lower noise levels (55–60 dB(A)).

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Estimated decline in NO₂ and noise exposure by speed limit reduction (1) and by traffic re-allocation (2) in the population living within 500 m of one of the involved highway sections.

Step 5: Estimation of the number of cases

The estimated changes were small for cardiopulmonary mortality, wheeze and ischaemic heart disease (table 1). In simulation 2, numbers remained the same or even slightly increased, despite lower exposure levels, because the population affected increased. In simulation 1, the intervention was estimated to result in approximately 200–300 less people being severely annoyed or disturbed in their sleep. In addition, the number of deaths and severe injuries by accidents were estimated to decline from 4 to 1 and 32 to 14, respectively, because of the speed limits (table 1).

Step 6: Disease burden

Figure 2 shows the pre- and post-intervention burden of disease related to road traffic for both simulations. Burden of disease was summarised by the type of exposure, with the presented burden for noise, for example, consisting of ischaemic heart disease, annoyance and sleep disturbance. Estimated changes in disease burden were generally small (<5%), which is in line with expected changes in numbers of cases, and the changes in population exposure. However, for deaths and injuries by accidents, estimated changes were more pronounced (60%), because of relatively high severity factors and longer durations in the DALY calculations (eg, people dying in a car accident generally lose a lot of life-years). Annoyance and sleep disturbance on the contrary, though affecting many people, did not show large intervention effects when expressed in DALYs, because of the relatively low severity factor. In simulation 2, the disease burden even showed a (very small) increase by the intervention, because of the increase in the population at risk, that is, the number of people living close to a highway. When adjusting for this increase in population size, the disease burden slightly decreased (pre-intervention: 24.4 DALYs per 1000; post-intervention: 23.5 DALYs per 1000) by the intervention.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

Estimated changes in burden of disease, expressed in disability-adjusted life years (DALYs) with prediction intervals and grouped by exposure, for speed limit reduction (1) and traffic re-allocation (2). Please note: the 95% prediction intervals with respect to the disease burden only represent a small part of the uncertainty in the outcomes, namely statistical uncertainty in the reported relative risks and in the severity factors applied in the DALY calculations.

Discussion

This article illustrates that general HIA models can be applied to local transport interventions. However, results should be interpreted with caution because of many uncertainties. In our simulations, uncertainties resulted mainly from the lack of local exposure and health endpoint measurement data and lack of scientific evidence, such as exposure-effect relations for subgroups and for effects of combined exposures. Given these uncertainties, simulated changes in population health by two local traffic interventions were estimated to be generally small, which is in line with previous investigations showing small predicted effects on exposures.13 14 35

Inevitably, in the course of quantifying health effects of policies, assumptions have to be made that are not comfortably supported by empirical evidence. It has previously been recommended to report these kinds of assumptions clearly.20 36 Table 2 gives an overview of assumptions made in the course of this HIA, including an evaluation of their effect. It is difficult to assess the relative importance of the sources of uncertainty, as many effects of uncertainties on the outcomes are not easily quantifiable. Important sources of uncertainties are discussed below and results of some alternative assumptions have been presented (ie, other impact zone or population at risk).

Uncertainty in population exposure

Local exposure information was difficult to establish, although we selected simulations for which at least changes in exposure had already been estimated. Variability in input data of exposure levels influenced overall uncertainty in the outcome measure, DALYs, only marginally in the Monte Carlo analysis and was therefore not included in the final analysis. Overall uncertainty, as expressed in the prediction intervals around the DALYs, was mainly driven by uncertainty in the exposure-effect relations. However, there is also uncertainty related to exposure reductions, which could not be easily included in the Monte Carlo analysis and which actually probably influenced the outcomes to a larger extent. To address these uncertainties, for instance relating to the assumption that traffic flow remained constant in simulation 1, we made them explicit in table 2.

The actual introduction of speed limit reduction caused traffic jams at some, but not all, locations. As traffic jams lead to more exposure to air pollution and noise, this would result in a negative health impact. However, data were not available to include these effects in the calculations. Population exposure also depends on the choice of the impact zone. Reducing the impact zone to 300 m instead of 500 m in simulation 1 (speed limit reduction) halved the population at risk of adverse health effects, and thereby, the traffic-related disease burden (not shown). On the other hand, the estimated change in disease burden is not affected by this assumption. Another important assumption was that the modelled changes in exposure to air pollution and noise are valid estimates of true, local-scale changes in exposure. The national air pollution models, for example, were developed to get a broad picture of air pollution “hot spots” in the Netherlands. Although these models are calibrated against daily measurement data at 25 spots all over the Netherlands,37 38 they are not developed to estimate exposure at a low-scale level.

