Seasonal Variations in Serum Cholesterol I.S. Ockene, etc

Transcription

1 Seasonal Variations in Serum Cholesterol I.S. Ockene, etc Introduction During the past half-century a number of small longitudinal and larger cross sectional studies have been published suggesting that cholesterol levels are higher in the fall and winter than in the spring and summer. [Thomas, 1961 #1; Fyfe, 1968 #2; Thelle, 1976 #3; Buxtorf, 1988 #5; Cucu, 1991 #6] The most striking of these studies suggested that in areas of extreme seasonal climatic variation, such as Finland, there may be as much as a 100mg/dl seasonal variation in serum cholesterol levels. [Keys, 1958 #7] A large scale longitudinal study of seasonal variation of plasma lipids carried out using the placebo group of the Lipid Research Clinics trial reported a mean seasonal change in plasma total cholesterol of 7.4 mg/dl, but the study was restricted to hyperlipidemic men with initial plasma total cholesterol levels averaging 274 mg/dl. [Gordon, 1988 #4] Råstam and colleagues reported a cross-sectional study of seasonal variation in plasma cholesterol levels drawn as part of the screening process for the Minnesota Heart Health Program. [Råstam, 1992 #8] Cholesterol levels were found to peak in January, with a statistically significant seasonal variation evident in men. In women there was a non-significant trend towards higher winter levels. Using the NCEP guideline for hyperlipidemia of >240 mg/dl, 25.4% of men were at or above this level in winter, whereas only 13.5% met this cutpoint in summer. Despite this accumulated evidence that seasonal variation of lipid levels may be important, such variation is not considered in the U.S. National Cholesterol Education Program guidelines, including the most recent update to those guidelines. [National Cholesterol Education Program, 1988 #10; Expert Panel on detection, 1993 #11] It is also rarely taken into account in the normal clinical management of hyperlipidemia, although the seasonal differences described above would result in large differentials in the frequency of patients being labeled as having high cholesterol at different times of the year. The described seasonal variation would also lead to a much lower percentage of patients reaching goal cholesterol levels if treatment were initiated in the summer and remeasurement carried out in the winter, than if the converse were true. In order to resolve these issues we carried out a longitudinal study of seasonal variation of lipids in healthy volunteer patients. We also measured all of those factors that were thought likely to be related to such variation. Methods The methodology of the SEASON study is fully described in a previously published paper, and is summarized here. [Merriam, 1999 #12] Further details of the analytic methodology can also be found on the study website at Subject selection Participants in the SEASONS study were recruited from the Fallon Healthcare System, a Health Maintenance Organization serving Worcester, Massachusetts and surrounding communities. Fallon Healthcare System members who were residents of Worcester County, aged 20 to 70 years, and who had telephone service were eligible to participate if they were not taking cholesterol-lowering medications and were not actively on lipid-lowering or weight-control diets, were free from possible causes of secondary hyperlipidemia (e.g. hypothyroidism, pregnancy), and were free of chronic life-threatening illness (cancer, or renal or heart failure). Subjects were recruited between December 1994 and February 1997 at a rate of approximately 25 subjects per month. Subjects were compensated for their participation in the study with a portable blood pressure monitor, blood lipid and dietary assessment results, and a monetary sum of $70. The

