In this issue, Rich et al present findings that strengthen the evidence for a causal role of air pollution in triggering arrhythmias among patients with implanted cardioverter defibrillators.¹ The use of objective health and exposure data in conjunction with the chosen case-crossover approach underscores the power of “quasi experimental” epidemiological research in this field. Based on only 139 ventricular arrhythmias (VA) recorded on implantable cardioverter defibrillators among 56 patients, the investigators found statistically significant associations between markers of acute exposure to air pollution and arrhythmias. Associations with the 24 hour mean concentrations prior to the event were stronger than with both shorter (6 and 12 hour) and longer (48 hour) time windows.

The mechanisms linking ambient pollution with arrhythmias remain to be elucidated. However, the plausibility and evidence for effects of pollutants on autonomic function are steadily increasing.² In a series of experiments, exposure of dogs to concentrated ambient particles led to cardiac and respiratory changes mediated via both the sympathetic nervous system and the vagus nerve.³ A recent controlled experiment in humans observed reduced cardiac vagal control after exposure to 200 ppb SO₂.⁴ These changes in autonomic function reflect increased susceptibility to cardiac arrhythmias.

The new study leaves open questions that reflect both limitations and challenges of the case-crossover approach. Case-crossover analyses (if based on a small number of events) come with the same inherent assumption as panel studies, namely that exposure of a specific individual—rather than the total population average—can be characterised for the time window prior to the event, as well as for the chosen control day(s). Fixed site monitors do not necessarily accomplish this on the individual level. Moreover, the validity of this assumption grossly varies for different pollutants and time windows.^5,6 This leaves us with the problem that comparisons of estimates across pollutants or time windows (see tables 2 and 3 in Rich et al¹) are of inherent interest but of questionable validity; for example, when measured at a single site, sulphur certainly reflects the exposure of a given subject to urban background pollution much better than more spatially heterogeneous pollutants such as NO₂ or elemental carbon, while in the case of ozone, the monitor station may entirely fail to characterise exposure.⁷ Thus associations between pollutants (or different time windows of exposure) and arrhythmias may differ not only for biological reasons but due to different degrees of misclassification.

Correlations between outdoor and personal concentrations give some insight into this differential misclassification. Janssen et al reported high outdoor-to-personal correlations (⩾0.90) for 24 hour mean sulphur mass concentration, but much more subject specific and lower correlations for black smoke and PM_2.5 mass.⁵ In the EXPOLIS study, the correlation of personal and outdoor 48 hour mean CO concentrations ranged between 0.33 and 0.77 across five European cities, whereas correlations for lead (as a marker of traffic related PM) and potassium (as a marker for crustal PM) were 0.53 and 0.21 respectively.⁸ Therefore, extensive pollution speciation at a single “super site” monitor will not provide better insight in the health relevance of pollutants (or sources). Personal or close-to-personal exposure measurements are needed for that purpose. Rich et al also discuss how effects depend on wind direction—another promising approach to understand the health relevance of pollution from specific sources or air sheds.

Comparisons of effects across pollutants (table 2 in Rich et al¹) need to rely on comparable scales. This is usually achieved by scaling pollutants to comparable contrasts, for example to the highest versus lowest or to the interquartile range (IQR). The comparable exposure scale in the case-crossover design is the difference of pollutant concentrations between event and control day(s) rather than the ambient concentrations.⁹ The two distributions are not necessarily the same, and Rich and colleagues’ paper is the first air pollution case-crossover study to reveal both the ambient and the design relevant distribution of differences. Comparisons (across pollutants) are valid on the latter scale only. As seen in table 1 of Rich et al, IQRs were similar on both scales except for O₃ where the variability of the relevant term was far smaller than across the ambient concentrations. Thus, the interpretation of their results in table 2—based on ambient scale—would not change much.

Forastiere et al conducted a case-crossover analysis of air pollution effects on cardiac arrest.¹⁰ The published results were also based on IQR of the ambient concentrations. Table 1 presents the IQR of the difference between case and control exposure. In the original table, effects (as odds ratios (OR) per IQR) appear clearly stronger for particle number count (PNC). However, if assessed on the scale that allows direct comparisons, the effect sizes of PNC, PM₁₀, and CO are similar. The example demonstrates that the use of the relevant exposure term is crucial to appropriately investigate, interpret, and compare effects.⁹

In summary, Rich et al further support the notion that air pollutants trigger arrhythmias. The relevance of different pollutants and time windows of exposure remain to be elucidated, and case-crossover studies may play an important role in this. The use of personal exposure assessment and of appropriate exposure terms will be critical for valid comparisons across pollutants, constituents, and time windows of exposure.