Outdoor temperature is associated with serum HDL and LDL

Abstract

Background

While exposures to high and low air temperatures are associated with cardiovascular mortality, the underlying mechanisms are poorly understood. The risk factors for cardiovascular disease include high levels of total cholesterol and low-density lipoprotein (LDL), and low levels of high-density lipoprotein (HDL). We investigated whether temperature was associated with changes in circulating lipid levels, and whether this might explain part of the association with increased cardiovascular events.

Methods

The study cohort consisted of 478 men in the greater Boston area with a mean age of 74.2 years. They visited the clinic every 3–5 years between 1995–2008 for physical examination and to complete questionnaires. We excluded from analyses all men taking statin medication and all days with missing data, resulting in a total of 862 visits. Associations between three temperature variables (ambient, apparent, and dew point temperature) and serum lipid levels (total cholesterol, HDL, LDL, and triglycerides) were studied with linear mixed models that included possible confounders such as air pollution and a random intercept for each subject.

Results

We found that HDL decreased −1.76% (95% CI: −3.17 – −0.32, lag 2 days), and −5.58% (95% CI: −8.87 – −2.16, moving average of 4 weeks) for each 5°C increase in mean ambient temperature. For the same increase in mean ambient temperature, LDL increased by 1.74% (95% CI: 0.07 – 3.44, lag 1 day) and 1.87% (95% CI: 0.14 – 3.63, lag 2 days). These results were also similar for apparent and dew point temperatures. No changes were found in total cholesterol or triglycerides in relation to temperature increase.

Conclusions

Changes in HDL and LDL levels associated with an increase in ambient temperature may be among the underlying mechanisms of temperature-related cardiovascular mortality.

1. Introduction

The health effects of outdoor temperature have been under vigorous investigation for the past decade. In addition to interest in the current effects of temperature on health, there is growing concern that climate change will increase the incidence of temperature extremes (Solomon et al., 2007). Historical and recent studies have shown that exposure to hot and cold temperatures increases mortality (Gover M, 1938; McMichael et al., 2008; Medina-Ramon and Schwartz, 2007; Ostro et al., 2009), especially from cardiovascular diseases (Basu and Ostro, 2008; Medina-Ramon et al., 2006). However, the mechanisms underlying the cardiovascular deaths related to temperature changes are not well established.

The development of cardiovascular disease is a multi-factorial process with many pathways including increased levels of cholesterol, inflammation markers, and blood pressure. Studies on the relation between temperature and blood pressure have reported inverse associations (Alperovitch et al., 2009; Barnett et al., 2007; Halonen et al., 2010b) and provided no evidence that increased blood pressure is a mechanism for heat-related cardiovascular deaths. On the other hand, a few studies have shown that a decrease in temperature increases the levels of inflammation markers (Halonen et al., 2010a; Schneider et al., 2008), suggesting that inflammation was related to mortality at low, but not at high temperatures.

Elevated level of total cholesterol is a predictor of coronary heart disease mortality in men (Houterman et al., 2000). High levels of low-density lipoprotein (LDL) and triglycerides are known risk factors for atherosclerosis (Farmer and Torre-Amione, 2002; Genoud et al., 2008), and a link between LDL and myocardial infarction among men has also been recently established (Noda et al., 2010). In addition, low levels of high-density lipoprotein (HDL, which is often considered as “good cholesterol”) have been found to be another independent risk factor for heart diseases (Despres et al., 2000). One reason why cholesterol levels might be influenced by changes in temperature is the possibility that cholesterol regulates the survival sensitivity of mammalian cells to high temperatures (Cress and Gerner, 1980). The mechanism for this phenomenon could be the ability of cells to take up more cholesterol at higher temperatures, which would lead to a decrease in circulating cholesterol. However, the possible link between temperature and serum lipid levels has rarely been investigated (Keatinge et al., 1986; Neild et al., 1994; Yamamoto et al., 2003), especially on a population level (Donaldson et al., 1997).

