Main

COVID-19 quickly spread globally at the beginning of 2020; on 11 March it was declared a global pandemic by the World Health Organization, which led to public health measures being implemented to mitigate the spread of COVID-19. Young adults are at low risk of being severely ill because of COVID-19, but it has been suggested that they are the most vulnerable to the collateral damages of lockdown. Several studies found an aggravation in the mental health of young adults during the initial lockdown, especially in young women1,2,3,4,5. However, the literature exploring whether this aggravation in mental health also manifested as changes in self-injury, suicidality and eating disorders (EDs) during lockdown is sparse and its findings are inconsistent. Most pre- and post-lockdown comparisons of self-injury and suicidality are based on health registries, which are compromised by the general reduction in patient contacts during lockdown and mostly refers to registrations of cases brought to hospital due to somatic injuries6. A Danish register-based study found a signal for increase in hospital-registered suicidal behaviour in young adults aged 18–29 years during the first lockdown but no change during the entire first year after the lockdown7. Register-based studies from other countries have documented a decrease in self-injury and suicide attempt among young adults8,9,10,11. Studies from Norway and Korea based on self-reported repeated cross-sectional data found no pre-to-post-lockdown change in suicide ideation and a decrease in suicide ideation and attempts, respectively12,13. On the other hand, a longitudinal study based on self-reported data in China demonstrated an increase in self-injury and suicide ideations and attempts during spring 2020 (ref. 14). An international register-based study, including preliminary suicide data from 21 countries in all age groups, showed no evidence of increased suicide rates during the first year of the pandemic15. However, other studies have suggested increased suicide rates during lockdown when including young adults only; preliminary and unvalidated data based on Danish registers revealed a signal of an increased number of suicides among young women aged 20–24 years16,17,18.

Regarding EDs, studies from the USA and Canada found that hospital admissions and new diagnosis for restrictive EDs were twice as high among adolescents during the first year of lockdown compared to previous years19,20,21. To our knowledge, no studies have compared pre- and post-pandemic self-reported data on EDs among young adults.

To mitigate the spread of COVID-19, the Danish government, like many other countries, implemented a national lockdown in March 2020 requiring a closure of schools, day-care centres, sport facilities, restaurants and shops; working from home was either mandatory or highly recommended in non-critical functions (Fig. 1)1,2,22. The restrictions were slowly lifted during spring 2020 but gradually reinforced during autumn. In December, a second national lockdown was declared. This second lockdown turned out to be more prolonged and was slowly lifted during spring 2021. In this study, we compare the proportion of young men and women reporting self-injury, suicide ideation and attempts and symptoms of EDs with similar pre- and post-lockdown data across the two national lockdowns in Denmark.

Fig. 1: Outline of the COVID-19 lockdowns in Denmark.
figure 1

The development of the COVID-19 pandemic, aligned with seminal events during the lockdown in Denmark, together with the timing of the data collections.

Results

Longitudinal data

Our study population were aged 19–23 years, with a median age of 20.9 years in spring 2021; more women than men responded in the Danish National Birth Cohort (DNBC) 18-year follow-up (DNBC-18) and wave 8 of the COVID-19 survey. Less than 10% were living without parents and 5% had another occupation than school at age 18 (Supplementary Table 1). In the analyses of the longitudinal data, the proportion of self-injury before lockdown was 13.9% (95% confidence interval (CI) = 13.4–14.4) and 5.7% (95% CI = 5.3–6.1) among women and men, respectively, and decreased post-lockdown in both women and men (Fig. 2a). The proportion reporting self-injury decreased with 5.7% points (95% CI = 6.6 to −4.8, P < 0.001, d.f. = 2) among women and 3.2% points (95% CI = −4.3 to −2.2, P < 0.001, d.f. = 2) among men (Fig. 3a). For suicide ideation, the absolute decrease in percentage points was similar; it was 7.4% points (95% CI = −8.7 to −6.0, P < 0.001, d.f. = 2) in women and 3.0% points (95% CI = −4.6 to −1.4, P = 0.002, d.f. = 2) in men. The pre-lockdown proportion of suicide ideation was 25.4% (95% CI = 24.7–26.1) and 18.4% (95% CI = 17.7–19.1), respectively (Figs. 2b and 3b). The pre-lockdown proportion of suicide attempts was 0.9% (95% CI = 0.8–1.0) in women and 0.5% (95% CI = 0.4–0.6) in men (Fig. 2c). For women, the proportion increased 0.1% points (95% CI = −0.2 to 0.4, P = 0.371, d.f. = 2), thus indicating that the data were compatible with no change. For men, the proportion increased 0.4% points (95% CI = 0.1–0.7, P = 0.019, d.f. = 2) but with wide confidence intervals due to the low absolute number of suicide attempts in men (Fig. 3c). Before lockdown, 14.1% (95% CI = 13.6–14.6) of women and 2.1% of men (95% CI = 1.8–2.4) reported symptoms of EDs within the last month (Fig. 2d). Post-lockdown these proportions decreased 2.3% points (95% CI = −3.2 to −1.4, P < 0.001, d.f. = 2) among women and 0.6% points (95% CI = −1.7 to 0.5, P = 0.258, d.f. = 2) among men (Fig. 3d). Thus, for men, the data were compatible with no change in symptoms of EDs. Bayes factors were calculated for the statistically non-significant results, that is, suicide attempts in women (Bayes factor = 0.021) and symptoms of EDs in men (Bayes factor = 0.10) and showed strong evidence of no effect of the lockdown (Table 1).

