Ambient air temperature and temperature variability affecting blood pressure—a repeated-measures study in Augsburg, Germany

Ambient air temperature and temperature variability are supposed to influence blood pressure (BP); however, findings are inconsistent. We examined the effects of short-term changes in ambient temperature and temperature variability on systolic BP (SBP), diastolic BP (DBP), mean arterial pressure (MAP), and pulse pressure (PP) in a repeated-measures study. Repeated BP measurements were available for 3184 participants from the German population-based Cooperative Health Research in the Region of Augsburg (KORA) S4 survey (1999–2001) and two follow-up examinations (2006–08 and 2013–14). Daily meteorological data were obtained from fixed measurement stations including air temperature and diurnal temperature range (DTR). We used confounder-adjusted additive mixed models to examine immediate (same-day, lag 0), delayed (lag 1 to lag 4), and cumulative (up to lag 0–13) exposure effects. Decreases in air temperature were associated with increases in SBP, DBP, and MAP, while we observed no effects for PP at all. For example, a 1 °C decrease in the 14-day moving average (lag 0–13) mean air temperature was associated with a 0.54% [95% confidence interval [95%CI]: 0.41%;0.68%] increase in SBP. Furthermore, decreasing DTR was linked to increasing SBP, DBP, and MAP measures. In the sensitivity analyses, results were found to be robust. Examination of exposure–response functions according to season revealed, that associations for summer and winter can be considered linear, while we detected non-linear functions in spring and autumn. Furthermore, exposure–response functions also differed in the three different surveys. As BP levels influence the risk of cardiovascular mortality, our results show the importance of considering temperature and its variation as potential risk factors. As ongoing climate change affects temperature variability, it is important to understand how the body adapts to changing ambient temperatures.


Introduction
Ambient air temperature is known to influence cardiovascular mortality. Low and high daily air temperature levels, as well as temperature variability, have been associated with an increased risk [1][2][3][4][5][6][7]. In particular, temperature variability might be an important meteorological indicator, considering that unstable weather patterns are predicted to occur more frequently in the future [6].

Covariate assessment
Anthropometric measures were taken at the KORA study center. Information about health status, smoking status, alcohol consumption, and medication was assessed by questionnaires. Waist-to-hip ratio (WHR): waist circumference [cm]/hip circumference [cm]. The definition regarding the intake of antihypertensive medication was taken from the recommendations of the German Society for Hypertension and Prevention [25]. The smoking status: regular; irregular; former; or never. Physical activity: very active: regularly > 2 h/week ; moderate active: regularly 1 h/week; little active: irregularly 1 h/week; non-active: no activity; during the whole year. Socioeconomic status: low: income per month <625€; medium-low: 625€ to <1250€; medium: 1250€ to <1875€; medium-high: 1875€ to <2500€; high: ⩾2500€.

Statistical analyses
For our analysis, we included all participants from the baseline examination with at least one follow-up visit. Descriptive analyses were performed for participant characteristics as well as for meteorological and air pollution data. Analysis of variance (ANOVA) was used to compare seasonal differences of SBP, DBP, MAP, and PP; Spearman's rank correlation coefficient was used to compare the different meteorological measurement sites as well as to calculate correlations between ambient temperature and air pollutants.
We used additive mixed models with random participant intercepts to assess the effect of ambient temperature (daily mean, minimum, maximum temperature, and DTR) on repeated measurements of SBP, DBP, MAP, and PP. The confounder model was chosen a priori and was identical for all outcomes. It included season, time trend (day of the year as counting variable), day of the week, relative humidity with the same lag as air temperature, barometric pressure with the same lag as air temperature, age, sex, WHR, physical activity, smoking status, alcohol consumption (g d −1 ), socioeconomic status, intake of antihypertensive medication (yes vs. no), and intake of statins (yes vs. no). Air temperature was included as a linear term. Time trend and relative humidity were included as penalized splines (P-splines). Immediate (current day, lag 0), delayed (lag 1 up to lag 4), and cumulative (mean of lags 0 and 1, mean of lags 0-4, mean of lags 0-6, and mean of lags 0-13) associations between air temperature and BP metrics were investigated.
As an additional analysis, we calculated two-exposure models, with mean air temperature and DTR as exposure variables. The confounder model was identical as for single exposure models.
All results were presented as %-change in the outcome mean and 95%-confidence intervals [95% CI] per 1 • C decrease in the respective temperature measure (supplement S1 and S2).
We assessed potential effect modification by including an interaction term between air temperature metrics and the effect modifier. The following effect modifiers were taken into account: (1)