We therefore investigated the associations between three measures of outdoor temperature (ambient, apparent, and dew point temperature) and levels of cholesterol on a range of timescales among a cohort of men who did not need medication for lowering cholesterol. We studied separately the changes in total cholesterol and its fractions, HDL, LDL, and triglycerides in association with an increase in the three temperature measures.

2. Material and Methods

2.1. Study Population

Our study population consisted of a cohort of men recruited into the Normative Aging Study which was established by the Veterans Administration in 1963 (Bell, 1972). Men living in the Greater Boston area were aged 21–80 years and healthy when entering the study. They were asked to return to the clinic every three to five years for a physical examination and to complete questionnaires on smoking, medical conditions, medication use and alcohol consumption. During each visit anthropometric data were collected, and the participants were interviewed to confirm the identity and purpose of medications used, and to report the diagnoses of new diseases. Informed consent was obtained from all participants at each study examination. This study has been approved by the Institutional Review Boards of the Harvard School of Public Health and the Veterans Affairs Boston Healthcare System.

Physical examinations by a physician took place after overnight fasting and overnight abstinence from smoking. Blood samples for the lipid analyses were drawn by a laboratory technician, and the samples were analyzed at the clinical laboratory of the Veterans Affairs Boston Healthcare System. Because of the long follow-up, 1995–2008, three different analyses were used for obtaining the measures of cholesterol levels: for total cholesterol, HDL and triglyceride analyses before November 2000, the laboratory used the BM/Hitachi 747–100 Automatic Analyzer (Roche Diagnostics Corporation, formerly Boehringer Mannheim Corp., IN, USA); from November 2000 to December 2006 the laboratory used the Olympus AU640/AU400 Chemistry Analyzer (Olympus America Inc., PA, USA); and from January 2006 to the present the laboratory has used the Cholesterol, HDL, and Triglyceride assays of Abbott Architect (Abbott Diagnostics, IL, USA). The methodologies for these serum assays have been described previously (Bucolo and David, 1973).

Using the levels of total cholesterol, triglycerides and HDL obtained from the laboratory analyses, we calculated the levels of LDL cholesterol using the Friedewald equation (Friedewald et al., 1972) as follows:

$$\text{Low-density}\text{lipoprotein}=\text{Total}\text{cholesterol}-(\text{High-density}\text{lipoprotein}+\text{Triglycerides}/5)$$

Measurements with concentrations of triglycerides greater than 3.4mmol/l (300mg/dl) were excluded (n=48) because the reliability of the LDL cholesterol estimations decrease considerably with increasing triglyceride concentrations (Warnick et al., 1990).

Because traffic-related air pollution is associated with cardiovascular events (Brook et al., 2010; Grahame and Schlesinger, 2010), we restricted our analysis to visits from 1995 on, when we had black carbon (BC) and fine particle (aerodynamic diameter <2.5μm, PM_2.5) measurements. After indentifying missing data and excluding subjects reporting the use of statin medication, we had data on 862 clinic visits for 478 subjects.

2.2. Environmental Measurements

The ambient temperature (°C), dew point temperature (°C), and relative humidity (%) measures were obtained daily from Boston Airport weather station, located eight km from the clinic. Because we wanted to identify the optimal exposure metric, we used measures of ambient, apparent and dew point temperature in the analyses. Dew point temperature is the temperature when condensation of water vapor first begins. It is closely related to relative humidity, and when relative humidity reaches 100% dew point temperature equals the ambient temperature. Apparent temperature is an exposure variable used to describe how people perceive the combination of temperature and humidity (Steadman, 1984). The values for apparent temperature were based on the measures of ambient and dew point temperature and calculated using a formula described by Steadman (Steadman, 1979):

$$-2.653+[0.994\ast 24\text{-hour}\text{mean}\text{air}\text{temperature}(°\mathrm{C})]+\text{[}0.0153\ast 24\text{-hour}\text{mean}\text{dew}\text{point}\text{temperature}{(°\mathrm{C})}^2]$$

The participants lived an average of 18 km from the Boston monitoring site. We used only one monitoring site for the temperature measures, but the correlation between 24-hour mean ambient temperatures at Boston Airport (mean 12.7°C, standard deviation 8.6) and at T.F Green Airport (Warwick, RI) airport (mean 13.1°C, standard deviation 8.6), which is closer to the study subjects who lived the furthest from Boston, was 0.96. The airports are 100 km apart from each other, but the high correlation and the same standard deviations suggest that a single monitoring site is adequate to capture temporal variability in the temperature exposure in our study participants.

