Impact of short-term aircraft noise on cardiovascular disease risk in the area surrounding London Heathrow airport: the RISTANCO epidemiological study

Limitations identified in the Aviation Environmental Design Tool model

We identified several limitations in the use of the AEDT model, particularly as we extended its use to look at short periods within a single day (it is usually used to provide long-term average noise exposures).

1. Atmospheric pressure, relative humidity and wind speed are set as meteorological constants that reflect the 30-year average at the airport. These simplifications are a limit of current modelling practices, when estimating sub-annual average aircraft noise exposures.

2. The headwind speed is maintained at 8 knots during the entire period of each operation. This may result in inaccurate aircraft performance parameters, such as climb and speed, which are related to the location and intensity of noise.

3. Wind speed or direction is not used by the AEDT sound propagation calculations (i.e. a uniform dispersion in all directions is assumed at all times).

4. The terrain model only accounts for elevation of natural landscapes and not manmade features. Therefore, containment and sheltering effects in urban locations are ignored.

5. The computational demands for creating sub-daily exposure surfaces:

   1. Limited the spatial resolution of the model outputs, returning a coarser exposure gradient, although we think that this still gave good exposure contrast for our epidemiological study.

   2. Dryer air temperatures were summarised into profiles that accounted for season and time of day across the 5-year study period. Therefore, the influence of unusual temperature events on sound propagation is not accounted for.

These factors are likely to lead to exposure misclassification bias. However, annual average aircraft noise surfaces are currently only routinely modelled by the CAA. This study has used several approaches to develop and enhance the existing approach to create sub-annual exposure surfaces:

1. Radar tracks of individual flights were provided by Heathrow airport, with a unique set of aircraft footprints constructed for each modelled period.

   1. The created AEDT surfaces cover 1826 days across the 5 years of 2014–18 (i.e. 14,608 flight activity-informed noise surfaces were created vs. 5 annual average surfaces for each noise metric).

   2. Actual flight paths were used rather than these being estimated by operational movements being estimated by headwind direction and performance parameters.

2. Unique temperature profiles were used, which correspond to the specified ‘season’ and ‘time of day’. Annual average AEDT models only use long-term averages.

We also had a very large dataset to work with, which may offset some of the lack of precision as a result of random bias.

Correlations

In Table 6, we present the pairwise Pearson correlation coefficients (r) between daily aircraft noise levels at eight specified time bands and standard four noise metrics (LAeq24, Lday, Lnight and Leve), and the daily number of flight events at eight time bands.

We found high to very high correlations (coefficient 0.68–0.90) between each pair of the four standard noise metrics (Lday, Lnight, Leve and LAeq24).

In comparison, daily aircraft noise during early morning (04.30–06.00 hours) and late night (24.00–04.30 hours), had a much weaker correlation (coefficient < 0.4) with noise levels during any other time bands. There were moderate to high correlations (coefficient 0.52–0.87) between each pair of daily aircraft noise levels during the day (07.00–15.00 hours), the afternoon (15.00–19.00 hours) and the early evening (19.00–22.00 and 22.00–23.00 hours).

There were weak to moderate correlations (coefficient 0.07–0.48) between the number of flight events and the actual noise levels during the eight time bands, with the exception of N60 06.00–07.00 hours, which had a moderate correlation with noise levels during 06.00–07.00, 07.00–15.00, 15.00–19.00 and 19.00–22.00 hours. Moreover, there were relatively weak correlations between each pair of noisy flight events except for N60 06.00–07.00 hours and N60 04.30–06.00 hours (coefficient 0.73), N65 19.00–22.00 hours and N65 15.00–19.00 hours (coefficient 0.96), N65 22.00–23.00 hours and N65 15.00–19.00 hours (coefficient 0.71), N65 22.00–23.00 hours and N65 22.00–23.00 hours (coefficient 0.55). This may show that the distribution of noisy flight events may differ from that of daily noise levels in dB.

The lower correlations of non-standard and noisy event metrics with the standard noise metrics of Lday, Lnight, Leve and LAeq24 (compared with high correlations between the standard metrics) raises the possibility that standard metrics may miss important characteristics of noise exposure, with potential relevance for impacts on biological systems.

Previous studies on relief/respite

We identified one study by Beghelli et al. that examined the effect of respite on medical costs. 24 This study used a natural experiment of the Early Morning Arrival Trial, which was implemented from 5 November 2012 to 31 March 2013 to provide noise respite to certain communities living near Heathrow airport. The trial identified four exclusion zones (two to the east and two to the west of Heathrow). This trial designated four exclusion zones (two to the east and two to the west of Heathrow). During the trial, each week, night and early morning (23.30–06.00 hours) aircraft movement was rerouted from one set of air traffic exclusion zones to non-exclusion areas. This reroute alternated weekly between two sets of air traffic exclusion zones. Beghelli et al. showed that the trial was associated with a 5.8% decrease in spending on central nervous system treatment, which included the treatment of sleep loss, concentration deficits and other stress-related illnesses in quiet set of zones, compared with control zones, which saw no change in flight movement during the period. 24 However, there was a non-statistically significant increase in medication for CVD during the trial. 24

We did not have information on flight trials to include in our study. However, the trial exclusion zones considered in Beghelli et al. were relatively small, with each zone measuring only 1 nautical mile (1.15 miles) in width25 (our study area was 97 × 47 km) and the period was relatively short (5 months – we had 5 years of data). Linking these trial areas is likely to provide an insufficient sample size for health analyses, especially when using binary outcomes indicating more severe disease (hospital admissions and mortality data).

