Reflections on burden of disease and health impact assessment methods for noise Martin Röösli 1 , Benedikt Wicki 2 , Danielle Vienneau 3 Swiss Tropical and Public Health Institute. Kreuzstrasse 2, 4123 Allschwil, Switzerland. University of Basel. Petersplatz 1, 4001, Basel, Switzerland.

ABSTRACT Health risk assessments (HRA) includes Burden of Disease (BoD) studies to assess the noise impact on health at a given time and in a given population, and Health Impact Assessment studies (HIA) conducted in the frame of decision-making to evaluate health effects of a policy, programme or pro- ject. In a 2020 report, the European Environmental Agency estimates transportation noise exposure in Europe annually results in 22 million highly annoyed, 6.5 million highly sleep disturbed, 48,000 cases of ischemic heart diseases and 12,000 deaths, and 12,400 children with cognitive impairments. These are based on END exposure assessment and exposure-response functions from the WHO noise guidelines 2018. Many more regional or national HRA studies have been conducted, and common methods across future assessments for consistent communication are desirable. Given the rapid de- velopments in noise research, current HRA are partly outdated. Critical methodological issues will be discussed using national and regional examples with a focus on Switzerland. This includes the selection of noise sources, relevant health outcomes and corresponding exposure-response functions. Derivation of the lowest effect noise threshold, potential double-counting, and communication of re- sults are also discussed with the aim toward a common understanding for the conduct of future HRA.

1. INTRODUCTION

Noise exposure has negative impacts on human health through various mechanisms. High levels of noise cause chronic sleep disturbances with well-established consequences for cardio-metabolic and mental health. Noise is also a stressor that can lead to the activation of the autonomous nervous system and the hypothalamus-pituitary-adrenal (HPA) axis. This results in changes of blood pressure, heart rate variability, glucose metabolism and lipid metabolism that then contribute to an increased risk of cardiovascular disease, metabolic syndrome (diabetes) and mental health. In the WHO Environmental Noise Guidelines for the European Region (ENG) noise research published until 2015 has been eval- uated to derive guidelines [1]. Specifically, in relation to road, rail, aircraft and wind turbine noise, systematic reviews and meta-analyses have been conducted for the following critical outcomes: inci- dence of ischemic heart disease (IHD), incidence of hypertension, percentage of highly annoyed (%HA), percentage of highly sleep disturbed (%HSD) and reading and oral comprehension. Further, permanent hearing impairment from leisure noise such as personal audio players was evaluated. In addition to these critical outcomes the following important outcomes were also evaluated in the WHO

1 martin.roosli@swisstp.ch

2 benedikt.wicki@swisstp.ch

3 danielle.vienneau@swisstp.ch

inter.noise 21-24 AUGUST SCOTTISH EVENT CAMPUS. ? O? ? GLASGOW

report: adverse birth outcomes (birth weight, pre-term delivery, small for gestational age), quality of life, well-being and mental health (emotional and conduct disorders in childhood, self-reported qual- ity of life and various measures of depression, anxiety and psychological distress) as well as metabolic outcomes (diabetes, overweight). In general, research on these outcomes was scarce and less conclu- sive. In a health risk assessment, the exposure distribution of the target population is combined with expo- sure response functions, which may be derived from a different context, to obtain the number of people affected by the exposure. In general, HRA can be differentiated in Burden of disease estimates (BoD) or health impact assessments (HIA): • BoD : The public health burden associated with current or projected levels of noise is estimated. This may be done for specific sources of noise and/or for selected economic sectors . • HIA : The human health impacts are estimated for a current policy or implemented action. It also may include human health benefits associated with changing noise policy or applying a more stringent noise standard.

In general, a HRA consists of four major steps:

1. Defining the scenario. This includes decisions about the noise sources to be included, the reference year, the geographic region, the choice of the counterfactuals and any other deci- sions relevant for conducting the HRA. The counterfactual for a BoD would typically entail a situation without any noise, i.e. the minimum noise level considered to have some health impact. In a HIA, the counterfactual could be the current policy situation or any other refer- ence situation. 2. Estimation of the exposure distribution for the target population for the selected noise sources. 3. Selecting the outcomes and evaluating the exposure-response association for the selected out- comes. 4. Quantify the impact in form of number of attributable cases/deaths, number years of life lost, years lost due to disability or DALYs etc.

