We used the Western Australian Rheumatic Disease Epidemiological Registry (WARDER) to access unit-record linked data from the Western Australian Hospital Morbidity Data Collection (HMDC). This data collection is maintained by the Western Australian Department of Health and linked through the Western Australian Data Linkage System (WADLS). WARDER contains all public and private hospital admissions between 1980 and 2014 for patients with principal and secondary discharge diagnosis of systemic autoimmune rheumatic disease, gout, or osteoarthrosis [14].

Metropolitan Perth in Western Australia (WA) is bounded by the Indian Ocean on the west and the Darling Range on the east. It is situated on relatively flat land spanning (at the time of writing) approximately 150km from north to south and up to approximately 45km, from east to west. This region experiences a hot-summer Mediterranean climate and cool to mild wet winters [15].

For this time series study, we used the HMDC dataset to identify daily hospitalisations to public and private hospitals where the principal diagnosis was gout (International Classification of Diseases, Ninth Revision, Clinical Modification [ICD-9-CM] code 274 and the International Classification of Diseases Tenth Revision, Australian Modification [ICD-10-AM] code M10) among residents of the Perth metropolitan area (identified from residential postcode) between 1 January 1980 and 31 December 2014 (observation period). We excluded patient transfers where the interval between discharge and subsequent admission for the patient was within one day consistent with our previous work [16, 17]. During this period, the estimated resident population was 1.02 million in 1980 and increased to 2.01 million in 2014.

We obtained daily maximum temperature data for the Perth Regional Office station (latitude=31.96°S; longitude=115.87°E; elevation=19.0 m above mean sea level) from the Scientific Information for Land Owners (SILO) dataset in Australia, hosted by the Science and Technology Division of Queensland Government’s Department of Environment and Science [18]. This was the only station in the metropolitan area with temperature data from 1980 to 2014. Data available from this weather station included daily maximum/minimum temperatures, rainfall, vapour pressure, vapour pressure deficit, evaporation, solar radiation, relative humidity at maximum temperature, relative humidity at minimum temperature, mean sea level pressure.

Data on WA public holidays were obtained from Nager.Date [19] and used to control for public holidays in the regression model. Estimated resident population data for population denominators were obtained from the Australian Bureau of Statistics [20].

We calculated descriptive statistics for daily maximum temperature (mean, standard deviation [SD], median, 1^st, 5^th 95^th, 99^th percentiles) and number of daily gout flare hospitalisations for the observation period. Colder days are defined as those ≤5^th percentile of daily maximum temperatures, while hotter days are those ≥95^th percentile of daily maximum temperatures.

In addition, for each day during our observation period with a particular temperature value (e.g. 16°C) we counted the number of gout flare hospitalisations per 1,000,000 population over lag days 0-21. We then averaged these counts/1,000,000 population for each temperature value.

We used a distributed lag non-linear model (DLNM) [21] to investigate the relative risk of gout flare hospitalisations with maximum daily temperature, allowing for potential non-linear associations as well as lagged effects. This model was selected to account for the potential lag period between temperature exposure and gout flare hospitalisations [7]. Specifically, a cross-basis function, derived as a special tensor product of natural cubic splines, was used to jointly model maximum daily temperature (°C) and the lag of the number of days since that exposure. Spline parameters were consistent with another Australian study [22]. Spline 1 of the cross-basis function was specified using three internal knots placed at equally spaced values of temperature, and spline 2 using an intercept plus three internal knots placed at equally spaced values of the logged lag days with a maximum lag of 21 days [22]. We used the mean of the daily maximum temperatures across all years 1980-2014 as the reference temperature (24°C) for calculating relative risks (RRs). Furthermore, a spline term was included to control for changes with time during the study period by modelling day number as an integer, from 1 (1 January 1980) to ‘12784’ (31 December 2014) using a natural spline with 6 degrees of freedom (df) per year [22]. Three additional model terms were included: daily humidity (%), modelled using a natural spline with 4 df; weekend day (yes or no); public holiday (yes or no). This was fitted as a quasi-Poisson Generalized Linear Model (GLM) with a log link function. We adjusted for weekends and public holidays to account for differences in health service utilisation and patient behaviours on these days [23].

We initially analysed all cases and then stratified by biological sex (hereafter sex) and age (<75 years, ≥75 years). We chose this age cut-off as it is commonly used in age-related analyses [8, 24, 25]. Furthermore, we explored using 54 and 64 years as age cut-offs, however there was only a small number of total hospitalisations in females aged 54-64 years (<90 hospitalisations) to produce a stable model fit and estimates. However, there were sufficient numbers of total hospitalisations for both males and females in the <75-year and ≥75-year age groups for the statistical modelling and convergence. Due to the small number of days with hot and cold extremes (<1% each), temperatures <15°C were recoded to 15°C while those >40°C were recoded to 40°C.

We used Stata for data handling and preparation and R for DLNM and GLM. DLNM was fitted using the dlnm package in R [26].