Sensitivity analyses
To test the robustness of our results, we performed the following sensitivity analyses: (1) We visually checked the exposure-response functions between air temperature and BP metrics for deviations from linearity. This was done by replacing the linear air temperature term with penalized splines. Degrees of freedom were chosen according to the Aikaike information criterion (AIC). (2) We additionally inspected exposure-response functions according to season and for each survey separately. (3) We used distributed-lag linear models (DLM) [26] instead to assess the association between mean air temperature or DTR and BP (supplement S17). (4) Confounder selection by maximizing the adjusted R 2 . Season and day of the year were forced into each model. We included continuous confounders linearly or smoothly as P-splines depending on the R 2 value. In the case of smooth effects, degrees of freedom were chosen according to R 2 . (5) Instead of mean air temperature, we included mean apparent temperature as the exposure variable. (6) We additionally adjusted for PM 2.5 , UFP, NO 2 , and ozone with the same lag as the analyzed temperature lag. (7) As BP levels are lower after treatment with antihypertensive drugs, we considered the effect of antihypertensive medication [27]. Hence, we artificially increased BP levels in treated individuals. First, raw residuals were calculated by subtracting the mean BP from the observed BP. In treated individuals, the raw residuals were adjusted by calculating an average of the original value and all larger residuals. The treatment-adjusted BP levels were then calculated by the observed value minus the raw residual plus the adjusted residual. BP levels of untreated individuals remained unchanged. Afterward, we re-calculated the air temperature effects using the treatment-adjusted SBP, DBP, MAP, and PP levels. (8) Temperature effects might depend on the average temperature of the previous weeks. To capture this effect, we examined the temperature excess by subtracting the daily mean temperature from the mean temperature of the previous 14 d.

Study population and exposure measures
For our analyses, we included 3184 individuals from baseline survey S4, 3079 participants from the first follow-up F4, and 2279 from the second follow-up FF4 with no missing values in BP measurements (S4-F4-FF4: 2150 participants with both follow-up visits; S4-F4: 3065 participants with only the first follow-up visit; S4-FF4: 2269 participants with only the second follow-up visit). Table 1 shows the characteristics of the study population. SBP, DBP, MAP, and PP did not significantly differ during seasons (ANOVA in supplement S3). Furthermore, mean BP levels decreased from wave to wave, with the highest levels being measured in baseline examination S4. A possible explanation for this finding might be that healthier participants were more likely to participate in the later follow-up visits.
Descriptive statistics of meteorological variables, air pollution measurements, and the corresponding correlation coefficients are provided in tables 2 and S4. Over the whole study period, we observed a mean daily mean temperature of 8.5 • C, and a mean DTR of 9.6 • C. Daily mean temperature and DTR were moderately positively correlated (Spearman rank correlation coefficient = 0.51), and correlations between temperature variables and relative humidity were all negative (weak to moderate). Figure 1 presents the associations between mean air temperature and SBP, DBP, MAP, and PP (a corresponding table is provided in supplement table S3; minimum and maximum temperature effects are provided in table S4). A decrease in daily mean air temperature was associated with increased SBP, DBP, and MAP. We observed immediate, delayed and cumulative effects. The strongest temperature effects were observed for the 14 day moving average (lag 0-13)-a 1 • C decrease was associated with a 0.54% [95%CI: 0.41%;0.68%] increase in SBP. For DBP and MAP, we observed a 0.42% [95% CI: 0.30%;0.55%] and 0.48% [95% CI: 0.35%;0.60%] increase, respectively.
For PP, we did not observe any significant effects at all. In the two-exposure models (figure 3, table S8), the associations observed in the single-exposure models remained nearly unchanged. An immediate, delayed and cumulative mutual exposure to mean air temperature and DTR was associated with a significant increase in SBP, DBP, and MAP.