ENG provided methodological guidance for conducting HRA in chapter 5.5 [1]. Since the publication of ENG most HRA studies are based on this guidance and used data from this report. However, data published in ENG represents the state of studies in 2015. Given the steep increase in noise research in the last decade, it is thus outdated to some extent. The aim of this paper is to critical discuss some methodological aspects relevant for HRA studies as a basis for a broader discussion.

2. RESULTS AND DISCUSSION

2.1. Exposure assessment

2.1.1 Noise sources In the WHO ENG scientific basis was considered to be sufficient to conduct HRA for road traffic, railway and aircraft noise as well as wind turbine noise [1]. The EEA also considered industrial noise in the HRA [2]. Other noise have been rarely considered in HRA studis. In principle, there are many additional noise sources such as military noise, neighbourhood noise, construction work, sporting and music events, fireworks, church bells, playing with noisy toys and video games, listening to personal music players and many more. The selection of noise sources to be included may depend on the overall aim of a HRA. The more prevalent a noise source, the more relevant it may be for the health of the population. For non-transportation noise source, a challenge is the fact that studies on health effects are scarce, in particular if it comes to somatic disease such as cardiovascular diseases, diabetes

inter.noise 21-24 AUGUST SCOTTISH EVENT CAMPUS. ? O? ? GLASGOW

or depression. Thus, additional assumptions about transferability of exposure-response functions are needed (see chapter 4.3). In case, several noise sources are considered in the same HRA, the question arises whether effects are additive. ENG concluded that in terms of the proposed outcomes and noise sources the effects are cumulative [1]. However, two aspects need to be considered. First, if the impact is expressed as number of people affected (e.g. highly sleep disturbed or highly annoyed), one needs to consider that the same person may be annoyed from more than one source. Thus, just adding up the number of highly affected people may be misleading if this aspect is not made transparent, although one may argue that being annoyed by several sources is an additional burden. This becomes particularly relevant with increasing proportion of affected people and with increasing number of noise sources included in a specific HRA. For relatively rare diseases like car- diovascular disease or diabetes this aspect is mostly negligible because in this case the exposure- response functions refers to a risk increase. Whether this risk increase accumulates in the same person or in different persons does eventually not matter for the number of attributable cases.

Second, noise exposure from various sources may be correlated because in urban areas it is general noisier than in rural areas. This may affect additivity of the impacts from various sources if the expo- sure-response association is derived from studies that have only considered one nose source. This aspect is for instance well recognized for HRA studies of air pollution and that is why usually only one indicator pollutant is used in these studies (PM10 or PM2.5), sometimes complemented with NO2 and O3, the latter typically not correlated with particulate matter. For noise, is not recommended to just use one indicator noise source, as correlation between various noise sources is typically much lower because of the higher spatial heterogeneity. In principle, the most preferred method to deal with this challenge is to derive exposure response functions from studies that have concurrently considered in their analysis more than one noise source, e.g. road, railway and aircraft noise. In this exposure- response associations are mutually independent. However, this would reduce the number of available studies and may interfere with the statistical precision. In the context of air pollution HRA, a method has been proposed to combine results from single and multi-pollutant models in the COMEAP (Com- mittee on the Medical Effects of Air Pollutants) study [3]. The basic idea is to compare exposure- response function from single-pollutant studies with multi-pollutant studies. If the latter have a lower slope due to the correlation, one could make a corresponding correction of the exposure-response functions derived from single pollutant studies. The advantage of this approach is that all available evidence is considered, whereas restriction to multi-pollutant studies may result in less powered ex- posure-response functions and potential selection bias in terms of study area. However, the compar- ison may be affected by other between study factors and/or random variability and that differences between single- and multi-pollutant may not be because of the underlying correlation of various noise sources yielding to an overcorrection.

