Selenium status and blood lipids: the cardiovascular risk in young finns study
Saverio Stranges, MD, PhD, Health Sciences Research Institute, University of Warwick Medical School, Gibbet Hill Campus, Medical School Building, Room A105, Coventry CV4 7AL, UK.
(fax: + 44 (0) 2476528375; e-mail: S.Stranges@warwick.ac.uk).
Abstract. Stranges S, Tabák AG, Guallar E, Rayman MP, Akbaraly TN, Laclaustra M, Alfthan G, Mussalo-Rauhamaa H, Viikari JSA, Raitakari OT, Kivimäki M (Health Sciences Research Institute, University of Warwick Medical School, Coventry; University College London, London, UK; Semmelweis University Faculty of Medicine, Budapest, Hungary; Johns Hopkins Bloomberg School of Public Health, Baltimore, MD, USA; National Center for Cardiovascular Research (CNIC), Madrid, Spain; University of Surrey, UK; National Institute of Health and Medical Research (Inserm) U888. F-34000 Montpellier, France; National Institute for Health and Welfare, Helsinki, Finland; Hjelt Institute, University of Helsinki, Helsinki; University of Turku and Turku University Hospital, Turku; University of Turku, Turku; Finnish Institute of Occupational Health and University of Helsinki, Helsinki, Finland). Selenium status and blood lipids: the cardiovascular risk in young finns study. J Intern Med 2011; 270: 469–477.
Background. Concern has been recently raised about possible adverse cardio-metabolic effects of high selenium status, such as increased risks of diabetes and hyperlipidaemia. However, most of the evidence comes from selenium-replete populations such as that of the United States.
Objectives. To examine cross-sectional and longitudinal associations of serum selenium with cardiovascular risk factors in Finland where selenium levels were amongst the lowest in the world until the early 1980s before the implementation of a nationwide selenium fertilization programme.
Methods. Serum selenium was measured in 1235 young Finns aged 3–18 years at baseline in 1980 (prefertilization) and in a subgroup (N = 262) at the 6-year follow-up (1986, postfertilization). During the 27-year follow-up, serum lipids, blood pressure, body mass index and smoking were assessed five times (1980, 1983, 1986, 2001 and 2007).
Results. Mean (±SD) serum selenium concentrations were 74.3 ± 14.0 ng mL−1 in 1980 and 106.6 ± 12.5 ng mL−1 in 1986 (average increase 32.3 ng mL−1; 95% CI: 30.3 to 34.3, P < 0.0001). In univariate and multivariable cross-sectional models in 1980 and 1986, increased serum selenium levels were consistently associated with increased total, HDL and Low-density lipoprotein (LDL) cholesterol. However, the average longitudinal changes in lipids were −0.20 mmol L−1 (95% CI: −0.30 to −0.10, P < 0.0001) for total cholesterol, 0.06 mmol L−1 (95% CI: 0.03 to 0.10, P < 0.0001) for HDL cholesterol, and −0.23 mmol L−1 (95% CI: −0.31 to −0.14, P < 0.0001) for LDL cholesterol. Selenium measured in 1986 was not associated with lipids assessed in 2001 and 2007.
Conclusions. Cross-sectional findings from the Young Finns study corroborate positive associations of selenium status with serum lipids. However, longitudinal evidence does not support the causality of this link.
Considerable research effort is currently focused on understanding the full range of effects of selenium status on chronic disease outcomes across different populations worldwide [1–3]. Selenium is a micronutrient at the active site of glutathione peroxidase and other selenoproteins involved in important enzymatic functions, such as redox homeostasis [4–5]. The primary emphasis of selenium research has been on evaluating the health effects of selenium deficiency, as well as the potential benefits of its antioxidant and anticancer effects [6–8]. Recent studies, however, have suggested an association between high selenium exposure and adverse cardio-metabolic effects, at least in selenium-replete populations such as that of the United States [9–16]. Nonetheless, most of these studies are cross-sectional [9,11,13–16], and there is little longitudinal data on the association of selenium status with cardiovascular risk factors.