Uncertainty in attributable cases and deaths

Another source of uncertainty is health effects of combined exposure to road traffic noise and air pollution, which are probably larger than the sum of the effects separately. In a traffic study in Oslo, for example, it was observed that the higher air pollution levels people are exposed to, the more likely they are to be annoyed by road traffic noise at a specified noise level.39 Unfortunately, to our knowledge, such relations have not been published yet for the other health outcomes in our study. In addition, exposure-effect relations are often only reported for a specific age range, and we did not apply them to other age or sex groups. If we applied the reported relative risk for wheeze to 5–18-year-old children instead of 7–12 year olds, the estimated number of cases before and after the intervention was doubled. However, logically, it did not affect relative changes in numbers in the simulations (not shown). Finally, our HIA does not make clear when positive health effects of the simulated interventions would occur. Logically, a reduction of traffic accidents could be expected immediately. It is more difficult to estimate when reductions in ischaemic heart disease or cardiopulmonary mortality — if substantial reduction would have been predicted — would occur.

This article presents numbers of deaths that can theoretically be attributed to exposure to air pollution in table 1. Some recent papers showed that it is more appropriate to calculate the average loss of life expectancy due to exposure to air pollution instead of attributable numbers of deaths,40 41 for reasons described in the supplementary textbox. In future HIAs, especially those involving long latency effects or if carried out in highly dynamic populations, we recommend using a life table approach.

Dealing with uncertainties

It is generally acknowledged that there are serious gaps in the evidence base required to carry out a rigorous HIA.12 Because of the many assumptions and uncertainties involved, epidemiologists have often been reluctant to become involved in HIAs like this. However, before concluding that quantification is not possible, it may be worthwhile to bear in mind that the perspective of a decision-maker differs from that of an epidemiologist. An expert’s guess may still be better than no guess at all2 and a step-by-step presentation of HIA results, though incomplete because of data gaps, has the potential to raise the awareness of policy-related health effects among policymakers. In addition, for the development of a HIA evidence base it is essential to be proactive in specifying what types of data are needed, and then explore the adequacy and quality of what is available,12 like we did in this article. Nevertheless, it is necessary to give proper information about the uncertainties and assumptions involved, and the way the results should (not) be interpreted. In this article, uncertainties have been addressed explicitly and if possible, they have been included in the Monte Carlo uncertainty analysis. Other methods, like sensitivity analysis or various qualitative approaches, can also be employed to obtain a wider and more appropriate representation of all uncertainties involved and their potential effect on decision-making.42 One might discuss whether DALYs are the proper output indicators for the HIAs presented in this article. Arguably, merely looking at changes in population exposure (step 3) leads to the same conclusions, because the small reductions in exposure will logically result in small reductions in numbers of cases or deaths and DALYs. Using changes in exposure as output indicator would have had the advantage that uncertainties of subsequent calculation steps would not have “accumulated” in the ultimate study outcome: DALYs. However, the calculation of DALYs facilitated the comparison of different health effects by air pollution and noise, and traffic accidents (the latter only in simulation 1). Another reason for continuing the HIAs was the descriptive goal of this paper: which problems do we encounter when passing through each HIA step? Logically, uncertainty would have been reduced if modelled study results could have been evaluated against measurement data. However, this kind of “direct evidence” of traffic interventions influencing population health is very scarce and therefore, intervention studies or before and after studies should be more encouraged.

Policy implications

This quantitative HIA was not part of the decision-making process, but contributed to the development of methodology to address future questions of policymakers. Will future assessments following this methodology be useful for policymakers, and could they support final decisions and implementation of policies? With regard to our simulation results, we are confident to state that they could have been useful for policymakers, because — despite uncertainties — they show the order of magnitude and range of health effects which may be expected after the interventions. This HIA also shows that the potential reduction of disease burden by local interventions is small compared with the total burden attributable to traffic-related air pollution and noise in the Netherlands. That is because the total health damage attributable to air pollution and noise is different from the health gain achievable by a change in levels from a policy intervention, unless the proposal is to remove the exposures altogether,12 35 which is hardly ever possible. For a substantial reduction in negative traffic-related health effects more fundamental interventions than the ones described here are required. According to Woodcock et al,43 policies promoting active transport in particular have the potential to improve health and reduce emissions.