key: cord-0273942-nmnul86h authors: Konstantinoudis, G.; Minelli, C.; Vicedo Cabrera, A. M.; Ballester, J.; Gasparrini, A.; Blangiardo, M. title: Ambient heat exposure and COPD hospitalisations in England: A nationwide case-crossover study during 2007-2018. date: 2021-10-20 journal: nan DOI: 10.1101/2021.10.19.21265213 sha: 5def1178bc3e3d367f601962cc46ce0123f1d0f8 doc_id: 273942 cord_uid: nmnul86h Background There is emerging evidence suggesting a link between ambient heat exposure and Chronic obstructive pulmonary disease (COPD) hospitalisations. Individual and contextual characteristics can affect population vulnerabilities to COPD hospitalisation due to heat exposure. This study quantifies the effect of ambient heat on COPD hospitalisations and examines population vulnerabilities by age, sex and contextual characteristics. Methods Individual data on COPD hospitalisation at high geographical resolution (postcodes) during 2007-2018 in England was retrieved from the small area health statistics unit. Maximum temperature at 1 kmx1km resolution was available from the UK Met Office. We employed a case-cross over study design and fitted Bayesian conditional Poisson regression models. We adjusted for PM2.5, O3, relative humidity and national holidays, and examined effect modification by age, sex, green space, average temperature, deprivation and urbanicity. Results After accounting for confounding, we found a 0.52% (95% Credible Interval 0.22% to 0.84%) increase in the hospitalisation risk for every 1oC increase in the lag 0-2 temperatures above 23.8oC. We reported weak evidence of an effect modification by sex and age. We found a strong spatial determinant of the COPD hospitalisation risk due to heat exposure, that was alleviated when we accounted for contextual characteristics. Assuming a causal effect, 6 280 (95% CrI 3 441 to 8 993) COPD hospitalisations were attributable to heat exposure. Conclusion Our study suggests that resources should be allocated to support the public health systems, for instance through developing or expanding heat-health alerts, to challenge the increasing future heat-related COPD hospitalisation burden. Chronic obstructive pulmonary disease (COPD) is the most prevalent chronic respiratory disease worldwide, with point prevalence varying from 1.56% in Sub-saharan Africa to 6.09% in Central Europe, eastern Europe, and central Asia in 2007 [1] . In England, COPD is a significant cause of morbidity and mortality, leading to 115,000 emergency admissions and 24,000 deaths per year [2] . The causes of acute exacerbation of COPD are established and include factors such as sex, age, COPD severity and comorbidities [3] . Environmental triggers of COPD hospitalisations such as air-pollution exposure have also been discussed extensively [4] . There is emerging evidence suggesting a link between heat exposure and COPD hospitalisation, either directly or through exacerbating the effects of factors such as ozone concentration that are associated with these events [5] . Several previous studies have examined the effect of high temperatures on COPD hospitalisations, reporting higher rates with heat exposure [6, 7, 8] and heat waves [9, 10] . The majority of these studies are based on aggregated data (at the city or regional level) [7, 10, 11, 9] , whereas only a few considered individual data [6, 8] . Lack of individual data can lead to inadequate adjustment for confounding due to individual factors such as age, physical activities, comorbidities etc., but also to ecological bias, as group level associations do not necessarily reflect individual ones [12] . Although previous studies have assessed the vulnerability related to individual factors, such as age and sex [8, 7] , contextual characteristics, such as green space, average temperature, deprivation and urbanicity, are still poorly characterised. Two of the previous studies have examined the spatial variation of the temperature effect on COPD hospitalisation, using however very coarse geographical resolution [8, 7] . In this nationwide study in England during 2007-2018, we investigated the effect of heat exposure on COPD hospital admissions using a semi-ecological framework. We took advantage of the individual data availability of the outcome and adopted a case-crossover study design that naturally accounts for time-constant variables at the individual patient level. Thus, we were able to account for factors like age, sex, comorbidities, deprivation as well as lifestyle characteristics such as physical activity through the study design. We also adjusted for time-varying confounders, such as air-pollution exposure and relative humidity and examined how the effect of temperature is modified by age, sex and in space. Last, we assessed the extent to which contextual characteristics, such as green space, deprivation, urbanicity and average temperature, contribute to the observed spatial variation of the effect of temperature. 