The research team, led by Tong Gong and Yi Lu of Karolinska Institutet, drew on the Child and Adolescent Twin Study in Sweden, a national cohort that has tracked participants at ages 9, 15, 18 and 24. Because the cohort was linked to national patient registers and prescription drug records, the researchers could examine not only self-reported symptoms of depression and anxiety but also clinically diagnosed disorders and antidepressant dispensing over follow-up periods extending more than a decade. This combination of symptom scales and hard clinical endpoints is rare in screen time research, which has often relied on small samples and coarse self-reports.

The headline results were striking at the population level. At ages 15 and 18, screen time showed a clear dose-response pattern with internalizing symptoms: the more hours per day young people spent on screens, the higher their depression and anxiety scores. Comparing heavy users, defined as those spending more than six hours per day, with restrictive users at two hours or less, the standardized differences in symptom scores ranged from 0.20 to 0.38 in the full sample. The associations were markedly stronger in girls than in boys for interactive activities such as gaming, chatting and browsing, with heavy use corresponding to symptom elevations of 0.37 to 0.51 standard deviations in girls, compared with 0.19 to 0.22 in boys.

The clinical picture was even more dramatic. Heavy interactive screen use at age 15 was associated with a roughly two- to threefold increased risk of being diagnosed with a depressive or anxiety disorder by age 28, with weekday heavy use showing an adjusted hazard ratio of 3.37. Even daily gaming at age 9, reported by parents, was linked to a 38 percent higher risk of clinical outcomes through age 22. Passive activities such as watching television, videos and DVDs showed far weaker and mostly non-significant associations, suggesting that what young people do on screens may matter as much as how long they spend there.

But the study’s most consequential findings emerged when the researchers turned to the twin structure of their data. Monozygotic twins share essentially all of their inherited genes, while dizygotic twins share on average half, and both types share their rearing environment. By comparing one twin who used screens heavily with their co-twin who did not, the co-twin control design effectively removes confounding from shared genetics and family background, factors that ordinary observational studies cannot measure. If the screen-mental health link were purely causal, it should persist within identical twin pairs.

It did not, at least not robustly. Within discordant dizygotic pairs, effect estimates attenuated modestly. Within discordant monozygotic pairs, the attenuation was far more pronounced: at age 15, estimates shrank by 32 to 95 percent relative to the full sample and were no longer statistically significant after correction for multiple testing. At age 18, the corresponding estimates were attenuated by 16 to 45 percent. In the longitudinal analyses, the strong cross-age associations between heavy interactive use at 15 and symptoms at 18, and between use at 18 and symptoms at 24, were reduced by roughly 30 to 60 percent within twin pairs and fell below the significance threshold. The pattern of progressive but incomplete attenuation is consistent with a mixture of genetic confounding, shared environmental confounding and possibly a small residual direct effect.

The clinical outcomes told a similar story. The threefold risk elevation associated with heavy weekday interactive use at age 15 dropped by 57 to 59 percent within discordant twin pairs, leaving a point estimate of a 65 percent increased risk that no longer reached significance after multiple-testing correction. Daily gaming at age 9, which showed a significant hazard ratio of 1.38 in the full sample, showed no comparable association within discordant pairs. The authors interpret this cautiously: the analyses had limited statistical power because informative discordant pairs are few, so small direct effects cannot be ruled out entirely, but the evidence for a clear causal effect of screen time on internalizing problems is weak.

The study also probed directionality, a persistent puzzle in this field. Screen use at age 9 did not predict symptoms at 15, but heavy interactive use at 15 and 18 predicted symptoms at 18 and 24 respectively. Intriguingly, the reverse pathway also appeared: adolescents with higher depressive and anxiety symptoms at 15 showed a small but measurable increase in interactive screen use at 18, with an odds ratio of 1.04 per standard deviation of symptoms. Although modest, this reverse association suggests that vulnerable young people may gravitate toward heavier digital engagement, potentially encountering more negative social comparison and mood impacts, a dynamic documented in recent studies of adolescents with mental health conditions.

The findings carry important implications for policy. Public health agencies in several countries have issued guidelines limiting recreational screen time for children and adolescents, but these rest largely on non-causal observational evidence. The new results align with earlier genetically informed work showing that screen behaviors themselves are moderately heritable, with twin-based estimates of 20 to 49 percent, and that genetic confounding accounts for a substantial share of the apparent association between media use and mental health. They also echo the influential analyses by Orben and Przybylski, who found that per-hour effects of screen use on adolescent well-being are small; the co-twin estimates here correspond to roughly 0.03 to 0.05 standard deviations of additional symptoms per extra hour of interactive use at age 18.