The Cardiovascular Risk in Young Finns Study provides a unique opportunity to examine cross-sectional and longitudinal associations of selenium status, as measured by serum selenium concentrations, amongst 1235 participants from different parts of Finland at the baseline examination (1980, aged 3–18 years) and in a subgroup of 262 participants at the 6-year follow-up, on cardiovascular risk factors. Overall, participants were followed up for an average of 27 years from childhood to adulthood [17–18]. In Finland, selenium intake was very low for geochemical reasons until the early 1980s . An early case–control study in a Finnish population showed potential associations of low selenium status with increased cardiovascular risk . In 1985, a nationwide selenium fortification programme (use of selenium-containing fertilizers) was implemented to increase dietary intakes of selenium in the country . In the current analysis of the Young Finns Study, we have been able to use for the first time longitudinal data from the study to determine the association of selenium status, including initial changes in serum selenium resulting from the fortification campaign, with cardiovascular risk factors in the study population.
Subjects and methods
The Cardiovascular Risk in Young Finns Study was launched in 1980 to assess risk factors underlying cardiovascular disease in children and young adults [17–18]. The first cross-sectional survey was conducted in 1980, when 3596 participants were randomly chosen from the national register (aged 3–18 years). Follow-up studies were conducted in 1983, 1986, 2001 and 2007 with 2991, 2779, 2283 and 2204 participants, respectively.
The study cohort for the present manuscript comprised those subjects (N = 1235) with available data on serum selenium and cardiovascular risk factors at the baseline examination in 1980 (prefortification), which were used for our cross-sectional analyses. In addition, serum selenium was measured in a smaller subgroup of participants (N = 262) in 1986 (postfortification). In this subgroup with two available selenium measurements, we performed additional cross-sectional analyses using data from the 1986 screening and longitudinal analyses with data from screenings in 1986 (N = 262), 2001 (N = 234) and 2007 (N = 208). Parents gave written informed consent, and the study was approved by local ethics committees.
We used data from study years 1980, 1983, 1986, 2001 and 2007. Height and weight were measured, and body mass index (BMI) was calculated as weight in kilograms divided by height in metres squared. Blood pressure was measured with a standard mercury sphygmomanometer in 1980 and 1983 and with a random-zero sphygmomanometer thereafter. The agreement between blood pressure measurements using the standard mercury and random-zero sphygmomanometers is generally high, with correlation coefficients ranging from 0.85 to 0.90 [17–18]. The average of three measurements was used in statistical analysis. Smoking was assessed with a questionnaire in subjects aged 12 years or older. Smoking on a daily basis was defined as regular smoking. Family history of hypertension and diabetes was ascertained by means of a questionnaire. Parental socio-economic status (SES) was assessed from occupational level and categorized into manual versus nonmanual. Where occupational level differed between parents, data on the parent with the higher level were used. For a sensitivity analysis, we created a composite parental SES measure including three components: parental occupational status in 1980 (1 = manual, 2 = lower-grade nonmanual and 3 = higher-grade nonmanual), parent’s household income in 1980 (1 = bottom quartile, 2 = 2 middle quartiles and 3 = top quartile) and parental education in 1983 (1 = comprehensive school; 2 = secondary education, not academic; and 3 = academic) and used this as a sum score of parental occupational status, household income and parental education (range, 3–9), as previously reported .
Venous blood samples were taken after the subject had fasted for at least 12 hours. Selenium determinations were made by flameless atomic absorption spectrophotometry (Perkin-Elmer 5000, HGA-400), and commercial standard reference material Monitrol® (Dade, Miami, FL, USA) was used to control analytical quality, as previously described in detail . Serum selenium was measured in 1,235 subjects at baseline in 1980 and in a smaller subgroup of participants (N = 262) in 1986.
Lipid determinations used standard methods in a consistent fashion throughout the whole period of follow-up. Details of analytical procedures have been reported previously . Low-density lipoprotein (LDL) cholesterol was calculated by the Friedewald formula for subjects with triglycerides of <4 mmol L−1. The coefficient of variation was 2.2% for total cholesterol, 2.3% for high-density lipoprotein (HDL) cholesterol and 3.8% for serum triglycerides.
Descriptive data are expressed as means and standard deviation, or as percentages. Participants were divided by quartiles of baseline serum selenium concentrations. In cross-sectional analyses, differences in cardiovascular risk factors were examined across selenium quartiles by linear (age, BMI, blood pressure and serum lipids) and logistic regression (sex, smoking, parental SES, parental hypertension and parental diabetes), adjusted for age and sex. In multivariable-adjusted models, further adjustments included BMI, smoking and parental socio-economic position.