3 . CC-BY 4.0 International license It is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint We included inpatient hospital admissions from COPD in England during 2007-2018 as retrieved from Hospital Episode Statistics (HES) data held by the UK Small Area Health Statistics Unit, provided by the Health and Social Care Information Centre. Age, postcode of residence at time of the hospitalisation, and date of hospitalisation were available for each record. We focused only on admissions with acute exacerbation of COPD as primary diagnosis. We investigated the following diagnostic groups: J40-44 according to the International Classification of Disease version 10 (ICD10) [13] . The analysis is restricted to June, July and August. Daily temperatures at 1 km×1 km resolution were available from the UK Met Office with methods described elsewhere [14] . In brief, the daily temperature in each grid is estimated based on inverse-distance-weighted interpolation of monitoring data, also accounting for latitude and longitude, elevation, coastal influence, and proportion of urban land use. To assign daily temperature to health records, the postcode centroids of each patient were spatially linked to the 1km× 1km grid cell, applying a 100m fuzziness to the postcode location to fulfil governance requirements. We focused on daily maximum temperature averaged over the day of hospitalisation and the preceding two days (lag0-2) to account for the cumulative health effects [15, 16, 17] . To adjust for air-pollution we used hourly concentration of Ozone (O 3 ) and atmospheric particulate matter that has a diameter of less than 2.5 µm (PM 2.5 ), as retrieved from the unified model produced by the Met Office measured in µg/m 3 [18] . The model outcome is then post-processed to correct for bias using observational data [18] . For O 3 we calculated the daily mean of the 8 hours of maximum O 3 , whereas for PM 2.5 the daily mean concentration. The geographical resolution of the air-pollutants is 12 km×12 km for the years 2007-2011 and 2km×2km during 2012-2019. We adjusted for relative humidity (daily and at 10 km×10 km grid) through a model that integrates MetOffice data on daily observations from the meteorological stations and monthly nationwide data as retrieved from HadUK [14] , see Online Supplement Text S1.1. All confounders were included at lag0-2. We also accounted for the effect of national holidays. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint As a measure of green space we used the proportion of a region that is covered by green land such as woodland, agricultural land, grassland and other natural vegetated land as classified in the Land Cover Map 2015 (LCM15) [19] . Deprivation is measured using the Index of Multiple Deprivation (IMD) 2015, as retrieved from the Ministry of Housing, Communities and Local Government [20] . IMD is calculated based on the following domains: a. income, b. employment, c. education, skills and training, d. health and disability, e. crime, f. barriers to housing and services and g. living environment deprivation [20] . Urbanicity (predominantly rural, urban with significant rural and predominantly urban) is based on the Office for National Statistics (ONS) classification in 2011 [21] . We also incorporated the average temperature during 2007-2018 per lower tier local authority (LTLA; Online Supplement Figure S1 ), as a measure of adaptation on higher temperatures [22] . We used a time-stratified case-crossover design, commonly used for analysing the effect of transient exposures [23, 24] . The temperature on the day of COPD hospitalisation (event day), is compared with the temperature on non-event days. This design automatically controls for individual level factors, such as age, sex and ethnicity, but also for confounding that do not vary or vary slowly over time, such as deprivation. We selected non-event days on the same day of week and calendar month as the event day to avoid the overlap bias [25] . Thus we could have maximum 4 non-event days per event day. We modelled the effect of temperature on event compared to non-event days by specifying Bayesian hierarchical conditional Poisson models, with a fixed effect on the event/non-event day grouping [26, 27] . As the effect of temperature on health is typically non-linear [27] , we