Real-time forecast of temperature-related excess mortality at small-area level: towards an operational framework

The development of innovative tools for real-time monitoring and forecasting of environmental health impacts is central to effective public health interventions and resource allocation strategies. Though a need for such generic tools has been previously echoed by public health planners and regional authorities responsible for issuing anticipatory alerts, a comprehensive, robust and scalable real-time system for predicting temperature-related excess deaths at a local scale has not been developed yet. Filling this gap, we propose a flexible operational framework for coupling publicly available weather forecasts with temperature-mortality risk functions specific to small census-based zones, the latter derived using state-of-the-art environmental epidemiological models. Utilising high-resolution temperature data forecast by a leading European meteorological centre, we demonstrate a real-time application to forecast the excess mortality during the July 2022 heatwave over England and Wales. The output, consisting of expected temperature-related excess deaths at small geographic areas on different lead times, can be automated to generate maps at various spatio-temporal scales, thus facilitating preventive action and allocation of public health resources in advance. While the real-case example discussed here demonstrates an application for predicting (expected) heat-related excess deaths, the framework can also be adapted to other weather-related health risks and to different geographical areas, provided data on both meteorological exposure and the underlying health outcomes are available to calibrate the associated risk functions. The proposed framework addresses an urgent need for predicting the short-term environmental health burden on public health systems globally, especially in low- and middle-income regions, where rapid response to mitigate adverse exposures and impacts to extreme temperatures are often constrained by available resources.


Data sources
The computation and mapping of the expected number of excess deaths was performed for each lower super output area (LSOA) in England and Wales.The LSOAs are census-based statistical units with approximately 1,500 residents.These areas correspond to the definitions used in the 2011 census, with a total of 34,753 LSOAs in the two countries.The computation was applied separately by age groups, specifically 0-64, 65-74, 75-84, and 85 and older.The data sources are described below.
(i) Temperature-mortality relationship: An exposure-response relationship is defined as a function that represents the risk across the exposure range using a specific reference value.In this context, exposure-response relationships are represented by non-linear dependencies that inform about the risk of mortality for all causes associated with daily mean temperature (°C).The risk usually increases at both high and low temperatures, with the lowest value corresponding to the minimum mortality temperature (MMT) used as a reference.Age-specific exposure-response relationships with a lag period of 0-21 days for all the LSOAs were estimated in a previous small-area study [1], using individual mortality records and temperature data for the period 2000-2019.Mortality data were provided by the Office of National Statistics (ONS, agreement MRP 2291/2013), while daily mean values of near-surface air temperatures on a 1x1 km grid across the United Kingdom were extracted from the HadUK-Grid database developed by the Met Office [2].Specific details are provided in the original publication [1].
(ii) Forecast temperatures were retrieved from the 3-hourly '2 metre temperature' (variable '2t') covering the period 17/07/2022 (03h UTC) -22/07/2022 (24h UTC) from the ECMWF high-resolution Open Data [3].The data are provided on a 0.4°x 0.4° spatial grid (~45km) and based on mediumrange high-resolution forecast models (HRES).These data are made publicly available by the ECMWF as part of their real-time meteorological products.The single-level temperature data in grib2 format (https://www.ecmwf.int/en/forecasts/datasets/set-i#I-i-a_fc,'2t', unit: °K) were accessed from the forecast issued on 17/07/2022 at 00h UTC using the open-source python package 'ecmwfopendata' (https://github.com/ecmwf/ecmwf-opendata).We next aggregated the 3-hourly gridded temperature fields to daily averages in °C in Python, and then spatially aggregated them to the 37,473 Lower Super Output Areas (LSOAs) in England and Wales.The spatial aggregation to LSOA boundaries was performed in R (version 4.2.3)[4] using the package 'exactextractr' (version 0.10.0,https://github.com/cran/exactextractr)[5], and polygon shape files from the Open Geography portal of ONS [6] (elaborated below).The LSOA-aggregated level forecast daily temperature for 17-22 July 2022 were finally matched to the estimates of age-specific exposure-response functions between temperature and all-cause mortality obtained from the earlier published analysis [1].
(iii) Baseline daily mortality counts were estimated by applying age-stratified mortality rates to populations in each LSOA.Both sources of information were retrieved from the latest updated data Figure S2

Figure S2 .
Figure S2.Regions in England and Wales used for aggregating population and excess-deaths in TablesS1 and S2.

Table S1 .
Number of expected excess deaths and rate (per 1,000,000 people) (95% eCI) in England and Wales stratified by region, age group, and date, as predicted during the heatwave of 17-19 July 2022.

Table S2 .
[7]parison of predicted expected excess deaths (95% eCI) and UKHSA-ONS official estimates (95% CI) by age groups and regions in England and Wales.Note: The predicted excess deaths are for 17-19 July 2022.The UKSHA-ONS estimates are extracted from Tables1 and 2in ref[7]and instead cover 10-25 July 2022.