Do causal concentration–response functions exist? A critical review of associational and causal relations between fine particulate matter and mortality

What is an exposure-response curve?

Exposure-response curves are among the most widely used tools of quantitative health risk assessment. However, we propose that exactly what they mean is usually left ambiguous, making it impossible to answer such fundamental questions as whether and by how much reducing exposure by a stated amount would change average population risks and distributions of individual risks. Recent concepts and computational methods from causal artificial intelligence (CAI) and machine learning (ML) can be applied to clarify what an exposure-response curve means; what other variables are held fixed (and at what levels) in estimating it; and how much inter-individual variability there is around population average exposure-response curves. These advances in conceptual clarity and practical computational methods not only enable epidemiologists and risk analysis practitioners to better quantify population and individual exposure-response curves but also challenge them to specify exactly what exposure-response relationships they seek to quantify and communicate to risk managers and how to use the resulting information to improve risk management decisions.

Re-assessing human mortality risks attributed to PM2.5-mediated effects of agricultural ammonia

How can and should epidemiologists and risk assessors assemble and present evidence for causation of mortality or morbidities by identified agents such as fine particulate matter or other air pollutants? As a motivating example, some scientists have warned recently that ammonia from the production of meat significantly increases human mortality rates in exposed populations by increasing the ambient concentration of fine particulate matter (PM2.5) in air. We reexamine the support for such conclusions, including quantitative calculations that attribute deaths to PM2.5 air pollution by applying associational results such as relative risks, odds ratios, or slope coefficients from regression models to predict the effects on mortality or morbidity of reducing PM2.5 exposures. Taking an outside perspective from the field of causal artificial intelligence (CAI), we conclude that these attribution calculations are methodologically unsound. They produce unreliable conclusions because they ignore an essential distinction between differences in outcomes observed at different levels of exposure and changes in outcomes caused by changing exposure. We find that multiple studies that have examined associations between changes over time in particulate exposure and mortality risk instead of differences in exposures and corresponding mortality risks have found no clear evidence that observed changes in exposure help to predict or explain subsequent changes in mortality risks. We conclude that there is no sound theoretical or empirical reason to believe that reducing ammonia emissions from farms has reduced or would reduce human mortality risks. More generally, applying CAI principles and methods can potentially improve current widespread practices of unsound causal inferences and policy-relevant causal claims that are made without the benefit of formal causal analysis in air pollution health effects research and in other areas of applied epidemiology and public health risk assessment.

Evaluation of a Meta-Analysis of Ambient Air Quality as a Risk Factor for Asthma Exacerbation

Background: An irreproducibility crisis currently afflicts a wide range of scientific disciplines, including public health and biomedical science. A study was undertaken to assess the reliability of a meta-analysis examining whether air quality components (carbon monoxide, particulate matter 10 µm and 2.5 µm (PM10 and PM2.5), sulfur dioxide, nitrogen dioxide and ozone) are risk factors for asthma exacerbation. Methods: The number of statistical tests and models were counted in 17 randomly selected base papers from 87 used in the meta-analysis. Confidence intervals from all 87 base papers were converted to p-values. p-value plots for each air component were constructed to evaluate the effect heterogeneity of the p-values. Results: The number of statistical tests possible in the 17 selected base papers was large, median = 15,360 (interquartile range = 1536–40,960), in comparison to results presented. Each p-value plot showed a two-component mixture with small p-values < 0.001 while other p-values appeared random (p-values > 0.05). Given potentially large numbers of statistical tests conducted in the 17 selected base papers, p-hacking cannot be ruled out as explanations for small p-values. Conclusions: Our interpretation of the meta-analysis is that random p-values indicating null associations are more plausible and the meta-analysis is unlikely to replicate in the absence of bias.

Emulating causal dose-response relations between air pollutants and mortality in the Medicare population

BACKGROUND

Fine particulate matter (PM2.5), ozone (O3), and nitrogen dioxide (NO2) are major air pollutants that pose considerable threats to human health. However, what has been mostly missing in air pollution epidemiology is causal dose-response (D-R) relations between those exposures and mortality. Such causal D-R relations can provide profound implications in predicting health impact at a target level of air pollution concentration.

