Secular trends in risk factors for adolescent anxiety and depression symptoms: the Young-HUNT studies 1995–2019, Norway

Data were obtained from more than 25,000 participants aged 13–19 years in three waves of the Young-HUNT (YH) study, Norway: YH1 in 1995–97 (n = 9140, 89.6% response rate); YH3 in 2006–08 (n = 8203, 78.4% response rate); and YH4 in 2017–19 (n = 8124, 75% response rate). The surveys included self-reported questionnaires covering a wide range of aspects of physical and mental health, quality of life, lifestyle, and behavioural factors. All detailed instruments, variables, and response options are on the webpage for the HUNT Research Centre (https://www.ntnu.edu/hunt).

We calculated descriptive statistics as frequencies, percentages, and averages. We specified two types of regression models to study associations between the exposures and HSCL-5.

First, linear regression models (OLS) were used to estimate absolute differences (beta coefficients, B) in (standardized) mean HSCL-5 between exposure categories, and we report 95% confidence intervals based on 2000 bootstrapped samples. Second, generalized linear models (GLMs) with a gamma distribution and a log-link function were used to estimate differences in (log of) means in HSCL-5 between exposure categories. Coefficients were exponentiated and is interpreted as a relative measure of difference (relative risk, RR). Estimates are reported with 95 percent confidence intervals (95% CI). The RR represents the ratio between means, whereas the beta coefficients represent the difference between standardized mean HSCL across exposure categories. To test for linear trends across survey waves, interaction terms between exposures and survey waves were specified. Analyses were conducted for each sex separately and we adjusted for continuous age centred at age 16 in each wave. Exposures that showed statistically significant changes (1995–2019) for both absolute and relative differences were included in the Discussion. All data management and analyses were conducted using Stata version 17 (Statacorp, 2021).

Overall, we found an increasing trend of several well-known risk factors from YH1 to YH4. For girls, the most prominent were fatigue (50.6–64.8%), bullying (1.9–6.7%), loneliness (7.8–14.3%), musculoskeletal pain and migraine (32.2–58.5%), overweight/obesity (20.3–28.6%), and GP visits (57.9% to 73.9%). In contrast, we observe a decrease in physical inactivity (77.7–67.6%) and excess alcohol use (27.2–17.1%).

For boys, the largest increases from YH1 to YH4 were found in bullying (2.2–8.4%), musculoskeletal pain and migraine (19.4–41.1%), overweight/obesity (22.0–34.1%), and GP visits (45.2–62.2%). As in girls, a decrease in excess alcohol use was also observed in boys (30.8–19.0%).

In contrast, self-report of poor self-perceived health, perceived family economy, low family cohesion, Adverse Life Events, and sleep problems remained relatively stable from YH1 to YH4, in both boys and girls.

Table 2 shows absolute differences (beta coefficients) in HSCL-5 for individual, relational, and contextual risk factors adjusted for age. For girls, the absolute difference in standardized HSCL between those reporting poor self-rated health compared with those reporting good self-rated health was 0.58 in YH1 which increased to 1.11 in YH4. Differences according to self-reported fatigue (0.58–1.05), musculoskeletal pain/migraine (0.42–0.76), physical inactivity (0.05–0.38), sleep problems (0.49–0.79), overweight/obesity (0.01–0.09), bullying (0.70–1.03), loneliness (1.18–1.59), GP visits (0.10–0.36), low family cohesion (0.64–0.75), and Adverse Life Events (0.32–0.47) were also increased.

For boys, we found evidence for small changes in absolute differences for fatigue (0.47–0.64), musculoskeletal pain/migraine (0.37–0.48), physical inactivity (0.05–0.14), sleep problems (0.39–0.54), GP visits (0.12–0.27), and low family cohesion (0.37–0.49). In contrast to other risk factors, we observed decreasing differences according to bullying and mental health problems from YH1 to YH4 (0.84–0.41).

The absolute differences were higher in all waves for all risk factors in girls, except for overweight/obesity, bullying, and GP visits in YH1. Differences were generally higher for all risk factors in all YH waves for girls, except for bullying, GP visits, physical inactivity, and loneliness in YH1.

Results for the relative differences (Relative Risk, RR) are displayed in supplementary Table 3. For girls, those reporting poor self-rated health had 1.23 times higher HSCL than those reporting good self-rated health in YH1. In YH4, the relative risk increased to 1.36. Similarly, we observed increased relative differences in HSCL from YH1 to YH4 in fatigue (1.25–1.41), musculoskeletal pain (1.17–1.27), physical inactivity (1.02–1.13), sleep problems (1.21–1.29), and GP visits (1.04–1.12).

For boys, smaller increases from YH1 to YH4 was observed for fatigue (1.23–1.30), physical inactivity (1.02–1.13), sleep problems (1.19–1.25), and GP visit(s) (1.06–1.12) were observed. On the contrary, reported bullying declined (1.38–1.18).