Continuous BC and PM_2.5 measurements were obtained during the time period from January 1995 to December 2008. Black carbon and PM_2.5 measurements were performed at the Harvard School of Public Health, located one km from the clinic. Black carbon was measured by using an aethalometer (Magee Scientific, Berkeley, CA), and PM_2.5 by using the Model 1400A Tapered Element Oscillating Microbalance (TEOM) (Rupprecht and Patashnick, Albany, NY, USA). Ambient ozone (O₃) levels were measured continuously at four monitoring sites in the Greater Boston area (in the cities of Boston, Chelsea, Lynn, and Waltham) all of which conformed to US Environmental Protection Agency (EPA) standards. We used the average of the four ozone measurements in our analyses.

2.3. Statistical Analyses

Because the distribution of the lipid concentrations were not quite normal, we log₁₀ –transformed the lipid concentrations. In addition, because we had repeated outcome measurements from the same participants, we used linear mixed models with random effects for each subject for the statistical analyses. This method takes into account the correlation of repeated measurements within each subject by use of a random intercept that is fitted for each subject; therefore, differences across subjects are controlled for and the estimates of associations are effectively from within-subject differences (Pinheiro and Bates, 2000). We do note that the estimated random intercepts can be noisily estimated if the participants had only one measurement; however, the inferences for the fixed effects, such as temperatures, are appropriate in these models. Even though this method is used especially for dependent measures, mixed models do not require repeated measures for all subjects, and having a few measures per subject can only cause some loss of power for the analyses.

Several possible confounders, potentially affecting blood lipid levels, were included in the models. We used a dummy variable for weekday, to capture day of the week effects. Fluctuation of the lipid levels over the year, which peak in the winter and fall in the summer (Al-Tamer et al., 2008), suggests a sinusoidal pattern and therefore we used sine and cosine terms to capture seasonality. A sinusoidal wave is uniquely defined by its amplitude, its frequency (set to one complete cycle per year), and its phase (the time of the year when the peak lipid concentration occurs), and can be written as:

$$\text{Asine}(2\mathrm{\pi }\text{fdate}+\mathrm{\varphi })$$

where A is the amplitude, f the frequency, and ϕ the phase (Sterling, 2005). Using the elementary trigonometric identity:

$$\text{sine}(\mathrm{a}+\mathrm{b})=\text{sine}(\mathrm{a})\text{cosine}(\mathrm{b})+\text{cosine}(\mathrm{a})\text{sine}(\mathrm{b}),$$

we can see that the first formula can be written as:

$$\mathrm{\alpha }\text{sine}(2\mathrm{\pi }\text{fdate})+\mathrm{\beta }\text{cosine}(2\mathrm{\pi }\text{fdate}),$$

where α=Acosine(ϕ) and β=Asine(ϕ).

Hence, including both a sine and a cosine term for date in the model will estimate the amplitude and phase of the sinusoidal wave. We set the frequency to one cycle per 365.24 days. These trigonometric terms have been used in a few epidemiological studies previously (Schwartz et al., 1996).

We also controlled for personal characteristics that could change with time: age, smoking status (never/former/current), use of alcohol for ≥2 drinks/day (yes/no), diabetes and hypertension (yes/no), body mass index (BMI), level of fasting blood glucose, and years of education. In addition, we controlled for race. In the models for ambient temperature, relative humidity was included as a possible confounder, because humidity was incorporated into the other exposure metrics of apparent and dew point temperature and was therefore not separately used in those models.

We chose a priori to study the effects of all temperature variables on single lag days 0 to 7 and the effects of cumulative exposures over 1, 2, 3 and 4 weeks. Cumulative exposure of one week included temperatures of lag days 0 to 6 prior to the clinic visit, cumulative exposure of 2 weeks included lag days 0 to 13 prior to the visit, etc. Using these lags for each temperature variable, and the co-variables described above, we ran separate models for each of the temperature measures as primary analyses.