Fig. 2: Self-injury, suicide ideation and attempt and symptoms of EDs in men and women pre- and post-lockdown (longitudinal data).
figure 2

ad, Weighted proportions based on longitudinal data collected in the DNBC-18 (n = 27,441) and wave 8 (n = 7,597) of the COVID-19 survey, approximately 1 year after the initial lockdown stratified by sex. Data are presented as the mean (%) on a scale ranging from 0% to 100% ± s.e.m. a, Self-injury within the last year. b, Suicide ideation within the last year. c, Suicide attempt within the last year. d, Symptoms of EDs within the last month.

Fig. 3: Change in self-injury, suicide ideation and attempt and symptoms of EDs in men and women from before to during lockdown (longitudinal data).
figure 3

ad, Weighted fixed-effect linear regression estimating change in the proportions based on the longitudinal data collected in the DNBC-18 (n = 7,597) and wave 8 of the COVID-19 survey (n = 7,597), approximately 1 year after the initial lockdown. Data are presented as the mean change in percentage points ± s.e.m. a, Self-injury within the last year in men (P < 0.001, d.f. = 2) and women (P < 0.001, d.f. = 2). b, Suicide ideation within the last year in men (P = 0.002, d.f. = 2) and women (P < 0.001, d.f. = 2). c, Suicide attempt within the last year in men (P = 0.019, d.f. = 2) and women (P = 0.371, d.f. = 2). d, Symptoms of EDs within the last month in men (P = 0.258, d.f. = 2) and women (P < 0.001, d.f. = 2).

Table 1 Calculation of Bayes factors for non-significant results

Repeated cross-sectional data

In the analyses of the repeated cross-sectional data, women had higher proportions of self-injury, suicide ideation and symptoms of EDs than men. The calendar periods showed no clear pattern of self-injury, suicide ideation and attempt and symptoms of EDs, and there was no year-to-year variation of the measures (Fig. 4). The linear regression analyses did not suggest any change in the proportions of self-injury, suicide ideation and attempt and symptoms of EDs from pre- to post-lockdown in men or women (Fig. 5). Further, the estimated change for suicide attempt was similar in men −0.2% points (95% CI = −0.5 to 0.1, P = 0.197, d.f. = 8) and women −0.2% points (95% CI = −0.5 to 0.2, P = 0.306, d.f. = 8). The CIs for all measures were narrow, indicating that our data are most compatible with changes of no practical importance. Bayes factors for both self-injury (0.00020), suicide ideation (0.00015), suicide attempt (0.00011) and symptoms of EDs (0.000097) further supported strong evidence of no effect of the lockdown (Table 1).

Fig. 4: Self-injury, suicide ideation and attempt and symptoms of EDs in men and women pre- and post-lockdown (repeated cross-sectional data).
figure 4

ad, Weighted proportions in different lockdown-related periods based on repeated cross-sectional data collected in the DNBC-18 (n = 24,625) in 2018–2021 and stratified by sex. Periods defined as: P1, 1 January to 11 March/1 January to 1 March in 2021; P2, 12 March to 26 May; P3, 27 May to 9 August; P4, 10 August to 4 November; P5, 5 November to 31 December. Data are presented as the mean (%) on a scale ranging from 0% to 100% ± s.e.m. a, Self-injury within the last year. b, Suicide ideation within the last year. c, Suicide attempt within the last year. It was not possible to stratify by sex due to low frequency. d, Symptoms of EDs within the last month.