A priori identification of aircraft noise variability based on respite criteria

Using our daily aircraft noise data, we experimented with using arbitrarily defined cut-off noise levels and numbers of days affected to define areas with detectable noise level changes compared with control areas with much more constant levels of noise, with the aim of seeing whether we could identify sufficient postcodes for health analyses.

We attempted to establish criteria for selecting postcodes (relief group) that were exposed to loud aircraft noise on a significant number of days per year but also had a significant number of relatively quiet days. The difference between loud and quiet noise levels was chosen to be 5 dB in order for people to be able to detect the change in sound level according to a previous report on respite and relief periods. 23 We then modified the selection criteria to identify two control groups of postcodes with varying numbers of days exposed to loud aircraft noise. The number of postcodes belonging to each group and their average noise levels were computed and presented in Tables 7–9.

Since our data split a day into eight time bands, we first focused on the morning shoulder period (06.00–07.00 hours), which was typically the noisiest period of the day, as shown in the preceding section. We arbitrarily identified a relief group that consisted of postcodes with over 100 days of exposure to noise levels above 55 dB and 100 days of exposure to noise levels below 50 dB per year. We chose 55 and 50 dB as our thresholds because a difference of at least 5 dB in sound level is required for people to consciously notice it and also because these are moderate levels of noise at or above mean levels for each time period (Table 7). The first control group consisted of postcodes that were exposed to aircraft noise during morning shoulder period for at least 100 days below 50 dB but between 100 and 200 days above 55 dB during the same period. The second control group includes postcodes that have at least 200 days that were exposed to above 55 dB during morning shoulder period. Table 7 displays the number of postcodes in each group per year, together with their average decibel levels. Between 23,439 and 25,679 postcodes meet the relief group criteria. However, these numbers represent only about 17% of the total number of postcodes. There were between 2084 and 5065 and 18,374 and 22,235 postcodes that met the criteria for control groups I and II. These numbers nevertheless remain small.

W experimented with various noise thresholds and days to identify postcodes with significant variations in aircraft noise levels. In panel 2 of Table 7, instead of 100 days as the relief group identification criterion, we used 50 days. Similarly, we used 100 days as the cut-off for control group II, as opposed to 200 days. In panel 3 of the same table, we adopted a 7 dB noise difference as opposed to the 5 dB difference used in panel 1. Results indicate that the number of postcodes that meet the criteria in panel 2 and 3 is comparable to that in panel 1.

We used the same method to identify the relief group and the two control groups, but with afternoon noise levels (15.00–19.00 hours). Table 8 shows the results, which demonstrate that similarly small number of postcodes met the criteria.

Finally, we linked morning shoulder (06.00–07.00 hours) noise levels with afternoon (15.00–19.00 hours) noise levels to identify a relief group with a relatively noisy morning shoulder period but a quieter afternoon relative to a control group with significantly noisier morning shoulder and afternoon periods. We identified relief postcodes as those exposed to noise levels above 55 dB during the morning shoulder period but below 50 dB in the afternoon on the same day for at least 100 days per year. The control group consisted of postcodes where morning shoulder period and afternoon noise levels exceeded 55 dB on the same day for at least 200 days per year. Table 9 (top half of panel 1) shows that between 2014 and 2018 there are between 3029 and 5647 postcodes in the relief group, but only between 1655 and 1896 in the control group. We then relaxed the criteria by selecting postcodes that were exposed to noise levels above 55 dB during the morning shoulder period and below 50 dB in the afternoon on the same day but for at least 50 days in a year. The control group comprised all postcodes that were exposed to noise levels above 55 dB during the morning shoulder period and afternoon of the same day for at least 100 days. The results shown in the bottom half of panel 1 of Table 9 display that the number of postcodes in the relief and control groups increased significantly compared with the previous analysis, but the total number remained relatively small. We reapplied the same strategy but increased the noise level difference from 5 to 7 dB to identify relief and control postcodes. The results were presented in panel 2 of Table 9.

In conclusion, using different selection criteria for relief and control postcodes and predefined noise cut-off points to identify areas with large changes in noise levels, we only identified a small proportion of postcodes in the study area (~5000–25,000 postcodes compared with 150,000 overall). Further, some of the predefined control areas had similar or lower noise levels to the areas presumed to have relief periods. We therefore did not conduct health analyses because the small sample size was likely to have been insufficient to detect effects.

