Health impact assessment of active transportation: A systematic review

Available online 18 April 2015


Introduction
Contemporary car-ownership, and the vast network of roadway systems to accommodate it, adversely impact public health through environmental pathways such as air pollution, noise, greenhouse gas emissions, and traffic hazards (Haines and Dora, 2012). The convenience of motorized transportation has reduced dependence on physically-demanding travel while simultaneously increasing sedentary time spent (González-Gross and Meléndez, 2013). Today, globally, more than 30% of all adults are estimated to perform insufficient physical activity (PA) (Hallal et al., 2012). A lack of PA is associated with all-cause mortality, cardiovascular diseases (CVD), type 2 diabetes, cancer and impaired mental health (Physical Activity Guidelines Advisory Committee, 2008), and together with an energy-dense diet, the driving force of the progressing obesity epidemic (Ng et al., 2014).
The promotion of walking and cycling for transportation complemented by public transportation or any other 'active' mode, i.e. active transportation (AT), presents a promising strategy to not only address problems of urban traffic strain, environmental pollution and climate change, but also to provide substantial health benefits (de Hartog et al., 2010). Despite associated risks of exposure to traffic and to a lesser extent air pollution , AT may overcome car dependence and increase PA levels (Lindsay et al., 2011).
In recent years, there has been growing interest in health impact assessment (HIA) as a method to estimate potential health consequences of non-healthcare interventions . HIA aims at identifying the direction and magnitude of potential health impacts of these interventions in order to mitigate harms and increase health benefits . As until now longitudinal pre-/post-intervention studies in AT are scarce, HIA has to make do with scenarios to study what health effects would be if changes in transportation behavior took place. To our knowledge, despite evidence of AT health benefits (Cavill et al., 2008;Teschke et al., 2012;Xia et al., 2013), a systematic review quantifying health benefits and risks of AT does not yet exist. Therefore, we systematically reviewed studies conducting quantitative HIA of a mode shift to AT.

Methods
This review was performed following the PRISMA guidelines for reporting of systematic reviews (Moher et al., 2009). Systematic database searches of MEDLINE, Web of Science and Transportation Research International Documentation were conducted. Keyword combinations of "health impact assessment", "active transportation", "physical activity", "traffic incidents", "air pollution" and "noise" were used (Appendix A.1). Limits were English, Spanish, Dutch, French, or German language and abstract availability. Manual bibliographic review, internet searches and expert consultation were conducted to ensure completeness of peer-reviewed studies. Two independent researchers (NM and DR-R) performed all levels of screening and discrepancies were resolved by consensus.

Eligibility criteria
For review inclusion the study had to (1) focus on prospective or retrospective interventions in transportation, built-environment, land-use, economy, or energy that produced a mode shift between motorized transportation and AT; (2) include a quantitative HIA methodology of comparative risk assessment, cost-benefit analysis, risk assessment or benefit assessment; (3) report a quantitative change in the exposure distribution of at least one health pathway; and (4) report a quantitative change in at least one health endpoint.

Outcome measure
The benefit-risk or benefit-cost relationship of the AT mode shift was the primary outcome of this review. If not reported by the study and if possible, the benefit-risk or benefit-cost ratio was calculated based on expected change in exposure distribution of health pathways resulting from a mode shift to AT.

Data extraction and synthesis
Essential data of eligible studies were extracted into a data extraction tool for descriptive and analytic synthesis (Appendix A.2). The literature search, study selection, data extraction and synthesis were performed between February 2, 2014 and December 9, 2014.

Search results
The literature search produced a list of 3594 articles. Initial title screening identified 333 candidate studies. Abstract screening identified 130 candidate studies and independent full-text reading resulted in 30 eligible studies (Fig. 1).

