The NHLBI FHS is a multicenter, population-based study that includes probands recruited from three parent studies: the Framingham Heart Study (Framingham, MA), the Atherosclerosis Risk in Communities Study (Forsythe County, NC and Minneapolis, MN), and the Utah Family Health Tree (Salt Lake City, UT). Between 1993 and 1995, 2,000 randomly selected participants and 2,000 with family histories of CHD enrolled in the 3 parent studies were mailed invitations to provide information regarding the health of themselves, their parents, their children, and their siblings. Of these, 3,150 returned responses, and family members of these responders were contacted. From the 22,908 persons who provided information, 588 families were randomly selected and an additional 657 with high family risk scores were also selected. Family risk score was calculated by comparing the actual incidence of CHD to the age- and sex-adjusted expected incidence of CHD within a family. Enrollees underwent study examinations during which height, weight and blood pressure were documented, and 12-hour fasting blood samples were taken on which lipids and other parameters were measured, and from which DNA was stored. Genotyping for the APOE allele was performed on a selection of FHS participants. More detailed description of the methods of the NHLBI FHS have been described previously . The FHS is supported by the National Heart, Lung, and Blood Institute (NHLBI) (grant numbers U01 HL56563, U01 HL56564, U01 HL56565, U01 HL56566, U01 HL56567, U01 HL56568, U01 HL56569, and K01-HL70444). Each participant in the FHS gave informed consent and the study protocol was reviewed and approved by the respective Internal Review Boards at each of the participating institutions.
For this study, we selected subjects with information on APOE genotype and with complete data on lipids, lifestyle factors, and co-morbidities. Due to small numbers of non-white subjects, participants of other races (54 subjects) were excluded from this study, as were 34 subjects who were using cholesterol modifying therapy. Finally 37 subjects were excluded with the ε2/ε4 genotype, as the ε2 allele has been associated with higher levels of HDL while ε4 carrier status has been associated with lower HDL levels .
APOE genotyping was performed using polymerase chain reaction (PCR) to amplify a 267-base pair fragment from exon 3 of the APOE gene. The PCR product was digested using the Hha-1 restriction endonuclease, which resulted in a specific banding pattern for the 3 isoforms of the apolipoprotein E protein when separated by polyacrylamide gel electrophoresis and silver stained . Subjects with genotypes of ε3/ε4 or ε4/ε4 were carriers; subjects with genotypes of ε2/ε2, ε2/ε3, or ε3/ε3 were non-carriers.
Height and weight documented during the study examination were used to calculate body mass index (BMI). BMI was treated both as a continuous and a categorized variable (BMI < 25 kg/m2, 25 ≤ BMI < 30 kg/m2, and BMI ≥ 30 kg/m2) in the analyses described below.
During the study visits, a fasting blood sample was collected in evacuated tubes without additives, spun at 3000xg for 10 minutes at 4°C, and stored at -70° until they were processed. HDL, total cholesterol, and triglyceride concentrations were measured using a COBAS FARA high-speed centrifugal analyzer (Roche Diagnostic Systems, Montclair, New Jersey).
Information on lifestyle factors including smoking (currently smoking versus not currently smoking), alcohol intake (consumption of 2 or more beverages daily), and physical activity (minutes of exercise per day) was obtained by interview during the study visit. Mono- and poly-unsaturated fat intake (MUFA and PUFA, in grams) was determined using a previously validated food frequency questionnaire administered by study staff during the study interview and a nutrient database from Harvard University supplemented by manufacturer's information . Participants provided self-reported information on co-morbities including angina, CHD, stroke, cancers, and hypertension. A subject was deemed to have diabetes mellitus if he was taking hypoglycemic agents, if a physician had told the subject that he had diabetes mellitus, or if fasting blood glucose was ≥ 126 mg/dL. In subjects with triglycerides < 400 mg/dL, LDL (LDL) was calculated using measured HDL, total cholesterol, and triglycerides. LDL was measured directly on EDTA plasma using ultracentrifugation on EDTA plasma in subjects with triglycerides ≥ 400 mg/dL.
The chi-square test was used to determine whether the APOE allele distribution was in Hardy-Weinberg equilibrium in this population prior to exclusions. We compared characteristics of the cohort by ε4 carrier status and by BMI category, using t-tests to compare continuous variables and Cochran-Mantel-Haenszel tests to compare categorical variables. We then used generalized estimating equations (GEE, using the LSMEANS option in PROC GENMOD in SAS) to determine adjusted means and standard errors for HDL across BMI categories, in subgroups by ε4 carrier status. GEE adjusts for familial clustering in the NHLBI FHS. Means were adjusted for age, categorized BMI, gender, current smoking status, alcohol consumption (≥ 2 beverages daily versus non or < 2 daily), minutes of exercise daily, mono- and PUFA intake in grams, angina, stroke, hypertension, coronary heart disease, diabetes mellitus, and LDL. We determined the parameter estimate and p-value for a categorized BMIxε4 carrier product term in a model including the main effects of categorized BMI and carrier status, plus age, gender, lifestyle factors, co-morbidities, and LDL. In secondary analyses, we examined whether a continuous BMIxε4 carrier product term was significantly associated with HDL after adjusting for the main effects of continuous BMI and carrier status and age, then additionally adjusting for gender, lifestyle factors, co-morbidities, and LDL. We did not include triglycerides in adjusted models due to collinearity between the outcome (HDL) and triglycerides as previously modeled by others .
In a sensitivity analysis, we examined whether there was a significant association between continuous BMIxε4 carrier product term and HDL in a sub-cohort consisting of one individual (randomly selected) per family, and in subgroups of participants defined by CHD risk (members of randomly selected families versus members of families with increased CHD risk). To examine whether there was a significant interaction between BMI and APOE genotype (in contrast to carrier status for the ε4 allele) we treated genotype as an ordinal variable (ε2/ε2, ε2/ε3, ε3/ε3, ε3/ε4, ε4/ε4) and included a continuous BMIxgenotype product term in the model. To examine whether a BMIxε4 carrier status interaction was significantly associated with HDL in subgroups by gender, we repeated the analysis including the main effects of BMI and ε4 carrier status and their interaction term in subgroups of men and women. All analyses were performed using SAS version 9.2 (SAS Institute Incorporated, Cary, North Carolina).