Coefficient of variation

The CoV (SD/mean × 100) is a measure of variability that can be used to quantify the day-to-day variability per postcode between 2014 and 2018. A high CoV suggests that daily noise levels within a postcode vary more in a given period.

The 2014–18 descriptive summary of the CoV for daily noise levels by postcode level is presented in Table 10. In this table, for each time band, we calculated the CoV for all seasons (winter, winter transition, summer and summer transition) and additionally by pooled seasonal data across 2014–18 (summer, summer transition, winter and winter transition). Our findings showed that daily noise levels by postcodes varied more during the night. Particularly, 24.00– 04.30 hours had the highest mean CoV (67.33–74.16) of all time bands, followed by 04.30–06.00 and 23.00–24.00 hours. The morning shoulder period (06.00–07.00 hours) had the highest mean daily levels, but its mean CoV (15.98–16.83) was the fifth highest among all time bands. Daytime aircraft noise tended to be less variable.

Deprivation

Given that deprivation has many different aspects, we focused on two: material and health deprivation. We measured material and health deprivation using three variables: Carstairs index of multiple deprivation [census output areas level (COA), 2011 only], fuel poverty rate [lower-layer super output areas level (LSOA), 2014–18] and avoidable death rate per 100,000 [local authority district level (LAD), 2014–18].

• Carstairs index is a commonly used area-level measure of material deprivation in health studies. 30 It was calculated using four variables from the 2011 Census, including male unemployment, low social class, non-car ownership and overcrowding. This variable has the highest spatial resolution among the three deprivation indicators chosen for this study, due to its geography being COA (the highest spatial resolution of English Census geography of average population of 310 individuals). This indicator is time invariant as only 2011 values were available. Data were obtained via the UK Data Archive (link: www.data-archive.ac.uk, accessed 23 November 2022).

• Annual fuel poverty rate is used to measure the percentage of households that were unable to maintain standard thermal comfort and safety. 31 Fuel poverty has been increasingly recognised as a distinct form of social and health inequality. 32 It has been hypothesised that cold may be associated with excess winter deaths. 33 A cold home due to fuel poverty has been linked to respiratory problems, arthritis and rheumatism in people of all ages, as well as mental health problems in adolescents. 34 This indicator is annual, covering the period 2014–18. The geographic level is LSOA level (Census geography category with average population of 1500 individuals) and covers the period 2014–18. We extracted fuel poverty from UK annual fuel poverty statistics. 35

• We used yearly avoidable death rate per 100,000 to measure health inequality. Mortality is an outcome that can be clinically quantified; avoidable mortality is amenable to policy intervention. 36 Avoidable death rate could therefore be used to capture the geographical disparity in health. 36 The definition of avoidable death rate is available from the Office for National Statistics. 37 The data were at LAD level (mean population of approximately 179,361.6 per LAD), covering each year 2014–18. We downloaded the data from the Office for National Statistics. 38

Confounders

We adjusted for the quintiles of percentage of non-white population per LAD, considering that ethnic concentration may be related to both deprivation and aircraft noise levels. These data were obtained from Nomis (www.nomisweb.co.uk, accessed 13 February 2023).

Statistical analyses

Since noise exposure levels were calculated daily, serial correlation is a concern. We specified a random effects model with autoregressive first-order autoregression model disturbance to estimate the association between daily noise levels and quintiles of deprivation.

The equations are specified as:

where i represents individual postcode, j represents individual output areas, k represents individual LSOA, l represents individual LAD, t represents year; noise_it, carstairs_j and ethnic_lt represents daily noise levels (continuous), quintiles of Carstairs index and quantiles of percentage ethnic minority population; year_t is the year fixed effect; u_i is the random heterogeneity and e_it is error term.

where i represents individual postcode, j represents individual output areas, k represents individual LSOA, l represents individual LAD and t represents year; noise_it, avoid_lt and ethnic_lt represent daily noise levels (continuous), quintiles of avoidable death rate and quantiles of percentage ethnic minority population; year_t is the year fixed effect; u_i is the random heterogeneity and e_it is error term.

where i represents individual postcode, j represents individual output areas, k represents individual LSOA, l represents individual LAD and t represents year; noise_it, carstairs_j, avoid_lt, fuelpov_kt and ethnic_lt represent daily noise levels (continuous), quintiles of fuel poverty rate and quantiles of percentage ethnic minority population; year_t is the year fixed effect; u_i is the random heterogeneity and e_it is error term.

where i represents individual postcode, j represents individual output areas, k represents individual LSOA, l represents individual LAD and t represents year; noise_it, fuelpov_kt and ethnic_lt represent daily noise levels (continuous), quintiles of Carstairs index, quintiles of avoidable death rate, quintiles of fuel poverty rate and quantiles of percentage ethnic minority population; year_t is the year fixed effect; u_i is the random heterogeneity and e_it is error term.

Each postcode uniquely belongs to an output area, a lower LSOA and a LAD, which enables us to link data.