Cholesterol levels are often elevated among the obese (Grundy, 1998), and it has been suggested that cholesterol levels vary by season (Al-Tamer et al., 2008; Ockene et al., 2004). We therefore investigated possible effect modification by season (warm: May-September - cold: August-April), and obesity (BMI>30 kg/m²) as secondary analyses, by adding an interaction term for ambient temperature and season or obesity to the models used for the primary analyses. We also studied possible effect modification and confounding by traffic-related air pollution and ozone by adjusting the models for BC, PM_2.5 or O₃. Traffic pollution is a known risk factor for cardiovascular morbidity and mortality (Grahame and Schlesinger, 2010; Perez et al., 2009; Zanobetti and Schwartz, 2006), and a possible risk factor for increased lipid levels (Tomao et al., 2002). There is also some evidence that ozone has adverse effects on cardiovascular health (Middleton et al., 2008). Possible confounding by pollutants was investigated by using the same day measure of temperature and pollutant concentration in the models, e.g. for the current day effect: temperature at lag 0 and BC at lag 0 were included in the model, for the previous day effect: temperature at lag 1 and BC at lag 1 were included in the model, etc.

As sensitivity analyses, we ran 1) models excluding the hottest and the coldest 2.5% of temperatures, and 2) models including a variable to control for possible confounding by serum albumin level. Albumins are blood plasma proteins that regulate blood volume by controlling oncotic pressure. They are involved in cholesterol transfer between cells and lipoproteins (Zhao and Marcel, 1996) and therefore possibly have an effect on the lipid levels. As a post hoc analysis, we also ran an HDL model that included two different lag terms for temperature, in order to find out if the short-term and cumulative effects were independent of each other. In this analysis, ambient temperature measures for lag 2 and for the 4-week cumulative exposure were included in the same model. In this way we obtained effect estimates for both acute and long-term exposures while controlling for the other. Finally, we tested the linearity of the association between temperature and lipid levels using penalized spline models in the generalized additive mixed model framework. All analyses were performed with statistical software R 2.10.1 (R Development Core Team, 2010), and all results are presented as a percentage change for a 5°C increase in temperature with 95% confidence interval (CI).

3. Results

A summary of the environmental variables is provided in Table 1. The study population was on average 74.2 years old (range 55–100 years). Other characteristics of the study population and the levels of serum lipids are provided in Table 2.

Table 1.

Summary of the environmental variables.

Variable	Min	25th percentile	Mean	75th percentile	Max	Correlation with ambient tempa
Ambient Temperature (°C)	−13.9	6.2	12.7	20.1	31.2	-
Apparent Temperature (°C)	−8.22	3.9	12.1	19.8	35.9	0.99
Dew Point Temperature (°C)	−28.1	0.16	6.2	14.3	23.0	0.92
Relative Humidity (%)	26.6	55.9	67.7	81.0	99.3	0.09
Black carbon (μg/m3)	0.12	0.52	0.84	1.1	2.68	0.34
Ozone (ppb)	0.003	0.01	0.02	0.03	0.09	0.50

^aSpearman Rank Correlation with ambient temperature

Table 2.

Descriptive statistics of the individual characteristics.