2 Institutional Review Boards of the Fallon Healthcare System and the University of Massachusetts Medical Center approved all subject recruitment and data collection procedures. Each subject signed an approved informed consent form prior to entering the study. Baseline data collection. Demographic data (e.g. age, gender, marital status, education, employment) and health habits (e.g. smoking) information were collected by self-administered questionnaire. Anthropometric data, including body mass (kg), height (m), and waist and hip circumferences (cm) were measured at the baseline clinic visit by the SEASONS staff. Blood Lipid Assessment Blood lipid values were assessed at baseline and then every three months to the one-year anniversary point within a three-week window on either side of the individual's quarterly appointment date (total of five assessments). Blood was collected from the antecubital vein after a 12-hour fast. Because there is known circadian variation of blood lipid levels, all blood draws were performed between 07:00 and 9:00 AM. Blood plasma was harvested by low-speed centrifugation at 4 C, aliquoted into individual tubes, and quickly frozen to - 70 C. On a regular basis, plasma samples were packed in dry ice and shipped for analysis via courier service to the Centers for Disease Control standardized laboratory at the University of Massachusetts at Lowell [Karge, 1989 #25]. Total cholesterol and triglyceride were measured in plasma by enzymatic methods using a Beckman System 700 autoanalyzer [Allain, 1974 #27; Bucolo, 1973 #24]. Highdensity lipoprotein (HDL) was measured in the resulting supernatant after heparin manganese precipitation of apo-b-containing proteins [Warnick, 1978 #26]. Low-density lipoprotein (LDL)- cholesterol values were calculated by the Friedewald equation using total cholesterol, triglyceride, and HDL in individuals with triglyceride values less than 400 mg dl-1 [Friedewald, 1972 #23] Nutritional data collection Because of its lower overall variance and reduced probability of bias in relation to other assessment methods, telephone-administered 24 HR recalls (24 HRs) were used as the definitive dietary assessment method in this study. [Posner, 1992 #15] Dietary intake was assessed by evening telephone interviews every 3 months. Three 24HRs were conducted on randomly selected days representing two week days and one weekend day during each trimester. All three 24HRs were administered within the 5-week period extending from two weeks prior to three weeks after the participant's quarterly clinic visit and were performed by registered dietitians who were trained to collect dietary data using our interview system. Participants also were asked questions on physical activity and light exposure during this interview. All dietary data for the 24HRs were entered and analyzed using the Nutrition Data System (NDS 2.3) of the University of Minnesota Nutrition Coordinating Center. [Buzzard, 1991 #16] Physical Activity data collection The physical activity information obtained during the 24HRs was derived from a standardized interview format in which the subject reported the amount of time they spent in light, moderate, vigorous, and very vigorous activities in each of household, occupational, and leisure-time activity domains. The 24 HRs were the primary assessment instrument used to measure seasonal change in activity. Data were entered directly into a computer using the Epiinfo data entry program. Estimates of physical activity energy expenditure (MET hours/day) were calculated using methods described in the compendium of physical activities developed by Ainsworth and colleagues. [Ainsworth, 1993 #18] We have compared the validity of this method with both the Baecke Physical Activity Questionnaire and the use of whole-body accelerometers and have found it to be quite comparable. [Baecke, 1982 #20; Matthews, #19]