Our dependent variables included the four noise metrics: LAeq24, Lnight, Leve and Lday.

We conducted four regressions per noise metric. The first regression of each metric included quintiles of Carstairs index and percentage non-white ethnicity. Models 2 and 3 replaced Carstairs index with avoidable death rate and fuel poverty rate, respectively. Model 4 included quintiles for all three measures of deprivation and percentage non-white ethnicity.

All analyses were conducted in Stata using module xtregar. 39

Results

Table 11 shows the descriptive summary of measures of deprivation and percentage non-white ethnicity in the analysis.

Table 12 illustrates the pairwise correlations between variables involved in analysis. The noise correlation coefficients between LAeq24, Lday, Leve and Lnight ranged between r = 0.68 and 0.98. The correlation between Lnight and Leve (r = 0.68) was the lowest among all pairs, whereas the correlation between Lday and LAeq24 (r = 0.98) was the highest. For deprivation variables, we used raw values rather than quintiles. The correlation between Carstairs index and both of avoidable death rate and fuel poverty was moderate (r ~ 0.4), while that between fuel poverty rate and avoidable death rate was particularly weak (r = 0.08). There was a fairly weak relationship between each pair of the three deprivation variables and area percentage non-white ethnicity in our data (r 0.08–0.49), with the correlation between the Carstairs index and the percentage non-white ethnicity being the strongest (r = 0.49).

Tables 13 and 14 demonstrate the main results from regressions. The dependent variables were Lday (07.00–19.00 hours) and Leve (19.00–23.00 hours) in Table 13, and Lnight (23.00–07.00 hours) and LAeq24 (24-hour average) in Table 14. In models 1–3, we separately regressed the association between one measure of deprivation and aircraft noise levels while adjusting for quintiles of percentage non-white ethnicity. The results of these models consistently demonstrated that almost all quintiles of the Carstairs index, avoidable death rate and fuel poverty rate (except for Q5 of the fuel poverty rate) had significant and positive coefficients, regardless of the noise metrics being examined. This evidenced that postcodes near Heathrow airport with the least material or health deprivation experienced the lowest daily noise levels between 2014 and 2018.

However, which quintile of deprivation (Q1 least deprived, Q5 least deprived) was the noisiest depended on the deprivation measures and noise metrics used. During the day (07.00–19.00 hours), Q2 of Carstairs index, Q4 of avoidable death rate and Q4 of fuel poverty rate had the highest levels of noise. In the evening (19.00–23.00 hours), Q2 of the Carstairs index, Q4 of the avoidable death rate and Q2 of fuel poverty rate had the highest levels of noise pollution. It is interesting to note, the quintile with the highest levels of aircraft noise exposure at night (23.00–07.00 hours) was Q5, for both the Carstairs index and the avoidable death rate, as well as Q3 for the fuel poverty rate.

There were two interesting exposure–response patterns in relation to this. First, among the three indicators of deprivation, avoidable death rate had the most noticeable gradient. Column (2) of Table 13 demonstrates Q2 and Q3 postcodes were exposed to slightly higher noise levels, whereas Q4 and Q5 postcodes were exposed to significantly higher noise levels during daytime (coefficients: Q2 – 0.08, Q3 – 0.48, Q4 – 0.76 and Q5 – 0.69). The second is that the exposure–response relationship between night-time aircraft noise levels and deprivation was more pronounced than during the day and evening. This gradient was particularly clear when we paired night-time noise with avoidable death rate (coefficients: Q2 – 0.41, Q3 – 1.22, Q4 – 1.59 and Q5 – 1.66). While the relationship did not appear to be linear, it supported an observation that postcodes in local authorities with higher avoidable death rates were more likely to be exposed to a higher level of aircraft noise at night.

We found a stronger association between deprivation and daily aircraft noise at night than during the day or evening in postcodes near Heathrow airport. In comparison with postcodes in Q1, those in Q2–Q5 of the Carstairs index were exposed to noise levels that were 2.66, 3.13, 2.79 and 3.18 dB higher at night, but only 1.02, 0.58, 0.09 and 0.40 dB higher during daytime, and 1.02, 0.68, 0.22 and 0.48 dB higher during evening. The same conclusion is supported by the results that LAeq24 (mean sound levels over the 24 hours) had very similar number of observations as Lnight (N observations: Lnight – 85,847,742 vs. LAeq24 – 82,265,795) but its relationship with deprivation was significantly smaller in size than that of Lnight. The evidence suggests that night-time aircraft noise exposure inequality is of particular concern in postcodes near Heathrow.

Lastly, we found that the fuel poverty rate had a weaker relationship with daily aircraft noise than the Carstairs index and avoidable death rate. There was a negative relationship between the fifth and fourth quintiles of fuel poverty and aircraft noise during some periods.

The above interpretation of the results was based on models that included only one measure of deprivation, but our conclusions held when all deprivation measures were included (model 4).