In sensitivity analyses to determine the robustness of our results, we increased the number of internal knots for Splines 1 and 2 from three to four. We also increased the df for ‘changes with time’ from 6 to 7 df per year, and for ‘daily humidity’ from 4 to 5 df.

Approval to conduct this study was obtained from the Human Research Ethics Committee of WA Department of Health (approval no. 2016/24) where a waiver of consent was granted as the research met the criteria outlined in the National Statement on Ethical Conduct in Human Research. Analyses were conducted according to relevant local and national guidelines and regulations.

During the study observation period of 12,784 days, daily maximum temperatures ranged from 12°C to 46°C, with mean and median of 24°C (SD=5.9) and 23°C, respectively (Figure 1). First, 5^th, 95^th and 99^th percentile daily maximum temperatures were 15°C, 17°C, 36°C and 40°C respectively. Hence in this study, colder and hotter days are when daily maximum temperatures are ≤17°C and ≥36°C, respectively.

There were 5,731 gout flare hospitalisations, and in general. the number of admissions increased with higher temperatures (Figure 1). For example, there were, on average, 5.8, 6.2, 6.8 and 6.5 hospitalisations per 1,000,000 population over lag days 0-21 (i.e. from lag day 0 to lag day 21) following exposure to daily maximum temperatures of 15°C, 17°C, 36°C and 40°C respectively. Similarly, the number of gout flare hospitalisations increased with higher temperatures amongst older (≥75years) and younger (<75years) males and females (Supplementary Table 1). However, these are crude counts and do not consider the complex exposure-time-outcome relationship. Findings hereafter are based on DLNM which provide a comprehensive picture of the time-course of the exposure-outcome relationship [27] and considers possible effect modification due to sex and age.

Significant exposure-time-outcome associations were observed for the risk of gout flare hospitalisations with daily maximum temperature (i.e., relative to the reference temperature of 24°C) in all data, and separately by sex and age group (<75 years, ≥75 years) (Figure 2). At lag day 7, (i.e. 7 days following temperature exposure), hotter days (36°C) were associated with increased risk of gout flare hospitalisations in all data (RR=1.04; 95% CI: 1.01-1.08), with a similar RR estimated for males only (RR=1.08; 95% CI: 1.01-1.15) but no significant association estimated for females. People aged ≥75 years had higher estimated RRs following both colder (≤15°C: RR=1.08; 95% CI: 1.01-1.16) and hotter days (36°C: RR=1.07; 95% CI: 1.02-1.12), relative to the reference temperature (Figure 2). However, those younger than 75 years had higher but non-significant estimated RRs on lag day 7 on hotter days (36°C: RR=1.03; 95% CI: 0.99-1.07), relative to the reference temperature. Given these findings, results presented hereafter are stratified by sex and age (i.e. males aged ≥75 years [1,465 hospitalisations], females aged ≥75 years [869 hospitalisations], males aged <75 years [2,936 hospitalisations] and females aged <75 years [461 hospitalisations]).

Males aged ≥75 years had higher risk of gout flare hospitalisations (Figure 3, Supplementary Figure 1, Supplementary Table 2) following hotter days (36°C) beginning immediately on lag day 0 (RR=1.39; 95% CI: 1.02-1.88) and again from lag day 4 (RR=1.10; 95% CI: 1.02-1.18) to lag day 8 (RR=1.06; 95% CI: 1.01-1.12). They also experienced higher risk of gout flare hospitalisations from lag day 7 (RR=1.11; 95% CI: 1.01-1.21) to lag day 10 (RR=1.09; 95% CI: 1.01-1.18) after colder days (≤15°C). Females aged ≥75 years did not experience increased risk of gout flare hospitalisations following hotter (≥36°C) days but had higher risk at 35°C from lag day 3 (RR=1.12; 95% CI: 1.02-1.23) to lag day 4 (RR=1.11; 95% CI: 1.01-1.22).

Males aged <75 years (Figure 4, Supplementary Figure 2, Supplementary Table 2) did not experience increased risk after hotter days (≥36°C) but had higher risks due to 35°C days from lag day 6 (RR=1.06; 95% CI: 1.01-1.12) to lag Day 8 (RR=1.05; 95% CI: 1.01-1.09). They had higher but non-significant risks following colder days (17°C) from lag day 1 (RR=1.04: 95% CI: 0.96-1.13) to lag day 3 (RR=1.01: 95% CI: 0.95-1.06), but significant lower risk from lag day 7 (RR=0.96: 95% CI: 0.92-0.99) to lag day 8 (RR=0.96: 95% CI: 0.93-0.99). Females aged <75 years did not experience higher risk of gout flare hospitalisations after hotter or colder days over lag day 0 to lag day 21, relative to the reference temperature.

In sensitivity analyses to determine robustness of our results, we observed similar shapes of exposure-time-outcome associations when we modified the number of knots for Splines 1 and 2 and df for ‘changes with time’ and ‘daily humidity’ (Supplementary Figures 3-6).