3 Light exposure data collection The light exposure data obtained during the 24HR are derived from a standardized question format in which the day's activities are reviewed with the participant and directly entered on a computer. Collection of information about light exposure was coordinated with the portion of the recall interview dealing with physical activity. The day is divided into parts: 24:00 (midnight) to 8:00, 8:00 to 12:00 (noon), 12:00 to 16:00, and 16:00 to 24:00. The participants are asked to recall information about outdoor and indoor exposure to natural lighting as well as exposure to artificial lighting. The validity of the light-exposure questionnaire was confirmed in a group of volunteers by comparing the 24HR estimates of light exposure provided with those obtained from a subset of 80 subjects who wore an Actillume TM light recording device for the week prior to the 24HR recalls. [Matthews, 1995 #21] Analytical Methodology We fit mixed models that include sine and cosine terms as fixed effects to estimate seasonal effects over a year for TC, HDL, LDL, and TG. Separate models were fit to each component of serum cholesterol for males and females. and to each component of serum cholesterol. The models considered subjects as random effects. Regression coefficients for the sine and cosine terms were transformed to estimate the amplitude and phase of seasonal effects(reference: Koopmans, L.H. (1974) The spectral analysis of time series, Academic Press, New Work 366p.). [also A first order Taylor series expansion was used to construct estimates of the variance of the amplitude and phase from the variance estimates of the sine and cosine coefficients. Statistical significance was assessed at the =0.05 level via likelihood ratio tests based on comparisons with reduced models. Initial models were fit solely with sine and cosine terms for seasonal effects. Subsequent models were fit that controlled for various time dependent covariates including body mass index, diet, physical activity, light exposure, and outdoor light exposure. We first identified a set of statistically significant independent variables that were related to one or more serum cholesterol fraction. These variables correspond to body mass index, daily estimates of total Kcal intake, saturated fat as a percent of Kcal, total fat as a percent of Kcal, total waking MET, and total time spent in outdoor light. We included this set of co-variables in each mixed models to examine the extent to which seasonal variation in other factors explains seasonal changes in serum cholesterol. We evaluate the impact of including the covariates by evaluating the extent to which the estimated amplitude of the seasonal effect changes with and without control for covariates. Finally, we assess the practical impact of the seasonality by tabulating the gender specific distribution of total cholesterol based on the seasonal model fit without covariates after applying the estimated seasonal effect to each subject, using the best linear unbiased predictor for individual subject effects. Preliminary Modeling of Co-variables For most subjects in the Seasons Study, three 24-hour recall measures were made of dietary intake, physical activity and light exposure in each quarter. The measures were made via 24 recall telephone interviews on two weekdays and one weekend day in each quarter (according to the protocol). Interviewers were trained nutritionists. Different interviewers conducted the interviews, and the actual day of the week varied from quarter to quarter, and subject to subject. For modeling quarterly lipids or other quarterly variables, we first estimated average covariate valuesneed a summary estimate of the intake (or activity) for each subject during the quarter. Rather than using simple three-day averages as the estimates, we used best linear unbiased

4 predictors (BLUP) from mixed models to estimate longer term (28 day) There are several ways estimate such a summary measure. If there is no systematic variation between interviewers, or between days of the week, the simple average will be an unbiased estimate of intake (or activity). However, since longer term (such as 28 day) average intake (or activity) is more likely to be related to changes in lipids (as opposed to 3 day averages), a better estimate of the longer term activity (in the sense of having smaller MSE) is the best linear unbiased predictor (BLUP) from a mixed model (Ref. Stanek et al 1999), assuming that the longer term average covariate values had stronger relationships with lipids.. We use this strategy to estimate quarterly intake. The preliminary gender specific covarite models used Based on the 24-hour data to, we first predict average nutrition, physical activity and light exposure values for subjects in each quarter accounting for possible day of the week effects and interviewer effects. via gender specific mixed models. IWe considered interviewer and day of the week were modeled as fixed effects, withand subjects and quarters nested in subjects as random effects. In nearly all models, effects of day of the week and interviewer were statistically significant. We estimated quarterly values for subjects with best linear unbiased predictors from the mixed models (SAS, 1996). These predictors were used as time dependent co-variables in subsequent mixed models for serum cholesterol. In order to properly account for potential differences between cross-sectional and longitudinal effects of covariates on serum cholesterol, we included in the final models the average value of the covariate for each subject (based on the quarterly BLUP estimates) in addition to the quarterly estimates (see Palta et al (1997), Neuhaus and Kalbfleisch (1998)). Palta, M., Lin, C.Y., and Chao, W.H. (1997). "Effect of confounding and other misspecification in models for longitudinal data," in Lecture Notes in Statistics: Modelling Longitudinal and Spatially Correlated Data. T. G. Gregoire, D.R. Brillinger, P.J. Diggle, E. Russek-Cohen, W. G. Warren, and R.D. Wolfinger (editors). Springer-Verlag, New York, pp Neuhaus, J.M. and Kalbfleisch, J.D. (1998). Between- and within-cluster covariate effects in the analysis of clustered data, Biometrics 54: Stanek, E.J.III, Well, A., and Ockene, I. (1999). Why not routinely use best linear unbiased predictors (BLUPS) as estimates of cholesterol, Percent Kcal from Fat, and Physical Activity, Statistics in Medicine, 18: Results (I think we need a basic table that summarizes the data- probably by gender and quarter- what definition should we use?). Discussion Primary outcome: seasonal variation of total cholesterol, triglycerides, HDL, TC/HDL ratio, looking at averages and at subgroups (age, gender), and also by ranks (quartiles or quintiles). Looking at those with the greatest variation may be of especial value. We also should attempt to identify counterbalancing effects, e.g., subjects who gain weight in the summer and offset those who lose weight. Factors contribution to the variation: weight fluctuation, diet (total calories and % calories as fat and saturated fat), activity,?light exposure Importance of the effect comparison with prior studies, when to measure lipids, when to repeat,?economic consequences.