METHODS

Using national Medicare cohort during 2000-2016, we simultaneously emulated causal D-R relations between chronic exposures to fine particulate matter (PM2.5), ozone (O3), and nitrogen dioxide (NO2) and all-cause mortality. To relax the contentious assumptions of inverse probability weighting for continuous exposures, including distributional form of the exposure and heteroscedasticity, we proposed a decile binning approach which divided each exposure into ten equal-sized groups by deciles, treated the lowest decile group as reference, and estimated the effects for the other groups. Binning continuous exposures also makes the inverse probability weights robust against outliers.

RESULTS

Assuming the causal framework was valid, we found that higher levels of PM2.5, O3, and NO2 were causally associated with greater risk of mortality and that PM2.5 posed the greatest risk. For PM2.5, the relative risk (RR) of mortality monotonically increased from the 2nd (RR, 1.022; 95% confidence interval [CI], 1.018-1.025) to the 10th decile group (RR, 1.207; 95% CI, 1.203-1.210); for O3, the RR increased from the 2nd (RR, 1.050; 95% CI, 1.047-1.053) to the 9th decile group (RR, 1.107; 95% CI, 1.104-1.110); for NO2, the DR curve wiggled at low levels and started rising from the 6th (RR, 1.005; 95% CI, 1.002-1.018) till the highest decile group (RR, 1.024; 95% CI, 1.021-1.027).

CONCLUSIONS

This study provided more robust evidence of the causal relations between air pollution exposures and mortality. The emulated causal D-R relations provided significant implications for reviewing the national air quality standards, as they inferred the number of potential early deaths prevented if air pollutants were reduced to specific levels; for example, lowering each air pollutant concentration from the 70th to 60th percentiles would prevent 65,935 early deaths per year.

Short-term exposure to sulphur dioxide (SO2) and all-cause and respiratory mortality: A systematic review and meta-analysis

BACKGROUND

Many studies have assessed the harmful effects of ambient air pollution on human mortality, but the evidence needs further exploration, analysis, and refinement, given the large number of studies that have been published in recent years. The objective of this study was to evaluate all the available evidence of the effect of short-term exposure to ambient sulphur dioxide (SO₂) on all-cause and respiratory mortality.

METHODS

Articles reporting observational epidemiological studies were included, comprising time-series and case-crossover designs. A broad search and wide inclusion criteria were considered, encompassing international and regional databases, with no geographical or language restrictions. A random effect meta-analysis was conducted, and pooled relative risk for an increment of 10 µg/m³ in SO₂ concentrations were calculated for each outcome. We analysed the risk of bias (RoB) in individual studies for specific domains using a new domain-based RoB assessment tool, and the certainty of evidence across studies with an adaptation of the Grading of Recommendations Assessment, Development and Evaluation approach. The certainty of evidence was judged separately for each exposure-outcome combination. A number of subgroup and sensitivity analyses were carried out, as well as assessments of heterogeneity and potential publication bias. The protocol for this review was registered with PROSPERO (CRD42019120738).

RESULTS

Our search retrieved 1,128 articles, from which 67 were included in quantitative analysis. The RoB was low or moderate in the majority of articles and domains. An increment of 10 µg/m³ in SO₂ (24-hour average) was associated with all-cause mortality (RR: 1.0059; 95% CI: 1.0046-1.0071; p-value: <0.01), and respiratory mortality (RR: 1.0067; 95% CI: 1.0025-1.0109; p-value: <0.01), while the same increment in SO₂ (1-hour max.) was associated with respiratory mortality (RR:1.0052; 95% CI: 1.0013-1.0091; p-value: 0.03). Similarly, the association was positive but non-significant for SO₂ (1-hour max.) and all-cause mortality (RR: 1.0016; 95% CI: 0.9930-1.0102; p-value: 0.60). These associations were still significant after the adjustment for particulate matter, but not for other pollutants, according to the results from 13 articles that evaluated co-pollutant models. In general, linear concentration-response functions with no thresholds were found for the two outcomes, although this was only evaluated in a small number of studies. We found signs of heterogeneity for SO₂ (24-hour average) - respiratory mortality and SO₂ (1-hour max.) - all-cause mortality, and funnel plot asymmetry for SO₂ (24-hour average) - all-cause mortality. The certainty of evidence was high in two combinations, i.e. SO₂ (24-hour average) - all-cause mortality and SO₂ (1-hour max.) - respiratory mortality, moderate in one combination, i.e. SO₂ (24-hour average) - respiratory mortality, and low in the remaining one combination.