Fig. 5: Change in self-injury, suicide ideation and attempt and symptoms of EDs in men and women before and during lockdown (repeated cross-sectional data).
figure 5

ad, Weighted linear regression estimating change in the proportions based on the repeated cross-sectional data collected in the DNBC-18 (n = 24,625) in 2018–2021. Data are presented as the mean change in percentage points ± s.e.m. a, Self-injury within the last year in men (P = 0.374, d.f. = 8) and women (P = 0.076, d.f. = 8). b, Suicide ideation within the last year in men (P = 0.727, d.f. = 8) and women (P = 0.142, d.f. = 8). c, Suicide attempt within the last year in men (P = 0.197, d.f. = 8) and women (P = 0.306, d.f. = 8). d, Symptoms of EDs within the last month in men (P = 0.485, d.f. = 8) and women (P = 0.398, d.f. = 8).

Sensitivity analysis

In our analyses of the severity scale, quantifying frequency of self-injury, suicide ideation and symptoms of EDs, on the longitudinal data, we found a decline in self-injury before and after lockdown in both men (change estimate = −0.05, 95% CI = −0.07 to −0.03, P < 0.001, d.f. = 2) and women (change estimate = −0.10, 95% CI = −0.12 to −0.08, P < 0.001, d.f. = 2) (Extended Data Fig. 1). In women, we observed a decline in suicide ideation (change estimate = −0.08, 95% CI = −0.11 to −0.05, P < 0.001, d.f. = 2) and symptoms of EDs (change estimate = −0.03, 95% CI = −0.04 to −0.02, P < 0.001, d.f. = 2) while in men we observed no change in suicide ideation (change estimate = 0, 95% CI = −0.04 to 0.04, P = 0.878, d.f. = 2) or symptoms of EDs (change estimate = −0.01, 95% CI = −0.02 to 0, P = 0.306, d.f. = 2). Like our primary analyses, we did not find any changes in any of the severity scales from before to after lockdown in the repeated cross-sectional set-up (Extended Data Fig. 2). When leaving out participants who replied ‘do not know’ to having had self-injury and suicide ideation and attempt within the last year, respectively, the results did not change notably in either the longitudinal or repeated cross-sectional data (Extended Data Figs. 3 and 4). However, the increase in suicide attempt among men (change estimate = 0.26% points, 95% CI = −0.06 to 0.57, P = 0.116, d.f. = 2) was lower and compatible with no change. Our sensitivity analyses in the longitudinal data restricted to participants aged 19–20 years when completing wave 8, that is, completing the DNBC-18 in 2019 or early 2020, resulted in similar results for suicide ideation and symptoms of EDs after lockdown (Extended Data Fig. 5). The decrease in self-injury in men (change estimate = 2.46% points, 95% CI = −3.92 to −0.99, P = 0.001, d.f. = 2) and women (change estimate = 4.42% points, 95% CI = −5.2 to −3.22, P < 0.001, d.f. = 2) were slightly smaller and the exclusion decreased the precision so that data were compatible with no change in suicide attempt among both men and women. Further, restricting self-injury and suicide ideation to within 4 weeks instead of 1 year in the repeated cross-sectional data did not change the results (Extended Data Fig. 6).

Discussion

In this study with tandem use of longitudinal and repeated cross-sectional data, we observed that during the lockdowns, including a more prolonged second lockdown during the winter season, there was no increase in self-injury, suicide ideation or symptoms of EDs. If anything, our longitudinal data indicate a post-lockdown reduction in self-injury and suicide ideation among both men and women and smaller reductions for symptoms of EDs in women. Further, we found no indication that the severity of self-injury, suicide ideation and symptoms of EDs worsened during lockdown. For suicide attempt in men, we found weak indication of an increase in the longitudinal data while no change was observed for women or for either men or women in the repeated cross-sectional data.

