Heat and Cold-related Hospitalization Risk in North-East of Iran: A Time-stratied Case Crossover Design

32 Background: This study aimed to estimate hospitalization risk/number attributed to air extreme 33 temperatures using time-stratified case crossover study and distributed lag non-linear model in a 34 region of Iran during 2015-2019. 35 Methods: A time-stratified case crossover design based on aggregated exposure data was used in 36 this study. In order to have no overlap bias in the estimations, a fixed and disjointed window by 37 using one-month strata was used in the design. A conditional Poisson regression model allowing 38 for over dispersion (Quasi-Poisson) was applied into Distributed Lag Non-linear Model (DLNM). 39 Different approaches were applied to estimate Optimum Temperature (OT). In the model, the 40 interaction effect between temperature and humidity was assessed to see if the impact of heat or 41 cold on Hospital Admissions (HAs) are different between different levels of humidity. 42 Results: The cumulative effect of heat during 21 days was not significant and it was the cold that 43 had significant cumulative adverse effect on all groups. While the number of HAs attributed to 44 any ranges of heat, including medium, high, extreme and even all values were negligible, but a 45 large number was attributable to cold values; about 10000 HAs were attributable to all values of 46 cold temperature, of which about 9000 were attributed to medium range and about 1000 and less 47 than 500 were attributed to high and extreme values of cold, respectively. 48 Conclusion: This study highlights the need for interventions in cold seasons by policymakers. The 49 results inform researchers as well as policy makers to address both men and women and elderly 50 when any plan or preventive program is developed in the area under study. 51


54
The exposure to weather parameters such as air temperature and humidity have been increasingly 55 paid attention due to many extreme events which have several health outcomes. The impact of heat 56 and cold on human health has been documented by several studies. For example, more recently, a 57 study conducted by multicountry data showed that heat waves had significant cumulative risk of 58 mortality in all countries involved in the study (Guo et al. 2017) . More importantly, climate change 59 is projected to increase weather extreme events such as heat and cold waves as well as extreme 60 temperature values that might highly endanger human health (Son et al. 2014). Hence, it is  The climate or weather change seems to be more serious problem in the Middle East countries, 73 especially in Iran. Iran is expected to experience an increase of 2.6 °C in mean temperatures and a 74 decrease of 35% in precipitation by the next decades(Mansouri Daneshvar et al. 2019). Thus, the 75 heat and cold continue to be big health problems in Iran. In this study, we aimed to assess the 76 association between air extreme temperatures and HAs using time-stratified case crossover study 77 and distributed lag non-linear model in a region located to the North-East of Iran. In addition, 78 Attributable Risk/Number were estimated in the study which can be useful for policymakers to 79 make decision about the problem. The Hospital Admissions (HAs) were obtained from a referral hospital in North Khorasan. The

84
HAs due to all causes during 2015-2019 were included in analysis. They were also categorized 85 based on sex and age group in order to determine high risk groups. The weather and pollutants 86 data were collected by national weather organization and department of environment. In this study, 87 daily mean temperature was used as the predictor of HAs because many study showed that it is 88 better predictor than minimum and maximum temperature for mortality or disability (Aboubakri for HAs, mean temperature and humidity was negligible (lower than 5 percent). However, there 91 were a higher number of missing data for air pollutants, but they were still low number (<20%), 92 thereby allowing us to impute the data by multiple imputation method.

94
A time-stratified case crossover design was used in this study. In order to have no overlap bias in 95 the estimations, a fixed and disjointed window was used in the design (Janes et al. 2005). In this 96 design, a given exposure for a case occurring on Monday, 8 July, for example, would be compared 97 to the exposure occurring on all other Mondays in July (i.e. 1 July, 15 July, 22 July, and 29 July).

98
Indeed, we developed one-month strata in which for a case that falls in a stratum the every seven 99 days before or after the case day in the same stratum were chosen as reference, thereby allowing 100 the reference period to be randomly selected over the months so that days of week and months  To determine the OT, we did sensitivity analysis and used two different approaches. In first 108 approach, a nonlinear function (loess spline) was fitted on the association between temperature 109 and ARIMA models' residuals. Indeed, arima (1, 1, 1), was fitted on HAs, and then the residuals  Median temperature has also been considered as OT in some studies (Guo et al. 2011) , though, it 125 had higher risk than the temperature in our study. In addition, the CI95% (in in figure S1, b) does 126 not included the median (14.9°C), showing unsuitable as OT in our study.

127
Cold and heat were the temperatures below and above the OT, respectively. Also, percentiles 5 th , 128 1 th , 95 th and 99 th were defined as cold, extreme cold, heat and extreme heat values. The ranges 129 between the values were, therefore, defined as different levels of cold and heat (i.e. the values from 130 the OT to percentile 5 th , the percentile 5 th to percentile 1 th , percentile 1 th to minimum value were 131 defined as medium, high and extreme values of cold, respectively).    In which, AF indicate the attributable fraction at time t associated to risk factor x in the past time 180 t-0… 21. Therefore, the AF in DLNM comes from cumulative relative risk. In this models, 181 Attributable Number (AN) in a specific day as to previous risk factor x can also be calculated by 182 multiplying the AF to number of events (HAs) in the same day (n t in equation 3). The latter is 183 more likely to be easily interpreted by policy makers. So, we calculated both AF and AN.    While the cumulative effect of heat during 21 days was not significant for any subgroups, cold had 214 significant cumulative adverse effect on all groups (   shown that cold has higher AF than heat. For example, Hang Fu et al (Fu et al. 2018) found that 240 cold temperatures contributed to higher AF of mortality than heat temperatures in India. In 241 addition, a multicountry study showed that more temperature-attributable fraction was caused by 242 cold (7·29%, 7·02-7·49) than by heat values (0·42%, 0·39-0·44) (Gasparrini et al. 2015). The cold 243 effect might be explained by some mechanisms leading to cardiovascular and respiratory diseases.   In our study, elderlies tended to be at higher risk for cold compared to heat values and young 262 people. The higher risk of elderly to cold may be due to chronic diseases and the weakened blood 263 circulation (Cui et al. 2019). We also showed that both men and women were high risk for cold 264 values. However, women were slightly more vulnerable to cold compared to men. This result is 265 similar to several previous evidences. For example, a study conducted in Korea found that women  Availability of data and materials 315 The dataset analysed during the current study are not publicly available due to ethical concerns 316 but are available from the corresponding author on reasonable request. Consent to participate 333 The data from hospital was collected with no personal identification. Relative risk (RR) of hospital admissions due to extreme temperature in different lags by different sex and age groups; Percentile 99th was compared to reference value (OT). The plots a, b, c, d, e, f provides the RR for total, male, females, young and elderlies HAs, respectively. Relative risk (RR) of hospital admissions due to extreme temperature in different lags by different sex and age groups; Percentile 1th was compared to reference value(OT). The plots a, b, c, d, e, f provides the RR for total, male, females, young and elderlies HAs, respectively.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download.