used piecewise linear threshold models, to allow more flexible fits, but retain ease of interpretation. We considered nationwide thresholds, specified as the 50th, 55th, . . . , 95th percentile of the temperature. We selected the threshold based on the model that minimises the WAIC [28, 29] . We then ran additional models allowing the effect of heat exposure (temperatures above the threshold) to vary by sex (male and female), age (0−64, 65−74, 75+) and space (LTLA). For the spatial effect modification, we used a modelling framework that borrows strength from the geographically adjacent LTLAs [30] . We fitted the latter with and without including the spatial effect modifiers, while adjusting for confounders. The model is described in details in the Online Supplement Text S1.2. Results are reported as medians and 95% Credible Intervals (CrI; 95% probability that the true values lies within this interval) of % increase in the hospitalisation risk [31] and posterior probabilities of a positive % increase. For the spatial varying risk we reported posterior 5 . CC-BY 4.0 International license It is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted October 20, 2021. ; https://doi.org/10.1101/2021.10.19.21265213 doi: medRxiv preprint probabilities that the % hospitalisation risk is larger than the average % hospitalisation risk. To calculate the population attributable fraction we extended [32] to incorporate the spatial dimension of the effect of heat exposure. We first calculated the cumulative heat exposure -COPD hospitalisation relative risk (RR s ) for the s-th LTLA. We could then calculate the attributable fraction: AF s = (RR s − 1)/RR s . Let n s be the number of hospitalisations at days above 23.8 o C and N s the total number of hospitalisation, then AF s (n s /N s ) is the population attributable fraction, i.e. the number of COPD hospitalisations attributable to summer heat exposure. The Bayesian framework adapted allows us to propagate all the random variable-specific uncertainty in our estimates. In a sensitivity analysis, we examined the robustness of the main results focusing on the period 2013-2018, when the confounders are available at the highest resolution. We also repeated the main analysis for the lags 0, 1 and 2 independently. Both sensitivity analyses are conducted using the temperature threshold suggested in the WAIC analysis while fully adjusting for confounders. All analyses are run in R-NIMBLE (Numerical Inference for Hierarchical Models Using Bayesian and Likelihood Estimation) [33] . The code for running the analysis is online available at https://github.com/gkonstantinoudis/ COPDTempSVC. The median maximum temperature across England has increased from 19. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted October 20, 2021. ; https://doi.org/10.1101/2021.10.19.21265213 doi: medRxiv preprint days, 20.78 for event and 20.45 for non-event days at lag 1, 20.49 for event and 20.47 for non-event days at lag 2 and 20.51 for event and 20.49 for non-event days at lag 0-2, Online Supplement Table S2 . The distribution of confounders across event and non-event days and the spatial distribution of the effect modifiers at the LTLA level can be found on the Online Supplement, Table S2-4 In the unadjusted models, the threshold minimising the WAIC was the 75th percentile of the temperature (22.6 o C), whereas the 85th percentile (23.8 o C) in the fully adjusted models, Online Supplement Table S5 . In the unadjusted model, we found 1.3% (95% CrI 1.0% to 1.5%) increase in the COPD hospitalisation risk for every 1 o C increase in temperatures above 22.6 o C. In the fully adjusted model, the effect was lower in magnitude (0.52%, 95% CrI 0.22% to 0.84%) for temperatures above 23.8 o C, Online Supplement Table S5 . All subsequent analyses were conducted using the 85-th percentile of the temperature as the threshold. In the unadjusted models, the percentage of risk increase in hospitalisations for every 1 o C increase in the heat exposure vary from 0.81% (95% CrI 0.12% to 1.49%) in males 64 years old or younger to 1.47% (95% CrI 0.78% to 2.12%) in females aged 65-74, Figure 2 and Online Supplement Table S6 . After adjusting for the selected confounders, the effects are smaller and the evidence weaker, Figure 2 . In the fully adjusted models, the effects vary from 0.27% (95% CrI -0.65% to 1.14%) in females 64 years old or younger to 0.75% (95% CrI -0.03% to 1.53%) in males 65-74 years old, Figure 2 and Online Supplement Table S6 . The spatial variation of the effect of heat exposure on COPD hospitalisations is shown on Fig. 3 . The risk of COPD hospitalisation is