Measures

BW for the twins as well as other relevant birth related covariates (see Covariates below) were derived from official birth records. For those born between 1926 and 1972, birth information was based on a nation-wide collection of information from original birth records recorded by midwives and/or doctors at the time of birth, while for those born after 1973 it was based on records from the Swedish Medical Birth Register. Birth record data were then linked to the collected twin data using the unique personal identification number assigned to each Swedish citizen. For details on the BW measures and matching procedure see Hogberg et al. (2013). BW effects were explored as continuous (birthweight in grams divided by 250 for ease in interpreting statistical results) as well as dichotomized coded as low (≤ 2500g) versus normal (> 2500g) based on commonly used classifications and past findings (Kramer, 2013). In addition, as there was some indication for a potential non-linear relationship between self-perceived health and BW, analyses were repeated with quadratic (continuous) BW.

Here we used three different self-perceived health outcome variables which were dichotomized to differentiate between healthy and less healthy individuals. The first self-perceived health question was ‘How do you estimate your general health?’ rated as excellent, good, versus average, not so good, and bad. This variable will subsequently be referred to as self-rated health (SrH). The second question was ‘How do you estimate your health compared to five years ago?’ rated as better, the same, versus worse and will be referred to as SrH-5years. Third, we used the question: ‘Do you think your health status prevents you from doing things you want to do?’ rated as not at all, versus to some extent, or a great deal. Subsequently, we will refer to this variable as Activity. Tetrachoric correlations between the three self-perceived health measures ranged between 0.59 and 0.74.

A number of factors that may also influence both BW and self-reported health were included as demographic covariates: sex of participant (if the sexes were analyzed together), age of participant at time health was rated; and birth information covariates: parity (the number of children previously born to the same mother), age of mother at birth, and gestational age (days of pregnancy since the first day of the last menstrual period until birth). To make sure we did not lose power in the corrected analyses (compared to uncorrected) due to missing values in covariates, missing values were replaced with the mean per zygosity (as the occurrence of DZ twins increases with maternal age and parity) for parity, age of mother, and gestational age.