CONCLUSIONS

Positive associations were found between short-term exposure to ambient SO₂ and all-cause and respiratory mortality. These associations were robust against several sensitivity analyses, and were judged to be of moderate or high certainty in three of the four exposure-outcome combinations.

Short-term exposure to ozone, nitrogen dioxide, and sulphur dioxide and emergency department visits and hospital admissions due to asthma: A systematic review and meta-analysis

BACKGROUND

Air pollution is a major environmental hazard to human health and a leading cause of morbidity for asthma worldwide.

OBJECTIVES

To assess the current evidence on short-term effects (from several hours to 7 days) of exposure to ozone (O₃), nitrogen dioxide (NO₂), and sulphur dioxide (SO₂) on asthma exacerbations, defined as emergency room visits (ERVs) and hospital admissions (HAs).

METHODS

We searched PubMed/MEDLINE, EMBASE and other electronic databases to retrieve studies that investigated the risk of asthma-related ERVs and HAs associated with short-term exposure to O₃, NO₂, or SO₂. We evaluated the risks of bias (RoB) for individual studies and the certainty of evidence for each pollutant in the overall analysis. A subgroup analysis was performed, stratified by sex, age, and type of asthma exacerbation. We conducted sensitivity analysis by excluding the studies with high RoB and based on the E-value. Publication bias was examined with the Egger's test and with funnel plots.

RESULTS

Our literature search retrieved 9,059 articles, and finally 67 studies were included, from which 48 studies included the data on children, 21 on adults, 14 on the elderly, and 31 on the general population. Forty-three studies included data on asthma ERVs, and 25 on asthma HAs. The pooled relative risk (RR) per 10 µg/m³ increase of ambient concentrations was 1.008 (95%CI: 1.005, 1.011) for maximum 8-hour daily or average 24-hour O₃, 1.014 (95%CI: 1.008, 1.020) for average 24-hour NO₂, 1.010 (95%CI: 1.001, 1.020) for 24-hour SO₂, 1.017 (95%CI: 0.973, 1.063) for maximum 1-hour daily O₃, 0.999 (95%CI: 0.966, 1.033) for 1-hour NO₂, and 1.003 (95%CI: 0.992, 1.014) for 1-hour SO₂. Heterogeneity was observed in all pollutants except for 8-hour or 24-hour O₃ and 24-hour NO₂. In general, we found no significant differences between subgroups that can explain this heterogeneity. Sensitivity analysis based on the RoB showed certain differences in NO₂ and SO₂ when considering the outcome or confounding domains, but the analysis using the E-value showed that no unmeasured confounders were expected. There was no major evidence of publication bias. Based on the adaptation of the Grading of Recommendations Assessment, Development and Evaluation, the certainty of evidence was high for 8-hour or 24-hour O₃ and 24-hour NO₂, moderate for 24-hour SO₂, 1-hour O₃, and 1-hour SO₂, and low for 1-hour NO₂.

CONCLUSION

Short-term exposure to daily O₃, NO₂, and SO₂ was associated with an increased risk of asthma exacerbation in terms of asthma-associated ERVs and HAs.

Risk of Bias Assessments and Evidence Syntheses for Observational Epidemiologic Studies of Environmental and Occupational Exposures: Strengths and Limitations

BACKGROUND

Increasingly, risk of bias tools are used to evaluate epidemiologic studies as part of evidence synthesis (evidence integration), often involving meta-analyses. Some of these tools consider hypothetical randomized controlled trials (RCTs) as gold standards.

METHODS

We review the strengths and limitations of risk of bias assessments, in particular, for reviews of observational studies of environmental exposures, and we also comment more generally on methods of evidence synthesis.