Health outcomes data

All hospital episodes and deaths due to primary CVD in the study area from 1 January 2014 to 31 December 2018 were included. We extracted postcode data on all hospital episodes and deaths from the Hospital Episode Statistics from NHS Digital and the mortality data from the Office for National Statistics held by the UK Small Area Health Statistics Unit at Imperial College London. Data were obtained for all events with primary cause of admission or death due to stroke [International Classification of Diseases version 10 (ICD-10) codes I61, I63–I64], coronary heart disease (CHD; ICD-10 I20–I25) and CVD (ICD-10 Chapter I) and linked to postcode-level noise estimates. Time of hospital episode and death were not available. The study was covered by national research ethics approval from the London – South East Research Ethics Committee (reference 17/LO/0846; date of opinion 29 June 2017). Data access to confidential patient information without consent was covered by the Health Research Authority Confidentiality Advisory Group under Regulation 5 of the Health Service (Control of Patient Information) Regulations 2002 (section 251); reference: 20/CAG/0028 (outcome date 24 March 2020, section 251 Register Index Sheet application number A02476).

Confounder data

The environmental confounders included in the models were mean temperature and particulate matter less than 2.5 µm in diameter (PM_2.5) concentration. Hourly dry air temperature measurements were captured at three National Oceanic and Atmospheric Administration Integrated Surface Database weather stations within 25 km of the study area. Hourly background measurements of fine PM_2.5 were captured by the six UK Automatic Urban and Rural Network sites within 25 km of the study area. Dry air temperature and background PM_2.5 concentrations were estimated at each residential postcode using a spatial interpolation technique known as inverse distance-squared weighting.

Individual-level ethnicity data were available for all hospital admissions in the Hospital Episode Statistics data and COA-level Carstairs Index quintile from the 2011 Census was linked to admissions and deaths data. The Carstairs index is a commonly used indicator of material deprivation in health studies. 30,44 All estimates were also adjusted for the effect of holidays.

Statistical analyses

Patients with multiple cardiovascular episodes (records) per day (n = 3,018; 0.07% of cases) had one record on the day randomly selected for inclusion. Each episode record represents a patient being seen by a new clinician, so these may relate to the same spell in hospital. Control periods were matched to case periods within the same year and month on the same day of the week, excluding control days on which an additional cardiovascular episode occurred (n = 15,856 controls); 528 cases with no suitable control days were also excluded from analyses. A flowchart of the exclusion criteria and how they affected the numbers of cases and controls is presented in Figure 3.

FIGURE 3.

Hospital episodes exclusion criteria.

Conditional logistic regression was used to estimate the OR and 95% CI per 5-dB increase for the metrics Lday, Leve, Lnight, Lden and LAeq24, as well as for eight distinct periods throughout the 24-hour period relating to aircraft flows. We considered all CVD, CHD only and stroke only for both hospital episode and deaths. Estimates were adjusted for mean temperature, PM_2.5 concentration and the effect of holidays, as these are variables that change rapidly in time, while long-term confounders were accounted for by the case-crossover study design. Analyses were also stratified by age, sex, ethnicity, deprivation and season to assess effect modification. All analyses were run in R statistical software45 using the Epi package. 46

Descriptive results

The descriptive summary of our sample is presented in Table 15 (note that the noise estimates relate to postcodes of cases only. This is a subset of all postcodes and descriptive statistics of noise levels differ from estimates in Table 5 relating to all postcodes; in fact, averages are lower than for all postcodes, likely reflecting spatial variability in characteristics of populations by postcode, such as age of individuals who become cases).

Results for short-term aircraft noise associations with hospital admissions

There was evidence of a small increase in risk for a 5-dB increment in noise during the evening (Leve OR = 1.005, 95% CI 1.000 to 1.010), particularly from 22.00 to 23.00 hours (OR = 1.006, 95% CI 1.002 to 1.010) for all CVD admissions (Table 16). A similar but not statistically significant pattern was estimated for admissions due to CHD. There was no evidence of an increased risk for hospitalisations due to stroke (Figure 4).

FIGURE 4.

Odds ratios and 95% CIs for hospitalisations due to all CVD, CHD and stroke per 5-dB increase LAeq at defined time points throughout the day, evening and night. Estimates adjusted for PM_2.5 concentration, mean temperature and holiday effect.

Stratified analyses for cardiovascular admissions can be seen in Figure 5. After stratifying by age and sex, the effect of aircraft noise on cardiovascular admissions was statistically significant in men over the age of 65 years during the daytime (Lday OR = 1.009, 95% CI 1.001 to 1.019) evening (Leve OR = 1.014, 95% CI 1.005 to 1.019) and, to a lesser extent, in women under the age of 65 years during the evening (Leve OR = 1.012, 95% CI 1.000 to 1.023; Figure 5). Similarly, the same figure shows, after stratifying by ethnicity, that an effect for hospitalisations due to all CVD was seen in patients who identified as South Asian during the evening hours 22.00–23.00 hours (OR = 1.004, 95% CI 1.002 to 1.028) and as other ethnicity (not South Asian or black) during the evening (Leve OR = 1.007, 95% CI 1.001 to 1.013). There was no significant effect modification by age, sex or ethnicity evident for CHD or stroke. There was also an increase in risk during late night hours among individuals residing in the third and fourth quintiles of deprivation (Figure 5).