Effect modification of temperature effects
Since we observed the strongest temperature effects for lag 0-13, effect modifications were calculated for this cumulative lag (figure 4), and for comparison for cumulative lag 0-1. Supplement tables S9, S10 and figure S11 present the temperature effects (mean temperature and DTR) on SBP, DBP, and MAP by season, sex, WHR, antihypertensive medication, age, and smoking status. Temperature effects on SBP, DBP, and MAP were modified by season. We observed significant effects of mean temperature during summer and winter compared to spring and autumn. For DTR, the interactive effect for winter was significant compared to other seasons. Intake of antihypertensive medication only modified results for DTR, while sex, WHR, age, or smoking did not modify the results.

Sensitivity analyses
Results of the sensitivity analyses are shown in figure 4 and the supplement (figures S12-S17, tables S16 and S18). Our results mainly seem to be robust; however, we found significant changes in the observed associations when examining season-specific effects.
Based on the visual inspection of the overall exposure-response functions using smooth terms for air temperature (figure S12), we concluded that exposure-response functions could be considered linear. However, the linearity assumption is affected by season (figures 5 and S13), particularly for daily mean temperature. While associations for summer and winter can be considered linear, we detected non-linear exposure-response functions for spring and autumn (figure 5). The non-linear exposure-response functions during spring and autumn limits the overall interpretability of temperature effects in 3.2., since in this analysis, temperature effects were considered linear. The visual inspection of the exposure-response functions according to the different surveys revealed that associations differed for DTR in F4, compared to S4 and FF4 (figure S15). When we used DLM to assess the association between temperature and BP, we observed significant effects for all exposure-outcome combinations except lag 0-1 for SBP and MAP (supplement S17). After selecting the confounder model for SBP, DBP, and MAP, air temperature effects remained nearly unchanged (table S18). The effects for using apparent temperature instead of air temperature as an exposure variable were not significant anymore. Adjustment for air pollution and inclusion of treatment effects of antihypertensive medication did not change the observed temperature effects. When examining temperature excess from the previous 14 d mean temperature, we could show that a 1 • C decrease in temperature excess is associated with a significant decrease in SBP and MAP, lags 2, 3, and 4. No significant associations were detected for DBP.
mean arterial pressure (MAP) and pulse pressure (PP) associated with a 1 • C decrease in mean ambient air temperature. All models were adjusted for season, time trend, day of the week, relative humidity with the same lag as temperature, barometric pressure with the same lag as temperature, age, sex, waist-to-hip ratio, physical activity, smoking status, alcohol consumption, socioeconomic status, use of antihypertensive medication, and use of statins.