2.1.2 Exposure assessment For noise exposure assessment in HRA, it is most critical that the noise exposure distribution of the target population is correctly estimated, whereas individual predictions are not needed. Current HRA assessments are mostly based on spatial noise modelling developed to predict individual exposure. A wide variety of calculation methods and approaches are used for deriving the noise exposure distri- bution of the population or to produce noise maps. An overview about applications in EEA country is given in Table 1.2 and Box 1.6 of the corresponding report [2]. Whereas the underlying physical propagation algorithms are established, systematic differences may occur due to various quality of input data such as traffic counts or the implementation of buildings in the model. To date, no study has systematically compared the different methods for noise modelling. It is very likely that the results

inter.noise 21-24 AUGUST SCOTTISH EVENT CAMPUS. ? O? ? GLASGOW

of different countries or years generated by different prediction methods are not fully comparable and thus should be interpreted with caution. As an example, in a recent BoD study, Khomenko et al [4] used the END dataset to estimate distribution of noise in 724 cities and 25 greater cities in 25 Euro- pean countries. In terms of the 25 European capitals substantial differences were found in the propor- tion of people exposed >55 dB L den ranging from 99.8 in Sofia, 93.3 in Luxembourg city and 81.6% in Riga to 33.8 in London, 30.3% in Brussels and 29.8% in Berlin. Although differences between cities are to be expected, the observed pattern suggests that a large part of the reported differences are due to methodological differences in exposure assessments. These differences in noise exposure assessment translate directly into local BoD estimates and are thus of critical influence of the esti- mated impact. It is important to realize that many countries outside Europe do not have any estimates of the noise exposure distribution of their population. Thus, novel methods may be developed to most accurately estimate the noise exposure distribution in a population, which is comparable across countries. Ide- ally, such a methods does not depend on the availability of detailed local data on noise sources and other characteristics as this may introduce systematic differences between countries. Methods, which may be easier to be applied on a global scale, may include land use regression techniques [5, 6] or modelling taking advantage of remote sensing data calibrated with local data.

2.2 Selection of outcomes ENG has made recommendations for the proposed outcomes and noise sources (see Table 1). Table 1: Quantifiable health outcomes in HIA from different noise sources according to ENG [1], p. 109

Road traffic Railway traffic Aircraft Wind turbines Annoyance     Sleep disturbance    - Ischemic Heart Diseases  - potentially - Stroke potentially - - - Diabetes potentially - - - Reading and oral com- prehension in children - -  -

Change in waist circum-

ference - - potentially -

Since the conduct of the systematic reviews by ENG, a substantial amount of new research on the health effects of noise has been published and thus more up to date systematic reviews and meta- analysis should inform the choice of outcomes. Potential additional priority outcomes to consider are diabetes, overweight and mental health problems including depression, partly already indicated in ENG as “potentially” to be included. In terms of cardiovascular disease there is emerging evidence that noise does not only cause ischemic heart but all kind of diagnosis. Thus, it may be more appro- priate to estimate the impact on all types of cardiovascular diseases. Other outcomes that may be affected by noise exposure, although research is not conclusive, includes behavioral problems in chil- dren and adolescents, cognitive impairment in children and adults (in children beyond aircraft expo- sure), additional metabolic outcomes, adverse birth outcomes, tinnitus and hearing impairment (re- lated to environmental noise). It is eye-catching that HRA studies in the field of noise typically use relatively narrow inclusion criteria for outcomes whereas for air pollution HRA broader outcomes definitions are used such as all natural cause mortality. This may partly also reflect the available data from original studies, which are also mostly based on narrow diagnosis criteria such as myocardial infarction or ischemic heart

inter.noise 21-24 AUGUST SCOTTISH EVENT CAMPUS. ? O? ? GLASGOW

disease. Noise studies on all natural cause mortality are still relatively rare [7]. However, stress reac- tion from noise are expected to be systemic and one may argue that many different type of diseases may be affected including cancer [8] or respiratory system [9, 10]. Thus, it may be worth to consider all natural cause mortality as an outcome for future HRA. This would have substantial implications as the baseline rate is much higher than for specific diseases. This implies that even with the same relative risk the impact would be considerably higher. Further, the results of cost-benefit analyses is mostly driven by the mortality estimates since the costs for deaths are substantially higher than for diseases. Thus, the decision which mortality outcomes to include, would have major implications for the result.