FIGURE 5.

Odds ratios and 95% CIs for hospitalisations due to all CVD per 5-dB increase LAeq, stratified by (a) age-sex, (b) ethnicity, (c) deprivation and (d) season. Estimates adjusted for PM_2.5 concentration, mean temperature and holiday effect.

There was also evidence of effect modification by season for all CVD and for CHD. The effect of night-time aircraft noise on all CVD was strongest in the summer (22.00–23.00 hours OR = 1.009, 95% CI 1.001 to 1.017; 23.00–24.00 hours OR = 1.009, 95% CI 1.002 to 1.016) and winter months (22.00–23.00 hours OR = 1.008, 95% CI 1.002 to 1.014; 23.00–24.00 hours OR = 1.009, 95% CI 1.002 to 1.015), and the effect of late afternoon and early evening aircraft noise was only evident in the winter (15.00–19.00 hours OR = 1.008, 95% CI 1.001 to 1.014; 19.00–22.00 hours OR = 1.012, 95% CI 1.004 to 1.020). The effect of early morning aircraft noise was strongest in the summer transition months (04.30–06.00 hours OR = 1.018, 95% CI 1.009 to 1.028; 06.00–07.00 hours OR = 1.010, 95% CI 1.003 to 1.017; Figure 6).

FIGURE 6.

Odds ratio and 95% CIs for hospitalisations CHD (right) per 5-dB increase LAeq, stratified by season. Estimates adjusted for PM_2.5 concentration, mean temperature and holiday effect.

In a sensitivity analysis including only the first hospital episode for the 60.8% of patients with more than one episode (Figure 7), similar patterns were seen as for the main analyses (see Figure 4 and Table 16).

FIGURE 7.

Sensitivity analyses: ORs and 95% CIs for hospital episodes due to all CVD, CHD and stroke per 5-dB increase LAeq at defined time points throughout the day, evening and night, including only the first hospital episode for the 60.8% of patients with more than one episode (n = 269,915). Estimates adjusted for PM_2.5 concentration, mean temperature and holiday effect.

Associations of aircraft noise with mortality

There was no evidence of an association between short-term aircraft noise and deaths due to CVD, CHD or stroke, with wide confidence intervals (Figure 8).

FIGURE 8.

Odds ratios and 95% CIs for deaths due to all CVD, CHD and stroke per 5-dB increase LAeq at defined time points throughout the day, evening and night. Estimates adjusted for PM_2.5 concentration, mean temperature and holiday effect.

Variability in risk estimates between areas with consistent patterns of noise exposure compared with those with changing patterns of noise exposure

As detailed previously, we used CoV to measure the variability of daily noise levels per postcode between 2014 and 2018. We reran regression analyses stratifying our samples by low (≤ mean) compared with high (> mean) CoV (daily noise levels between 2014 and 2018), to assess the difference in risks of hospitalisation for CVD across regions with low and high availability in noise. The mean CoV for each noise metrics can be found in Table 10. As seen in Table 17, there were increased risks of hospitalisation for CVD throughout the evening hours (19.00–22.00, 22.00–23.00 and 23.00–24.00 hours) in the low CoV group. However, there was a non-significant risk at any time of day in high CoV group.

To explore whether low CoV areas were those with higher noise levels (potentially suggesting high noise and less relief periods), we examined mean noise levels (Table 18). For the latter two periods, mean noise levels were higher in the low CoV postcodes (41 vs. 37 dB for 22.00–23.00 hours; 31 vs. 24 dB for 23.00–24.00 hours), However, for the period 19.00–22.00 hours, the mean noise levels were 41 dB in low CoV areas compared with 43 dB in high CoV areas. We therefore could not readily infer that a lack of relief periods (or at least some periods of lower noise exposure) was related to the association with hospitalisation.

Descriptive summary of daily aircraft noise data

We found that the morning shoulder period (06.00–07.00 hours) was the noisiest among all periods (mean 50.92 dB; 90th percentile 52.93 dB). Daytime (07.00–15.00 hours) aircraft noise levels were typically only slightly lower than the morning shoulder period (mean 49.87 dB; 90th percentile 51.50 dB). Night quota periods (23.30–04.30 hours) are typically times when people are sleeping, but we found that the average noise levels across postcodes from 23.00 to 24.00 hours and 24.00 to 04.30 hours were 41.06 and 29.81 dB, respectively.

We found that during 07.00–15.00 hours, postcodes within the study area experienced an average of eight flight events. Morning shoulder (06.00–07.00 hours) had an average of three events, while night-time (04.30–06.00 hours) had an average of one flight event.