Variable	Min	25th percentile	Mean (sd)	75th percentile	Max
Total cholesterol (mmol/l)	2.59	4.60	4.85 (0.98)	5.79	8.40
High-Density Lipoprotein, HDL (mmol/l)	0.46	1.06	1.28 (0.33)	1.55	6.41
Low-Density Lipoprotein, LDL (mmol/l)	1.09	2.75	2.93 (0.84)	3.77	6.41
Triglycerides (mmol/l)	0.27	0.87	1.40 (0.63)	1.65	3.37
Albumin (g/dl)	3.10	4.10	4.33 (0.33)	4.60	5.40
Age (years)	55	69	74.2 (6.7)	79	100
Body mass index (kg/m2)	17.2	25.1	28.1 (4.2)	29.8	52.6
Education (years)	6.0	12.0	14.6 (2.8)	16.0	25.0
	N subjects (%)
Obese (BMI >30)	124 (25.9)
Smoking status
Never	150 (31.3)
Former	300 (62.7)
Current	28 (5.8)
Hypertension	344 (71.9)
Diabetes	52 (10.9)
Race
Non-Hispanic white	458 (95.8)
Non-Hispanic black	10 (2.1)
Hispanic white	7 (1.4)
Hispanic black	3 (0.6)
	N observations (%)
Weekday of the visit
Monday	40 (4.6)
Tuesday	203 (23.5)
Wednesday	470 (54.5)
Thursday	149 (17.3)
Alcohol intake (≥2 drinks/day)	168 (19.5)

We observed negative associations between a 5°C increase in ambient temperature and the level of HDL throughout all lags (Figure 1), with significant findings for 2 days lag and cumulative exposures (Table 3). For LDL, we found positive associations with ambient temperature (Figure 1), with the strongest associations at lag days 1 and 2 (Table 3). The results for HDL and LDL only slightly changed when the extreme temperatures were excluded from the analyses (Table 3).

Figure 1.

The percentage change in high- and low-density lipoproteins for a 5°C increase in mean ambient temperature.

Model adjusted for relative humidity, season, weekday, age, smoking, hypertension, diabetes, body mass index, alcohol use (≥2 drinks/day), years of education and race

Table 3.

The percentage change in high-density lipoprotein and low-density lipoprotein levels for a 5°C increase in ambient temperature among men not taking statin medication. Results for basic models and models excluding extreme temperatures.

	Basic model a			Extreme Temperatures a
	% Change	95% CI		%Change	95% CI
High-Density Lipoprotein
Lag 0	−0.74	−2.07	0.60	−0.42	−1.94	1.13
Lag 1	−1.27	−2.64	0.12	−1.47	−3.06	0.14
Lag 2	−1.76	−3.17	−0.32	−1.64	−3.18	−0.07
Lag 3	−1.08	−2.33	0.19	−0.72	−2.12	0.70
1 week	−2.47	−4.44	−0.47	−1.92	−4.22	0.43
2 weeks	−4.50	−6.93	−2.00	−4.27	−7.13	−1.32
3 weeks	−5.16	−8.03	−2.21	−5.12	−8.40	−1.73
4 weeks	−5.58	−8.87	−2.16	−6.02	−9.76	−2.14
Low-Density Lipoprotein
Lag 0	0.23	−1.36	1.84	0.64	−1.18	2.51
Lag 1	1.74	0.07	3.44	1.15	−0.73	3.07
Lag 2	1.87	0.14	3.63	1.92	0.05	3.83
Lag 3	1.50	−0.03	3.05	1.11	−0.56	2.80
1 week	2.27	−0.19	4.79	2.79	−0.01	5.66
2 weeks	0.65	−2.43	3.83	1.72	−1.82	5.39
3 weeks	0.31	−3.33	4.09	1.34	−2.77	5.63
4 weeks	1.75	−2.50	6.18	2.24	−2.53	7.24

^aModel adjusted for relative humidity, season, weekday, age, smoking, hypertension, diabetes, body mass index, race, alcohol use (≥2 drinks/day), and years of education

We found no associations between ambient temperature and total cholesterol or triglycerides (Online supplement Table 1). The results for apparent and dew point temperature were similar to those for ambient temperature (Online supplement Figures 1 and 2), and we therefore used only ambient temperature for the further analyses.

