Impact of displacement on mental health in war‐exposed Ukrainian adolescents: a longitudinal study

The Russian invasion continues to put strain on Ukrainian people1. The burden on mental health of adolescents is particularly devastating2. Many people are being displaced: as of March 2024, 3.7 million of the Ukrainian population were estimated to be internally displaced and 6.5 million externally displaced3. Evidence on the associations between displacement and mental health is mixed and limited, especially for children and adolescents. Displacement can substantially increase the mental health burden of affected populations4, but being displaced could help people avoid exposure to war, which has devastating mental health consequences2. Furthermore, people can be internally or externally displaced, which involves different experiences and health consequences5. It is imperative to better understand the relationship between war-induced displacement and mental health, especially in a large-cale conflict such as the Russian-Ukrainian war. Though this is challenging – given the possibility of reverse causality as well as time-varying confounding due to war exposure – longitudinal data, coupled with causal inference methods that account for time-varying confounding6, may help overcome these challenges. We took advantage of the longitudinal design of the Adolescents of Ukraine During the Russian Invasion (AUDRI) Cohort to investigate the impact of displacement on the mental health of war-exposed Ukrainian adolescents. We conducted a prospective, observational study using longitudinal data from the first (May-June 2023) and second (November-December 2023) waves of the AUDRI Cohort. We recruited youths attending any secondary school in Ukraine (in person or online) who were aged 15 years or older at the time of wave 1. The cohort also included youths residing abroad while attending Ukrainian secondary school online2. The study was approved by the Ethics Committees of the Institute of Psychiatry at the T. Shevchenko National University of Kyiv and of the University of Tokyo. Informed consent was obtained from all participants and their parents/legal guardians, and data were anonymized prior to analyses. Students agreeing to participate were sent invitations via email in the follow-up waves. All questionnaires were self-administered and presented in Ukrainian on Qualtrics. Participants were asked about their displacement status: not displaced; displaced within Ukraine; displaced to another country; or displaced but returned back. We used well-validated tools to screen for post-traumatic stress disorder, PTSD (the Child and Adolescent Trauma Screen, CATS); depression (the Patient Health Questionnaire-9 for Adolescents, PHQ-A); anxiety (the Generalized Anxiety Disorder 7, GAD-7); substance use disorders (the Care, Relax, Alone, Forget, Friends, Trouble Test, CRAFFT 2.1); and eating disorders (the Sick, Control, One stone, Fat, and Food Questionnaire, SCOFF). We performed causal inference on the associations of displacement with a positive screen for each psychiatric condition using g-estimation of additive structural nested mean models7, which seeks to minimize bias from time-varying confounding. We calculated the risk differences (RDs) by displacement status (current displacement vs. no displacement) at wave 1 of positive screens for psychiatric conditions at wave 2. Linear regression models were adjusted for demographic variables (age, gender, education level of parents, and region of residence in wave 1), as well as baseline developmental traits and difficulties (having had learning difficulties as a child, having attended special education classes as a child, diagnosis of autism spectrum disorder or attention-deficit hyperactivity disorder). War exposure was included in the models as a time-varying covariate. We also ran two sets of regressions similar to the main regressions as above, with internal displacement compared to no displacement, and external displacement compared to internal displacement. Standard errors were computed using sandwich variance estimators and were clustered on region. Missing data were imputed using missForest8. Analyses were performed using R statistical software, version 4.1.1. We sent invitations to 8,339 individuals who provided informed consent as well as valid email addresses in wave 1 of the AUDRI cohort. Of those, a total of 1,020 agreed to participate in wave 