Yet the authors caution against dismissing the issue entirely. Within age-18 identical twin pairs, weekend heavy interactive use was still associated with 0.26 standard deviations of additional symptoms compared with restrictive use, an effect size comparable to established risk factors such as bullying victimization and childhood maltreatment, which typically fall in the 0.2 to 0.3 range. The study also has limitations: data collection began in 2004, before smartphones and social media became ubiquitous; the crude categories of interactive and passive use do not capture the complexity of modern digital life; and self-reported screen time is known to be underreported. Sensitivity analyses adjusting for physical activity and bullying victimization left the main findings essentially unchanged. The researchers conclude that heavy recreational screen use remains a legitimate public health concern, particularly during adolescence, but that shared genetic and familial factors partly explain the link to depression and anxiety, and that future research should focus on why and how different types of digital engagement affect vulnerable young people, rather than on screen time alone.

Subject of Research: Longitudinal twin study of recreational screen use and internalizing problems from preadolescence to young adulthood

Article Title: Recreational screen use and internalizing problems from preadolescence to young adulthood in a longitudinal twin cohort with co-twin comparison

Article References: Gong, T., Frans, E., Ohlis, A., Yao, S., Zhang, R., Nilsson, A., Molero, Y., Mosing, M. A., Brikell, I., Larsson, H., Lichtenstein, P., Thorell, L. B., Magnusson, P. K., Kuja-Halkola, R., & Lu, Y. (2026). Recreational screen use and internalizing problems from preadolescence to young adulthood in a longitudinal twin cohort with co-twin comparison. Nature Human Behaviour. https://doi.org/10.1038/s41562-026-02607-0

Image Credits: AI Generated

DOI: 10.1038/s41562-026-02607-0

Keywords: screen time, mental health, depression, anxiety, twin study, co-twin control design, adolescence, genetic confounding, internalizing problems, gaming, Nature Human Behaviour, epidemiology

The study, led by Marianne Nygaard of the University of Southern Denmark together with colleagues at the University of Bristol and other institutions, drew on the Danish GEMINAKAR cohort, which was established between 1997 and 2000 through the population-based Danish Twin Registry. Participants, aged 18 to 67 at baseline, underwent detailed examinations including fasting blood sampling, and were followed up more than a decade later, in 2010 to 2012. From this resource, the team assembled 132 complete monozygotic twin pairs with food frequency questionnaire data, genome-wide DNA methylation measurements, and covariate information at both time points, yielding a dataset of 264 individuals tracked across roughly twelve years of adult life.

The technical machinery of the study reflects the current state of the art in epigenetic aging research. DNA was extracted from whole blood, bisulfite converted, and profiled on Illumina Infinium MethylationEPIC BeadChips, with stringent quality control and fixed effects functional normalization to minimize batch artifacts. From these methylation data, the researchers computed principal component-based versions of four widely used epigenetic clocks, PCHannum, PCHorvath1, PCPhenoAge, and PCGrimAge, along with DunedinPACE, a measure trained on longitudinal data that estimates the pace of aging rather than a static biological age. Age acceleration measures were derived as the residuals of each clock regressed on chronological age, and all outcomes were standardized to allow comparison across clocks.

Dietary intake was assessed with validated semiquantitative food frequency questionnaires covering 192 foods and recipes, originally developed for the Danish Diet, Cancer and Health study. Adherence to national dietary guidance was quantified with the Danish Dietary Index, a score from zero to six built on the official Danish food-based dietary guidelines and the Nordic Nutrient Recommendations. The index rewarded higher intakes of fruits and vegetables, fish, whole grains, and fiber, and penalized high intakes of meat, saturated fat, and added sugar. Each component was scored as the ratio of actual to recommended intake, with a score of one indicating full compliance for that item.

The design’s decisive feature was the co-twin control approach. Monozygotic twins share practically all of their segregating DNA and, in most cases, a highly similar childhood environment. When twins within a pair differ in their diet scores, those differences cannot be explained by their genes or by the family they grew up in. Comparing individuals across the whole sample captures both the effect of diet and everything else that makes healthy eaters different, including inherited tendencies. Comparing twins within pairs strips away the familial confounding. Because both dietary habits and epigenetic aging are themselves partly heritable, the authors argued that genetic confounding was a plausible and testable threat to the observational literature.