In the secondary analyses we found that BC was a confounder for the association between ambient temperature and HDL, but not for the associations between ambient temperature and LDL. After adjusting for BC, the effect estimates for HDL became slightly greater than in those in Table 3, but the effect estimates for LDL were virtually unchanged (Table 4). No significant interaction was found between ambient temperature and BC in the HDL or LDL models. Ozone also confounded the association between ambient temperature and HDL: the effect estimates for nearly all associations became slightly greater, but only the associations for the 2- and 3-week cumulative averages remained statistically significant (Table 4). The confounding effect of O₃ was greater in the model for LDL, and no significant associations were observed in the O₃ adjusted analysis (Table 4). The interaction term for the continuous variables for ambient temperature and O₃ was positive and significant (p=0.039) in the HDL model, but non-significant in the LDL model. Adjusting the HDL model for fine particles had little effect on the results. In the LDL model, the associations became greater after adjusting for PM_2.5 and significant effect estimates were observed for lags 1 to 3 and for the one-week cumulative average (Table 4). Interaction terms between ambient temperature and PM_2.5 in the HDL or LDL models were not significant.

Table 4.

The percentage change in high-density lipoprotein and low-density lipoprotein levels for a 5°C increase in ambient temperature among men not taking statin medication when adjusting for black carbon, ozone and PM_2.5.

	Adjusted for BC a			Adjusted for O3b			Adjusted for PM2.5c
	% Change	95% CI		% Change	95% CI		% Change	95% CI
High-Density Lipoprotein
Lag 0	−1.24	−2.63	0.16	−2.14	−4.40	0.17	−1.26	−2.70	0.21
Lag 1	−1.48	−2.88	−0.05	−1.88	−4.25	0.54	−1.34	−2.85	0.18
Lag 2	−2.11	−3.55	−0.66	−2.22	−4.66	0.28	−1.83	−3.37	−0.27
Lag 3	−1.06	−2.34	0.23	−1.45	−3.65	0.79	−1.37	−2.76	0.05
1 week	−3.64	−5.68	−1.55	−2.73	−6.13	0.80	−2.97	−5.18	−0.72
2 weeks	−5.39	−7.91	−2.81	−5.74	−9.91	−1.37	−4.13	−6.82	−1.35
3 weeks	−6.16	−9.11	−3.11	−5.41	−10.3	−0.29	−4.54	−7.69	−1.27
4 weeks	−6.73	−10.1	−3.25	−4.61	−10.2	1.22	−5.19	−8.71	−1.53
Low-Density Lipoprotein
Lag 0	0.26	−1.42	1.97	−0.29	−2.64	2.11	0.11	−1.68	1.93
Lag 1	2.04	0.32	3.79	0.72	−1.76	3.28	2.65	0.77	4.58
Lag 2	1.78	0.01	3.58	0.87	−1.70	3.51	2.29	0.36	4.26
Lag 3	1.33	−0.23	2.91	1.60	−0.71	3.96	1.89	0.13	3.67
1 week	1.88	−0.73	4.55	1.64	−1.98	5.39	3.03	0.17	5.97
2 weeks	0.01	−3.20	3.33	−0.22	−4.70	4.48	0.88	−2.59	4.47
3 weeks	−0.73	−4.49	3.18	−1.43	−6.55	3.96	−0.25	−4.27	3.94
4 weeks	0.83	−3.54	5.40	1.84	−4.12	8.16	1.42	−3.18	6.23

^aModel adjusted for relative humidity, season, weekday, age, smoking, hypertension, diabetes, body mass index, race, alcohol use (≥2 drinks/day), years of education and black carbon

^bModel adjusted for relative humidity, season, weekday, age, smoking, hypertension, diabetes, body mass index, race, alcohol use (≥2 drinks/day), years of education and ozone

^cModel adjusted for relative humidity, season, weekday, age, smoking, hypertension, diabetes, body mass index, race, alcohol use (≥2 drinks/day), years of education and PM_2.5

We found no significant interactions between ambient temperature and season or between ambient temperature and obesity. Adjusting the HDL and LDL models for serum albumin only slightly affected the results: HDL decreased −1.37% (95 CI, −2.73 – 0.01), and −4.64% (95% CI, −7.84 – −1.33), at lag 2 days and over the 4-week cumulative average, respectively, and LDL increased 2.04% (95% CI, 0.31 – 3.81) at lag 2 days.

In the HDL model with two different lag terms for temperature, the strength of the association at lag 2 decreased (−1.16%, 95% CI −2.65 – 0.35), but the association for the 4-week cumulative average changed only slightly (−4.47%, 95% CI −8.22 – −1.12). The correlation between temperatures at lag day 2 and the 4-week cumulative average was 0.89.