2.3 Exposure-response relationship Most HRA risk assessments conducted after the publication of the WHO ENG used corresponding exposure-response functions, which represent pooled exposure-response functions from various ge- ographic regions. This assumes that exposure-response relationships are transferable between various geographic regions and between different time periods. Transferability of exposure-response func- tions is a general source of uncertainty in HRA studies of environmental exposures such as air pollu- tion or ionizing radiation since difference in the population at risk, the exposure circumstances or other aspects may prevent from transfer an exposure-response function from one area to another. For noise, the corresponding uncertainty may be even more pronounced than for other environmental pollutants. Exposure-effect relationships in large epidemiological noise studies on chronic diseases are based on the modelled noise exposure at the house facade (typically the maximum value) as a surrogate for the critical physiological measure, which is the indoor noise level at the ear of the study participants. Thus, the transferability of study results is therefore hampered by differences in noise characteristics (e.g. vehicle fleet mix, regulation for night noise, road surfaces, etc.) or the attenuation properties of residential buildings in addition to differences in noise modelling as discussed in chapter 4.1.2. Further, socio-cultural differences are also expected to play a role in particular for self-reported outcomes such as high annoyance or high sleep disturbance. For instance in the systematic review of the WHO ENG on high noise annoyance exposure-response functions were found to considerably differ depending whether Asian and Alpine studies were considered or not [11]. According to the exposure-response function derived from the full dataset including Alpine and Asian study the pro- portion of highly annoyed people at low exposure levels was higher than in a curve derived from a dataset excluding these studies. However, the exposure-response relationship for the full dataset was less steep and thus the proportion of highly annoyed people becomes smaller above an Lden level of about 70 dB compared to the restricted curve. Given these indications that transferability of exposure- response functions may be critical, HRA would ideally use exposure-response curve derived from a similar context in terms of geography and time. Whereas for self-reported outcomes the number of studies is large that may allow to consider the most appropriate subset, the situation for somatic dis- ease may be more challenging. There are less studies and the uncertainty for rare disease is usually high. Thus, to obtain a meaningful statistical precision of the exposure-response function, several studies need to be combined. Related is the issue of transferability of exposure-response functions between different noise sources. The characteristic and the diurnal pattern of noise exposure from different sources may vary and thus it is plausible that this translates into differences in the exposure-response functions, which are usu- ally related to Lden. In the WHO ENG meat-analysis were conducted separately for road, railway and aircraft noise. However, number of studies on somatic diseases and railway noise is scarce and basically absent for all non-transport related noise sources. Thus, the exposure-response curve for these corresponding sources is very uncertain. For this reason, the EEA assessment for ischemic heart

inter.noise 21-24 AUGUST SCOTTISH EVENT CAMPUS. ? O? ? GLASGOW

disease was conducted by means of the same exposure-response functions for road, rail, air, and in- dustry noise. This curve was derived from the studies on road traffic noise. For high annoyance and high sleep disturbance the number of studies and their precision allowed to calculate the impact by means of source specific curves. The form of the exposure-response relationship may vary depending on the outcome and the exposure source. To date, polynomial functions or quasi-linear exposure response functions have been used in HRA. The form of the exposure-response function is mostly informed by the original studies.

2.4 Counterfactuals In HIA studies the counterfactual is determined by a reference policy, action or program and the corresponding change in noise exposure. For BoD studies the counterfactual is typically the “no ex- posure” scenario, which is not straightforward to define. Some previous HRA used an Lden of 53 dB as the “no exposure” cut-off and only quantified health impacts above this threshold. The threshold was chose as this is the guideline value for road traffic noise proposed in the WHO ENG. It needs to be noted that the noise guidelines do not represent the no-effect threshold but were set for the levels where the accepted risk was exceeded for the critical outcomes. Accepted risk increase was set to 5% relative excess risk for IHD incidence, 10% relative excess risk for incidence of hypertension, 10% increase in the proportion %HA people, 3% increase in the proportion %HSD people and one-month delay in terms of reading age (cognition). Thereby, from all derived critical thresholds, the lowest level was chosen for the guidelines [1]. Further, several older studies included in the meta-analysis of the WHO ENG did not estimate noise exposure at low levels and thus did not provide any effect estimates for low to moderate noise levels. Thus, by just averaging the lowest exposure category, as done in the WHO ENG, the effect threshold may have overestimated. Indeed several newer studies, applying more sophisticated noise modelling covering the whole range of transportation noise expo- sure indicates that the exposure response-relationship is approximately linear across the whole expo- sure range for somatic disease. From that point of view, it would also be defendable to take a lower threshold for quantifying the health effects. Since the proportion of people exposed to moderate to low levels of air pollution is large, the choice of the cut-off has substantial implications for the esti- mated impact. 3. CONCLUSIONS