To further explore the association between selenium and serum lipids, additional analyses were performed in a subgroup of participants (N = 262) with selenium measured both at baseline (1980, prefertilization) and at the 6-year follow-up visit (1986, postfertilization). We computed Pearson correlation coefficients contrasting values in 1980 with values in 1986 for both selenium and serum lipids. We examined changes in the serum selenium–lipid associations across these two points of time by fitting a ‘time × lipid’ interaction term in random-intercept repeated-measures regression models with both selenium and lipids treated as time-dependent variables. This method takes into account the intra-individual correlation of repeated measurements in the same participant. To avoid multicollinearity, we analysed total cholesterol, HDL cholesterol and LDL cholesterol in separate models using maximum likelihood method. To examine the effect of postfertilization selenium concentrations on subsequent lipid concentrations, we performed random-intercept regression with selenium in 1986 (postfertilization) treated as a fixed variable and serum lipids in 1986, 2001 and 2007 as time-dependent outcome variables.
All analyses were computed using sas, version 9.1 (SAS Institute, Cary, USA), or stata, version 11.0 (Stata Corp, College Station, TX, USA). All reported P-values are two sided and have not been adjusted for multiple comparisons. Statistical significance was inferred at a two-tailed P < 0.05.
In the total cohort of 1235 Finns, the mean (SD) serum selenium concentration at the baseline examination was 74.2 (13.9) ng mL−1, which was somewhat higher than that reported in previous studies from Finland [19–20].
Table 1 shows age- and sex-adjusted means/percentages for cardiovascular risk factors by baseline serum selenium quartiles. Higher selenium concentrations were associated with older age, with no sex difference across selenium quartiles. Levels of serum lipids (except triglycerides) increased linearly across increasing selenium quartiles (P < 0.05), whereas regular smoking was inversely related to selenium status. BMI and blood pressure were not associated with selenium concentrations. The baseline cross-sectional associations between selenium and total, HDL and LDL cholesterol persisted after multivariable adjustment for age, sex, BMI, smoking and dichotomous parental SES (Table 2).
Table 1. Baseline characteristics of the study population – The Young Finns Study (1980–2007)
|Selenium, ng mL−1||1235||74.2 (13.9)||57.1 (0.3)||69.3 (0.3)||78.1 (0.3)||92.3 (0.3)|| |
|Age, y||1235||10.7 (5.1)||8.7 (0.3)||10.1 (0.3)||11.1 (0.3)||12.7 (0.3)||<0.0001|
|Sex, % men||1235||46.2||44.1||47.9||44.3||48.5||0.74|
|BMI, kg m−2||1235||17.9 (3.2)||18.1 (0.1)||17.8 (0.1)||17.9 (0.1)||18.0 (0.1)||0.27|
|Systolic blood pressure, mm Hg||1226||113.0 (12.6)||112.7 (0.6)||113.6 (0.6)||113.3 (0.6)||112.2 (0.6)||0.38|
|Total cholesterol, mmol L−1||1235||5.26 (0.87)||5.12 (0.05)||5.20 (0.05)||5.31 (0.05)||5.42 (0.05)||0.0001|
|HDL cholesterol, mmol L−1||1227||1.48 (0.29)||1.43 (0.02)||1.48 (0.02)||1.50 (0.02)||1.51 (0.02)||0.006|
|LDL cholesterol, mmol L−1||1227||3.33 (0.73)||3.24 (0.04)||3.27 (0.04)||3.35 (0.04)||3.45 (0.04)||0.002|
|Triglycerides, mmol L−1||1235||0.52 (0.39–0.70)b||0.53c||0.53c||0.52c||0.51c||0.50|
|Regular smoking, %||1235||12.8||9.9||9.4||4.8||4.1||0.004|
|Parental SES, % of manual||1235||38.0||36.2||37.0||36.3||42.5||0.34|
|Parental diabetes, %||1231||2.2||2.6||2.4||2.8||1.0||0.46|
|Parental hypertension, %||1224||12.7||13.2||14.1||10.4||13.2||0.54|
Table 2. Multivariable-adjusted mean values for blood lipids by baseline selenium categories. The Young Finns Study, baseline examination at prefertilization period (1980)
|Total cholesterol, mmol L−1||1203||5.14 (5.04 to 5.24)||5.19 (5.10 to 5.29)||5.32 (5.22 to 5.42)||5.42 (5.32 to 5.52)||0.10 (0.06 to 0.14)||<0.0001|
|HDL cholesterol, mmol L−1||1195||1.44 (1.41 to −1.47)||1.47 (1.44 to 1.50)||1.50 (1.47 to 1.53)||1.51 (1.48 to 1.54)||0.02 (0.01 to 0.03)||<0.0001|
|LDL cholesterol, mmol L−1||1195||3.25 (3.17 to 3.34)||3.27 (3.19 to 3.35)||3.36 (3.27 to 3.44)||3.45 (3.37 to 3.54)||0.07 (0.04 to 0.10)||0.002|
A sensitivity analysis with adjustment for age, sex, smoking, BMI and a composite measure of parental SES repeated these findings: each 10 ng mL−1 increase in selenium was associated with 0.09 mmol L−1 (95% CI: 0.05 to 0.13, P < 0.0001) increase in total cholesterol (n = 1003), 0.02 mmol L−1 (95% CI 0.006 to 0.03, P = 0.005) increase in HDL cholesterol (n = 997) and 0.07 mmol L−1 (95% CI: 0.03 to 0.10, P = 0.0001) increase in LDL cholesterol (n = 997).