RESULTS

Although RCTs may provide a useful starting point to think about bias, they do not provide a gold standard for environmental studies. Observational studies should not be considered inherently biased vs. a hypothetical RCT. Rather than a checklist approach when evaluating individual studies using risk of bias tools, we call for identifying and quantifying possible biases, their direction, and their impacts on parameter estimates. As is recognized in many guidelines, evidence synthesis requires a broader approach than simply evaluating risk of bias in individual studies followed by synthesis of studies judged unbiased, or with studies given more weight if judged less biased. It should include the use of classical considerations for judging causality in human studies, as well as triangulation and integration of animal and mechanistic data.

CONCLUSIONS

Bias assessments are important in evidence synthesis, but we argue they can and should be improved to address the concerns we raise here. Simplistic, mechanical approaches to risk of bias assessments, which may particularly occur when these tools are used by nonexperts, can result in erroneous conclusions and sometimes may be used to dismiss important evidence. Evidence synthesis requires a broad approach that goes beyond assessing bias in individual human studies and then including a narrow range of human studies judged to be unbiased in evidence synthesis. https://doi.org/10.1289/EHP6980.

Short-term exposure to particulate matter (PM10 and PM2.5), nitrogen dioxide (NO2), and ozone (O3) and all-cause and cause-specific mortality: Systematic review and meta-analysis

BACKGROUND

Air pollution is a leading cause of mortality and morbidity worldwide. Short-term exposure (from one hour to days) to selected air pollutants has been associated with human mortality. This systematic review was conducted to analyse the evidence on the effects of short-term exposure to particulate matter with aerodynamic diameters less or equal than 10 and 2.5 µm (PM_10, PM_2.5), nitrogen dioxide (NO₂), and ozone (O₃), on all-cause mortality, and PM₁₀ and PM_2.5 on cardiovascular, respiratory, and cerebrovascular mortality.

METHODS

We included studies on human populations exposed to outdoor air pollution from any source, excluding occupational exposures. Relative risks (RRs) per 10 µg/m³ increase in air pollutants concentrations were used as the effect estimates. Heterogeneity between studies was assessed using 80% prediction intervals. Risk of bias (RoB) in individual studies was analysed using a new domain-based assessment tool, developed by a working group convened by the World Health Organization and designed specifically to evaluate RoB within eligible air pollution studies included in systematic reviews. We conducted subgroup and sensitivity analyses by age, sex, continent, study design, single or multicity studies, time lag, and RoB. The certainty of evidence was assessed for each exposure-outcome combination. The protocol for this review was registered with PROSPERO (CRD42018087749).

RESULTS

We included 196 articles in quantitative analysis. All combinations of pollutants and all-cause and cause-specific mortality were positively associated in the main analysis, and in a wide range of sensitivity analyses. The only exception was NO₂, but when considering a 1-hour maximum exposure. We found positive associations between pollutants and all-cause mortality for PM₁₀ (RR: 1.0041; 95% CI: 1.0034-1.0049), PM_2.5 (RR: 1.0065; 95% CI: 1.0044-1.0086), NO₂ (24-hour average) (RR: 1.0072; 95% CI: 1.0059-1.0085), and O₃ (RR: 1.0043; 95% CI: 1.0034-1.0052). PM₁₀ and PM_2.5 were also positively associated with cardiovascular, respiratory, and cerebrovascular mortality. We found some degree of heterogeneity between studies in three exposure-outcome combinations, and this heterogeneity could not be explained after subgroup analysis. RoB was low or moderate in the majority of articles. The certainty of evidence was judged as high in 10 out of 11 combinations, and moderate in one combination.

CONCLUSIONS

This study found evidence of a positive association between short-term exposure to PM₁₀, PM_2.5, NO₂, and O₃ and all-cause mortality, and between PM₁₀ and PM_2.5 and cardiovascular, respiratory and cerebrovascular mortality. These results were robust through several sensitivity analyses. In general, the level of evidence was high, meaning that we can be confident in the associations found in this study.

