Ancillary health effects of climate mitigation scenarios as drivers of policy uptake: a review of air quality, transportation and diet co-benefits modeling studies

Motivations for the studies ranged from estimating current co-benefits of a specific policy proposed for a city or country to estimating future co-benefits globally under different mitigation scenarios. Therefore, the geographic and temporal scales differed. The choices of geographic and temporal scale further influenced the specificity of the scenario analyzed, with more detailed scenarios generally assessed at smaller scales.

The level of detail in each step in the analytic chain (i.e. estimating emissions related to a mitigation policy or scenario, modeling resultant changes in air quality, estimating the health impacts based on concentration-response functions) varied, making synthesis of studies challenging. Some studies started with detailed models developed to generate insights into the costs of mitigation policies (e.g. Rao et al 2013); the resultant changes in air pollutants were then coupled with a limited number of health concentration-response functions to estimate co-benefits. These studies had more depth in exploring mitigation options but less in exploring the wide range of possible health co-benefits, and thus may underestimate the extent of health co-benefits. Other studies started with detailed models of how a range of health outcomes can be affected by exposure to air pollutants. For example, West et al 2013 focused on several causes of premature mortality. These models were generally developed for other purposes, such as estimating the costs and benefits of air pollution regulation (e.g. BenMAP). A limited mitigation scenario was coupled with these models to estimate co-benefits. These studies explored a narrower range of mitigation options, but provided a more detailed assessment of health co-benefits.

Studies examined policies relevant at local, national, or international scales. The specificity of some policy scenarios, while useful for examining specific policy choices, restricted their generalizability to other technologies or contexts. For example, Gilmore et al (2010) examined a 500 MW sodium–sulfur battery charged during off-peak times of the day and discharged during peak times to replace four hours of electricity generation from two types of electricity generating peaking plants in New York, US. Eight studies focused on national level policies aimed at overall reductions in CO₂ to achieve national mitigation goals, energy efficiency measures, improvements in electricity generation, inter alia; for example comparing the US Clean Energy Standard with cap and trade policies (Saari et al 2015, Thompson et al 2014). Other studies focused on informing international negotiations on mitigation policies. One such study estimated the health co-benefits in the US of implementing a global carbon tax to achieve radiative forcing levels of 3.7 or 4.5 W m⁻² in 2100 (Garcia-Menendez et al 2015). Rao et al (2013) compared multiple possible air pollution, energy access, and climate policies to identify which would be associated with the largest health co-benefits.

Baselines to the policy scenario are important for co-benefit estimates because they determine background air quality and the level of additional achievable reductions via counterfactuals. Several near-term studies used current legislation (or a scenario of legislation) to ensure comparability in the baseline and estimated health co-benefit (e.g. West et al 2012). RCP 8.5 (Representative Concentration Pathway of GHG emissions that results in 8.5 W m⁻² in 2100, approximately the current emissions pathway), for example, was used as the baseline for several studies to represent future conditions under a no mitigation policy scenario (e.g. Schucht et al 2015). There are limited differences in global mean surface temperature across the RCPs until 2050, so RCP 8.5 is often used as a baseline for projections later in the century.

Temporal scales in the study sample ranged widely from present to 2100. The purpose of the study dictated the period of interest. The 2030s and 2050s were frequently used in projections of the magnitude and pattern of the risks of climate change (IPCC 2014), although longer simulations are typically needed to assess climate stabilization.

Studies varied in the sources and species of GHG covered by the policy scenario. Sources ranged from the full economy to fossil-powered electricity, buildings, agriculture, and transportation. As mentioned, these sources were targeted through specific policies or more general scenarios. Balbus et al (2014) took a different approach, analyzing ten options in the transportation, buildings, and power plant sectors that would account for one US 'wedge' of GHG reductions amounting to 19 GtCO₂ cumulative reduction over 50 years. Three studies focused on sources of methane and black carbon (Sarofim et al 2017, Anenberg et al 2012, West et al 2012), while others focused on CO₂ or CO₂e.

Studies varied in their modeling approaches including their level of sophistication at different points in the pathway from policy to air pollution. Some use detailed analyses to estimate the effect of policies on emissions (e.g. Crawford-Brown et al 2013), and emissions on concentrations (e.g. Shindell et al 2012). The effects of policies on emissions were estimated by selecting blanket reductions (e.g. Markandya et al 2009), using engineering calculations (e.g. Balbus et al 2014), or employing scenarios developed with models of electricity, transportation, or economic systems (e.g. Anenberg et al 2012), or a combination thereof, e.g. within an integrated assessment model (e.g. Rao et al 2016). Atmospheric response to emissions was estimated by chemical transport models to estimate co-benefits from co-emissions, or coupled with climate inputs or feedbacks to capture direct benefits from reducing the climate change penalty. Such methodological variety is often deliberate and appropriate for a specific policy application. However, it introduces an additional difficulty when attempting to draw comparisons across studies by adding variability related to model choice in addition to other sources of variability.