The initial results looked reassuringly familiar. In cross-sectional, individual-level analyses at baseline, every beta coefficient was negative, meaning that people who adhered more closely to the dietary guidelines showed decelerated epigenetic aging. The signal was strongest for the second-generation clocks: each one-unit increase in the Danish Dietary Index was associated with roughly 0.55 to 0.67 years lower PCGrimAge and a 3.1 percent lower pace of aging on DunedinPACE, corresponding to standardized effect sizes of about 0.16 to 0.23 standard deviations. The same directional tendency appeared in the follow-up data, echoing previous cross-sectional studies that linked higher diet quality to younger epigenetic ages, particularly for clocks trained to predict morbidity and mortality rather than chronological age.

Then the picture changed. When the same association was tested within twin pairs, the strong signals for PCGrimAge and DunedinPACE attenuated dramatically, leaving no robust association between dietary adherence and epigenetic age acceleration. The implication is uncomfortable: the individuals who eat well may simply be people whose genes and upbringing favor both better diets and slower biological aging, rather than people whose diets are actively slowing their clocks. Because the cohort contained only monozygotic pairs, the study could not disentangle genetic confounding from shared environmental confounding, but either way, the familial factors appear sufficient to explain much of the apparent benefit.

The longitudinal analyses added a further twist. When the researchers modeled the change in diet score from baseline to follow-up against the change in epigenetic age measures, the estimates were primarily positive, unexpectedly suggesting that improvements in dietary adherence were associated with increased, not decreased, age acceleration, although the confidence intervals were wide and the authors caution against overinterpreting this pattern. Subgroup analyses that might have clarified the result, for example comparing people who stayed at low diet scores with those who improved from low to high, were impossible because the relevant subgroups contained too few participants, around seventeen individuals, to support meaningful inference.

The authors are careful about what their findings do and do not show. The study remains observational, so reverse causality and confounding by non-shared factors cannot be excluded. Blood cell composition was deliberately not adjusted for, since doing so could remove part of the biological signal of interest, meaning that variation in cell mixture may contribute to the results. Dietary data were self-reported and retrospectively collected, which may attenuate estimates, particularly within twin pairs where the differences between co-twins are small and measurement error from both twins compounds. The sample, drawn from a cohort selected to be free of diabetes and cardiovascular disease at baseline, is healthier than the general population and relatively small, limiting statistical power and generalizability. Nearly half of eligible monozygotic pairs were excluded for missing data, although participants and non-participants were broadly comparable.

Even so, the study’s strengths are substantial, and its conclusion is a sobering one for the epigenetic anti-aging boom. No strong independent association between dietary adherence and epigenetic aging survived the twin comparison, and the authors conclude that apparent benefits of healthy eating on epigenetic aging may partly reflect genetic and shared environmental factors rather than diet itself. This does not mean diet is irrelevant to aging; randomized trials of dietary interventions have reported slowing of Horvath clock and GrimAge measures, and the broader evidence that high dietary quality reduces cardiovascular disease, cancer, and all-cause mortality remains intact. What the twin data challenge is the specific causal story written into methylation-based age estimates. As epigenetic clocks move from research labs toward commercial wellness products, the Danish twins offer a reminder that correlation, even molecular correlation measured across hundreds of thousands of CpG sites, is not automatically causation, and that the family you come from may shape both your plate and your methylation patterns in ways no questionnaire can fully untangle.

Subject of Research: The association between adherence to Danish dietary guidelines and epigenetic aging measures in monozygotic twins

Article Title: Dietary pattern and epigenetic aging: a longitudinal twin study

Article References: Nygaard, M., Elliott, H. R., Smith, G. D., Relton, C., Kyvik, K. O., & Dalgård, C. (2026). Dietary pattern and epigenetic aging: a longitudinal twin study. GeroScience. https://doi.org/10.1007/s11357-026-02550-y

Image Credits: AI Generated

DOI: 10.1007/s11357-026-02550-y

Keywords: epigenetic aging, DNA methylation, epigenetic clocks, twin study, dietary patterns, Danish Dietary Index, DunedinPACE, GrimAge, co-twin control design, confounding, GeroScience, longitudinal study