In conclusion, many aspects in HRA of noise needs to be evaluated. In the following a few recom- mendations are formulated for future HRA studies as a basis for further discussion among the noise research community.

1. Exposure assessment: There is indications that differences in noise modelling introduces ar-

tificial differences in estimated noise exposure between countries, which is a problem for multi-country comparison. It is thus recommended to develop noise exposure assessment methods, which are comparable across countries and feasible to implement in different set- tings. This does not necessarily needs to be done with propagation models but could also be achieved with novel methods based on statistical modelling. 2. Relevant noise sources: We advice to focus on the health effects of transportation noise as

data availability and evidence is most elaborate. In principle, there is high plausibility that other noise sources may have similar health effects. However, the data situation is not yet satisfying to accurately estimate the population exposure and to derive exposure-response re- lationships. We encourage to work towards improvement of this situation for the most rele- vant noise sources, which may include construction work, church bell, military noise and community noise from restaurants and partying.

inter.noise 21-24 AUGUST SCOTTISH EVENT CAMPUS. ? O? ? GLASGOW

3. Additivity of impact from v a r i o us noise sources: In general, effects from various noise sources

can be considered do be additive. However, since a person can be highly disturbed from more than one noise sources, careful communication is needed when talking about the total number of disturbed people from noise. Further, we recommend to have evaluate whether results from single- and multi-pollutant studies do systematically differ. If so, this could be an indication that additivity of the effects is hampered by the underlying correlation structure. 4. Selection of outcomes: It is highly recommended to systematically evaluate the health effects

from noise and to revisit the selection of outcomes. Potential additional outcomes to consider are diabetes, overweight and mental health problems including depression. Since noise is an established physiological stressor with anticipated effects on a broad range of diseases, noise HRA studies may also consider all natural cause mortality as an outcome, as it is standard for HRA studies in the field of air pollution. For cost-benefit analyses, impact on all-cause mor- tality is usually the most relevant cost contributor and thus the decision which mortality out- comes to include has major implications for the result. 5. Transferability of exposure re s p onse functions in space and time: There are indications that

the transfer of exposure-response functions across time and geographical region introduce substantial uncertainties, in particular in relation to self-reported outcomes. We advise that future HRA should use exposure-response functions from a similar context in terms of time period and geographic region. This is most critical for self-reported outcomes, whereas for somatic disease a vigilance balance is needed between statistical precision and most adequate transferability. 6. Source specific exposure-re s po ns e functions: Differences in noise characteristics and diurnal

pattern is a likely explanation that exposure response-functions differ between road, rail and aircraft noise. Whereas the use of source specific curve in HRA is already standard for high annoyance and high sleep disturbance, we advice that in the future source specific curves should also be derived for somatic outcomes if the number of studies permit. This may be mainly the case for cardiovascular diseases. For outcomes with less available research, the same exposure-response function may be used for all transportation noise sources. For non- transportation noise source, little research on somatic diseases may be available and if such sources are considered, exposure-response functions would be mainly derived from transpor- tation noise studies. In this case, one may choose either road, railway or aircraft noise to derive the curve, depending which is most similar in term of noise characteristics and diurnal pattern. 7. Threshold for quantification : The counterfactual in BoD studies is usually defined as the “no

exposure” scenario. Often HRA used the WHO road traffic noise guideline value of 53 dB Lden as a threshold for the impact quantification. However, this represents accepted risk and there is solid evidence that health effects occur below this value. It is thus advised to critical re-evaluate the threshold for BoD studies taking into account newer studies with reliable noise modelling in the low and moderate exposure range. Since a substantial proportion of the pop- ulation is exposed to moderate levels of noise exposure, the decision on the quantification threshold is influential for the estimated impact.