Study characteristics
The 30 eligible studies were published between September 2001 and January 2015 (Table 1). Interventions that produced a mode shift, and of which health impacts were estimated, included measures which make AT more attractive (e.g. bike-sharing system), or discourage private vehicle use (e.g. fuel price increase). Eighteen studies assessed health impacts of AT within Europe. One study compared London (UK) and Delhi (India). Seven studies estimated health impacts of AT in the United States. Five studies assessed health impacts in Australia and New Zealand. The studies covered a range of populations consisting mostly of driving-aged adults, partially stratified by age, sex, ethnicity or population density.
Twelve studies conducted comparative risk assessment, comparing estimated health benefits and risks of changed health pathway exposure distribution resulting from a mode shift to AT (Table 2). Twelve studies used cost-benefit analysis to estimate economic impacts. Of these, seven studies compared estimated benefits to intervention costs, while the other four compared savings and costs of expected health benefits and risks. Four studies were benefit assessments in which risks or costs were not considered. Two studies conducted risk assessment exclusively of traffic safety.

Physical activity
All studies, except Stipdonk and Reurings (2012) and Schepers and Heinen (2013), assessed the health impacts of increased PA resulting from a mode shift to AT. PA risk estimates for associated health outcomes used across the studies were taken predominantly from meta-analyses (Appendix B.1). The majority of studies assumed a linear association between PA and health. The World Health Organization's (WHO) Health Economic Assessment Tool (HEAT) was applied in seven studies and uses a log-linear dose-response function (DRF) between PA and all-cause mortality by applying a 22% risk reduction per 29 min of daily walking (World Health Organization, 2011), and a 28% risk reduction per 3 h of cycling per week (Andersen et al., 2000). HEAT caps a risk reduction at a threshold of 50%. Likewise, three studies used a linear DRF with either a threshold (Woodcock et al., 2009;Jarrett et al., 2012), or a square-root function for higher PA levels (Maizlish et al., 2013). Four studies modeled PA exposure with a continuous non-linear DRF with the consideration of baseline PA levels (Rabl and de Nazelle, 2012;Dhondt et al., 2013;Woodcock et al., 2013Woodcock et al., , 2014. Six studies used PA categories assigned with distinctive relative risks (RR) (Mooy and Gunning-Schepers, 2001;Saelensminde, 2004;Boarnet et al., 2008;Cobiac et al., 2009;Holm et al., 2012;Xia et al., 2015). All 28 studies obtained estimates for PA with a mode shift to AT that resulted in reductions in all-cause mortality, CVD, type 2 diabetes, weight gain, cancer, falls, or impaired mental health.

Traffic incidents
Twenty-one studies estimated health impacts of exposure to traffic with regard to fatality and injury risk, and one study with regard to the feeling-of-insecurity (Saelensminde, 2004). In all 21 studies, traffic incidents were estimated directly based on local or national statistics by including travel exposure data (Appendix B.1). The majority of studies modeled traffic incident risk linearly by mode-specific distance or time traveled. Eight studies, however, assumed non-linearity of risk by including risk components of a disproportional increase in traffic incidents ('safety in numbers'), changes in traffic volume, modal split, conflict types and kinetic energies, speed and road type traveled, as well as age and sex effects (Gotschi, 2011;Lindsay et al., 2011;Maizlish et al., 2013;Schepers and Heinen, 2013;Woodcock et al., 2013Woodcock et al., , 2014Macmillan et al., 2014;Xia et al., 2015). Fourteen studies estimated overall increases in traffic fatalities and injuries with increased levels of AT, while six studies estimated overall decreases in fatalities and injuries. Gotschi (2011) assumed no change in absolute number of traffic fatalities.

Air pollution
Seventeen studies estimated the health impacts of air pollution exposure. Air pollution risk estimates used across the studies were taken predominantly from longitudinal studies, but also from time-series analyses (Appendix B.1). While ten studies estimated health benefits to the general population from reduced car use and associated exposure reductions, three studies estimated the active traveler's individual exposure risk. Four studies included both estimations for the benefits to the population and the risk to the active traveler. Most frequently, PM 2.5 (particulate matter less than 2.5 μm) was used as a proxy for air pollution. Other traffic-related air pollution (TRAP) components considered included ozone, carbon monoxide, or elemental carbon. All studies,   Woodcock et al. (2009) only partially, used a linear DRF to describe the relationship between air pollution and health, with no modification of the DRF at higher exposure levels. All air pollution estimates for the general population obtained with a mode shift to AT resulted in reductions of all-cause mortality, respiratory disease, CVD, cancer, adverse birth outcomes, activity-restriction days, and productivity-loss. Air pollution estimates for the active traveler, however, resulted in increases of described health outcomes.