Thus, our findings overall do not support that the aggravation in mental health documented in several countries after lockdown have yet resulted in young adults having a higher risk of self-injury, suicidality and ED symptoms. As outlined in the introduction, findings from previous studies are inconsistent. The mixed findings in the literature may result from methodological differences. Cross-sectional studies using health register data all found a decline in self-injury- and suicidal behaviour-related hospital contact during lockdown7,8,9,10,11. This reduction might in part be due to bias caused by the general reduction in healthcare use and it is important to note that self-injury and suicidality with hospital contact only covers the most severe cases. However, cross-sectional studies using self-reported data, including this study, also found either a reduction or no change in self-injury and suicide ideation and attempts during lockdown12,13. In contrast, a longitudinal study in China with self-reported data suggested an increase in both self-injury and suicide ideation and attempts during lockdown14. The longitudinal study only included 4 months of lockdown, had a small population (n = 1,241) and participants were aged 9–16 years, which could, together with differences in the lockdowns, explain why the findings differ from ours even though the methodology is very similar. We were not able to include suicides from the cause-of-death register to explore the potential increase in suicide among especially young women as suggested by preliminary unvalidated Danish data and current international literature15,16,17,18. Regarding EDs, previous studies used data on new ED diagnosis or hospital admissions, which increased during lockdown contrary to our findings19,20,21. Because there was a general reduction in healthcare use due to lockdown, this increase may even be understated but differences in healthcare systems and thereby the registers may also affect comparability between studies. In this study, we measured the prevalence of ED symptoms while register studies measure the incidence of new diagnoses. Further, these health register measures of EDs cover more severe cases compared to our self-reported symptoms of EDs. However, our sensitivity analyses, which included symptoms that fulfilled our definition of threshold EDs (weekly symptoms), likewise indicated no signal of an increase after lockdown (Extended Data Figs. 1c and 2c).

In general, the previous literature has varied with regard to follow-up time, origins from different countries, the course of the pandemic, the public health precautions, for example, the extent of lockdown; thus, the impact on self-injury, suicidality and EDs may be different. Further, even small age differences in study populations focusing on young adults may explain the inconsistent result. Importantly, our population mainly consisted of young adults studying and still living at their parental home and our findings cannot necessarily be generalized to either younger nor older populations, who are more on their own and therefore could be more vulnerable to the collateral damages of lockdown.

Possible reasons for the decrease in self-injury, suicide ideation and symptoms of EDs in the longitudinal data is that the social distancing actually has been beneficial for some young adults23. Being closer to their families, having more time and fewer obligations to activities such as sports, part-time jobs and parties, that is, generally a reduction in the pressure of living up to the social norms for young adults could have impacted mental health in a positive direction. Further, it is possible that the impact of lockdown has been positive in some groups while negative in other groups, which equalizes the impact. A similar DNBC study found slight interim deterioration in mental health in young adults without pre-existing depressive symptoms after lockdown while no differences were observed in young adults with pre-existing depressive symptoms1. Furthermore, the initial decline mental health quickly attenuated and may have been a shock effect that did not manifest as self-injury, suicidality and symptoms of EDs.

A strength of this study is the use of self-reported data that capture more subtle cases than register data and still provide relevant information for screening and prevention purposes24,25. Further, our study populations are from a large cohort consisting of relatively healthy and well-functioning young adults.