Attributable Risk to Assess the Health Impact of Air Pollution: Advances, Controversies, State of the Art and Future Needs

Despite the increased attention given to the health impact assessment of air pollution and to the strategies to control it in both scientific literature and concrete interventions, the results of the implementations, especially those involving traffic, have not always been satisfactory and there is still disagreement about the most appropriate interventions and the methods to assess their effectiveness. This state-of-the-art article reviews the recent interpretation of the concepts that concern the impact assessment, and compares old and new measurements of attributable risk and attributable fraction. It also summarizes the ongoing discussion about the designs and methods for assessing the air pollution impact with particular attention to improvements due to spatio-temporal analysis and other new approaches, such as studying short term effects in cohorts, and the still discussed methods of predicting the values of attributable risk (AR). Finally, the study presents the more recent analytic perspectives and the methods for directly assessing the effects of not yet implemented interventions on air quality and health, in accordance with the suggestion in the strategic plan 2020-2025 from the Health Effect Institute.

Air Pollution and Suicide in Mexico City: A Time Series Analysis, 2000-2016

The association between air pollution and suicide has recently been under examination, and the findings continue to be contradictory. In order to contribute evidence to this still unresolved question, the objective of the present study was to evaluate the association between air quality and daily suicides registered in Mexico City (MC) between 2000 and 2016. Air quality was measured based on exposure to particulate matter under 2.5 and 10 micrometers (µm) (PM_2.5 and PM₁₀, respectively), ozone (O₃), nitrogen dioxide (NO₂), and sulfur dioxide (SO₂), adjusting for weather variables (air temperature and relative humidity), and holidays. To this end, an ecologic time series analysis was performed using a Poisson regression model conditioned by time and stratified by gender and age groups. Models were also generated to explore the lagged and accumulative effects of air pollutants, adjusted by weather variables. The effects of the pollutants were very close to the null value in the majority of the models, and no accumulative effects were identified. We believe these results, in this case, no evidence of a statistical association, contribute to the current debate about whether the association between air pollution and suicide reported in the scientific literature reflects an actual effect or an uncontrolled confounding effect.

The economic benefits of fulfilling the World Health Organization's limits for particulates: A case study in Algeciras Bay (Spain)

Algeciras Bay is an important industrial and port zone in the south of Spain whose pollution by particulate matter surpasses the threshold levels recommended by the World Health Organization (WHO) in its 2005 Guide on Air Quality. This study analyses the mortality avoided and the economic benefit which would be derived from a reduction of the pollution of PM2.5 and PM10 to the levels recommended by the WHO in Algeciras Bay in the period 2005-2015. The analysis carried out shows that the industrial zones, such as Los Barrios and San Roque, are those which have greater levels of pollution and in which the relative risk is greater. The calculations for Algeciras Bay between 2000 and 2015 show 182 deaths which would be avoided if the particulate matter pollution were reduced to the levels recommended by the WHO. Likewise, the economic valuation which this impact has on health is carried out through two concepts: the cost of illness and the Value of Statistical Life (VSL). The result shows that the economic benefit that would come out with the cost of illness valuation is 5,329,110€ and from the VSL is 414,787,113€. Implications: PM_2.5 has a greater concentration in industrial localities and is linked to the industrial activity. When the particulate matter pollution is reduced to the levels recommended by the WHO in an industrialised area such as Algeciras (Spain), 182 deaths which would be avoided. The result shows that the economic benefit that would come out with the cost of illness valuation is 5,329,110€ and from the value of statistical life is 414,787,113€.

Causal Modeling in Environmental Health

The field of environmental health has been dominated by modeling associations, especially by regressing an observed outcome on a linear or nonlinear function of observed covariates. Readers interested in advances in policies for improving environmental health are, however, expecting to be informed about health effects resulting from, or more explicitly caused by, environmental exposures. The quantification of health impacts resulting from the removal of environmental exposures involves causal statements. Therefore, when possible, causal inference frameworks should be considered for analyzing the effects of environmental exposures on health outcomes.