Discussion
This repeated-measures study investigated the association between ambient air temperature or temperature variability and different BP metrics. A decrease in mean air temperature was associated with an increase in SBP, DBP, and MAP, whereas we did not see any significant changes in PP. Regarding temperature variability, we saw increased SBP, DBP, and MAP for a decreasing DTR. The inverse association between air temperature and SBP or DBP found in our study is consistent with previous studies and has already been reported in detail [12][13][14][15][16]. We observed the inverse associations for all lags with the strongest effects for the 14 d moving average temperature. Whereas most previous studies of Figure 2. Percent change and 95% confidence intervals in (A) systolic blood pressure (SBP), diastolic blood pressure (DBP), (B) mean arterial pressure (MAP) and pulse pressure (PP) associated with a 1 • C decrease in mean DTR. All models were adjusted for season, time trend, day of the week, relative humidity with the same lag as temperature, barometric pressure with the same lag as temperature, age, sex, waist-to-hip ratio, physical activity, smoking status, alcohol consumption, socioeconomic status, use of antihypertensive medication, and use of statins. Figure 3. Two-exposure model: percent change and 95% confidence intervals in (A) systolic blood pressure (SBP), diastolic blood pressure (DBP), (B) mean arterial pressure (MAP) and pulse pressure (PP) associated with a 1 • C decrease in mean air temperature and DTR in a two-exposure model. All models were adjusted for season, time trend, day of the week, relative humidity with the same lag as temperature, barometric pressure with the same lag as temperature, age, sex, waist-to-hip ratio, physical activity, smoking status, alcohol consumption, socioeconomic status, use of antihypertensive medication, and use of statins. Temp: mean air temperature. DTR: diurnal temperature range. December)), sex (female vs. male), waist-to-hip ratio WHR (females: normal < 1 vs. high ⩾ 1; males: normal < 0.85 vs. high ⩾ 0.85), (B) antihypertensive medication intake (yes vs. no), age (<60 years vs. 60-74 years vs. >74 years), smoking status (regular vs. irregular vs. former vs. never). All models were adjusted for season, time trend, day of the week, relative humidity with the same lag as temperature, barometric pressure with the same lag as temperature, age, sex, waist-to-hip ratio, physical activity, smoking status, alcohol consumption, socioeconomic status, use of antihypertensive medication, and use of statins. The respective effect modifier was excluded from adjustment. acute temperature effects and BP have examined shorter periods of up to 7 d, we provide evidence that the observed effects are even more pronounced when more extended periods of up to 14 d are considered.
Decreasing temperatures activate the sympathetic nervous system, leading to increased heart rates and endothelial-mediated higher vascular resistance, both resulting in increasing BP [28,29].
As MAP is a predictor for cardiovascular disease in general [30,31] and is considered a relevant contributor to increasing ischemic stroke risk [32], it is important also to take temperature effects on MAP into account. In a web-based cohort, increasing MAP measurements were found per a decrease in hourly ambient temperature [16]. Similar results were reported in a study with hourly measurements [33]. This association can be explained due to the compound structure of MAP and similar effects of temperature on SBP and DBP.
There are a few studies investigating the influence of temperature variability represented by DTR on cardiovascular mortality [2,[4][5][6][7]34], but little is known about how temperature variability affects BP. A longitudinal study did not find significant associations between DTR and SBP or DBP [18]. A Chinese study showed an increase in SBP and PP for an increasing DTR. However, this study also found an increased DBP for a decreasing DTR [19]. This is only partly consistent with our findings, as we saw a decrease in DTR being associated with an increase in DBP, but also in SBP and MAP. Opposing to our results, a cohort study showed increasing BP measures being associated with an increasing DTR of up to 5 d and hourly temperature variation [20]. However, as we have looked at more extended periods of up to 14 d, with the effects being more pronounced at longer time lags, the results might be only partly comparable. Furthermore, as far as we know, this is the first study outside Asia to investigate this association.
Our effects significantly differed when including an interaction term for season in the effect modification analyses. We observed the strongest effects of mean temperature during summer and winter compared to other seasons, while DTR mainly affects BP during winter months. This result is reflected in our exposure-response functions for the different seasons, where we found a linear relationship for mean temperature only during summer and winter. Exposure-response functions during spring or autumn were W-shaped, indicating differing effects on BP in the respective sections of the curve. The non-linear associations detected subsequently also impair interpretability. This affects the comparison to winter and Figure 5. Exposure-response functions for the association between daily mean temperature in cumulative lag 0-13 with 95% confidence intervals, and changes in systolic blood pressure (SBP), diastolic blood pressure (DBP), and mean arterial pressure (MAP) on the y-axis, by different seasons. Daily mean temperature (Lag 0-13) was included as penalized spline with 6 degrees of freedom. All models were adjusted for time trend, day of the week, relative humidity with the same lag as temperature, barometric pressure with the same lag as temperature, age, sex, waist-to-hip ratio, physical activity, smoking status, alcohol consumption, socioeconomic status, use of antihypertensive medication, and use of statins.