The results using longitudinal and repeated cross-sectional data were not completely consistent, although both suggested that the proportion of young adults with self-injury, suicide ideation and symptoms of EDs did not increase during lockdown. Different strengths and limitations in the study design could have resulted in the different results: (1) because the longitudinal data include a specific COVID-19 related survey, participants may have overstated or understated their answers in accordance with their feelings of lockdown making the comparison with the pre-lockdown measure biased. This is less of a concern in the repeated cross-sectional data only using the DNBC-18 because it was an ongoing survey that did not mention the COVID-19 pandemic in any way; (2) when using longitudinal data there is a risk of differential attrition because young adults with mental health problems are less likely to participate in follow-ups26. This may have resulted in bias as the observed decline in self-injury, suicide ideation and symptoms of EDs in wave 8 may be explained by loss to follow-up rather than the lockdown; however, we used sampling weight to limit this bias. Further, the indication of an increase in suicide attempts among men indicates that people with severe mental health problems also participated. The repeated cross-sectional data are only vulnerable to differential attrition if the participation in DNBC-18 systematically changed over year of birth. There were no notable differences in characteristics of participants completing the DNBC-18 in each year from 2018 to 2021 (Supplementary Table 1); (3) a major strength of the longitudinal data is that we analysed the same young adults pre- and post-lockdown and thereby all time-invariant factors are adjusted for. In the repeated cross-sectional data, we compared groups of young adults based on birth year and the results may reflect factors related to birth year rather than lockdown. However, the level of self-injury, suicidality and symptoms of EDs were stable during the entire pre-lockdown period and thus it is unlikely that a sudden change would have happened in the absence of lockdown; (4) a strength in comparing different young adults aged 18 years pre- and post-lockdown is that the age is adjusted for. In the longitudinal data, the pre-lockdown measures were collected at age 18 while the post-lockdown measures were collected at age 19–23. Thus, the time interval between pre- and post-measure varied; as many major events, such as graduating from high school, moving away from parents and starting to shape the future happens in this age range, the results may be biased. However, sensitivity analyses showed similar results when restricting the time interval to 2 years, indicating that the varying time interval and thereby age cannot explain the entire observed decrease in self-injury, suicide ideation and symptoms of EDs during lockdown (Extended Data Fig. 5). The sensitivity analyses showed no change in the proportion of suicide attempts among men after lockdown, which could be explained by data insensitivity as the frequency was very low; (5) the measures of self-injury and suicidality were defined as being within the last year. Thus, in the longitudinal set-up the measures were restricted to an entire year of lockdown and thereby adjusted for seasonal differences even though the pre-lockdown data were collected at different times of the year. In the repeated cross-sectional data, the post-lockdown measures will somewhat overlap the pre-lockdown period until the last lockdown period that covers an entire year of lockdown. However, analyses where self-injury and suicide ideation were restricted to being within 4 weeks instead of 1 year did not change the results (Extended Data Fig. 6). Symptoms of EDs were defined as being within the last month. Seasonal differences in ED symptoms could bias the results in the longitudinal data because the pre- and post-lockdown data were mainly collected at different calendar periods. However, the repeated cross-sections did not reveal any clear seasonality because the predefined lockdown-relevant periods represent different periods of the year; (6) the self-reported measures of self-injury, suicidality and symptoms of EDs are prone to misclassification. For self-injury and suicidality, this is not a great concern because the possible misclassification is assumed to be the same before and during lockdown. The participants replying ‘do not know’ to these measures may be misclassified as not having the behaviour. Our sensitivity analyses excluding these participants did not change the results. For symptoms of EDs, the before and during lockdown measures may not be completely comparable because the lockdown led to more meals being prepared and eaten at home and limited the exercise opportunities (closed fitness centres and sports activities), which is measured in our ED items. However, from the beginning of the pandemic the Danish health authorities had campaigns on the importance of continuing physical activity outdoor. Because our study population is older than the peak age of incident EDs, it may not be possible to generalize these results to younger age groups27. Further, because we measured prevalence rather than incidence in an age group with low incidence, our study population may not have been large enough to detect a possible increase. Self-reported symptoms of EDs are distinct from clinical diagnoses, which is why our frequency of EDs is higher than in studies using clinical diagnosis. Estimates indicate that only approximately 30% of people affected with EDs (fulfilling threshold diagnostic criteria) seek treatment; it may even be as low as 20% among adolescents25,28. Since substantial cross-over between threshold and subthreshold variants of EDs exists, it is a strength that we could distinct severity, that is, weekly versus monthly frequency of symptoms29.

In conclusion, this study suggests that during the lockdown, there was no increase in the proportion of young adults with self-injury, suicide ideation or symptoms of EDs. Findings from longitudinal analyses even indicate that the proportion of self-injury and suicide ideation decreased slightly post-lockdowns in both men and women, while a minor decrease in symptoms of EDs were observed only in women. An indication of increase in suicide attempts was observed only among men in the longitudinal data, while no indication of sex-specific changes was supported in the repeated cross-sectional analyses.