We acknowledge that this list of recommendation is not exhaustive and additional important method- ical recommendations may be relevant for future HRA. HRA is always based on assumption yielding to uncertainties, which may produce uneasiness. However, the alternative would be not to make as- sumptions and to refrain from doing an HRA, which implicitly means that there is no health impact of noise. This is most likely more erroneous than any uncertain decision. As a guidance for making

inter.noise 21-24 AUGUST SCOTTISH EVENT CAMPUS. ? O? ? GLASGOW

assumption, we would recommend as realistic as possible but in case of uncertainty the conservative decision may be preferred to not create spurious health impact assessments. 6. REFERENCES

1. WHO Environmental noise guidelines for the European Region ; WHO Regional Office for Europe: Copenhagen, 2018. 2. EEA, E. E. A. Environmental noise in Europe — 2020 ; Copenhagen, 2020. 3. Gowers, A. M.; Walton, H.; Exley, K. S.; Hurley, J. F., Using epidemiology to estimate the impact and burden of exposure to air pollutants. Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences 2020, 378, (2183), 20190321. 4. Khomenko, S.; Cirach, M.; Barrera-Gomez, J.; Pereira-Barboza, E.; Iungman, T.; Mueller, N.; Foraster, M.; Tonne, C.; Thondoo, M.; Jephcote, C.; Gulliver, J.; Woodcock, J.; Nieuwenhuijsen, M., Impact of road traffic noise on annoyance and preventable mortality in European cities: A health impact assessment. Environ Int 2022, 162, 107160. 5. Aguilera, I.; Foraster, M.; Basagana, X.; Corradi, E.; Deltell, A.; Morelli, X.; Phuleria, H. C.; Ragettli, M. S.; Rivera, M.; Thomasson, A.; Slama, R.; Kunzli, N., Application of land use regression modelling to assess the spatial distribution of road traffic noise in three European cities. Journal of exposure science & environmental epidemiology 2015, 25, (1), 97-105. 6. Raess, M.; Brentani, A.; Ledebur de Antas de Campos, B.; Flückiger, B.; de Hoogh, K.; Fink, G.; Röösli, M., Land use regression modelling of community noise in São Paulo, Brazil. Environmental research 2021, 199, 111231. 7. Thacher, J. D.; Hvidtfeldt, U. A.; Poulsen, A. H.; Raaschou-Nielsen, O.; Ketzel, M.; Brandt, J.; Jensen, S. S.; Overvad, K.; Tjonneland, A.; Munzel, T.; Sorensen, M., Long-term residential road traffic noise and mortality in a Danish cohort. Environmental research 2020, 187, 109633. 8. Sorensen, M.; Poulsen, A. H.; Kroman, N.; Hvidtfeldt, U. A.; Thacher, J. D.; Roswall, N.; Brandt, J.; Frohn, L. M.; Jensen, S. S.; Levin, G.; Raaschou-Nielsen, O., Road and railway noise and risk for breast cancer: A nationwide study covering Denmark. Environmental research 2021, 195, 110739. 9. Wallas, A. E.; Eriksson, C.; Ogren, M.; Pyko, A.; Sjostrom, M.; Melen, E.; Pershagen, G.; Gruzieva, O., Noise exposure and childhood asthma up to adolescence. Environmental research 2020, 185, 109404. 10. Eze, I. C.; Foraster, M.; Schaffner, E.; Vienneau, D.; Heritier, H.; Pieren, R.; Thiesse, L.; Rudzik, F.; Rothe, T.; Pons, M.; Bettschart, R.; Schindler, C.; Cajochen, C.; Wunderli, J. M.; Brink, M.; Röösli, M.; Probst-Hensch, N., Transportation noise exposure, noise annoyance and respiratory health in adults: A repeated-measures study. Environ Int 2018, 121, (Pt 1), 741-750. 11. Guski, R.; Schreckenberg, D.; Schuemer, R., WHO Environmental Noise Guidelines for the European Region: A Systematic Review on Environmental Noise and Annoyance. Int J Environ Res Public Health 2017, 14, (12).

inter.noise 21-24 AUGUST SCOTTISH EVENT CAMPUS. ? O? ? GLASGOW