The studies considered a range of adverse health outcomes, including premature mortality from cardiorespiratory diseases, lung cancer, and acute respiratory infections. Studies of morbidity estimates included hospital admissions, long-term health care, asthma admissions, and restricted activity days.

When no-effects thresholds were applied, they ranged from 7.5–50 µg m⁻³ for PM_2.5, although there is evidence for risk below this range (Lepeule et al 2012). Thresholds were typically derived from studies conducted in high-income countries, so there are questions regarding the appropriateness of applying the same functions and thresholds in low- and middle-income countries, where concentrations of air pollutants can be much higher. Further, some studies assumed concentration-response relationships were linear and others used non-linear functions (e.g. Rao et al 2013). Additionally, the health benefits of reductions in ambient air pollution can be difficult to model without knowing the background contribution of household air pollution or secondhand smoke to total exposures (Burnett et al 2014).

Health co-benefits were reported as reductions in disability adjusted life years (DALYs), years of life lost (YLL), and mortality. The broad range of policy scenarios limits more detailed statements than mitigation policies would result in health co-benefits, with the extent of co-benefits varying by policy specifics, air pollutants considered, and analytic choices of geographic and temporal scale, demographic and socioeconomic changes over the study period, and health outcomes included.

Fourteen of the included studies then monetized the estimated health co-benefits to estimate the extent to which these benefits could offset the costs of implementing the policy. Although all showed some degree of offset of the policies assessed, the differences across the studies in how this was calculated makes general statements challenging.

Nonetheless, studies that compared the effects of co-emitted pollutants to that of the climate change penalty suggest that the first is the most significant, at least by mid-century, for ozone (Selin et al 2009) and fine particulate matter (Garcia-Menendez et al 2015). In addition, several studies reported an estimate of the dollars of air quality co-benefits per ton of CO₂ avoided. The range across these studies was $2–380 ton⁻¹ of CO₂ avoided, with a maximum nearly double that of earlier reviews yielding $2–196/ton CO₂ (Nemet et al 2010), and ranges even higher to $700–5000 ton⁻¹ for CH₄ (Sarofim et al 2017, Shindell et al 2012).

While these studies presented overall air quality benefits, several indicated the potential for dis-benefits of climate policy. For example, increased health risks occurred in localized areas due to NO_x titration of ozone (Thompson et al 2014). In addition, changes in health-related exposure due to regulation or lack of regulation are not constrained to that area, and 'leakage' of health risks was observed in specific unregulated regions or sectors under specific policies (Thompson et al 2016). Thus, distributional considerations could identify dis-benefits for certain stakeholders even if a policy yielded overall health co-benefits.

Using behavior-focused approaches, most transportation studies developed hypothetical scenarios involving replacement of a proportion of personal automobile use with walking, cycling, and use of public transportation (Stevenson et al 2016, Xia et al 2015, Macmillan et al 2014, Woodcock et al 2013, Rojas-Rueda et al 2011, Grabow et al 2012, Lindsay et al 2011). Three studies started from an emissions-focused premise evaluating the required shifts in transportation modes (e.g. Creutzig et al 2012, Woodcock et al 2009, Shindell et al 2016). The remaining two studies evaluated both types of scenarios (e.g. Sabel et al 2016, Maizlish et al 2013). Research design increased in sophistication over time, starting from simple models based on assumptions of policy or behavior implementation and general distribution of benefits, to more sophisticated approaches including stratified designs that acknowledged the age-dependency of co-benefits and co-harms (e.g. Woodcock et al 2013).

Not all included studies had a policy assessment; however, many of them had policy implications and/or laid the groundwork for future policy analyses. For instance, some of the studies could provide support for the enactment of Complete Streets policies, establishment of a bikeshare program, or introduction of congestion charges and other price interventions to reduce use of motor vehicles. Complete Streets policies support active transportation through the routine design, maintenance, and operation of streets and communities that are safe and accommodating for all people, regardless of age, ability, or mode of transport (Carlson et al 2017). On the other hand, some of the scenarios were policy packages developed with stakeholder engagement (e.g. Creutzig et al 2012, Woodcock et al 2009 and 2013). Shindell et al (2016) estimated the climate and health benefits of reducing US emissions consistent with a 2 °C increase in global mean surface temperature; these analyses assumed transportation reductions avoiding 0.03 °C warming in 2030 and 0.15 °C in 2100.