In examining the shape of the association between temperature and lipids, the generalized additive mixed model (gamm) chose the optimized degrees of freedom for temperature using generalized cross validation. The gamm models were adjusted for the same possible confounders as the model used for the primary analyses. As a result of the gamm analysis, we observed two and one degrees of freedom for ambient temperature at lag 2 days and over the 4-week moving average, respectively, in both HDL and LDL models, and the resultant curves of the associations between temperature and HDL and LDL appear linear (Online supplement Figure 3). These findings support linear relationships between ambient temperature and the lipid levels.

4. Discussion

We investigated the associations between outdoor temperature and serum lipid levels in a cohort of men who were not using cholesterol lowering medication. We found negative associations between ambient temperature and the level of HDL with a few days delay and cumulatively up to a month, and positive associations between temperature and the level of LDL with a few days delay. There were no associations between temperature and the levels of total cholesterol or triglycerides. The results were similar whether we used ambient, apparent or dew point temperature as the exposure variable.

We found a decrease in the level of HDL and increase in the level of LDL in association with each 5°C increase in ambient temperature with two days delay. The observed changes in the lipid levels are important because of their influence on the risk of coronary heart disease and other cardiovascular events (Despres et al., 2000; National Cholesterol Education Program, 2002; Schwartz, 2009). To our knowledge, the associations between outdoor temperature and the levels of HDL or LDL have not been previously reported in a population study. Results from chamber studies have been only partly consistent with our findings. A study in London showed that six hours of exposure to a high temperature (41°C) increased the levels of HDL and LDL (Keatinge et al., 1986), but a more recent study reported a decrease in LDL levels after one hour of exposure to a high temperature (39.8 °C) (Yamamoto et al., 2003). These differences in findings may be partly due to the fact that cholesterol decreases with age, and while these two studies included younger participants, we studied a cohort of men aged 55 to 100 years (Ettinger et al., 1992; Ferrara et al., 1997). Additionally, the chamber studies examined extreme temperatures rather than the mild temperatures we evaluated. We also observed a cumulative negative association between temperature increase and the level of HDL that was stronger than, and independent of the short-term effect. This suggests that the longer term exposure to increasing temperatures may also have adverse effects on cardiovascular health, even though mortality related to temperature increases is often observed shortly after exposure (Basu and Samet, 2002; Zanobetti and Schwartz, 2008).

The association between cardiovascular mortality and temperature has not only been described to be acute, but also J- U- or V-shaped (Basu and Samet, 2002). Our finding, however, indicated that the associations between ambient temperature and HDL and between ambient temperature and LDL were linear. Linear inverse associations between temperature and inflammation markers have been reported (Halonen et al., 2010a; Schneider et al., 2008), which together with the current findings suggest that the mechanisms of cardiovascular morbidity and mortality due to exposure to cold and warm temperatures may be different. Changes in the levels of HDL and LDL cholesterol may be among the underlying mechanisms of cardiovascular mortality related to increases in ambient temperature, but not of mortality due to decreases in temperature.

We found no association between temperature and the level of total cholesterol, which is consistent with the results of a time-series study from London (Donaldson et al., 1997). However, in that study the effects of temperature on specified fractions of cholesterol were not examined, and our data which include HDL and LDL provide a more complete assessment of the temperature-lipid relationship.

