Health Impact Assessment - Stage 4

Assess where you are in health impact assessment to determine which stage you are in and identify the key activities you need to undertake as an air  quality manager to go to the next stage. 

The guidance below is for Stage 4. Stage 1, Stage 2, Stage 3 and Stage 5 are also available.

 

01 Update the stage 3 plan to refine the scope of the HIA, focusing in particular on how the assessment will characterise impacts over space and time

Consider first the policy questions the team or decision-maker is asking and what additional data, new methods or alternative tools will be needed to address this question that were not available to you in Stages 1-3. In those earlier stages, you will have been able to estimate the burden to health of recent air quality levels, the impacts associated with a current policy, and the benefits of reducing air pollution.  If the decision-maker is satisfied with an HIA whose scope and complexity is matched to Stage 3, investing in improved data and more advanced techniques may not be warranted (Bhat et al., 2014). Alternatively, if the policy question requires an HIA whose scope and complexity exceeds the current expertise or resources, then you may consider: working with the decision-maker to re-scope the question; or, request additional time and support to construct an HIA that can address the policy question.  

After confirming that these resources are accessible, develop a project plan that specifies milestones for: gathering and incorporating these data into your selected HIA tool; performing the analysis; quality-assuring and synthesising the results; and, briefing key stakeholders (Government of Western Australia, 2010). Finally, this guidance applies to HIAs that address common air pollutants like PM2.5 and not air toxics like benzene or formaldehyde; the latter follows an approach that differs significantly from the one described here.   

02 Plan the HIA with subject matter experts, including risk assessors, epidemiologists, air quality and emissions modellers and/or demographers

Form your project team, taking care to identify subject matter experts who can match the data, methods and tools with the policy question being asked; these include risk assessors, epidemiologists, air quality and emission modellers, and demographers (National Research Council, 2011). Create, and build upon, connections with the health sector to identify appropriate background rates of death and disease. Considering the policy question, your team will determine the appropriate HIA method (a single year analysis, a life table1 or another method), health endpoint(s) to be quantified, epidemiological studies supplying concentration-response estimates, and air quality data (retrospective, prospective, or both) (Bhat et al., 2021; WHO, 2025).  

When selecting, configuring and applying input data, ensure that each input dataset is temporally and spatially harmonised; for example, if you are estimating air pollution-attributable deaths from a time-series study, be sure to use daily death rates to quantify effects and not an annual death rate (the latter of which will result in attributable counts that are overstated by a factor of 365) (Fann et al. 2011a). If the team economist is estimating the value of the cases of mortality and morbidity, ensure that she or he can match the endpoints of interest with the available cost-of-illness or willingness-to-pay measures (Howard et al. 2019; U.S. EPA, 2024b). 

03 Characterising exposure: collect monitoring data

If your HIA is focused on recent-year or historical changes in air quality, consider gathering monitoring data. By design, air quality monitors reflect pollutant concentrations in a given location and at a specific moment in time (WMO, 2024). Hence, if you use monitored air quality data to estimate population exposure, consider whether the existing monitoring network provides sufficient temporal and spatial coverage to support creating an interpolated surface of concentrations; in general, the larger the number of observations, the more reliable the air quality surface (U.S. EPA, 2018).  

A variety of techniques exist for creating an interpolated surface, including Voronoi Neighbourhood Averaging (VNA), kriging and Machine Learning, among others (Sacks et al. 2018; WMO, 2024). As described in the air quality modelling section below, you may wish to “fuse” the monitored air quality data with model-predictions (U.S. EPA, 2011).  

To the extent that the monitoring network is partly, or largely, comprised of low-cost sensors, determine whether and how to integrate these observations with regulatory (or reference-grade) monitors (WMO, 2024). Regardless of the source of the monitor data, confirm that a Quality Assurance/Quality Control (QA/QC) documentation exists. Finally, consult the team epidemiologist to ensure that the temporal and spatial scale of the monitored input data are compatible with the air quality concentrations used in the epidemiological study supplying the concentration-response estimates (U.S. EPA, 2010). 