less than 0.49% for every 1 o C increase in heat exposure in South West, top left panel We found weak evidence that populations in areas with higher proportions of green space, larger average temperature and higher level of urbanicity are more resilient to COPD hospitalisations due to heat exposure, is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint Assuming a causal relationship, there were 6 280 (95% CrI 3 441 to 8 993) COPD hospitalisations attributable to heat exposure during 2007-2018 in England. This accounts for 2.2% (95% CrI 1.2% to 3.2%) of the total COPD hospitalisations during summer months. The proportion of COPD hospitalisations attributable to heat exposure has a clear spatial structure and is more than 3% in East Midlands, East of England, London and South East, while it is below 1% in the South West (Fig. 4 ). Focusing on the years 2013-2018, when the confounders are available at higher geographical resolution, the effect of heat exposure in the fully adjusted model remained consistent with the main analysis (0.52%, 95% CrI 0.14% to 0.87%). The lag with the highest influence was lag 1 with the risk of COPD hospitalisation being 0.73% (95% CrI 0.48% to 0.98%) for every 1 o C increase in heat exposure. For lag0 and lag2 the point estimate was positive, but the interval included zero, 0.03% (95% CrI -0.20% to 0.29%) and 0.23% (95% CrI -0.01% to 0.50%) respectively. This is the first nationwide case-crossover study in England investigating the short-term effects of heat exposure on COPD hospitalisation. After accounting for confounding, the results indicate that for every 1 o C increase in heat exposure the COPD hospitalisation risk increases by 0.53% (95% CrI 0.22% to 0.84%). We found weak evidence of an effect modification by sex and age. The attributable burden of heat exposure has a clear spatial is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted October 20, 2021. ; https://doi.org/10.1101/2021.10.19.21265213 doi: medRxiv preprint available for linkage with the exposure and confounding factors. Such geographical resolution is expected to minimise misclassification resulting from any spatial misalignment between the outcome and exposure/confounder. The availability of individual data for the outcome also minimises ecological bias [12] , while guaranteeing high statistical power due to the population-based nature of the study. We ascertained hospital records from NHS digital covering almost all hospitalisation occurred in the public sector in England during 2007-2018. Our study has some limitations. First, residential temperature does not reflect the actual temperature exposure of an individual, as individuals are exposed to different temperatures in the course of the day. In addition to this, the outdoor temperature, as provided by Met Office does not reflect the actual temperature exposure inside the house. Nevertheless, in line with most of the studies in this field and given the lack of more precise individual exposure data, we used residential temperature outdoors as a proxy for the individual exposure. The resulting misclassification is expected to be non-differential and bias the effect estimates towards the null. To allow for flexible fits, we used a linear threshold model. More complex relationships may need multiple thresholds, however the WAIC analysis suggested that the linearity assumption suffices. Although we adjusted for the main COPD hospitalisation environmental contributors, we could not evaluate other potential confounders (eg, seasonal allergies and pollen counts) due to the lack of available data. Additionally, exposure to other air-pollutants, such as NO 2 , SO 2 , might also confound the observed relationship; we decided to adjust for PM 2.5 and O 3 as they seem to have a larger impact on COPD hospitalisation and to avoid potential collinearity with other pollutants. Our results can be compared with studies examining COPD hospital admissions and ambient temperatures during the hottest months [7, 11, 34, 35] . Our study is in line with a US study including 12.5 million participants that found a 4.7% (95% CrI 3.9 to 5.5%) increase in the COPD hospitalisation rate at lag 0 for every 5.6 o C increase in the average daily temperature during May-September [7] . Our study is also in line with a case-crossover study in Brazil that reported a 5% (95% CrI 4% to 6%) increase in the hospitalisation odds for every 5 o C increase in the average temperature (0-3 lags) during the 4 hottest months [11] . In contrast, a study in New York reported a 7.64% increase in the risk of COPD admissions for each 1 o C increase in daily mean apparent temperature above 32 o C [34] . A study in 12 European cities, reported a 4.5% (95% CrI 1.9 to 7.3) and 3.1% (95% CrI 0.8 to 5.5) increase in total respiratory admissions (the majority being COPD) in Mediterranean and North-Continental cities, respectively, for every for each 1 o C increase in the maximum apparent temperature (lag 0-3 days) above the 90th percentile [36] . A study in Taiwan reported negative correlation between the average daily temperature and emergency admissions with COPD, but a 14% increase in the emergency COPD admissions when the diurnal temperature range is larger than 9. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint We found weak evidence of an effect modification by age and sex, but discrepancies in vulnerability in space. A previous study in Brazil, reported higher COPD hospitalization odds for women and the elderly [11] . In our study, we observed some evidence that women and elderly are more vulnerable, but only in the unadjusted models. In the fully adjusted models, in line with a previous study in the US [7] , the age group 65-74 was the most vulnerable. Some spatial variability by regions or counties was also observed in previous studies in Brazil and the US, potentially due to socioeconomic characteristics or exposure to higher average summer temperatures [11, 7] . In our study, green space, average temperature, deprivation and urbanicity explained some of the observed variation in the observed spatial vulnerabilities, the evidence of an effect was however inconclusive. Our results are lower in magnitude compared to all previous studies [11, 7, 34, 36, 35] . Differences in the adjustment either for individual (through the study design; only [11] is a case-crossover study) or environmental confounding might explain the observed discrepancy. While [36, 34] and [7] have adjusted for air-pollution, the study in Brazil has not [11] . All previous studies had available coarser geographical resolution (city or county level), leading to inadequate adjustment for confounding, as confounders, such as air-pollution, vary in high geographical resolution [37] . Differences in the definition of the outcome can also lead to the observed discrepancy as previous studies have used the apparent temperature (a metric that combines air temperature, relative humidity and wind speed) [34, 36] or diurnal temperature range [35] , while others, more in line with our approach, the daily mean [11, 7] . Last, decisions regarding the selection of the temperature threshold, the warm-season months and the lags to be considered can also partly explain the observed difference in the effect estimates. Acute COPD episodes are associated with airways and systemic inflammation but also with cardiovascular comorbidity and may be triggered by exposures to heat [36] . Exposure to ambient heat can lead to heat dissipation through hyperventilation and may trigger dynamic hyperinflation and dyspnoea in patients with pre-existing COPD [11, 7] . The higher risk of COPD hospitalisation in the 65-74 age group observed in our study could be explained by the inability of this frail population to dissipate excess heat through circulatory adjustment, and exposure to extreme temperatures increases their risk of developing pulmonary vascular resistance secondary to peripheral pooling of blood or hypovolemia [36] . In addition, elderly populations are of higher risk to have cardiovascular comorbidities, which are hypothesised to increase the risk of COPD hospitalisations associated with heat exposure. Nevertheless, such evidence is inconclusive [36] . We also reported a weak protective effect of higher average temperatures, arguing towards protective adaptation to heat, possibly related to differences in housing stock or behavior during hot weather [11] . We observed weak evidence of increased resilience in populations in more deprived areas and in areas with higher degrees of urbanicity. Although this evidence is 10 . CC-BY 4.0 International license It is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted October 20, 2021. ; https://doi.org/10.1101/2021.10.19.21265213 doi: medRxiv preprint inconclusive, potential factors that could confound the observed effect include differences in demographics, for instance ethnicity. Previous studies examining future trends in COPD, population demographics and temperature changes have predicted a higher COPD prevalence, a raise in the average age of the population and increased global temperatures [38, 39, 40] . Resources should be allocated to support the preparedness and resilience of public health systems, for instance through developing or expanding heat-health alerts, to challenge the increasing heat exposure related COPD hospitalisation burden. We thank Hima Daby, Gajanan Natu, Eric Johnson and Bethan Davies for their help with data acquisition, storage, preparation and governance. All authors acknowledge infrastructure support for the Department of is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted October 20, 2021. The authors declare no competing interests. Not required. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint Tables Table 1 . Median, 95% credible intervals of the percentage change of the heat exposure related spatial hospitalization risk due to green space, average temperature, index of multiple deprivation and urbanicity and probability that this percentage change is higher than 0. 17 . CC-BY 4.0 International license It is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted October 20, 2021. ; https://doi.org/10.1101/2021.10.19.21265213 doi: medRxiv preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted October 20, 2021. ; https://doi.org/10.1101/2021.10.19.21265213 doi: medRxiv preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted October 20, 2021. ; https://doi.org/10.1101/2021.10.19.21265213 doi: medRxiv preprint Prevalence and attributable health burden of chronic respiratory diseases, 1990-2017: a systematic analysis for the global burden of disease study Overview of potential to reduce lives lost from chronic obstructive pulmonary disease (copd) Risk factors for all-cause hospital readmission following exacerbation of copd: a systematic review and meta-analysis Outdoor air pollution and copd-related emergency department visits, hospital admissions, and mortality: a meta-analysis Climate change, extreme weather events, air pollution and respiratory health in europe Joacim Rocklöv, and Bertil Forsberg. The effect of heat waves on mortality in susceptible groups: a cohort study of a mediterranean and a northern european city Heat-related emergency hospitalizations for respiratory diseases in the medicare population Ambient heat and hospitalisation for copd in brazil: a nationwide case-crossover study Excess mortality and morbidity during Cause-specific risk of hospital admission related to extreme heat in older adults Effects of ambient temperature on acute exacerbations of chronic obstructive pulmonary disease: results from a time-series analysis of 143318 hospitalizations Ecologic studies revisited Recording of hospitalizations for acute exacerbations of copd in uk electronic health care records Haduk-grid-a new uk dataset of gridded climate observations The effect of high temperatures on cause-specific mortality in england and wales Mortality displacement of heat-related deaths: a comparison of delhi, sao paulo, and london Heat-related and cold-related deaths in england and wales: who is at risk? Occupational and environmental medicine Application of a statistical postprocessing technique to a gridded, operational, air quality forecast Department foe Communities and Local Government. The english indices of deprivation 2015 statistical release Defra Rural Statistics. 2011 rural/urban classification Changes in population susceptibility to heat and cold over time: assessing adaptation to climate change The case-crossover design: a method for studying transient effects on the risk of acute events On the equivalence of case-crossover and time series methods in environmental epidemiology Overlap bias in the case-crossover design, with application to air pollution exposures Conditional poisson models: a flexible alternative to conditional logistic case cross-over analysis Vulnerability to the mortality effects of warm temperature in the districts of england and wales Understanding predictive information criteria for bayesian models Asymptotic equivalence of bayes cross validation and widely applicable information criterion in singular learning theory Bayesian image restoration, with two applications in spatial statistics Risk set sampling for case-crossover designs Population attributable fraction Programming with models: writing statistical algorithms for general model structures with nimble Extreme high temperatures and hospital admissions for respiratory and cardiovascular diseases Diurnal temperature range and emergency room admissions for chronic obstructive pulmonary disease in taiwan High temperature and hospitalizations for cardiovascular and respiratory causes in 12 european cities Long-term exposure to air-pollution and covid-19 mortality in england: A hierarchical spatial analysis Projecting the copd population and costs in england and scotland