5

6 Tables Table 1: Lipids: means and quartile values (n = 331m; 310f) gender 25 th % 50 th % 75 th % mean TC m f HDL m f TG m f LDL m f Table 2. Summary of Model Fitting Statistics and Amplitude and Phase Estimates Cholesterol Subjects # Obs # Subj. -2Log(L)+ Subject Residual Mean Amplitude Phase++ Variance Variance TC Overall *** *** 358 TC Male *** *** 349 TC Female ** ** 3 HDL Overall *** *** 45 HDL Male *** *** 32 HDL Female *** *** 51 LDL Overall * ** 350 LDL Male LDL Female * * 350 TG Overall ** ** 306 TG Male *** ** 337 TG Female ** ** Seasonal effect test relative to null model ++ Day of year, numbered consecutively * p<0.05 ** p<0.01 *** p<0.001

7 Table 3. Summary of Model Fitting Statistics and Amplitude and Phase Estimates for Seasons Study of Cholesterol after control for BMI, Kcal, %Kcal from SFat, %Kcal from Fat, Total Waking Activity (MET), and Proportion of Daily Outdoor Light Cholesterol Subjects # Obs # Subj. -2Log(L)+ Subject Residual Mean Amplitude Phase Variance Variance TC Overall *** *** 349 TC Male *** ** 334 TC Female *** * 1 HDL Overall *** *** 44 HDL Male *** *** 32 HDL Female *** *** 51 LDL Overall *** LDL Male *** LDL Female *** TG Overall *** ** 286 TG Male *** * 308 TG Female *** * 261 * p<0.05 ** p<0.01 *** p< Co-variable test relative to sine/cosine model Table 4: Amplitude of seasonal variation in lipids above and below the median, and in the highest quartile Below median Above median 4 th quartile TC Overall 3.0* 6.1*** 6.8** m *** 4.0 f ** 9.3** HDL Overall 2.1*** 2.7*** 3.1*** m 1.4*** 2.0*** 2.8*** f 3.1*** 3.5*** 3.7** TG Overall ** 23 m * 47 f ** 24* LDL Overall * 4.4* m f * p<0.05 ** p<0.01 *** p<0.001

8 I wouldn't use Table 5, and instead report the percent change directly based on the change in amplitude between Table 2 and Table 3. (EJS) Table 5: Percent of the Residual Variance Attributed to Seasonal Effects that is Explained by Seasonal changes in body mass index, daily estimates of total Kcal intake, saturated fat as a percent of Kcal, total fat as a percent of Kcal, total waking MET, and total time spent in outdoor light. Males Females TC 50.7% (reduced) 50.1% (reduced) HDL 63.7% (enhanced) 10.3% (reduced) TG 55.4% (reduced) 46.4% (reduced) LDL 100% (reduced) 71.1% (reduced) Source: snc99p164.sas I'd use Table 6&7. Source:sne99p163.SAS SEASON 12/24/99 EJS Table 1. Male TC percentiles by date Pct Pct Pct Pct Pct >230 mg/dl >235 mg/dl >240 mg/dl >245 mg/dl >250 mg/dl DAYX DAY (G230) (G235) (G240) (G245) (G250) Jan Apr Jul Oct Source:sne99p163.SAS SEASON 12/24/99 EJS Table 2. Female TC percentiles by date Pct Pct Pct Pct Pct >230 mg/dl >235 mg/dl >240 mg/dl >245 mg/dl >250 mg/dl DAYX DAY (G230) (G235) (G240) (G245) (G250) Jan Apr Jul Oct