In the 262 participants with serum selenium levels measured at baseline (1980) and at the 6-year follow-up (1986), serum selenium concentrations increased substantially over this period (74.3 ± 14.0 ng mL−1 vs. 106.6 ± 12.5 ng mL−1). In contrast to selenium, there was little change in serum lipids: 5.20 ± 0.85 vs. 5.00 ± 0.93 for total cholesterol, 1.46 ± 0.28 vs. 1.52 ± 0.29 for HDL cholesterol and 3.29 ± 0.71 vs. 3.06 ± 0.86 for LDL cholesterol. The Pearson correlation coefficients between 1980 and 1986 levels were 0.23, 0.55, 0.55 and 0.61 for selenium, total, HDL and LDL cholesterol, respectively (all P < 0.001).
Table 3 presents cross-sectional associations between serum selenium and serum lipids at baseline (1980, prefortification) and 6-year follow-up (1986, postfortification) based on random-intercept repeated-measures analysis. Despite a major increase in selenium concentrations during this period, the cross-sectional associations of selenium with total cholesterol, HDL cholesterol and LDL cholesterol were little attenuated postfortification (P for time × selenium interaction 0.88 for total cholesterol, 0.77 for HDL cholesterol and 0.92 for LDL cholesterol). Thus, each 10 ng mL−1 increase in serum selenium was associated with a 0.12 mmol L−1 increase in total cholesterol, 0.02 mmol L−1 in HDL cholesterol and 0.09 mmol L−1 in LDL cholesterol at prefortification assessment and 0.11, 0.02 and 0.08 mmol L−1 increases at postfortification assessment (for the full model, see Table S1).
Table 3. Age- and sex-adjusted cross-sectional associations between serum selenium and serum lipids measured before (1980) and after (1986) selenium fertilization. Random-intercept repeated-measures analysis
|Total cholesterol, mmol L−1|
(N = 262; 522 observations)
| Prefortification (1980)||0.12 (0.05 to 0.19)||<0.0001|
| Postfortification (1986)||0.11 (0.04 to 0.19)||0.003|
|HDL cholesterol, mmol L−1|
(N = 262; 521 observations)
| Prefortification (1980)||0.02 (0.00 to 0.04)||0.03|
| Postfortification (1986)||0.02 (−0.00 to 0.05)||0.09|
|LDL cholesterol, mmol L−1|
(N = 262; 518 observations)
| Prefortification (1980)||0.09 (0.03 to 0.14)||0.003|
| Postfortification (1986)||0.08 (0.01 to 0.15)||0.02|
The within-participant change in serum selenium was 32.3 ng mL−1 (95% CI 30.3 to 34.3, P < 0.0001) between 1980 and 1986. Assuming that the cross-sectional associations between selenium and lipids in Table 3 represented a causal effect, the increase in lipid levels that would have been expected in this population by an increase in selenium intake of that size is 0.37 mmol L−1 for total cholesterol, 0.06 mmol L−1 for HDL cholesterol and 0.27 mmol L−1 for LDL cholesterol. The observed within-participant changes were −0.20 mmol L−1 (95% CI: −0.30 to −0.10, P < 0.0001) for total cholesterol, 0.06 mmol L−1 (95% CI: 0.03 to 0.10, P < 0.0001) for HDL cholesterol and −0.23 mmol L−1 (95% CI: −0.31 to −0.14, P < 0.0001) for LDL cholesterol.
