• No se han encontrado resultados

US transportation agencies are expressing increased interest in integrating health considerations

into decision-making (USDOT, 2014; USDOT, 2012). Additionally, the use of HIA has been growing

rapidly in the transportation sector (Dannenberg et al., 2014). While a handful of existing policy

mechanisms exists to consider health impacts in transportation decision-making, applications of these

mechanisms are limited. In air quality nonattainment areas (areas not meeting the requirements of the

Clean Air Act), funds from the Congestion Mitigation and Air Quality Improvement Program may be

used to fund transportation projects that improve air quality and improve public health (USDOT, 2016).

Additionally, new highway and transit projects that may increase emissions from diesel vehicles in

nonattainment and maintenance areas are now required to conduct quantitative PM hot-spot analyses at

the project scale (EPA, 2015). Similar stipulations exist for CO emissions from a wider range of projects

in nonattainment and maintenance areas (EPA, 2015). Finally, many transportation agencies seek to

reduce fatal crash rates on a per-VMT basis, failing to consider that increases in VMT offset the benefits

of programs that reduce per-VMT health risks (Litman, 2014). While increasing physical activity is

sometimes considered a potential benefit of transportation systems, the health risks of built environments

that discourage active transportation behaviors are rarely conceptualized as a health impact of

119

multiple transportation health risks and offers a substantially more rigorous approach than is offered by

current tools and approaches.

As interest in the health impacts of transportation has grown, so has the development of decision-

support tools seeking to bridge the gap between transportation and public health agencies. However, these

emerging tools often provide data at low spatial resolution and provide little insight into the relationships

between transportation systems, exposures to transportation health risks, and health impacts. Without

sufficient spatial resolution or linkages between exposure and health outcomes, such tools fail to

adequately inform transportation decision-makers how their decisions may impact public health. Lacking

tools to estimate the health impacts of specific transportation decisions, a number of transportation

agencies have recently included health metrics into established decision-making processes as an approach

to make progress towards public health goals. At the state level, a number of state departments of

transportation have including health-related metrics in structured decision-making processes, such as the

inclusion of bicycle and pedestrian mode share as a recommended evaluation metric in Caltrans’ Smart

Mobility framework (USDOT, 2014). Local-level efforts have also focused on including health metrics in

structured decision-making, such as awarding points to transportation projects that address identified

health disparities in prioritizing project funds in Nashville, TN (USDOT, 2012). A critical gap in these

existing frameworks is the lack of modeling tools to translate transportation-related exposures to

population health impacts. Improved modeling of transportation health impacts helps clarify the pathway

from exposure to health impacts and could be a valuable tool in integrating health considerations into

routine transportation decision-making.

The modeling framework employed in this dissertation is well-suited to integrate with recent

innovations in transportation demand modeling. Traditional four-step travel demand models first generate

trips at the household level, distribute these trips across space, assign modes to these trips, and finally

assign these trips onto the transportation network. Four-step models can be used to support health impact

assessments of transportation air quality impacts (Mansfield et al., 2014). However, the usefulness of

urban area into “transportation analysis zones” (TAZs). Four-step models trips estimate between, but not within, TAZs; thus, active trips with short distances (i.e., occurring entirely within a TAZ) are not

modeled. Four-step transportation demand models are increasingly being replaced with activity-based

transportation demand models, which offer much finer geographic resolution and provide detailed

estimations of travel behaviors at the individual level (TRB, 2015). A necessary step when building an

activity-based travel demand model is the generation of a synthetic population for which the model will

estimate travel behaviors. An emerging literature explores the development of synthetic populations in

urban areas for this purpose (Zhu & Ferreira, 2014). Critically, the microsimulation framework used in

this research could easily use the same synthetic population as activity-based transportation demand

models. Detailed predictions of individual-level travel behaviors (including trip modes, distances, and

locations in an urban area) could easily be used to characterize individual-level exposure in the model

used in this dissertation. Thus, this work provides a framework that could support the integration of

detailed population-level health impacts into routine travel demand modeling activities as activity-based

travel demand models gain prominence in the field.

Even without complete integration with transportation demand models, the model developed in

this dissertation could support the integration of health considerations into a range of decisions about the

built environment. The model is modular and scalable, enabling its application in many routine

transportation decision-making practices. Metropolitan planning organizations could use the model

developed in this dissertation to support a variety of planning efforts. Integrated land use and

transportation planning efforts could be compared based a range of health metrics. The model could also

be used to include health outcomes when transportation agencies make funding decisions under budget

constraints (e.g., project prioritization). The model could be downscaled further to compare health

outcomes between alternatives at the project or corridor scale, offering transportation agencies a means to

quantify public health outcomes as a part of the National Environmental Policy Act (NEPA) process. In

addition, the model could be used to consider transportation health risks alongside other health risks that

121

modeling framework is modular and can be modified to include additional intermediate disease pathways

and additional exposures. In sum, the modular and scalable design of the model developed here presents

an opportunity to integrate health considerations into transportation decision-making in a much more

rigorous manner than is currently practiced and to provide robust decision-support as the fields of

transportation and public health continue to converge.

5.3. Limitations

While providing a more rigorous approach to estimating the population health impacts of

transportation systems, the model developed in this dissertation has substantial data requirements.

Notably, dynamic models require detailed age-specific functions characterizing baseline death rates,

disease prevalence, and disease incidence. Because mortality is a rare event for younger age groups, state-

level data were needed to develop baseline death rate functions. Additionally, incidence data are not

readily available for many diseases in the US; thus, disease incidence was estimated using the World

Health Organization’s DisMod II tool. However, baseline model calibration revealed reasonable model performance despite estimations in underlying epidemiological data. Finally, epidemiological evidence

linking chronic exposure to PM2.5 to morbidity is limited. Thus, the model developed in this dissertation

does not consider morbidity related to air pollution exposure. Practically, the model used in this work is

very computationally intensive due in large part to the size of the transition matrices within the Markov

model. While executable on a typical desktop workstation, the computational demands present practical

limitations to performing uncertainty analysis using techniques such as Monte Carlo simulation.