04 Characterising exposure: collect or perform air quality modelling 

When your analysis is considering air quality changes occurring in the future—perhaps as a result of attaining a new air quality standard or reducing emissions from an industrial facility—you will need to use an air quality model. When selecting among models, determine first whether you are simulating an inert pollutant (e.g. black carbon) for which a Gaussian-plume model may be appropriate or a chemically reactive pollutant (e.g. ozone) that requires photochemical modelling. The formation of ozone in particular is subject to complex non-linear chemistry, such that reducing precursors to ozone (including NOx and VOCs) can in some circumstances increase ozone concentrations (Simon et al., 2012).  

Modelling, monitoring and remote sensing (e.g. satellite) data can also be combined to create a fused surface (Wesson et al. 2010). This approach offers a number of advantages over using the modelled surface directed, including: bias-correcting the modelling; predict pollutant concentrations in locations with a sparser monitoring network; predicting concentrations at a spatially resolved scale (Fann et al., 2011b; Hunt et al., 2021). On the other hand, air quality modelling is computationally- and resource-intensive and requires detailed estimates of source emissions. Finally, note that some literature finds that the model grid resolution can influence the size of the estimated health impacts (Thompson et al., 2014). 

05 Characterising exposure: collect detailed population data 

If your HIA is estimating impacts occurring in a recent or historical year, select population counts from an authoritative source (e.g. national census). If you are instead quantifying the impacts of a prospective population change, select projected population counts that reflect: births, changes in longevity, in-migration, out-migration and, internal relocation. Resources including SEDAC, WorldPop and IPUMS provide projected population counts reported at varying geographic scales and for different future years. Population counts should also be reported at a spatial scale the same as, or finer than, the air quality surface created above.

When selecting among population counts, take care to ensure the population age, sex, race and ethnicity strata match those specified in the epidemiological study supplying the concentration-response relationship (Fann et al., 2011a). Ensuring that the population counts are disaggregated in this way will allow you to address more specific policy questions regarding impacts among population groups by age, sex and other demographic attributes.   

06  Characterising exposure: collect background health data 

If you have not already done so, form a working relationship with experts in the health sector including departments of health and vital statistics offices. Because health data in most locations are governed by restrictive data use agreements, you will need to build trust with your counterpart in the health sector to ensure they are confident in your ability to protect the privacy and confidentiality of the individuals identified in these records. To that end, clarify that you need de-identified (or anonymised) counts of death and disease that are aggregated up from the location of residence to a coarser spatial unit. Alternatively, you may ask your counterpart to calculate the rates, ensuring that you both understand and agree upon the minimum count of cases needed to perform this calculation.  

Ideally the rates of death and disease will be expressed at the same geographic scale as the air quality modelling predictions. The rates should also be reported for the same ICD-9/10 codes as specified in the epidemiological studies supplying the concentration-response relationships (Fann et al., 2011a). Because rates of death and disease vary from year to year, consider requesting three or more years of rates.  

More detailed, and spatially resolved, background health data will support HIAs that address differences in risk among population groups—allowing decision-makers to better understand how a given policy is affecting vulnerable populations (Fann et al. 2011b). Moreover, as compared to a conventional strategy, air quality management plans designed to reduce pollutant concentrations among these vulnerable groups can both: (1) yield a more equitable distribution of exposures and risk; (2) generate larger overall monetised benefits (Fann et al., 2011b; Wesson et al., 2010).Consider whether characteristics beyond baseline health status—perhaps including temperature—could be used to define vulnerability, given the role of temperature extremes in compounding air pollution-related risks.  

07 Select concentration-response parameters 

At Stage 4, air pollution epidemiology studies may have been performed for your region or country of interest, in which case you will preferentially select concentration-response relationships from these studies (Hubbell et al., 2009). However, you can also complement this evidence with concentration-response relationships from HRAPIE, HRAPIE-2 and the Global Burden of Disease (Lim et al., 2012; WHO 2013, 2025). These and other authoritative sources will indicate when enough evidence exists to infer a causal relationship between a pollutant and a given health outcome; you should not need to do this yourself. When selecting among concentration-response relationships, ensure that you are choosing non-overlapping health-endpoints—for example, you could estimate all-cause mortality and cardiovascular mortality, but you cannot add the two because doing so would double-count effects.  