Noise
Three studies considered health impacts of noise exposure to the general population. Noise associations used came from technical reports.
While James et al. (2014) assessed noise exposure by changes in traffic volume, Creutzig et al. (2012) and Rabl and de Nazelle (2012) used an indirect economic assessment of traffic-related noise exposure, including health costs; relying on a cost function dependent on vehicle-kilometers traveled, mode-type, time of day and urbanization. Noise costs were estimated to decline with a mode shift to AT, however, the noise health impact was not quantified independently (Appendix B.1).

Health endpoints
Health endpoints summarizing the overall estimated health impact of the studies were (1) all-cause or disease-specific mortality, including traffic fatalities; (2) morbidities, including CVD, respiratory disease, adjusted life years (DALYs); (5) activity-restriction days and; (6) monetized health impacts, including health care costs, feeling-of-insecurity costs, activity-restriction costs, or productivity loss.

Health impacts
Estimated benefit-risk or benefit-cost ratios ranged from −2 to 360 (median = 9). Twenty-seven studies estimated health benefits of a mode shift to AT to outweigh associated risks or costs, irrespective of geographical context or baseline setting (Appendix B.2). The three studies that did not estimate an overall beneficial health impact were distinctive in their assessment approaches. Cobiac et al. (2009) calculated investment costs of their AT information and merchandise intervention to be excessive compared to the small change in AT behavior that the intervention produced. Stipdonk and Reurings (2012) and Schepers and Heinen (2013) assessed exclusively the risk of traffic incidents, to give a predicted overall increase in fatalities and injuries with a mode shift to AT.
Overall, however, net health benefits were estimated (Fig. 2). In all studies with multiple health pathways, except for Dhondt et al. (2013), health benefits of increased PA clearly outweighed estimated detrimental effects of traffic incidents and air pollution (Appendix B.3). These benefits contributed positively to at least 50% of all estimated health impact of AT. Dhondt et al. (2013) estimated the greatest benefits (52%) from reduced traffic incidents, but assumed a mode shift predominantly to safer transportation modes of public transportation and car-sharing (as passenger) and only a small proportion (2%) to walking and cycling (high risk modes).

Susceptible populations
Patterns of intra-population benefit differences were recognizable. The larger body of studies estimated older people (typically N45 years) to benefit more overall from a mode shift to AT than younger people (de Hartog et al., 2010;Rojas-Rueda et al., 2011, 2013Dhondt et al., 2013;Woodcock et al., 2014;Edwards and Mason, 2014;Xia et al., 2015). Albeit, when assessing only traffic safety, younger people (typically b30 years) were estimated to experience a road safety gain with a mode shift to AT (de Hartog et al., 2010;Stipdonk and Reurings, 2012;Dhondt et al., 2013;Schepers and Heinen, 2013). Nevertheless, in settings where AT increases the incident risk, AT appears especially hazardous for younger people, relative to the proportional change in baseline mortality (Edwards and Mason, 2014;Woodcock et al., 2014).
In spite of Edwards and Mason (2014) finding no sex differences, overall males were estimated to benefit more from AT than females (Olabarria et al., 2012;Dhondt et al., 2013;Woodcock et al., 2014). Assessing only traffic safety, Stipdonk and Reurings (2012) found male cyclists to be at increased injury risk, while contradictorily Woodcock et al. (2014) found female cyclists to be at increased injury risk. Finally, disadvantaged ethnic sub-populations were estimated to benefit more from AT than the general population (Lindsay et al., 2011). c Benefit-cost ratio or benefit-risk ratio was taken as reported directly from the study. Reported cost-benefit ratios may include environmental and economic impacts and may compare to investment costs. If the study did not report a ratio and if possible, a benefit-risk or benefit-cost ratio was calculated based on change in health pathway exposure distribution, except for Cobiac et al. (2009) Fig. 2. Health pathway contribution to estimated health impact of a mode shift to active transportation a,b . (S) = scenario. a The health pathway contribution was calculated based on estimated change in health pathway exposure distribution and is comparing health benefits with health risks. Each health pathway contribution is expressed as a proportion of the overall estimated health impact of the scenario. If the study estimated multiple active transport scenarios, the health impact was calculated for the most conservative scenario (scenario with the smallest benefit-risk ratio or benefit-cost ratio). b The health pathway contribution could not be calculated for studies that assessed only one health pathway; for studies where the health impact could not be untangled from environmental and economic impacts; for studies where the individual health pathway contributions were expressed in different units. Therefore excluded: Mooy and Gunning-Schepers, 2001;Saelensminde, 2004;Boarnet et al., 2008;Cobiac et al., 2009;Guo and Gandavarapu, 2010;Gotschi, 2011;Olabarria et al., 2012;Stipdonk and Reurings, 2012;Creutzig et al., 2012;Mulley et al., 2013;Schepers and Heinen, 2013;Deenihan and Caulfield, 2014;James et al., 2014.