Loneliness showed the highest relative risk in all YH waves for both sexes, but remained stable across waves for both girls (1.47–1.52) and boys (1.53–1.60).

As proposed by Geoffrey Rose [43], even relatively small increases in the prevalence of risk factors can lead to substantial effects on the population risk factor distributions, particularly distributional changes that increase the prevalence of high-risk groups—making them of population health importance.

In keeping with other international studies [44, 45], we observe similar prevalences and gender differences in poor self-perceived health (10–20%) Further, our data indicate that, among girls, self-rated health is also more strongly associated with psychological distress in YH4 compared to YH1.

Fatigue and tiredness are commonly reported in the U.S and Europe [46, 47], and affect between 7 and 45% of the adolescent population [48]. Our results along with an Australian study [49] indicate an even higher prevalence of fatigue. The increase in fatigue, and the strength of association between fatigue and mental health problems, is more prominent in girls than boys, consistent with other studies [48, 50].

Musculoskeletal pain (MSP)/migraine are relatively common in adolescence and particularly in girls [51, 52]. We observe similar increases in MSP in both genders, yet a stronger association with mental health problem in girls compared to boys from 1995 to 2019.

Problems initiating sleep affect about one in four adolescents in Europe and Canada [53], and are more problematic among girls than boys [53, 54]. Further, in contrast to many European and Scandinavian countries [55, 56], our findings show a relatively stable prevalence of sleep problems over 2 decades in both sexes. Nevertheless, associations between sleep and mental health problems increased over time in our study.

In contrast to several European countries [57], our study indicates an increase in bullying in both sexes. However, bullying was relatively infrequent overall, and more prevalent in boys than girls [57, 58]. However, being bullied was increasingly related to symptoms of anxiety and depression from 1997 to 2019 in girls, while these associations decreased in boys.

Feeling of loneliness is now considered a public health problem [59], and prevalences range between 9 and 14% worldwide [60]. Some Scandinavian studies also suggest an increase in adolescent loneliness the last decades [61, 62], particularly in girls [63, 64], consistent with our findings. Further, we observe an increasing association between loneliness and mental health problems over time in both boys and girls.

Parallel to the increases in psychological distress, more adolescents had visited their GP from YH1 to YH4 in our study. Similar estimates have been reported in Australian and Irish adolescents [65, 66], and underlines worsening self-perceived health [67]. Although variations are expected due to differences in health care systems, a Canadian cross-sectional study found similar prevalence rates, but a decrease over time in GP visits and an increase over time for MHP (mental health professionals) visits [68]. Finally, according to our data, GP visits show stronger associations with anxiety and depression symptoms over time.

In sum, prominent risk factors are associated with trends in adolescent mental health, particularly in girls, and represent a cluster of modifiable individual, relational, and contextual health characteristics and behaviours.

Several methodological issues in our study are worth considering [69]. When it comes to selection bias, all three HUNT studies used similar designs and sampling, and were collected 10–12 years apart, and showed high participation rates, which improve the generalisability of these findings. Although the HUNT catchment area is confined to small towns and rural areas, it is fairly representative of Norway [70].

Measurement bias is also worth mentioning when investigating cross-sectional studies over time. The HSCL-5 (outcome) has been used in all three HUNT studies. However, over 20 years, societal and cultural changes may have influenced awareness surrounding mental health, and, in turn, responses to mental health questions. These changes influence willingness to report and therefore could lead to inflated estimates over time. This may also apply to adolescents’ responses to exposure variables. Missing values were low (< 5%) for near all variables in all three YH waves, with higher missing rates among boys (e.g., 6.9% for boys compared to 4.8% for girls on HSCL-5 in YH4). Higher missing values (5–10%) were observed for more sensitive variables in both genders (e.g., bullying and trauma). While many of the questions/variables originate from well-known instruments and scales, such as Health Behaviour in School-aged Children (HBSC), Interpersonal violence, brief lifetime trauma screen, and HSCL-5, this study also used some less well-validated questions. Although the latter were repeated across YH1-4, they often make direct comparison to other studies difficult. Further, the YH questionnaires were developed and adapted over 3 decades; therefore, there are some minor differences in some of the scales and response options on the exposure variables/covariates used from 1995 to 2019. This was considered carefully during the operationalization and categorization of the variables. Unfortunately, we did not have comparable data/instruments from all three YH studies for school-level variables, social media/online gaming, and behaviour/externalizing problems—factors of great interest and linked to mental health problems in adolescents.

To the best of our knowledge, this is one of the largest studies to study decennial trends in correlates and risk factors for adolescent mental health problems. Finally, in an exploratory study of many possible risk factors for anxiety and depression, it is not feasible to thoroughly describe trends for each factor. We argue that well-designed qualitative studies of changes in mental health stigma and awareness, and semantic issues related to mental health, are needed to further understand these phenomena.