summer, as well as the main associations, where temperature has been considered linear. Compared to other studies, Alperovitch et al found higher BP values during winter when examining outdoor temperature [12], the same applies to a large Chinese study [15], while Halonen et al did not find differences among seasons [13].
The seasonal temperature changes to which the body is exposed require appropriate acclimatization. An experimental study examining thermophysiological adaptions to mild heat found that SBP and DBP decrease in the course of the adaption process [35]. The opposite effect has been shown for cold adaption, which is associated with increasing BP measures [36]. Given the temperature instability occurring in spring and autumn according to the change between warm and cold seasons [37], this might explain the non-linear exposure-response curves we observed in our analysis.
Interestingly, we also found differing exposure-response functions for the three surveys. The negative linear exposure-response function for mean temperature detected during the baseline (S4, 1999-2001) and the first follow-up (F4, 2006-2008) examination is not present during the second follow-up (FF4, 2013-2014). However, effects estimates for the surveys did not significantly differ. Temporal variations regarding ambient temperature had already been described before for the Augsburg region when examining the effects on myocardial infarction [38]. Due to increasing mean daily temperatures, from 7.0 • C during S4 to 11.4 • C during FF4, effects for cold temperatures might play a less important role and could explain the altered exposure-response functions.
Anthropogenic climate change is causing a transformation in temperature variability and stability [21]. It is all the more important to improve the understanding of the corresponding adaptation of cardiovascular markers such as BP. Assessments of meteorological data over the past decades have shown a decrease in DTR, mainly due to a substantial increase in the nighttime minimum temperature [39]. A modeling study predicts a similar development for the coming years, but taking into account extensive regional differences in this respect [40]. This is particularly important considering our results that could show increasing BP levels for decreasing DTR. In the light of the inconsistency of results compared to other studies [18][19][20], and under reconsideration of the temporal variation in the observed exposure-response functions, further research is needed to better understand and characterize how this relates to BP levels.
When examining the temperature excess in the sensitivity analysis, our results show that higher excess temperatures are significantly associated with increasing BP measures. Rapid but recurring temperature changes affect the body by requiring an adaptive response each time. This leads to increased blood viscosity, an impaired immune system, and higher BP measures [41,42]. Given that, temperature variability is assumed to be an independent risk factor for increased mortality [6,7].
A strength of our study is that our data comprised 8542 observations from 3184 individuals in a time range of 15 y. Combined with a strong study design that accounted for repeated measures and included different exposure metrics and covariates, our results are supposed to reflect the exposure-response relationship very well. Furthermore, detailed information on participant characteristics was available, enabling appropriate adjustment for potential confounders. Finally, several sensitivity analyses were performed to test the robustness, including additional adjustments for air pollution and the intake of antihypertensive medication.
Several limitations should also be acknowledged. First, we used air temperature data from fixed monitoring sites rather than personal exposure measurements, which can cause exposure measurement errors. This non-differential error likely biases the estimates downward [43]. Second, this study was performed in a temperate climate; therefore, the results may not be relevant to other populations living in different climatic conditions. Furthermore, stress levels are an important contributing factor when examining BP. As we could not adjust our models accordingly, there is the possibility of unobserved bias. A further limitation that has to be acknowledged is the missing information about indoor temperature of the building, were BP measurements took place, which might result in a measurement error.

Conclusions
Although temperature effects on BP are well known, there has been little information concerning the effects of temperature variability. The present repeated-measures study found that BP metrics were associated with decreasing DTR. As this is the first study to examine this association outside Asia, it provides new evidence on the topic. The same applies to the so far rarely examined temperature effects on MAP. As our study investigates longer time periods of up to 14 d, the results indicate that acute effects of ambient temperature on BP last for a prolonged time-a fact that plays an important role, especially regarding preventive measures.
In the course of climate change, temperature variability is becoming increasingly important. In particular, the body's adaptive mechanisms need to be better understood. As far as BP is concerned, our study makes an important contribution to this issue.

Data availability statement
The data cannot be made publicly available upon publication because they contain sensitive personal information. The data that support the findings of this study are available upon reasonable request from the authors.