Air pollution is a known risk factor of cardiovascular health (Brook et al., 2010), and exposure to traffic-related pollutants may also have an impact on the serum lipid levels (Tomao et al., 2002), though the evidence is limited. Particulate air pollution levels are often correlated with temperature (Gietl and Klemm, 2009; Hussein et al., 2005), which is why we controlled for the possible confounding by combustion-related and particulate ambient pollutants (BC, PM_2.5) in our analyses. We found that the association between temperature and HDL was slightly stronger when adjusted for BC, but slightly weaker when adjusted for PM_2.5. The effect of BC adjustment on the association between temperature and LDL was minor, but PM_2.5 did strengthen the association between temperature and LDL. These findings suggest that combustion-related and particulate air pollution should be considered as possible confounders when estimating the effects of outdoor temperature on cardiovascular health. However, we note that these analyses had limitations because we used only one measurement station to estimate the exposure to BC or PM_2.5, which may have led to exposure misclassification. Even though the exact exposure of the participants to these pollutants cannot be estimated, the single monitor does capture the general day-to-day variation in the levels of BC and PM_2.5. While there may be geographic differences, resulting in measurement error in BC and PM_2.5, for these to confound the effect of temperature, the day-to-day difference between BC/PM_2.5 at a given address and BC/PM_2.5 at the central monitor would have to be correlated with the day-to-day variations in temperature at the monitoring site. We believe such correlation is unlikely. The use of only one measurement station may lead to an underestimation of the health (or confounding) effects, because most of the measurement error would be so-called Berkson error, which reduces power to reveal significant effects (Zeger et al., 2000).

Ozone confounded the association between temperature and LDL and adjusting for it attenuated the effect of temperature. Adjusting the HDL model for ozone had little effect on the results; however, we found that ozone has the potential to modify the effect of temperature on HDL. The positive interaction between ozone (as a continuous variable) and temperature suggests greater effects of temperature on HDL in the presence of ozone, which is consistent with the findings of a study on temperature and cardiovascular mortality (Ren et al., 2008). However, the link between ozone exposure and serum lipid levels is not well studied. One recent investigation found no association between ambient ozone and cholesterol (Chuang et al., 2010), and the mechanisms through which ozone may confound the association between temperature and LDL, or modify the association between temperature and HDL requires further research. While our findings for the confounding effects of ozone in HDL and LDL models go in opposite directions, these results suggest that ozone should also be considered as a possible confounder and an effect modifier when estimating the health effects of temperature.

There are some limitations to our study. One is that the analytic sample included only men aged over 55 years, limiting the generalizability of the results. More studies using varieties of populations are needed to corroborate our findings. We also had only one temperature measurement site, thus measurements may not accurately reflect individual exposures. The high correlation (0.96) between temperatures at Boston and T.F. Green (Warwick, RI,) airports 100 km apart from each other suggests that possible exposure misclassification is non-differential. However, some exposure misclassification may be due to the lack of information on air conditioning use, a variable that has been found to modify the effects of temperature on mortality (Anderson and Bell, 2009). It is possible that inability to control for the use of air conditioning led to weaker than expected associations especially between higher temperatures and lipid levels. Future studies should consider personal temperature monitoring to better assess the health effects of temperature. Another source of misclassification may be the uncontrolled heat island effect, where dense urban locations with little vegetation get hotter, and cool off more slowly than the surrounding areas. The heat island effect intensifies the exposure to temperature within cities, and it has been found to increase heat-related mortality (Tan et al., 2010). However, as Boston is located by the Atlantic Ocean, winds from the ocean possibly make this effect less potent than in inland cities. Another limitation is that we report associations between temperature and the levels of serum lipids over timescales of days to weeks. It is generally long-term increases (on timescales of years) in cholesterol that have been associated with adverse cardiovascular outcomes, and the implications of short-term changes in cholesterol due to short-term changes in temperature are unclear. In vitro, it has been shown that HDL can inhibit platelet activation, and LDL, from persons with moderate hypercholesterolemia or high cholesterol, can increase platelet activity within an hour from the beginning of the experiment (Beitz et al., 1993). Platelet activation promotes arterial thrombosis and further increases the risk of myocardial infarction and ischemic stroke that are conditions that have been associated with temperature increase (Basu and Ostro, 2008; Dawson et al., 2008). Thus, the observed short-term changes in HDL and LDL levels have a hypothetical pathway to increase the risk of cardiovascular events.

In conclusion, this is the first population study to find that an increase in ambient temperature is associated with a decrease in the level of serum HDL and an increase the level of serum LDL among men. Both of these associations may be detrimental for cardiovascular health, and our results suggest that changes in HDL and LDL in association with an increase in temperature may be among the underlying mechanisms of temperature-related cardiovascular mortality.