The participants are from a longitudinal follow-up cohort of subjects born preterm at VLBW between 1978 and 1985 and discharged alive form the neonatal intensive care unit of Children's Hospital, Helsinki University Central Hospital, Finland. Of the original 335 survivors, 255 VLBW subjects lived in the greater Helsinki area at the time of follow-up and were invited to a clinical study together with a sex-, age-, and birth hospital-matched comparison group of 314 term-born subjects who were not small for gestational age (birth weight more than −2 SD) ; 166 VLBW and 172 controls participated. Of these, 4 and 3, respectively, were excluded because of not having fasted overnight, or being pregnant. Thus, the study finally included 162 VLBW and 169 term-born control subjects. None of these were treated with lipid lowering medication. One subject had type 1 diabetes. All subjects underwent a 75 g 2-hour oral glucose tolerance test (OGTT), based on which none had type 2 diabetes .
After an overnight fast of at least 8 hours, the clinical examination included weight and height measurement, and a 2-hour 75 g OGTT . From both baseline- and 2-hour-sample, blood was drawn for lipoprotein subclass analysis. Moreover, the participants completed questionnaires that covered their medical history, smoking habits, and educational level of their parents.
Total lipid (L), triglyceride (TG) and cholesterol (C) concentrations of 14 lipoprotein subclasses were analyzed by proton NMR spectroscopy in serum samples. The details of this methodology have been described [17, 18] and this platform has recently been applied in large-scale epidemiological and genetic studies [19, 20]. The lipoprotein subclass data are as follows: chylomicrons and largest (XXL) very-low-density lipoprotein (VLDL) particles (average particle diameter at least 75 nm); five different VLDL subclasses: very large (XL) VLDL (average particle diameter of 64.0 nm), large (L) VLDL (53.6 nm), medium (M) VLDL (44.5 nm), small (S) VLDL (36.8 nm), and very small (XS) VLDL (31.3 nm); intermediate-density lipoprotein (IDL) (28.6 nm); three LDL subclasses: L-LDL (25.5 nm), M-LDL (23.0 nm), and S-LDL (18.7 nm); and four HDL subclasses: XL-HDL (14.3 nm), L-HDL (12.1 nm), M-HDL (10.9 nm), and S-HDL (8.7 nm). Due to resolution and concentration issues TG and C are not available for every subclass . The mean size for VLDL, LDL, and HDL particles was calculated by weighting the corresponding subclass diameters with their particle concentrations. IDL particles were included in the LDL measure. Altogether 44 lipoprotein measures were used.
Statistical analyses were conducted using SPSS 17 and 19 software (IBM SPSS, Chicago, IL). Comparison of basic characteristics between VLBW and term was performed using t-test for continuous and chi-square -test for categorical variables. Because of skewed distributions lipoprotein subclass measurements were transformed as follows: 0.1 was added to avoid zero-values and natural logarithm was used to achieve more symmetrical distributions. Distributions were checked after logarithm transformation and those two that were not sufficiently normal were transformed again. To find a situation where the residuals of XXL-VLDL-TG and residuals of XXL-VLDL-L were fairly normal, we utilized the ‘Box-Cox transformations for linear models’ in the ‘car’-package downloaded in April 4th, 2013, to R-software, version 2.15.2. Oct 26, 2012. To avoid non-positive dependent values we added 0.1 to the dependent prior to the Box-Cox iterative procedure. The lambdas with maximal log likelihood were -7.3 and -5.3, for the respective outcomes, yielding to our best transformations for the full models: y’ = (y + 0.1)(lambda-1)/(lambda), where y = original variable. With these y’ as dependent in the full models, the resulting in symmetric distributions of residuals but, unfortunately, with less pronounced aggregation of mid values.
We used multiple linear regression to compare individual lipoprotein subclass measurements between VLBW and control subjects. We adjusted for confounding factors in different models. The fully adjusted model included age, sex, height, highest parental education, body mass index, mother’s smoking during pregnancy, and daily smoking of the participant (yes/no) as covariates. Additionally we also adjusted for alcohol use in the fully adjusted model; dicotomized values for whether the subject used alcohol weekly to get drunk or not, and whether the subject ever used alcohol or not. We also adjusted for fasting and 2-hour glucose concentrations in additional models. We tested group differences in 44 outcome variables. Many of the lipoprotein measures are highly correlated and thus we also utilized the principal component analysis to reduce the number of outcomes tested (the FACTOR-program in SPSS, version 19, was used). We included the five first components as they explained 95% of the variation and only their eigenvalues were greater than 1.0. We then tested group differences (VLBW versus term) of each individual's Varimax-rotated component scores using adjusted linear regression models.
The study was performed according to the declaration of Helsinki. The study protocol was approved by the Ethics Committee at the Helsinki and Uusimaa Hospital District. Written informed consent was obtained from each participant.