Study region and period
The study region was the Melbourne greater metropolitan area in the state of Victoria, Australia (2011 population ~ 4 million). Data for the year 2011 were collected from the population census to correspond with the 2011 road traffic noise model. This is the most recent road noise data available. The study region was defined to incorporate all LGAs in which noise exposure-model data were estimated.
Road traffic noise model
The EPA provided detailed maps of noise levels (measured as dB) that were developed from a project that used data from 45,000 road segments covering 7500 km of roads hosting over 2,000,000 buildings in a 10,000 km2 study region [15]. Noise maps of Melbourne were generated using the NPM96 with road traffic data from the VITM 2011. The algorithms of the NPM96 calculate A-weighted equivalent continuous sound pressure levels (LAeq) from traffic flows of light and heavy vehicles, and consider speeds, distances to road center lines, heights of roads, heights, positions and thicknesses of barriers, types of ground surface and locations of exposed individuals. The model was verified by attended and unattended noise measurements at over 300 locations using ARL 315, ARL 316 and NTi XL2 environmental noise loggers during 2011. These noise loggers were used in accordance with the Australian Standard—AS 2702:1984. Noise levels were recorded as LAeq and A-weighted maximum noise levels (LAmax) at 15- or 60-min intervals. A-weighting was applied to correct noise volumes relative to those perceived by the human ear, as the ear is less sensitive to very high and low audio frequencies.
Noise levels were modelled using SoundPLAN, which is an environmental noise modelling software suite from SoundPLAN GmbH. SoundPLAN was also used to develop detailed 3D models including ground contours, buildings, ground absorption, pavement surface types, noise barriers and various source data, such as percentages of traffic during day, evening and night-time periods that are predictive of road noise emissions. Predicted noise levels were verified against measured data, revealing 90% confidence between modelled and measured road traffic noise values to within ± 4–5 dB across day, evening and night periods. Using the verified noise model, noise levels were then predicted both at the façades of all floors for sensitive buildings in the study region, and as a noise grid covering the entire study area at a reference height of 1.8 m above ground level.
Noise estimates for the noise grid were aggregated to 10 × 10-m pixels, which were then averaged over MB areas using the boundaries from the ABS 2011 census. MBs contain about 90 people. Although noise levels vary substantially within MBs, these noise exposures are likely underestimates of those in homes that are close to noisy roads. Moreover, noise levels from road traffic were modelled using data from the VITM, which include traffic volumes for major arterial and feeder roads only. The limited coverage of small roads likely contributes to underestimates of noise exposures in MBs.
Health data
LGA level IHD rates were obtained from the public mortality database of the Australian Institute of Health and Welfare: Mortality Over Regions and Time (MORT) books Local Government Area (LGA), 2011–2015 Table 2: Leading causes of death by sex, 2011–2015 [19]. To accommodate the limitations of LGA level IHD mortality data, prior to calculating attributable excess risks, we performed statistical downscaling computations using meshblock (MB; ~ 90 people per MB) population data from the Australian Bureau of Statistics. Specifically, mortality rates were downscaled to MBs using MB population data after adjusting for LGA-specific population weighted exposures (described below).
Health impact function
To estimate noise-attributable excess risks of IHD deaths for all MBs in Melbourne, we calculated odds ratios (OR) of IHD in all MBs by inserting noise levels into the nonlinear polynomial that is recommended for relating traffic noise with IHD by the World Health Organization [11], as follows:
$$\begin{array}{l} {{\text{For}}\;\;L_{{{\text{day}},16{\text{h}}}} \ge 55\;{\text{dB,}}} \\ {} \\ \quad \quad \quad {OR\left( {L_{{{\text{day}},16{\text{h}}}} } \right) = 1.63 - 6.13 \times 10^{-4} \times (L_{{{\text{day}},16{\text{h}}}} )^{2} + \; 7.36 \times 10^{-6} \times (L_{{{\text{day}},16{\text{h}}}} )^{3} } \\ {} \\ {{\text{Else}}\;\;OR\left( {L_{{{\text{day}},16{\text{h}}}} } \right) = 1,} \\ \end{array}$$
where \(L_{{{\text{day}},16{\text{h}}}}\) represents average noise levels between 0700 and 2300 h.
Excess risks are computed as follows:
$$ER_{i} = \left( {OR\left( {L_{{{\text{day}},16{\text{h}},i}} } \right) - 1} \right) \times P_{i} \times \widehat{{I_{k} }},$$
where \(ER_{i}\) is the excess risk of deaths in MBi and \(L_{{{\text{day}},16{\text{h}},i}}\) represents the average exposure level in MBi. \(OR\left( {L_{{{\text{day}},16{\text{h}},i}} } \right)\) is the odds ratio at that noise level and is estimated by inserting the MB estimated \(L_{{{\text{day}},16{\text{h}}}}\) into the polynomial function above. \(P_{i}\) is the population of MBi and the baseline mortality incidence rate is represented by \(\widehat{{I_{k} }}\). Our approach uses MB population numbers to calculate excess risks at the highest spatial resolution.
To estimate \(\widehat{{I_{k} }}\) we used the regional average annual IHD mortality rate \(I_{k}\) (for each region e.g. in \(LGA_{k}\)) and divided this by the population-weighted average OR of all MBs within region k, as follows:
$$\widehat{{I_{k} }} = \frac{{I_{k} }}{{\overline{{OR_{k} }} }},$$
where \(\overline{{OR_{k} }}\) represents the population-weighted average OR within region k and is calculated using the following equation:
$$\overline{{OR_{k} }} = \frac{{\mathop \sum \nolimits_{i = 1}^{n} P_{i} \times OR\left( {L_{{{\text{day}},16{\text{h}},i}} } \right)}}{{\mathop \sum \nolimits_{i = 1}^{n} P_{i} }},$$
These formulae represent the hypothetical underlying cause-specific mortality rate for region k and approximate the health outcomes that would be observed in a counterfactual unexposed population. Hence, multiplication of the resulting mortality rate by the population and the OR yields the attributable excess risk in person-years given the observed level of exposure. Figure 1 illustrates our risk assessment approach.