Approaches to identifying respite and/or relief period

We did not have information on flight trials so attempted to identify relief periods (caused by e.g. wind direction changes as well as trials) in other ways. A priori definition of appreciable changes in noise did not identify enough postcodes. We therefore examined CoV of daily noise levels at postcodes. Highest variability was seen in night-time periods, with 24.00–04.30 hours having the highest mean CoV, followed by 04.30–06.00 and 23.00–24.00 hours.

Daily aircraft noise and material and health inequality

We examined the relationship between aircraft noise Lday, Leve and LAeq24 and three different deprivation/inequality measures. While postcodes near Heathrow with the least material or health deprivation experienced the lowest daily noise levels between 2014 and 2018, the relationship with deprivation measures, including the Carstairs index and fuel poverty, was complex and did not appear to have a clear gradient. A gradient was more evident between Lnight and avoidable death rate. The fuel poverty rate had a weaker relationship with daily aircraft noise than Carstairs index and avoidable death rate.

Short-term impact of aircraft noise on cardiovascular morbidity and mortality

We included all recorded hospitalisations (n = 442,442) and deaths (n = 49,443) in 2014–18 due to CVD. We used conditional logistic regression to estimate the OR and adjusted for PM_2.5 concentration, temperature and holidays. We estimated an increase in risk for a 5-dB increment in noise during the evening (Leve OR = 1.005, 95% CI 1.000 to 1.010), particularly from 22.00 to 23.00 hours (OR = 1.006, 95% CI 1.002 to 1.010) for all CVD admissions. We found some evidence for effect modification by age, ethnicity, deprivation and season, but patterns were not consistent. The findings provide support for a potential mechanism through which aircraft noise may disturb sleep and elevate blood pressure, contributing to increased risk of cardiovascular hospitalisation.

Additionally, we found that an increased risk of CVD hospitalisation for increases in noise during the night-time hours (19.00–22.00, 22.00–23.00 and 23.00–24.00 hours) was only seen in postcodes with lower CoV. The average noise levels in the lower CoV postcodes were higher than other postcodes in two of the three periods – only giving partial support to the hypothesis that areas with higher noise levels and fewer relief periods have higher CVD admission risks. The impact of relief periods needs further research, ideally looking at other relevant outcomes (e.g. sleep disturbance, blood pressure, heart rate variability).

Delays

The dissemination of results was originally scheduled for 31 August 2020. The delays were caused by (1) relocation of the project from Imperial College to the University of Leicester, (2) complexities in the modelling process, which had not been anticipated by the noise consultancy conducting the modelling and (3) the COVID-19 pandemic, which among other impacts, slowed the recruitment of a researcher to conduct the analyses. The NIHR was supportive and responsive to communications about the delays throughout the project and a no-cost extension was agreed.

Strengths

We created and analysed a large dataset, which included daily average noise levels for approximately 155,000 postcodes near Heathrow airport. The AEDT version 3b was used to assess aircraft noise levels at each of the 155,000 postcodes in the vicinity of Heathrow airport from 1 January 2014 to 31 December 2018. These data are probably those with the highest resolution regarding Heathrow airport’s daily aircraft noise levels. Our analysis of noise data concludes that exposure to daily aircraft noise remains an environmental problem for some communities near Heathrow airport.

The study area encompassed around 6.3 million people who resided near Heathrow airport in 2011. The high temporal resolution of our daily aircraft noise exposure data was an additional strength. Thirdly, our noise data were estimated for eight time bands, allowing us to distinguish between different daytime and night-time periods.

In the subsection where we examined the association between daily aircraft noise exposure and deprivation, we used a range of deprivation domains, including not only poverty but also health disparities.

Our study focused on the impact of short-term aircraft noise on cardiovascular morbidity and mortality and, to our knowledge, is one of only three such studies to date. In our analysis of population health, we included nearly all hospitalisations and deaths attributable to CVD, providing sufficient statistical power to detect an effect. By design, the case-crossover design controlled for significant measured and unmeasured confounders, such as lifestyle factors, ethnicity and age. Differentiating the effects of noise during specific times of the day, evening and night provided evidence for certain biological mechanisms observed in previous research.

Patient and public involvement

The methods and results were discussed at meetings of the study scientific advisory board, which includes representatives from the non-governmental Aviation Environment Federation and the Heathrow Association for the Control of Aircraft Noise, a community group set up over 50 years ago to represent people living under the Heathrow flight path.

Equality, diversity and inclusion issues

Our study included 442,442 naturally occurring hospital admissions and 49,443 fatalities from CVD, after removing some duplicate reports and records with no control variables. Males made up 58.0% of hospital admissions, with 56.7% being above the age of 65 years. Among those who stated their ethnicity, 9.4% were black and 11.4% were South Asian.

Limitations

Among the limitations are that the study area was designed to capture outer bounds of the CAA annual average aircraft noise contours in 2011. Some postcodes outside the study area could still be affected by aircraft noise but were not included in the analysis. However, given the spatial and temporal resolution and size of the data, this was a reasonable compromise.

