Population sample
The Older Finnish Twin Cohort is a longitudinal study of Finnish twins who belonged to pairs of the same gender born before 1958, and who were still alive in 1975 [18, 19]. As these pairs were selected from the Central Population Registry of Finland in 1974, the Cohort is population-based and its overall mortality and cancer incidence does not differ from that of the general population [18, 20]. Cohort members have responded to four questionnaires in which sleep duration was asked. The first questionnaire was mailed in 1975, and the response rate was 89% (N = 30,915 twin and non-twin individuals). The second questionnaire in 1981 yielded a response rate of 84% (N = 24,506 individuals). The third questionnaire survey in 1990 was mailed to pairs born in 1930–1957, both co-twins of whom were resident in Finland in 1987 (response rate 77%, N = 12,502 individuals). The fourth questionnaire was sent in 2011/2012 to all twins in the cohort who were alive with an address in Finland according to the Finnish national population register, and who were born in 1945–1957. The response rate was 72% (N = 8510 individuals). The questionnaires included about 100 questions on demographic, social, health/illness and lifestyle variables. The number of sleep-related variables in the different questionnaires varied, but sleep length was elicited in all four questionnaires [4, 21, 22].
Table 1
Characteristics of subjects included in the present study at each of the four measurement points. Persons with missing data on sleep length are excluded.
Year of questionnaire | N of individuals | Women (%) | Age (mean ± 1.96 × SD) in years Range of age (birth years) |
1975 | 30,915 | 51.4 | 35.7 ± 14.6 18–95 (1880–1957) |
1981 | 24,535 | 51.7 | 40.5 ± 13.7 24–101 (1880–1957) |
1990 | 12,450 | 54.4 | 43.9 ± 7.8 33–60 (1930–1957) |
2011 | 8334 | 55.4 | 59.7 ± 3.7 54–66 (1945–1957) |
Characteristics of subjects included in the present study at each of the four measurement points are given in Table 1. Age distributions of the respondents in each of the four points of measurement are given as Supplement (Figure S1).
Questionnaire data
Information on sleep length was obtained by asking "How many hours do you usually sleep per 24 hours?" In 1975 seven response alternatives were used (less than 4 hours, 5, 6, 7, 8, 9, and 10 hours or more), and in 1981, 1990 and 2011 nine alternatives (6 hours or less, 6.5, 7, 7.5, 8, 8.5, 9, 9.5, and 10 hours or more).
Statistical methods
Standard statistical methods were used in data analysis to obtain descriptive statistics such as frequencies and proportions when comparing sleep duration groups and implemented in the Stata package (version 13.1, Stata Corp, College Station, Tx, USA, www.stata.com). In this study self-report of sleep duration is the dependent variable, and as explanatory variables we used birth cohort, age of the respondent at each measurement, and gender.
We used a Weibull regression model as a parametric regression model for survival data [23]. Because the data was composed of repeated measurements on the same persons, we modelled the covariance structures using random effects. We further took into account the lack of statistical independence within twin pairs due to sampling of twin pair candidates in 1975 by modelling the twin pairs as random effects. Finally, we did not have sleep length as a continuous measure (e.g. hours and minutes) but rather as time intervals, so interval censoring was used [24]. The detailed statistical approaches are described in Haasio [25].