Table 4 shows results from random-intercept repeated-measures analysis for serum lipid levels during the 21-year follow-up at 1986, 2001 and 2007 as the outcome and serum selenium in 1986 as a time invariant predictor. We found no association of postfertilization selenium in 1986 with total, HDL or LDL cholesterol at any of the three measurements (full model in the Table S2).
Table 4. Age- and sex-adjusted beta coefficient for selenium per 10 ng mL−1 in 1986 (postfertilization, max N = 262) in relation to serum lipids in 1986, 2001 (max N = 234) and 2007 (max N = 208) as the outcome. Random-intercept repeated-measures analysis
|Total cholesterol, mmol L−1|
(N = 262; 702 observations)
| 1986||0.09 (−0.01 to 0.18)||0.06|
| 2001||0.01 (−0.09 to 0.11)||0.89|
| 2007||0.06 (−0.04 to 0.16)||0.26|
|HDL cholesterol, mmol L−1|
(N = 262; 702 observations)
| 1986||0.02 (−0.01 to 0.04)||0.27|
| 2001||−0.01 (−0.04 to 0.02)||0.59|
| 2007||0.01 (−0.02 to 0.04)||0.47|
|LDL cholesterol, mmol L−1|
(N = 262; 692 observations)
| 1986||0.05 (−0.03 to 0.13)||0.21|
| 2001||0.02 (−0.07 to 0.11)||0.64|
| 2007||0.05 (−0.04 to 0.14)||0.28|
In this analysis, we examined cross-sectional and longitudinal associations of selenium status with cardiovascular risk factors amongst participants of the Young Finns Study, who were followed up for an average of 27 years from childhood to adulthood. In agreement with previous studies [13,15–16], we found positive cross-sectional associations between selenium status and serum lipids in our data. However, a marked increase in serum selenium after the implementation of the nationwide fortification programme was not associated with parallel increases in lipid levels. We can speculate that if the cross-sectional associations between selenium and lipid levels had been causal, we would have expected an increase of 0.37 and 0.27 mmol L−1 in total and LDL cholesterol levels, respectively, between 1980 and 1986. By contrast, we observed decreases of 0.20 and 0.23 mmol L−1 in these lipids, respectively. The magnitude of the within-person change in selenium concentrations between 1980 and 1986 was not associated with the change in lipid levels during that period. Furthermore, selenium concentrations after the nationwide fortification programme in 1986 were not predictive of any of the investigated lipid levels during the 21-year follow-up period. These results do not support a lipid-raising effect of selenium, although we cannot exclude the possibility that downward secular trends in lipid levels in Finland may have modified the effects of selenium fortification .
Fortification of virtually all multi-nutrient fertilizers with 16 mg selenium per kg as sodium selenate was started in Finland in 1985 to raise the selenium intake of the entire population . The mean selenium intake per capita increased from 38 μg per day in 1984 to a maximum level of 120 μg per day in 1990–1991. Thereafter, three amendments to the amount of selenium in fertilizers were made: from 1990, 6 mg selenium per kg, from 1998, 10 mg selenium per kg and the latest, 15 mg selenium per kg. Thus, the mean dietary intake of selenium varied from 60 to 80 μg per day between 1994 and 2009. The mean serum selenium concentration of adults reflected the changes in intake, varying from 90 to 110 ng mL−1 over this period . Compared to prefertilization levels, selenium intake increased by an average of 70%, and serum selenium increased by an average of 40%.
Cross-sectional studies have consistently identified strong, graded, positive associations between selenium status and blood lipids in different populations and across a very wide range of selenium concentrations worldwide [13,15–16,26–31]. To our knowledge, this is the first population-based study that attempted to address the longitudinal association of selenium status with cardiovascular risk factors from childhood to adulthood. A strength of the study is that we enrolled participants randomly selected from a community-wide population, alongside high standardization of data collection. Furthermore, we had an opportunity to exploit a natural experiment (nationwide fortification programme) relating to a population-level increase of ∼30 ng mL−1 (40%) in serum selenium.