Discussion
Consistently, the vast majority of the reviewed HIAs estimated substantial net health benefits with a mode shift to active transportation (AT). Estimated benefits were largely due to increases in PA levels, which greatly outweighed associated detrimental effects of traffic incidents and air pollution exposure. Noise impacts were only considered secondary. The large range of benefit-risk and benefit-cost ratios observed may be attributable to distinctive HIA approaches, different assumptions on health pathways, scenario design and baseline population parameters.

Physical activity
Estimated gains in PA from AT constituted at least half of the total health impact, except in Dhondt et al. (2013). Uncertainties remain, however, regarding assumptions on possible PA substitution (from another domain, with AT). There remains limited understanding on the relationship between transportation PA and total PA (Cavill et al., 2008). On the one hand, studies show independent health benefits from PA gained by AT, even after adjusting for other domains of PA (Andersen et al., 2000;Matthews et al., 2007;Hamer and Chida, 2008;Kelly et al., 2014). On the other hand, studies have shown uncertainty as to how much AT adds to total PA (Forsyth et al., 2008;Thomson et al., 2008;Wanner et al., 2012). This uncertainty is attributed to two things: (1) the failure of detecting significant associations; and (2) the argument that total PA is predetermined by the social environment as people who do more leisure time PA do less for other purposes and vice versa. Nonetheless, recent longitudinal studies estimated significant contributions of PA from AT to overall PA, without reducing participation in other PA domains Goodman et al., 2014). Thus, the assumption of a 1:1 gain in overall PA (i.e. no substitution) by all reviewed HIA studies appears plausible.
The shape of the applied DRF significantly impacts the PA benefit magnitude. As done by a few studies, a more biologically plausible approach is the application of a non-linear DRF which implies that health benefits vary in magnitude for different PA levels. Non-linearity coheres with results of a meta-analysis showing a strongly curvilinear relationship between PA and all-cause mortality, with the greatest benefits occurring for inactive people becoming moderately active (Woodcock et al., 2011). However, to apply a non-linear DRF, knowledge on baseline PA is essential. Given that in most cases data on baseline PA was not available, a linear DRF was used in which case no assumptions about baseline PA are required. Nevertheless, a linear DRF assumes equal changes in health benefits for active and non-active people; this assumption can lead to under-estimations of health benefits of PA for non-active people and to over-estimations for active people (Appendix C.1) (Woodcock et al., 2011;Rojas-Rueda et al., 2013).