Study settings varied widely, from cities to regions to national assessments, and there was little commonality in the interventions themselves, which consisted of various combinations of walking, cycling, and taking public transport. The baseline against which policy scenarios were measured was 'business as usual' (e.g. Macmillan et al 2014, Sabel et al 2016).

Studies focused on the health benefits of increasing physical activity by replacing a portion of car trips with active transportation either today or within the next few decades. One study (Shindell et al 2016) projected health co-benefits to the end of the century, but the majority presented results for a decade in the first half of the century. The baselines were generally present day or a recent period.

Burning fossil fuels is the primary source of transportation emissions; over 90% of the fuel used for transportation is petroleum-based, including gasoline and diesel. Studies reviewed included operational GHG emissions from motor vehicle road traffic, in particular commuter motor vehicles.

Over time, studies employed increasingly sophisticated approaches to estimate the co-benefits and co-harms of increasing active transport and considered the benefits of increased physical activity alongside the risks of injury and road traffic fatalities, and exposure to air pollution, noise, well-being, and other factors. Increasing method sophistication included use of age, gender, and fitness stratification (Woodcock et al 2013), consideration of social and cultural factors (e.g. the extent to which biking is considered normal or even favored), and inclusion of infrastructure parameters (e.g. the extent and safety of bike lanes). More recent studies used system dynamics modeling to address the complexity inherent to transportation policy including system feedbacks, interacting variables as drivers of system behavior, and time-dependency of cause-and-effect relationships that may produce trade-offs between long-term and short-term policy effects (Macmillan et al 2014).

The studies relied on quantifications of the adverse health consequences of exposure to air pollution. The studies analyzed cardio-respiratory diseases, other chronic diseases, road traffic fatalities, obesity, well-being (e.g. mental health), and other health outcomes, with mortality the most frequent endpoint considered. The studies were conducted predominantly in high income countries; it is uncertain the extent to which they would be applicable to low- and middle-income countries where infrastructure, density of settlement, traffic conditions, and vehicle speeds vary greatly. Stevenson et al (2016) is an example of a transportation modeling study that investigated co-benefits in a range of lower and higher income countries.

Transportation health co-benefits modeling studies focused on classic health outcomes such as cardiovascular disease, overweight, and all-cause mortality. Other important health-related effects of reduction in motor vehicle transport may be included in future studies. For instance, with one exception (Maizlish et al 2013), these models have not yet incorporated social-emotional wellbeing and mental health outcomes affected by social severance (e.g. diminished social interactions in neighborhoods divided by roads with high volumes of motor traffic).

Significant reductions in DALYs, YLL, and/or mortality were reported from active transport even when potential injuries were considered (see supplementary table 3). A more detailed synthesis is not possible because of differences across the studies in baselines and time slices. Detailed longitudinal assessments, such as those incorporated in the ITHIM tool (Woodcock et al 2013), are needed to fully capture the cumulative benefits of increased physical activity.

Lindsay et al (2011), Macmillan et al (2014), Grabow et al (2012), and Shindell et al (2016) estimated the economic benefits. Lindsay et al (2011) estimated a savings of over NZ$1 million per 1000 commuter cyclists per year in New Zealand, and Shindell et al (2016) estimated near-term benefits of about US$250 billion annually in the US for implementing ambitious policies promoting clean energy and vehicles, depending on assumptions and the discount rate used; these benefits would likely exceed implementation costs. Grabow et al (2012) also estimated economic benefits of $8.7 billion annually over the course of the months when it would be the most feasible for active transportation in the 11 largest cities in the Upper Midwest of the US. Macmillan et al (2014) provided a detailed cost-benefit analysis incorporating implementation costs of commuter cycling policies and their health-related savings due to reduced mortality, hospitalizations, and disease incidence.

One concern of increased active transport is potential exposure to cycle-car and pedestrian-car crashes. Injuries and crashes were considered along dimensions of striking vehicle mode, exposure per distance traveled, vehicle or non-vehicle occupant type (e.g. cyclist, pedestrian, heavy vehicle, etc.), road type and severity of injury (Creutzig et al 2012, Xia et al 2015, Lindsay et al 2011, Woodcock et al 2009). Transportation and land use are tightly connected such that the balance of benefits and co-harms in a shift to more active transport depends on infrastructure. A 50% increase in cycling in a city with an extensive cycling infrastructure would make only a small difference in injury rates compared to cities where bicycle lanes are less common. In cities in which cycling is uncommon, the safety-in-numbers effect of reduced injury incidence with increasing bicycling prevalence is likely to result principally from growing pressures to invest in safer bicycling infrastructure (Macmillan et al 2014, Jacobsen 2003).