Method

Our research of human participants complies with all relevant ethical regulations. This study was approved by the Danish Data Protection Agency via a joint notification to the Faculty of Medicine and Health Sciences, University of Copenhagen (ref. no. 514-0497/20-3000, ‘Standing together at a distance: how are Danish National Birth Cohort participants experiencing the corona crisis?’). The cohort was approved by the Danish Data Protection Agency and the Committee on Health Research Ethics (case no. (KF) 01-471/94). Data handling in the DNBC has been approved by the Statens Serum Institut (SSI) under ref. no. 18/04608 and is covered by the general approval (Fællesanmeldelse) given to the SSI. The DNBC-18 was approved under ref. no. 2015-41-3961. DNBC participants were enrolled by written informed consent.

The DNBC

The DNBC is a nationwide cohort established in the mid-1990s, which includes about 30% of all children born in Denmark in 1996–2003 (ref. 30). Data from prenatal life through early adulthood has been collected with the latest collection being the 18-year follow-up (DNBC-18). DNBC-18 data collection started in 2016 and was completed in December 2021 when the last participant reached 18 years and 3 months, which was the age of invitation. Further information about the cohort and DNBC-18 is available at www.dnbc.dk. For our purpose, we used two different study populations; one with longitudinal data collected pre- and post-lockdown and one consisting of repeated cross-sections of lockdown-appropriate periods in 2020–2021, as well as similar periods in 2018–2019 (Fig. 1)1,2. To create sampling weights, both study populations were restricted to DNBC participants with information on the following characteristics collected during pregnancy of the participant’s mother: household and socio-occupational status; parity; and maternal smoking during pregnancy. Moreover, we used information on maternal age at childbirth when generating the sampling weights.

Longitudinal data

In April 2020, during the initial lockdown, a COVID-19 survey was launched and consisted of seven weekly waves, in which questions about self-injury, suicidality or symptoms of EDs were not included22. In April–May 2021, when the second national COVID-19 lockdown was gradually lifted, the DNBC invited participants to complete an additional wave, that is, wave 8 including questions about self-injury, suicidality and symptoms of EDs. All DNBC participants with an active social security number, who had not actively withdrawn their participation and provided either their private postal address or phone number were invited; the overall response rate was 44%. From a population of 67,346 participants born into the DNBC, for whom we had information obtained from their mothers during pregnancy, 27,441 participants completed the DNBC-18 before 11 March 2020 and provided complete data on self-injury, suicidality and symptoms of EDs (Supplementary Fig. 1). This resulted in a retention rate of 37% for men and 49% for women. The population was further restricted to 7,597 participants with information on self-injury, suicidality and symptoms of EDs from wave 8, that is, pre- and post-lockdown measures. The response rate in wave 8 was 25% for men and 39% for women.

Repeated cross-sectional data

Participants in the DNBC were born over a period spanning 8 years. Thus, per design, they completed the DNBC-18 in different years and we exploited this feature to perform cross-sectional comparisons of participants completing the DNBC-18 between 1 January 2018 and 11 March 2020 with those completing thereafter until 1 March 2021. The study population included 24,625 participants with data on self-injury, suicidality and symptoms of EDs subdivided into the years 2018–2021; this was 38% and 51%, respectively, of the invited men and women (Supplementary Fig. 2). Based on the date completing the DNBC-18, participants were assigned to 1 of 16 different periods; 11 periods represented pre-lockdown (January 2018–11 March 2020) and 5 periods represented post-lockdown (12 March 2020–1 March 2021) (Fig. 1)1,2.

Measures of self-injury, suicidality and symptoms of EDs

To measure self-injury, suicide ideation and suicide attempt, items in the DNBC-18 and wave 8 of the COVID-19 survey were used to measure whether these behaviours had occurred within the last year (yes versus no) (Supplementary Table 2). Self-injury was worded as ‘have you harmed or hurt yourself on purpose within the last year’ and suicide ideation was worded as ‘have you thought about taking your own life (even though you would not do it) within the last year’. Suicide attempt was worded as ‘have you tried to take your own life within the last year’. If participants answered ‘do not know’ to items on self-injury or suicidality, they were categorized as not having the behaviour. Symptoms of EDs were collected in the DNBC-18 with items adapted from the McKnight Risk Factor Survey on weight and shape concerns and items from the Youth Risk Behavior Surveillance System survey on binge eating, self-induced vomiting and use of laxatives31,32. We defined symptoms of EDs in accordance with definitions used and described by Micali et al.33 to classify threshold (weekly) and subthreshold (monthly) anorexia, bulimia, purging disorder and binge eating disorder (Supplementary Tables 3 and 4). Because of the low frequency of threshold EDs in men, we chose to combine threshold and subthreshold EDs into one measure reflecting symptoms of EDs within the last month (yes versus no). If participants answered ‘do not know’ to the items included in the definition of symptoms of EDs, they were categorized as not having the specific symptom; because the final ED symptom variable was based on 15 items, it was still possible for an individual to be categorized as having symptoms of EDs if answering ‘do not know’ to some items.