Traffic incidents
Estimated health risks by traffic incidents are minor compared with health benefits gained by PA. Generally, an increase in traffic incidents resulting in fatalities or injuries was estimated with increases in walking and cycling. Shifting to active modes may increase incident risk as these are considered high-risk modes (Teschke et al., 2012;Wegman et al., 2012;Zegeer and Bushell, 2012). Moreover, an increase in singlemode incidents ('slipping') is projected (Schepers and Heinen, 2013).
Several studies, nevertheless, estimated their AT mode shift scenarios to lead to reduced incidents. These findings are due to three assumptions: (1) overall reduced motorized traffic volume; (2) a mode shift to safer transportation modes such as public transportation and carsharing (as passengers) may reduce incidents (Dhondt et al., 2013); and (3) the concept of 'safety in numbers' assumes a less than proportional increase in incidents, with increased walking and cycling share and acquired modal co-existence (Jacobsen, 2003;Elvik, 2009). In this context, one study found that the risk for cycling casualties decreased in communities with a higher cycling proportion (Vandenbulcke et al., 2009). However, uncertainties remain regarding the location-specific threshold level until a 'safety in numbers' effect may occur (Macmillan et al., 2014). Thus, there are suggestions that secure infrastructure measures must precede traffic safety and increases in AT ('numbers in safety') (Bhatia and Wier, 2011).
Generally, the injury burden of AT might be underestimated due to potential under-reporting of minor injuries. Two studies found that only 7% of all cycling incidents were reported in police statistics and chances for reporting increased with injury severity (Aertsens et al., 2010;de Geus et al., 2012). Another study found that single-mode incidents accounted for 40% of all bicycle incidents with 70% resulting in minor injuries (Tin Tin et al., 2010). Moreover, the incident risk is dependent on many setting-specific variables not currently comprehensively considered (Mindell et al., 2012;Wegman et al., 2012). Distance or time traveled, infrastructure provisions, traffic volume, modal split, conflict types, speed and road type traveled, kinetic energies as well as age and sex effects all affect incident risk.

Air pollution
Air pollution exposure was estimated to have small health impacts, with small benefits to the general population and small risks to the active traveler. Only two studies estimated larger air pollution improvements, but their studies assumed substantial reductions in motorized traffic volume (Grabow et al., 2012;Dhondt et al., 2013). While population health benefits emerge from reductions in motorized traffic volume and associated emission reductions, the risk to the active traveler is more complex to assess. On the one hand, walkers and cyclists may experience lower direct TRAP exposure than vehicle occupants, especially while traveling on segregated sidewalks or bike lanes (Boogaard et al., 2009;MacNaughton et al., 2014). On the other hand, increased ventilation rate resulting from physical strain increments the uptake of pollutants at least twofold (Zuurbier et al., 2010;de Nazelle et al., 2012). Taking into account ventilation rate, lung deposition and potential increases in travel time while substituting motorized transportation, estimations need to be revised upwards (Briggs et al., 2008;Int Panis et al., 2010).
Using air pollution risk estimates from elsewhere involves uncertainty because air pollution components are location and source specific (Stevens et al., 2014). PM 2.5 is a commonly-used proxy for exposure to all fossil fuel combustion sources. It has been suggested to be the most health relevant pollutant and is used in the Global Burden of Disease Study (Lim et al., 2012). Nevertheless, PM 2.5 cannot be differentiated by components, source or toxicity (Burnett et al., 2014). Thus, there is concern that PM 2.5 underestimates the health effects of incomplete fuel combustion (Janssen et al., 2011). All studies applied linear associations for air pollution, except Woodcock et al. (2009) for Delhi. Instead, a log-linear DRF for PM 2.5 was used as yearly average concentrations in Delhi exceeded 40 μg/m 3 and a linear DRF would predict implausible risks. Recent new evidence suggests that the relationship between PM 2.5 and excess mortality does not necessarily follow a linear function for the entire exposure range (Burnett et al., 2014).

Noise
So far, health impacts of traffic noise have mostly been neglected in HIA, despite assumptions of reductions in motorized traffic volume decreasing noise exposure to the general population. However, there remains inconclusive evidence to what extent traffic noise and TRAP are correlated (Foraster, 2013), given that both exposures are associated with CVD and diabetes (Babisch, 2014;Dzhambov, 2015). As the majority of risk estimates used for air pollution has not been adjusted for noise, attempting to include noise as an independent health pathway may confound health impact estimations.