In our analysis of deprivation, we used area-level not individual-level noise and deprivation estimates therefore the ecological fallacy may apply. Heathrow airport is situated close to highly populated areas, some of which are very wealthy, so may not be representative of other airports.

We were only able to examine outdoor noise levels and were not able to take account of housing characteristics, including double glazing, which may have affected indoor noise exposures.

Misclassification bias may also have been introduced because we used noise exposure at small geographical level rather than individual level. Another source of potential misclassification is that individuals may move outside of the postcode to which their exposure has been assigned at different periods throughout the day. We expect less exposure misclassification in the evening and night-time hours because individuals are more likely to be at their postcode of residence during these times. We also expect less misclassification among older individuals throughout the day and night, as they are less likely to travel away from home for work or school during the day. This may partially explain why effect estimates were highest during evening and night-time hours, and among individuals over the age of 65 years. Lastly, exposure misclassification may have been introduced because data on exact time of admission and death were not available, and we were therefore unable to define the precise window of exposure before an event occurs. To compensate for this, we used the average of lag 0 and lag 1 before the event day to ensure the defined exposure window captures the true exposure. Lastly, misclassification bias may be introduced due to moving home; according to English Housing Survey, between 9% and 11.3% of households in England moved home per year between 2011 and 2018. 47

Carstairs index of deprivation (England and Wales, 2011)

The Carstairs index is an indicator of relative deprivation that is commonly used in spatial epidemiology to identify socioeconomic confounding. 48,49 This deprivation index is constructed from four unweighted UK Census variables, which describe the level of male unemployment, overcrowding, private vehicle ownership and social composition in each community.

A revised form of the Carstairs index was constructed for census output area and lower layer super output areas in England and Wales, using the 2001 classification of low social classes devised by Norman. 48,49 The revised low social class variable approximates its counterpart from the 1991 Census, developed to account for Office for National Statistics methodology and classification changes in later censuses.

Datasets from the 2011 Census were obtained from Nomis (www.nomisweb.co.uk), the official online delivery service of labour market statistics provided by the Office for National Statistics. Tables KS602EW, QS409EW, KS404EW and QS607EW contained the necessary information to create the Carstairs index for 2011 across England and Wales.

Table 19 lists the variable names, descriptions and formulas.

Each of these variables were z scored (mean-centred and divided by their SD) and all four z scores were summed to return an index value measuring the relative level of deprivation in each community. A value of 0 identifies communities that follow the national average of England and Wales, with negative values identifying increased affluence, and positive values identifying increased levels of deprivation.

Carstairs index values for the 181,408 COA communities in England and Wales were then categorised in quintile groups, to control for any outliers within the data. This procedure was repeated for the 34,753 LSOA communities in England and Wales.

Nitrogen Dioxide Concentration

Nitrogen dioxide (NO₂) concentrations from all sources of pollution in µg/m³ were obtained from a land use regression-modelled raster surface with a 50-m grid resolution, developed by Gulliver et al. 50 This pollution model is based on a multivariate regression equation that describes the relationship between sample locations and environmental variables (i.e. rural–urban land cover classifications and road network data within several proximity buffer zones).

Postcode centroids (i.e. points representing the population-weighted centre of each postcode unit) were intersected with the continuous NO₂ raster surface, to extract an exposure value at each location.

Road transport noise (annual, 2013)

Modelled road-transport noise estimates were calculated in accordance with the CNOSSOS-EU common framework for noise assessment methods developed by the European Commission (2002/49/EC). For the purposes of this study, the CNOSSOS-EU model algorithms were implemented in PostgreSQL via the PostGIS v2.1 extension, following the protocol described by Morley et al. 51

Annual average daily traffic counts and traffic speeds across the UK road network in 2013 enter the model, together with information relating to the surface roughness of land cover, building heights, wind profiles and average temperatures in 2013.

The CNOSSOS-EU model ran on the ALICE high-performance computing facility at the University of Leicester. Figure 9 (adapted from Gulliver et al. 52), describes the workflow of the CNOSSOS-EU model.

FIGURE 9.

Workflow of the CNOSSOS-EU model.

In brief, the coordinates of each receptor (residential postcode centroid) are assigned to the closest building. The building façade that is likely to experience the most noise levels is identified (i.e. traffic count on a nearby road/road distance), and a receptor (point) is placed 1 m away from the building. A geographical information systems operation locates all major roads within a 1000-m radius and all minor roads within a 100-m radius of each receptor.

Noise exposure from road transport was reported in accordance with five noise metrics that are ‘A’ frequency weighted. The ‘A’ weighting is a standard weighting of the audible frequencies designed to reflect the response of the human ear to noise (between 500 Hz and 6 kHz):

Hospitalisation and death registration data are held by and available through NHS Digital to researchers meeting relevant governance standards (https://digital.nhs.uk/services/data-access-request-service-dars).

All other data are available online. The avoidable death rate can be obtained from government statistics (gov.uk). The Carstairs index is accessible via the UK Data Archive. The Office for National Statistics provides the fuel poverty rate and percentage of non-white population per local authority district.