A multidisciplinary European expert team rated the likelihood of causal relationships between ambient UFP exposure and selected health endpoints, including all-cause mortality and cardiovascular and respiratory hospital admissions. The experts also rated the likelihood of potential underlying causal mechanisms that may explain an effect of UFP on cardiac events. The likelihood of a causal relationship between short-term UFP exposure and all-cause mortality, cardiovascular and respiratory hospital admissions, aggravation of asthma symptoms and lung function decrements was rated as medium to high by most of the experts. The ratings for long-term exposure related effects were only slightly lower compared to short-term related effects for mortality and cardiovascular hospital admissions; long-term exposure effects on respiratory morbidity and lung cancer were mainly rated medium. Divergence of opinions among experts was larger for respiratory morbidity and lung cancer than for total mortality and cardiovascular morbidity. There does not appear to be much difference between the ratings of epidemiologists, toxicologists and clinicians, although our sample size is too small to make any definitive statements about potential interdisciplinary variation. Moreover, even though most experts were nominated in one specific category, they might in reality represent the views of multiple disciplines.
From the evaluated causal pathways relating UFP exposure to cardiac events, the pathway involving respiratory inflammation and subsequent thrombotic effects (pathway 1a) was rated most likely. All experts rated the likelihood of this pathway as high or very high. This pathway is most often described in the literature, and the specific steps of the causal pathway were all considered plausible. The lowest ratings were given to the pathway 2b describing translocation of particles affecting the autonomic nervous system (ANS), which in turn may lead to changes in HR, HRV or arrhythmia. For this specific pathway, evidence was considered missing or even contradictory, and the plausibility was thought to be limited. Other pathways involving effects on the ANS (1c, 2a and 2c) were given slightly higher ratings. However, overall, the experts expressed the route via plaque rupture/thrombosis to be more likely to cause a cardiac event than the route via the effects on the ANS. Although we confined the elicitation to mechanisms potentially explaining cardiac effects, (part of) the same or similar mechanisms may also explain the onset and progression of other adverse health effects. Similarly, parts of the proposed mechanisms might also play a role in explaining adverse health effects related to exposure to coarser fractions of PM, other components or air pollution or engineered nanomaterials [1, 11, 19, 51].
The experts were explicitly asked to rate the likelihood of a causal relationship, given the available evidence (i.e. they were not asked to rate the availability of evidence). The general lack of consistent evidence made the process of rating a challenging task for most experts. Complicating factors included that information had to be extrapolated: from animal to humans; from larger particles (PM2.5 and PM10) to UFP; from high doses as frequently used in experimental studies to concentrations similar to those in ambient air ; from (time-series) monitoring measurements to personal exposure; and from the studied sample of the population to the general population. Additionally, experts indicated the uncertainty related to lag times and presence or absence of a threshold of effect, accuracy of death certification and hospital admissions, susceptible groups, as well as variation in the analytical methods employed . The presence of these uncertainties supports the use of formal expert elicitation to assess the likelihood of causality and causal mechanisms.
The likelihood of a causal association between UFP and mortality was assessed as being lower than in a recent US expert panel assessing causality of PM2.5 and mortality . In that study, 10 out of 12 experts rated the likelihood of causality as very high, one as high and one as medium (quantitative numbers translated into our schema using the nearest number). In our study, likelihood of a causal association between UFP and mortality was rated high or very high by 4 out of 12 experts, medium by 7 experts, and low by one. Besides being based on higher evaluations of the likelihood of a causal relation for PM2.5 as compared to UFP, the differences may also be partly due to the composition of the panel or other methodological differences between the studies. Furthermore, it is likely that the larger database of studies on long-term exposure of PM2.5 compared to UFP played an important role in the higher rating in the US study. There was no overlap between experts participating in the US study and those participating in our study.
The overall medium to high likelihood rating of causality of health effects of UFP exposure and the high likelihood rating of at least one plausible mechanism explaining associations between UFP and cardiac events, support the potential usefulness of inclusion of UFP in future Health Impact Assessments (HIA) involving air pollution. The elicited concentration-response functions that were derived on the second day of our expert meeting (Hoek et al. submitted) can be used as input for these assessments. HIAs usually assess the health effects of particulate matter air pollution based on concentration-response functions as derived for e.g. PM2.5 or PM10,  which as such function as a proxy for other (correlated) components of the air pollution mixture. However, PM2.5 and PM10 concentrations do not capture variations of UFP very well; hence the potential effects of UFP have been largely ignored. Especially in the assessment of transport-related air pollution this may be a limitation, as motorized transport affects UFP concentrations more than PM2.5 or PM10. Therefore, in order to provide improved estimates of the health effects of air pollution, UFP effects should be assessed separately. As such we have made a first attempt to fill an important hiatus in current HIA of air pollution. Compared to the evidence for PM2.5, there is still considerable uncertainty, calling for additional research. The results of this expert elicitation give a starting point for evaluating which aspects of the pathways could be focused on in further research.
We have only presented the final likelihood ratings, which were given after plenary discussion, while the individual initial ratings were given before any discussion on the particular question. In a post-hoc comparison of final and initial ratings, we saw that – although most experts stuck to their initial judgments – some did occasionally adjust their rating after group discussion. In those cases, the ratings were usually not changed more than one point (rating class) and mostly in the direction of the mean. We have no reason to believe this to be an effect of 'peer pressure'. Rather, adjustments to initial ratings were brought about by considering new arguments, as indicated in the written motivations, or by harmonization of interpretation of the question within the expert panel, which we think both have increased the value of the final ratings. We did not aim at reaching consensus among the experts. Given the limited state of knowledge in the field, the scientific debate should consider the full spectrum of reasonable hypotheses, and forcing conversion to a consensus view may lead to putting more weight to one of the hypotheses than warranted. Closure of the scientific debate should be based on empirical evidence and not in a negotiated consensus amongst expert judgments. Also, different estimates are often based on completely different motivations, and a central 'consensus' estimate might no longer relate clearly to anyone's viewpoints. Finally, we realize that we have been able to obtain a multidisciplinary sample of respected scientific opinions, but not a representative set of all opinions available on the subject matter in the scientific community. Possibly, the nomination procedure has biased the selection towards 'believers' of UFP as a causal agent. Nonetheless, the set of expert judgments we have assembled is expected to collectively enhance scientific understanding of the (likelihood of) health effects related to UFP exposure and helps to clarify the current state of knowledge.