2, and were included in the analyses. Among these participants, 25.4% self-identified as male, 71.4% as female, and 0.8% as transgender, while 2.4% did not self-identify as belonging to any of those groups. At the time of wave 1, 38.6% were 15 years old, 33.4% were 16, 17.0% were 17, 3.8% were 18, 0.5% were 19, and 6.7% were 20 or older. In wave 1, 60.4% reported having been exposed to war and 29.2% were displaced; in wave 2, 61.5% reported having been exposed to war and 27.3% were displaced. The proportion of adolescents screening positive for each psychiatric condition was higher among those who were displaced than those not displaced. In wave 1, 53.8% vs. 41.3% screened positive for clinically relevant psychological trauma (as a proxy for PTSD), 47.3% vs. 41.9% for depression, 29.7% vs. 23.9% for anxiety, 18.3% vs. 16.9% for substance use disorders, and 40.8% vs. 34.7% for eating disorders. In wave 2, 52.9% vs. 43.8% screened positive for clinically relevant psychological trauma, 53.8% vs. 47.5% for depression, 33.2% vs. 27.3% for anxiety, 25.8% vs. 18.2% for substance use disorders, and 39.4% vs. 38.5% for eating disorders. Displacement in wave 1 was associated with a higher risk of a positive screen for clinically relevant psychological trauma (RD: 8.24 percentage points (pp); 95% CI: 0.22-16.26) and eating disorders (RD: 8.00 pp; 95% CI: 0.23-15.78) in wave 2. Individuals who experienced displacement in wave 1 were more likely to screen positive for anxiety (RD: 6.03 pp; 95% CI: −0.63 to 12.69) and less likely to screen positive for substance use disorders (RD: −4.38 pp; 95% CI: −10.78 to 2.01) in wave 2, though the estimates were imprecise, due to wide CIs. The association was close to null for depression (RD: 2.27 pp; 95% CI: −5.63 to 10.17). Compared to no displacement, internal displacement in wave 1 was associated with greater risk of positive screens for clinically relevant psychological trauma (RD: 11.00 pp; 95% CI: 0.93-21.08) and anxiety (RD: 8.63 pp; 95% CI: 0.22-17.03) in wave 2. Individuals who experienced internal displacement in wave 1 were less likely to screen positive for substance use disorders in wave 2 (RD: −5.81 pp; 95% CI: −13.82 to 2.19), although the estimate was imprecise. Associations of internal displacement with eating disorders (RD: 2.69 pp; 95% CI: −6.95 to 12.34) and depression (RD: −0.86 pp; 95% CI: −10.76 to 9.03) were close to null. Relative to internal displacement, individuals who experienced external displacement in wave 1 were less likely to screen positive for most psychiatric conditions (RD: −22.83 pp, 95% CI: −44.89 to 2.23 for clinically relevant psychological trauma; RD: −20.07 pp, 95% CI: −41.28 to 1.14 for substance use disorders; RD: −3.81 pp, 95% CI: −28.06 to 20.44 for depression; RD: −14.03 pp, 95% CI: −37.64 to 9.58 for eating disorders), but were more likely to screen positive for anxiety (RD: 10.09 pp; 95% CI: −12.02 to 32.20), though the estimates were imprecise, due to wide CIs. Our findings indicate that displacement was modestly associated with greater risk of positive screens for psychiatric conditions six months later. However, at the population level, even modest effect sizes may result in a large mental health burden9. The differences between internal and external displacement also warrant attention: individuals who experienced internal displacement had a greater risk of positive screens for most psychiatric conditions, which may be due to factors such as the constant insecurity and the risks associated with continuing to live in a war-affected country, as well as the limited access to mental health support services due to consequences of the war. The results of this study should be interpreted in the light of some limitations. First, some CIs in the RD estimates were wide, and larger samples may be necessary for more precise estimates. However, with a sample of over 1,000 individuals, this is among the largest longitudinal studies in adolescents ever conducted in a conflict setting. Second, there may have been unmeasured confounders biasing the results. Third, a relatively large proportion of those who participated in wave 1 were lost to follow-up. A key contribution of this study is the use of rigorous causal inference methods on longitudinal data to control for time-varying confounding. Therefore, it adds significantly to the available research evidence on the impact of displacement on mental health in adolescents.