If you are constructing new concentration-response parameters, take care to select peer-reviewed epidemiological studies that: report measures of association (e.g. Relative Risk, Odds Ratio, Hazard Ratio, etc); indicate what epidemiological model used (e.g. Generalised Additive Model, Cox Proportional Hazards model, etc) clearly define the endpoint using ICD-9/10 codes; report a standard error around the mean estimates of effect; indicate the unit change in air pollution for which the risk is expressed (i.e., per 10 ppb, per IQR, etc) (Fann et al., 2011a). In general, prioritise multi-pollutant and multi-city epidemiological studies over single-pollutant and single-city studies (U.S. EPA 2024a). Finally, decide whether you want to assume a “counterfactual” pollutant concentration, below which you will not estimate health impacts (Navaratnam 2026).  

08 Run the analysis

After each member of your team carefully reviews the input parameters—air quality, population, background health and concentration-response functions—you are ready to perform the analysis. A variety of peer-reviewed and open-source tools exist to perform these analyses, including AirQ+, BenMAP-CE, AQBAT and others (Anenberg et al., 2016). When generating results, perform QA checks along the way, confirming that: the estimated counts of effects are directionally correct (reducing air pollutant concentrations yields smaller counts of deaths and illnesses); are in the order of magnitude that you would anticipate given the air quality change; are occurring in the location, and among the population groups, expected.  

When quantifying attributable counts of each endpoint, report counts rounded to two significant figures to account for uncertainties in the input data. If your tool allows it, perform a Monte Carlo analysis by sampling the distribution around the input parameters expressed as a distribution; for example, some tools will use the standard error reported in the epidemiological study to construct a distribution around the effect estimate that can then be sampled (Anenberg et al., 2016).  Consider reporting results as total counts, a rate (e.g. cases per 100k), or as the Attributable Fraction (the percentage of all cases associated with the pollutant). In contrast to Stages 1-3 of the Health Impact Assessment Guidance, you are now able to generate results that are more comprehensive—which can help answer more policy questions, but may also overwhelm decision-makers with detail.

09 Characterise key sources of uncertainty

To act upon the results of the HIA, decision-makers must also understand what uncertainties they are subject to. As a team, document each source of uncertainty that may affect the magnitude and distribution of the results. Separately categorise those sources that are to be addressed qualitatively and those to be addressed quantitatively (National Research Council, 2011; 2002). If you have performed a Monte Carlo analysis, clarify which sources of uncertainty the confidence intervals account for and which they do not.  

Common sources of uncertainty in an HIA include: emissions inventory, which may not accurately reflect emission rates, locations and release parameters; the air quality modelling, which may not account fully for the non-linear chemistry governing the formation of pollutants like PM2.5 and ozone; and, the epidemiological literature, which is subject to bias and confounding (US E.P.A., 2009). Separately identify key sources of sources and direction of bias in the analysis. For example, the chemical transport model may under-predict concentrations of PM2.5 in some situations, which could in turn bias-low the estimated counts of PM2.5-attributable effects. Providing for decision-makers both the results, and the sources and magnitude of uncertainty, will help them interpret and apply the results appropriately (U.S. EPA, 2011).  

10 Communicate the results

Prior to beginning the analysis, your team will have considered the audience for the HIA and the decision it was designed to support. Some audiences will find certain health outcomes to be more accessible than others; for example, may find counts of avoided cases of aggravated asthma to be more persuasive than estimated premature deaths.  When creating maps and tables, report results at a spatial scale no finer than the most coarsely resolved input. Identify key results that can be used for public communications materials; these can include aggregated counts of effects and total economic values, if quantified. If your analysis did monetise effects, you may want to highlight that promoting clean air is among the highest-returning investments available (UNEP/CCAC, 2026).​     ​

Use these results to develop a concise (perhaps 1-page) brief that summarises the results by policy option and notes key uncertainties and limitations. If similar HIAs were performed, consider comparing and contrasting your results against theirs, noting where the estimates diverge and providing a possible rationale.  You may also want to show what percentage of the impacts occur above or below national air quality standards or the WHO air quality guidelines. Regardless of the audience, ensure that a final report documents in detail the scope, data, methods and assumptions. Document all input data, including metadata, and provide these in a publicly-available location (e.g. GitHub). As you decide how to relay results to each audience, decide whether to deploy an epidemiological surveillance system that links air pollution to health outcomes; this may help create for your audience a clear and more immediate link between poor air quality and adverse impacts to health (see Public Engagement and Communications Guidance Stage 4, Step 1).