Statistical analysis

Sampling weights were estimated to account for differential attrition in DNBC-18 and wave 8 of the COVID-19 survey. For the inverse probability weighting, we used logistic regressions with participation, that is, having data as outcome and the following predictors: sex; household and socio-occupational status; maternal age at childbirth; parity; and maternal smoking, which were collected during pregnancy, categorized as shown in Supplementary Table 1. These predictors were included because they were important predictors of loss to follow-up in previous DNBC data collections34,35. Separate analyses were performed for DNBC-18 and wave 8 of the COVID-19 survey and performed on the relevant baseline populations described in Supplementary Figs. 1 and 2. The relevant sampling weights were used in all analyses in this study, except in the calculation of Bayes factors.

Sex-specific proportions of self-injury, suicidality and symptoms of EDs were estimated with corresponding 95% CIs pre- (DNBC-18) and post-lockdown (wave 8), with the longitudinal data and for each period in the repeated cross-sections to illustrate any differences. Subsequently, we estimated changes in the proportion of young adults with self-injury, suicide ideation and attempt as well as symptoms of EDs from pre- to post-lockdown by fixed-effect linear regression on the longitudinal data and linear regression on the cross-sectional data. We used linear regressions, instead of logistic regressions, to get an estimate of the absolute change (that is, in percentage points) of the proportions reporting self-injury, suicidality and symptoms of EDs from pre- to post-lockdown. Separate regressions were conducted for self-injury, suicide ideation and attempt, and symptoms of EDs, respectively. To examine changes during the lockdown on any of the outcomes in men and women, we added an interaction between lockdown (pre versus post) and sex. In the linear regressions on the repeated cross-sectional data, we initially tested if changes during lockdown varied across periods by including an interaction between lockdown (pre versus post) and period; if non-significant, this interaction was omitted. Bayes factors were calculated to support non-significant results with the equation: Bayes factor = n^((k0ka)2) × (RSS0/RSSa)^(n/2), where H0 = model (unweighted) without lockdown variable and Ha = model (unweighted) with lockdown variable, n = size of the study population, k = number of parameters in model and RSS = the residual sum of squares36,37,38. Thus, we tested if the model with versus without the lockdown variable were most suitable. We interpreted Bayes factors as in Jeffreys39. All analyses were performed unadjusted and with SAS v.9.4 (SAS Institute) using the commands proc surveyfreq, proc genmod and proc glm and by applying sampling weights and the absorb statement. The level of statistical significance was set at P < 0.05.

Sensitivity analysis

To investigate if the severity of self-injury and suicide ideation changed during lockdown, we used an item about how many times the participants had experienced self-injury and suicide ideation, respectively, within the last year. The 5 category items were transformed into a scale ranging from 0 to 4 (at no time = 0, once = 1, 2–5 times = 2, 6–10 times = 3, more than 10 times = 4). For symptoms of EDs, we used threshold EDs (weekly symptoms) and subthreshold EDs (monthly symptoms) to create a scale ranging from 0 to 2 (no symptoms = 0, monthly symptoms = 1, weekly symptoms = 2). In sensitivity analyses, these scales replaced the binary variables in the linear regression models. We also conducted sensitivity analysis where participants answering ‘do not know’ to items on self-injury and suicidality were categorized as missing because they might differ from those without self-injury and suicidality. This was not necessary for symptoms of EDs because it was possible to be categorized as having ED symptoms even when answering ‘do not know’ to some items.

The results in the longitudinal set-up may be biased because all participants were aged 18 years at the pre-lockdown measure but between 19 and 23 years at the post-lockdown measure. In the sensitivity analyses, we addressed this by restricting to participants aged 19–20 years when completing wave 8 to limit the time gap and thereby in-built age difference. Further, we did sensitivity analyses in the repeated cross-sectional data where self-injury and suicide ideation were restricted to being within 4 weeks instead of 1 year using the items presented in Supplementary Table 2.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.