Susceptible populations
Uncertainties persist in intra-population benefit differences. Overall, older people are estimated to benefit more from a mode shift to AT than younger people (Appendix C.2). Increased benefits from PA for older people are seen mainly because older people are at increased risk for chronic degenerative disease and PA can substantially reduce the absolute risk for disease development (Chodzko-Zajko et al., 2009;Vogel et al., 2009). Therefore, in older people the benefits of PA are estimated to greater outweigh the detriments of traffic incidents and air pollution exposure (de Hartog et al., 2010;Dhondt et al., 2013;Edwards and Mason, 2014;Woodcock et al., 2014;Xia et al., 2015). However, it remains inconclusive whether older people benefit differently from the same PA exposure than younger people, despite recent research findings indicating the latter. A systematic review found a larger mortality risk reduction for older people compared to younger people (RR = 0.78 vs RR = 0.81; 11 MET-hrs/week) (Woodcock et al., 2011). Assumptions that health benefits of PA are in fact long-term benefits support the argument that older people benefit more overall from AT (Edwards and Mason, 2014).
When assessing exclusively traffic safety of a mode shift to AT, in settings with low injury rates, younger people are estimated to experience a traffic safety gain, while older people are more vulnerable (de Hartog et al., 2010;Stipdonk and Reurings, 2012;Dhondt et al., 2013;Schepers and Heinen, 2013). Nevertheless, in settings where substituting AT for driving substantially increases the risk for incidents, there might be more relative harm for younger people, as injury and death at younger ages translate into a larger burden of disease due to lower baseline mortality and higher statistical life-expectancy (Edwards and Mason, 2014;Woodcock et al., 2014).
While one US study did not find sex differences in benefits (Edwards and Mason, 2014), three European studies estimated males to benefit more overall than females from a mode shift to AT: (1) males are less likely achieve PA recommendations (Olabarria et al., 2012); (2) the two different sexes are predicted to have distinctive disease risks (Woodcock et al., 2014); and (3) males benefit more from reduced motorized traffic incident risk (especially while switching to low risk modes of public transportation and car-passenger) (Dhondt et al., 2013). Despite cycling being a high risk mode for both sexes, males are said to have a higher injury risk as drivers, cyclists and pedestrians compared to females (Mindell et al., 2012). Nonetheless, one study estimated female cyclists to be at increased fatality risk in London, but also expressed local-specificity of their results given the typically lower risk faced by females (Woodcock et al., 2014).
As for older people, pronounced benefits for socially disadvantaged or ethnic sub-populations can be related to increased chronic disease incidence (Lindsay et al., 2011;Fang et al., 2012). However, differences in intrinsic motivations for AT engagement and intention-behavior relationships among different social classes need to be considered (Conner et al., 2013).