In addition to limitations of the health impacts model itself, estimation of individual-level

exposures to transportation health risks presents challenges. The most recent estimates of PM2.5

concentrations across the region were for 2011, while other exposures were estimated for 2013.

Additionally, it is assumed that PM2.5 concentrations are constant over time; however, PM2.5

concentrations will change in the future as the vehicle fleet changes and travel behaviors shift. While the

model is capable of incorporating exposures that vary over time, limitations in available estimates of

model used to estimate PM2.5 concentrations included most road segments in the study region, traffic

levels on many smaller roadways are not routinely collected as part of the Federal Highway

Administration’s National Highway Performance Program. Contributions to ambient PM2.5 from these

roadway segments were not included in estimated PM2.5 concentrations used in this study. Finally,

chemical pathways leading to secondary formation of PM2.5 were not modeled in the work used to

characterize PM2.5. Secondary formation of other pollutants in ambient, such as ozone, are also not

considered. However, secondary formation of air pollutants likely occurs after some atmospheric mixing

and is thus likely to vary less over space and, in turn, vary less in relation to neighborhood-scale built

environment variables. Despite these limitations, use of an advanced line-source dispersion model to

provide high-resolution estimates of PM2.5 across the study region offered a much more detailed

characterization of exposure to air pollution than used in previous studies. Further, the modeling

framework developed in this work could easily incorporate new exposure information as advancements in

high-resolution air quality modeling continue.

Additionally, the model used in this dissertation does not consider increased inhalation rates

during active transportation. Active commuters may be exposed to more air pollution while walking or

biking alongside roadways and may inhale greater amounts of pollutants in ambient air due to increased

respiration rates while being active. Individuals may also be more active during times of the year when

photochemistry is more active. However, using annual average pollutant concentrations masks potential

seasonal effects on total inhalation doses of airborne pollutants. Further, the modeling framework used

here does not predict where active transportation behaviors will occur—while walking trips may occur in

the same block group as an individuals’ home, biking trips and walk trips from public transit may occur in other block groups. Annual average PM2.5 concentrations are also used in this work to characterize

exposure while acute exposure for active commuters may vary significantly within and between days.

Additionally, it is unclear how increased acute exposure to air pollution may modify risks estimated in

long term cohort studies (Pope et al., 2002). However, Woodcock et al. assume an increase in air

123

health impacts (2014). Thus, the magnitude of underestimation in health risks for active commuters due to

this limitation is likely minor.

Estimations of transportation physical activity are largely based on individual commute modes,

while walking and biking trips may be taken by individuals who typically drive to work. While built

environment variables—population density and percentage of rental units—are considered and have

significant effects on transportation physical activity, the magnitude of these effects is small relative to

the magnitude of the commute mode to work variable (Figure 3.3). Thus, the transportation physical

activity predictions used in this work may under-predict walking and biking trips for individuals who live

in walkable environments but commute to work using a private vehicle. Additionally, fixed-route and

demand-responsive (paratransit) services are both considered public transit and riders who access transit

via park-and-ride lots are combined with users who walk or bike to access transit. Thus, additional factors

that may influence walking and biking associated with public transit use, but are not reported at the

Census block group geography in the American Community Survey, may lead to over-predictions of

active travel related to transit usage in areas with paratransit service only (e.g., rural areas) and for transit

users who use park-and-ride lots to access public transit. Finally, downscaling national level data to

estimate transportation physical activity in the study region may result in some upwards or downwards

bias due to unobserved variables. Despite these limitations, the regression models used to predict

transportation physical activity model performed well when validated in the study region.

The modeling framework used in this dissertation also does not consider area-level built

environment factors or the safety-in-numbers phenomenon in estimating fatal crash risk. Epidemiological

evidence used in this work conceptualizes fatal crash risk on the individual level, which limits the ability

of the model to consider area-based risk modification. Because this framework also does not consider the

specific locations of walking and biking trips, the ability to consider the role of specific built environment

variables in modifying fatal crash risk is constrained. Additionally, the model developed here considers

only mortality from motor vehicle, pedestrian, and bicycle trips in part due to methodological difficulties

environment. Despite these limitation, this work still finds the lowest total fatal crash risk in the most

walkable neighborhoods—neighborhoods that would be most likely to have built environments that

support active transportation and have a higher number of commuters. Incorporating risk reductions from

area-level factors and the safety-in-numbers phenomenon would likely strengthen the revealed association

between neighborhood walkability and fatal crash risk.

Finally, the model developed in this dissertation is applied in only one case study region. The

relationship between the built environment and air quality is influenced by exogenous policy variables,

such as regulatory regimes for automobile emissions. Additionally, fatal crash risks may be modified by

vehicle-level safety standards, cultural norms regarding driving, and other factors. Transportation physical

activity levels may also vary relative to built environment characteristics in different ways in different

cultural contexts. Thus, the magnitude of risk tradeoffs between air pollution, transportation physical

activity, and fatal crashes may differ in cities in less developed countries and in emerging mega-cities

with weak environmental regulation (e.g., Asian mega-cities). However, while the conclusions regarding

risk tradeoffs discussed here are not widely generalizable, the modeling framework developed in this

dissertation could be applied in other contexts to explore risk tradeoffs within different regulatory and

cultural contexts.

Documento similar