All six diet studies explicitly included both climate and health benefits of changing diets, but with different approaches. Three of the studies began with a simultaneous consideration of both climate and health impacts of diets (i.e. behavior-focused). Aston et al assumed that the food system accounted for one third of UK GHG emissions, and that animal products are especially emissions intensive. They calculated the proportion of the UK population with different diets based on consumption of red and processed meat, then calculated the changes in GHG emissions and relative risk (RR) of NCDs if high consumers of animal foods had diets of low consumers. Springmann et al (2016) began with the assumption that animal based foods are both a major source of GHG emissions and NCDs, and that diet change could be 'more effective than technological mitigation options for avoiding climate change.' They then estimated the GHG emissions and RR of NCDs for four dietary scenarios that progressively excluded more animal-sourced foods. Hallström (2017) created counterfactual healthy alternative diets statistically associated with changes in the RR for three NCDs, calculating the GHG emissions of producing these diets and even of the health care system related GHG emission savings.

The other three studies began with climate change mitigation strategies (i.e. emissions focused). Friel et al 2009 modeled the effect of four strategies (technological changes and a 30% reduction in production in UK livestock industry) needed to reduce GHG emissions in the UK to meet official mitigation targets for 2030. They then estimated changes in population level intake of saturated fat and cholesterol, and the resulting effect on prevalence of ischemic heart disease and stroke. Scarborough et al (2012) used the UK CCC carbon budget diet scenarios, designed as a climate intervention, and then estimated population level health impacts. Springmann et al (2017) assumed that 'GHG emissions related to food production will have to become a critical component of policies aimed at mitigating climate change,' and modeled a climate change mitigation policy of consumption taxes on all food commodities based on their GHG emissions. They then calculated dietary and weight related RR of health outcomes.

Of the diet studies, only Springmann et al (2017) discussed specific policies for achieving the modeled diet changes. Acceptability of the modeled diets at the individual and population level was not dealt with in these studies. In general, the objective was to 'explore a range of possible environmental [climate] and health outcomes...to encourage researchers and policymakers to act' (Springmann et al 2016). Exploration of policies for achieving diet change and their acceptability of modeled diets is an area of active research and policy development (Garnett et al 2015), but no studies addressing these questions met our inclusion criteria.

The policy scenarios in the included diet studies were often compared with actual dietary patterns as the baseline. For example, policy scenarios included sex-specific doubling of the proportion of vegetarians, a wider adoption of eating habits approximating the diets of those in the existing lowest quintile of meat consumption, or observed vegetarian and vegan intakes (Aston et al 2012, Springmann et al 2016). Another approach was to consider published dietary guidelines or food exposures for which there was strong evidence of a correlation with disease, or a combination of the above approaches (Hallström et al 2017). Baselines typically reflected existing diets or were based on UN Food and Agriculture Organization (FAO) forecasted diets.

Diet modeling studies presented results for a single year (or in one case, two years) between 2010 and 2050.

The studies all used life cycle assessments (LCAs) from other sources, but selected and adjusted values to some extent to make them more appropriate. The health benefits per CO₂e were highly dependent on the assumptions, methods, and data in the LCAs used as sources of CO₂e per unit of the foods in the baseline and counterfactual diets. The temporal, spatial, and structural boundaries used in LCAs can have large effects on CO₂e per unit of food, such as whether to include land use change or food waste. For example, only three of the studies (Springmann et al 2016, 2017, Halström et al 2017) incorporated wasted food; given that an estimated 30% of the global food supply is wasted, the impact on estimates can be considerable (FAO 2013).

There was considerable variation related to diet definitions, the relationship between diet and health outcomes, and the influence of diet on GHG emissions in the modeling choices and underlying assumptions of the reviewed studies. Policy scenario diets were defined de novo by using assumptions about health or climate benefits or both, or by modifying existing diets based on assumptions about effects on GHG emissions, or health, or both. Some studies defined them prior to the study according to climate change mitigation policies. There was also diversity in the extent of actual food intake exposures included in the models regarding components of meat or of meat products and inclusion of refined sugars and pulses. The most extensive model used was DIETRON; that is parameterized by total energy, fruit, vegetables, fiber, total fat, monounsaturated fatty acids, polyunsaturated acids, saturated fatty acids, trans fats, dietary cholesterol, and salt (Scarborough et al 2012).