Uncertainties in health impact estimations
The reviewed HIA studies carry uncertainties in the estimations of quantitative health impacts, which emphasizes that HIA remains an indicative rather than an empirical research tool (Parry and Stevens, 2001). Benefit-risk and benefit-cost ratios can only be interpreted as an indication of the magnitude of expected health impacts, as underlying HIA modeling assumptions vary largely across studies. As typically local risk estimates for PA, air pollution and noise were not available, a multitude of risk estimates taken from elsewhere were applied. This limits comparability across studies. Likewise, uncertainties about strengths of associations and shapes of DRFs limit comparability of studies, despite the significant influences on the benefit magnitude.
Benefit estimations are sensitive to the contextual setting and population parameters. Health impact estimations depend on baseline prevalence of AT, baseline exposure to health pathways and the general health status of the population. Assumptions of a 'healthy-walker/cyclist effect' minimize benefit estimations by assuming that only healthy people with a low baseline disease risk choose AT (Macmillan et al., 2014). Moreover, it is uncertain to what extent the mode shift scenarios reflect reality as individuals' intrinsic motivations for AT engagement have not been considered yet (Kroesen and Handy, 2013). Despite a recent metaanalysis finding no significant effect for efficacy of behavioral interventions for transportation behavior change (Arnott et al., 2014), another recent systematic review suggests that a combination of behavioral and structural (workplace, built-environment, AT facilities) interventions may best increase AT engagement (Scheepers et al., 2014). In this regard, culture can reinforce AT behavior where it is common, but has opposing effects where it is uncommon (Pucher et al., 2010). There is also concern for decay of behavioral effects over time (Cobiac et al., 2009;Hoffman et al., 2012).
To estimate longevity of AT health effects one needs to consider time-lags in health benefits and risks. PA benefits are predominantly long-term in nature (Reiner et al., 2013;Chevan and Roberts, 2014), whereas injuries from traffic are immediate detriments. Taking timelags into consideration can substantially alter benefit estimations. Delayed receipt of health benefits from PA makes AT less appealing for younger people, but reinforces the importance for older people (Edwards and Mason, 2014). AT, however, might not be the most convenient choice of transportation for older people. Making AT safe and convenient (and normal) may be the key to reaching this population.
The effects of AT on health equity remain uncertain. On the one hand, a study found higher uptake of walking and cycling infrastructure by socio-economically advantaged individuals . On the other hand, two studies found that children from lower income households were more likely to use AT, suggesting that AT may be able to narrow health inequity (D'Haese et al., 2014;Gray et al., 2014). Supporting the latter, two studies in adults found greatest AT health benefits for disadvantaged ethnic sub-populations (Aytur et al., 2008;Lindsay et al., 2011). Yet, AT land-use improvements and facilities are mostly implemented in high income areas which also report more traffic safety and less crime (Aytur et al., 2008;Sallis et al., 2011).
Future research on the health impacts of AT should aim to better consider acute impacts on quality of life, including physiological and mental indicators (e.g. less back pain, increased mobility, mental wellbeing and happiness) and integrate impacts outside of the health domain, such as the effects on social capital, crime, or productivity. Future studies should also look more in-depth into effects of age, sex and social class, in times of global shifting in population age, gender equality and social equity. All studies, except Woodcock et al. (2009), were exclusively conducted in high income settings, leaving uncertainty about how results can be transferred to low and medium income settings. Moreover, children have been underrepresented, even though AT is accessible for children and estimated health impacts presumably affect them as well. Currently, no studies exist that estimate health impacts of other modes of transportation that involve PA, such as skates or e-bikes. However, studies may soon be appropriate for e-bikes, given their rapid market growth and importance for AT.
While care is needed when interpreting the results of HIA, the reviewed studies show net health benefits of a mode shift to AT, irrespective of geographical context and varying HIA modeling assumptions. HIA is valuable to improve the understanding of the interrelationship between transportation and health and can assist in optimizing health gains of non-healthcare interventions (Thomson et al., 2008).

Limitations and strengths
For the first time, studies conducting quantitative HIA of a mode shift to AT were systematically reviewed. We provide evidence of net health benefits of AT. However, publication bias is plausible as HIA in transportation is frequently conducted for intervention planning outside the peer review framework, such as the gray literature (Appendix D.1). Studies with negative findings may also less likely be published. Despite such limitations, the systematic search strategy and comprehensive inclusion criteria limit selection bias. The review of both public health and transportation databases, the absence of a time restriction and limited language constraints ensure that the existing body of evidence was captured.

Conclusions
We conclude that net health benefits of AT are substantial, irrespective of geographical context. Projected health gains by increases in PA levels exceed detrimental effects of traffic incidents and air pollution exposure. Thus, we encourage the promotion of AT, as associated health risks are minor.

Role of the funding source
This work was supported by the European project Physical Activity through Sustainable Transportation Approaches (PASTA), which has partners in London, Rome, Antwerp, Orebro, Vienna, Zurich, and Barcelona. PASTA (http://www.pastaproject.eu/home/) is a four-year project and funded by the European Union's Seventh Framework Program under EC-GA No. 602624. The sponsors had no role in the study design; in the collection, analysis, and interpretation of data; in the writing of the report; and in the decision to submit the article for publication.