The main result of the present study was that the mean sleep duration slightly decreased in the adult study population over the 36-year follow-up. The change was seen in all adult age groups and in both genders. Overall the changes are quite small. The mean decrease in men was about 18 minutes and in women about 32 minutes (about 0.5 and 0.9 minutes per year of follow-up, respectively). In age-groups the largest change was in middle-aged persons: about 23 minutes or 0.6 minutes per year of follow-up. Although the sleep duration was longer in the earlier years, the probable main cause for the change in this study population is the effect of aging.
In a systematic review on secular trends on sleep duration from 15 countries from the 1960s until the 2000s showed also small changes . Self-reported average sleep duration of adults had increased in 7 countries (from 0.1 to 1.7 minutes per night each year, e.g. Canada and Britain) and decreased in 6 countries (from 0.1 to 0.6 minutes per night each year, e.g. Finland and Germany). The Finnish data was based on reanalysis of all available data from surveys carried out from 1972 to 2005, including about 440,000 self-reports of sleep duration , to which the present study contributed data from the first three surveys (1975, 1981, and 1990). The decrease in Finland was 18 minutes during a period of 33 years or about 0.5 minutes per year of follow-up. In all, the present results are well in accordance with the results from previous studies both from Finland and other countries. It has been pointed out that there is no validated measure of self-reported sleep duration and the different ways of asking (for example rounding sleep to within 15 or even 60 minutes) may affect the estimate . As the changes in each country and the differences between the countries are small, it is possible that they are at least partly due to methodological causes.
The strong association between health and sleep has also practical implications. It has long been known that many patients with somatic or psychiatric disorders also suffer from impaired sleep. Not until the last decades have we learned that poor sleep is not only comorbid with many disorders, but may also be a contributing causative factor, as shown in, for example, cardiovascular disease [1, 28-30]. Evidence for a causal role in breast cancer  and type 2 diabetes [32, 33] is less convincing. Genetic analyses suggest shared genetic liability to sleep duration and a variety of other medically relevant traits and diseases , but the causal paths require more studies. Thus, in clinical practice, it is important to evaluate sleep, as poor quality may increase the risk of many common diseases, decrease the quality of life, and impair patients’ treatment responses.
The present study has several strengths. First, the follow-up time of 36 years is exceptionally long. Second, the study cohort is population-based and therefore the results can probably be generalized to the population at large, at least in countries with a very high human development index. Compared to general population, the Twin Cohort shows no differences in e.g. overall mortality  or cancer incidence . Sleep lengths in twins have been similar as in other Finnish studies . In addition, our twin cohort has taken part in multiple genome-wide association studies, with cohort specific effect size estimates that do not differ from the meta-analysis results [35,36]. The results are also valuable from an ontogeny point of view. Thirdly, longitudinal cohort studies may give a more reliable picture of age-related changes than a cross-sectional study or combinations of these, as suggested by our earlier 36-year follow-up study on the quality of sleep in the same cohort .
There are also some limitations that must be kept in mind. Information on sleep length is derived from self-reports. As we have discussed earlier , objective assessments of sleep length for thousands of participants three to four decades ago was logistically and even technologically not possible. Large-scale studies continue to rely on self-reported sleep length, even though it has less accuracy, which affects in particular those with fragmented sleep (e.g. insomniacs). The 1990 and 2011 questionnaires did not cover age groups 65+ years, and 2011 did not include subjects < 35 years, thus the conclusions must be especially cautious regarding them. Because of the structure of the study population only limited conclusions of the possible intergenerational changes in sleep duration can be made. Our sample consists of twins but it is representative of general population. In our 1975 questionnaire we had both twins and persons who were not twins as a result of our ascertainment procedure of identifying as twins pairs of persons born on the same day, in the same local community, of the same sex and with the same surname at birth . The twins and non-twins did not differ in sleep duration or sleep quality. We are not aware of studies indicating significant differences in sleep and its disorders between twin individuals and the rest of the general population. Additionally, we cannot rule out that some fraction of the apparently age-related changes is in fact due to contemporaneous secular trends, such as the 24/7 society and changes in working life including reduced physical strain and the increasing use of portable screen-based media devices. These trends probably affect more younger adults (< 35-year-old) who were not represented in the last survey made during time when the use of this technology rapidly increased. The three first surveys were made during the era with no smartphones and social media. Even at the time of the fourth survey, internet and social media use was much less than now, a decade later. The last wave was collected in 2011/2012 among older adults born 1945 to 1957, who have adopted smart phones and tablets at a slower rate than adolescents and young adults.