Statistical analysis Directory UMM :Data Elmu:jurnal:A:Atherosclerosis:Vol152.Issue2.Oct2000:

types of alcoholic beverages consumed in Spain, which presumably covered all types of alcoholic beverages usually consumed. For each item, the questionnaire included seven frequency categories, and requested in- formation on the number of glassesday. For each item, glassesday were calculated by adding the weekend consumption to the workday consumption. For each alcoholic beverage, mean of daily ethanol consumption in grams was calculated by multiplying the amount consumed in ml by the percentage of ethanol supplied by each specific beverage. From the reported alcoholic beverages, alcohol consumption was categorized as a drinker variable according to the amount of alcohol consumed. Men and women were classified into three groups according to the population tertiles: No con- sumption alcohol consumption = 0 gday, Moderate B 10.5 g alcoholday for men, and B 5.5 gday for women, and High consumption \ 10.6 g alcoholday for men, and \ 5.6 g alcoholday for women. In some analyses, only two categories were considered: non- drinkers alcohol consumption = 0 and drinkers sub- jects with any amount of alcohol consumed. Physical activity was estimated by questions about regular leisure-time physical sports aerobics, basket- ball, bicycling, gymnastics, running, soccer, squash, swimming, tennis, volleyball, and others, as well as the average number of h per week spent in each activity. According to the type and time [23], subjects were categorized as Sedentary no physical exercise, Moder- ate one sport less than 3 hweek and High one sport more than 3 hweek or more than two sports. For regression analyses, physical exercise was also dichoto- mized into Sedentary versus Active Moderate plus High. In addition to the type and time spent per week in physical exercise, another question about regular daily walking more than 20 min, with two possible answers ‘yes’ and ‘no’, was also included. Marital status, classified as married, single and di- vorced, was dichotomized as being Single living alone and divorced or Living with a partner. Education was initially coded into five categories: Non-schooling, Pri- mary school, Secondary school, University-1 short cy- cle and University-2 long cycle, and afterwards recoded into three levels: Primary, Secondary and Uni- versity short + long cycles. 2 . 4 . Laboratory analyses Plasma total cholesterol and tryglicerides were deter- mined by a Technicon Chem 1 assay Technicon Instru- ments, Tarrytown, NY, and HDL-C was measured in the supernatant after precipitation of apolipoprotein B — containing lipoproteins with heparin-manganese chloride. LDL-C was calculated according to the equa- tion of Friedewald et al. [24] for samples with serum triglyceride concentrations below 400 mg per deciliter. 2 . 5 . DNA extraction and genotyping Genomic DNA was extracted from white blood cells by phenol extraction. A fragment of 535 base pairs in the intron 1 of CETP gene was amplified by polymerase chain reaction PCR. Each amplification was per- formed using 500 ng of genomic DNA in a volume of 50 ml containing 40 pmol of each oligonucleotide U: CACTAGCCCAGAGAGAGGAGTGCC and L: CT- GAGCCCAGCCGCACACTAAC, 0.2 mM dNTPs, 1.5 mM MgCl 2 , 10 mM Tris pH 8.4 and 0.25 U of Taq polymerase Gibco BRL, Paisley, UK. The PCR con- ditions were 95°C for 5 min, and subsequently 28 cycles at 95°C for 30 s, 60°C for 30 s, and 72°C for 45 s, and finally at 72°C for 5 min. The PCR products were subject to restriction enzyme analysis by digestion with 4 units of the restriction endonuclease TaqI for 16 ml of PCR sample at 65°C for 2 h, in the buffer recom- mended by the manufacturer Pharmacia Inc. Sweden.