Evaluation of a meta-analysis of air quality and heart attacks, a case study

It is generally acknowledged that claims from observational studies often fail to replicate. An exploratory study was undertaken to assess the reliability of base studies used in meta-analysis of short-term air quality-myocardial infarction risk and to judge the reliability of statistical evidence from meta-analysis that uses data from observational studies. A highly cited meta-analysis paper examining whether short-term air quality exposure triggers myocardial infarction was evaluated as a case study. The paper considered six air quality components - carbon monoxide, nitrogen dioxide, sulphur dioxide, particulate matter 10 μm and 2.5 μm in diameter (PM10 and PM2.5), and ozone. The number of possible questions and statistical models at issue in each of 34 base papers used were estimated and p-value plots for each of the air components were constructed to evaluate the effect heterogeneity of p-values used from the base papers. Analysis search spaces (number of statistical tests possible) in the base papers were large, median = 12,288 (interquartile range = 2496 - 58,368), in comparison to actual statistical test results presented. Statistical test results taken from the base papers may not provide unbiased measures of effect for meta-analysis. Shapes of p-value plots for the six air components were consistent with the possibility of analysis manipulation to obtain small p-values in several base papers. Results suggest the appearance of heterogeneous, researcher-generated p-values used in the meta-analysis rather than unbiased evidence of real effects for air quality. We conclude that this meta-analysis does not provide reliable evidence for an association of air quality components with myocardial risk.

Modernizing the Bradford Hill criteria for assessing causal relationships in observational data

Perhaps no other topic in risk analysis is more difficult, more controversial, or more important to risk management policy analysts and decision-makers than how to draw valid, correctly qualified causal conclusions from observational data. Statistical methods can readily quantify associations between observed variables using measures such as relative risk (RR) ratios, odds ratios (OR), slope coefficients for exposure or treatment variables in regression models, and quantities derived from these measures. Textbooks of epidemiology explain how to calculate population attributable fractions, attributable risks, burden-of-disease estimates, and probabilities of causation from relative risk (RR) ratios. Despite their suggestive names, these association-based measures have no necessary connection to causation if the associations on which they are based arise from bias, confounding, p-hacking, coincident historical trends, or other noncausal sources. But policy analysts and decision makers need something more: trustworthy predictions - and, later, evaluations - of the changes in outcomes caused by changes in policy variables. This concept of manipulative causation differs from the more familiar concepts of associational and attributive causation most widely used in epidemiology. Drawing on modern literature on causal discovery and inference principles and algorithms for drawing limited but useful causal conclusions from observational data, we propose seven criteria for assessing consistency of data with a manipulative causal exposure-response relationship - mutual information, directed dependence, internal and external consistency, coherent causal explanation of biological plausibility, causal mediation confirmation, and refutation of non-causal explanations - and discuss to what extent it is now possible to automate discovery of manipulative causal dependencies and quantification of causal effects from observational data. We compare our proposed principles for causal discovery and inference to the traditional Bradford Hill considerations from 1965. Understanding how old and new principles are related can clarify and enrich both.

Effects of exposure estimation errors on estimated exposure-response relations for PM2.5

Associations between fine particulate matter (PM2.5) exposure concentrations and a wide variety of undesirable outcomes, from autism and auto theft to elderly mortality, suicide, and violent crime, have been widely reported. Influential articles have argued that reducing National Ambient Air Quality Standards for PM2.5 is desirable to reduce these outcomes. Yet, other studies have found that reducing black smoke and other particulate matter by as much as 70% and dozens of micrograms per cubic meter has not detectably affected all-cause mortality rates even after decades, despite strong, statistically significant positive exposure concentration-response (C-R) associations between them. This paper examines whether this disconnect between association and causation might be explained in part by ignored estimation errors in estimated exposure concentrations. We use EPA air quality monitor data from the Los Angeles area of California to examine the shapes of estimated C-R functions for PM2.5 when the true C-R functions are assumed to be step functions with well-defined response thresholds. The estimated C-R functions mistakenly show risk as smoothly increasing with concentrations even well below the response thresholds, thus incorrectly predicting substantial risk reductions from reductions in concentrations that do not affect health risks. We conclude that ignored estimation errors obscure the shapes of true C-R functions, including possible thresholds, possibly leading to unrealistic predictions of the changes in risk caused by changing exposures. Instead of estimating improvements in public health per unit reduction (e.g., per 10 µg/m³ decrease) in average PM2.5 concentrations, it may be essential to consider how interventions change the distributions of exposure concentrations.