Effects of diets, component foods, food compounds, or diet-related proximal risk factors (e.g. overweight and obesity) on health were evaluated in terms of the effect of change in the relative risk of non-communicable diseases (most frequently CHD, cancers, type 2 diabetes), or change in mortality, YLL, or DALYs. Climate impact was used to define diets alone or in combination with other parameters, and/or was modeled in terms of GHG emissions per diet, food unit, or production unit (e.g. livestock). Most studies did not include indirect feedbacks from diet change on climate in their models, for example the effect of land use change resulting from decreased consumption and production of animal foods (see below: co-harms). Hallström et al (2017) included the reduction in GHG emissions due to reductions in health care costs as a result of healthier diets.

Population-attributable fractions (PAFs)/population impact fractions (PIFS) are used to estimate changes in morbidity and mortality due to scenario diets, based on diet or body weight risk factors from observational (correlational, cross-sectional) or experimental (randomized controlled trial) data from epidemiological studies, or meta-analyses of these data (Scarborough et al 2012, Springmann et al 2016, Springmann et al 2017, Hallström et al 2017).

There was great variability in populations, definitions of diet components and diseases, risk estimates, and methods, making it difficult to directly compare health outcomes per CO₂e avoided. Overall, however, the scenarios modeled in these studies yielded considerable reductions in chronic disease and mortality, and in GHG emissions. For example, Springmann et al (2016) estimated that a 25%−190% increase in fruit and vegetable consumption, and 56%–78% reduction in meat consumption could result in 5.1 million global deaths avoided from coronary heart disease, stroke, cancers, and type 2 diabetes, and a reduction of 11.4–8.1 Gt year⁻¹ of food-related GHG emissions. They further estimated that shifting the global population to vegetarian diets, and increasing produce consumption by 54% would result in avoiding 8.1 million deaths and a reduction of 11.4–3.4 Gt year⁻¹ of food-related GHG emissions by 2050. They estimated the economic benefits at $1–31 trillion, or 0.4%–15% of global GDP in 2050.

It is important to note that while there are co-benefits for many foods, and actual and model diets, this is not universal. There can be tradeoffs, or co-harms, between climate and health effects. Whether diets generate co-benefits or co-harms depends on how they are defined, and their definitions in terms of climate and health vary greatly. For example, compared with the existing US diet, the 2010 diet recommended by the USDA for improved nutrition increased GHG emissions 12% (Heller and Keoleian 2015). The main reason was that reduced meat consumption was balanced by an increase in dairy consumption, and to a lesser extent by an increase in seafood, fruit, oils and vegetables. However, the health benefits of dairy can be obtained in foods without the potential health costs of dairy (not considered by the USDA), and the USDA-recommended vegan diet with excess caloric intake eliminated reduced emissions from the current diet by 53%, suggesting that reducing or eliminating dairy may be critical for avoiding co-harms from some recommended diets.

Existing diets that have co-benefits in terms of some component foods or compounds may have co-harms in terms of others, and this varies between diets. For example, Payne et al (2016) reviewed 16 studies, including 100 existing dietary patterns, almost all in the Global North. They evaluated the relationship between diets containing nutrients (which they independently estimated) that are bad for health (e.g. saturated fat, salt), and those that are good for health (e.g. micronutrients), and the GHG emissions of the diets. They found that the majority of diets with lower GHG emissions had higher sugar and lower micronutrients, and that these diets had equal or higher levels of mortality and non-communicable diseases.

Of the studies included in this review, only the Springmann et al 2016 analysis considered co-harms of diet change directly. That study considered the ways that the projected dietary change might result in undernutrition among vulnerable populations, concluding that the corresponding health benefits outweigh these harms. They recognized there were disparities in who would benefit and be harmed.

An important class of co-harm is rebound, an issue particularly pertinent to dietary considerations, and which was dealt with in very few of the studies reviewed. Rebound can result when savings in the food or health care systems due to diet change are invested in activities that generate GHG emissions or disease that take back some of the benefits of the modeled scenarios. Rebound can also be negative and function as a co-benefit by reinforcing the intended effect of the scenario by investing health care savings from improved diets in improving access to fruits and vegetables, or revegetation of rangeland no longer needed for animal production, to increase carbon sequestration. While rebound is, as noted, outside the scope of this review, it is an important consideration for future co-benefits studies.