3. Statistical analysis

Questionnaires, biochemical and genetic data were key-entered twice and processed to avoid errors in coding and interpretation. Normal distribution for all continuous variables was checked by graphical methods and by hypotheses tests. Triglycerides and alcohol in- take were markedly skewed, and these variables were logarithmically and squared root transformed, respec- tively, to improve normality for statistical testing. To assess mean differences of lipid and anthropometric variables between genders, Student’s t test for indepen- dent samples was used after determining the homogene- ity of variances by the Levene statistic. Chi square tests were conducted to compare differences in percentages. This statistical test was also performed to examine whether the genotype frequencies were in Hardy – Wein- berg equilibrium [25]. For multiple comparisons of means between genotypes, one way analysis of variance ANOVA was performed. P values for linear trends between categories were also calculated using the ANOVA analysis by partition of the between-groups sums of squares into trend components. Once estab- lished that differences exist among means, to determine which means differ with correction for multiple com- parisons, Bonferroni test was applied. To test the null hypotheses of no association between CETP genotypes and HDL-C, controlling by one or more potential environmental and genetic confounders or effect modifiers, multivariate modeling was used. Analyses of covariance ANCOVA were conducted as exploratory analysis and to evaluate the statistical significance of multiple categorical terms or interaction terms into the multivariate model. To estimate the extent, direction and strength of the association between predictors and CETP genotypes, multiple linear regression analysis with dummy variables for categorical and interaction terms was applied. In order to improve statistical power, data from men and women were analyzed to- gether, including a dummy variable for gender into the regression model. A core model with gender, age, BMI, CETP TaqIB genotypes, and a set of variables for tobacco smoking and alcohol consumption, was first identified using forward and backward procedures. The selection criterion for choosing the core model was the following: a variable or interaction term was retained into the model if the global term was statistically sig- nificant P B 0.05. Regression diagnostics analysis of residuals, influence of outliers and colinearity were employed to check the assumptions and to assess the Table 2 CETP TaqIB polymorphisms in the population by gender a Women All Men n n n Genotypes 117 41.3 B1B1 222 43.2 105 45.5 100 43.3 223 43.4 123 43.5 B1B2 69 13.4 26 11.3 B2B2 43 15.2 514 283 100.0 Total 231 100.0 100.0 Allele frequency B2 Allele 0.329 0.351 0.369 0.329–0.398 0.286–0.372 95 C.I. 0.322–0.380 a Differences by gender across CETPTaqIB polymorphisms were nonsignificant. Chi square P value = 0.332. Table 1 Demographic, biochemical and life-style characteristics of the popula- tion Men n = 231 Women n = 283 Mean SD P Mean SD 0.102 Age years 35.7 10.4 37.4 10.1 Body mass index B 0.001 23.9 4.2 26.4 4.8 kgm 2 5.3 1.0 Total-C mmoll 4.9 0.9 B 0.001 3.1 0.8 3.5 0.9 LDL-C mmoll B 0.001 1.1 0.3 HDL-C mmoll 1.4 0.3 B 0.001 Triglycerides B 0.001 1.3 1.2 1.1 0.7 mmoll B 0.001 127.2 15.9 Systolic blood 117.5 17.7 presure mmHg 0.017 73.8 11.6 70.7 12.4 Diastolic blood presure mmHg 44.6 0.998 44.6 Current smokers Past smokers 0.002 25.7 12.9 B 0.001 Alcohol users 88.9 59.9 Physical exercise 36.5 61.7 Sedentary Moderate 32.7 39.6 5.6 0.005 23.9 High 74.3 Daily walkers 71.9 0.872 Education 0.0 Non-schooling 0.0 26.8 34.7 Primary 29.6 Secondary 31.4 University short 19.6 32.3 cycle University long 9.8 16.1 0.002 cycle Marital status 66.7 52.1 Married 41.8 30.8 Single 2.5 6.3 Divorced 0.016 P value in the comparison between men and women. Student’s test for comparison of means, and Chi square tests for percentages. accuracy of computations. Once the core model was identified for the total sample, separate models with the same variables were computed for men and women. Finally, additional life-style variables physical activity, marital status and educational level were included into the core model defined in the previous stage. A variable or interaction term was maintained in the final regres- sion model as independent term or as control variable on the basis of statistical significance P B 0.05 or P B 0.20, respectively. All analyses were done using the Statistical Package for Social Sciences SPSS, version 8.0 for windows.

4. Results