Recent evidence from observational studies and randomized clinical trials from the United States has raised concern about possible associations between high selenium exposure and adverse cardio-metabolic effects, primarily an increased risk of diabetes [9–15]. Given the increasing use of selenium-enriched foods, supplements and fertilizers in many countries [21,32–34], these findings have called for a thorough evaluation of potential risks and benefits associated with selenium status, across a wide range of concentrations. Indeed, health benefits of additional selenium intake above the physiological range for optimal selenoprotein activity  are still questionable. Furthermore, recent findings from the Selenium and Vitamin E Cancer Prevention Trial (SELECT), in the United States, do not corroborate previous expectations of selenium as a potential chemopreventive agent in the Northern American population .
In agreement with our longitudinal results on serum lipids, the few randomized trials that examined the effect of selenium supplementation, alone or in combination with other micronutrients, on cardiovascular disease or mortality end-points have yielded null findings. [3,36–37] With regard to the effect of selenium supplementation on lipid profiles, two small trials conducted in Finland and China found no significant differences between treatment groups [38–39]. However, recent findings from the UK PRECISE Pilot trial showed selenium supplementation to be associated with a mildly beneficial effect on the total-to-HDL cholesterol ratio in a relatively low selenium population .
Potential mechanisms that might explain consistent cross-sectional associations of selenium with blood lipids, as observed in this and previous reports [13,15–16,26–31], are unclear at the present time, although a number of pathways involving selenium or selenoproteins are known to interact with both lipids and lipoproteins [3,41–43]. Altogether, the available cross-sectional evidence linking higher selenium status to blood lipids is unable to determine whether lipid levels rise as a consequence of increased selenium intake (not supported by our longitudinal analyses nor by data from the UK Pilot study ) or whether a common metabolic pathway, or common co-exposures, or reverse causality might explain the association between selenium status and lipid levels .
Several limitations need to be considered when interpreting our findings. First, without an unfortified control group, we cannot exclude the possibility that an overall declining trend in lipid levels, as observed in Finland , could partially mask the effect of increasing selenium status. Second, the amount and proportion of selenium in fertilizers and thus fortified foods changed over follow-up , which means that our measurement of postfortification selenium levels may not be representative of long-term follow-up. As a consequence, a null result cannot be directly interpreted as a lack of effect of selenium on lipids. Third, the six-year follow-up covered a period of adolescence when many participants underwent physical changes related to puberty. Puberty is known to affect lipid concentrations and could therefore obscure the association between selenium and subsequent lipid levels . Fourth, forty per cent of the original cohort was lost during the 27-year follow-up. However, we obtained very similar results with all available data and with a restricted set of participants with selenium measured at baseline (1980) and at the 6-year follow-up (1986), indicating that major bias is unlikely. Finally, the study population was limited to persons of white European ancestry, thus limiting the generalizability of our findings.
Potential differences in the effects, either beneficial or detrimental, of selenium on chronic disease end-points might be explained by the variability of selenium status and selenium dietary intakes across different countries and population subgroups [1–3]. In this view, it is likely that the association between selenium and cardio-metabolic outcomes is U-shaped with potential harm occurring at selenium levels both below and above the physiological range for optimal activity of selenoproteins [3, 45–46]. Whilst our results suggest that selenium fortification did not contribute to increased lipid levels in Finland, further epidemiological and randomized studies should be conducted to investigate the link between selenium and lipids and cardiovascular risk factors in more detail across different ranges of concentration. This would help determine the optimal level of selenium intake in the general population that maximizes the health benefits whilst avoiding potential subclinical toxic effects.
Conflict of interest statement
This study was financially supported by the Academy of Finland (grant no. 77841, 117832, 201888, 121584), the Social Insurance Institution of Finland, Turku University Foundation, Kuopio, Tampere and Turku University Hospital Medical Funds, Emil Aaltonen Foundation, Juho Vainio Foundation, Finnish Foundation of Cardiovascular Research and Finnish Cultural Foundation. MK is supported by the National Heart, Lung, and Blood Institute (R01HL036310-20A2), the National Institute on Aging (R01AG034454), US, the BUPA Foundation, UK, and the Academy of Finland. The sponsors had no role in preparing the manuscript.
We thank Ms Jenny Head for her help in mixed effect modelling. Ville Aalto and Irina Lisinen are acknowledged by skilful data management and/or statistical analysis in the Young Finns Study.