Peer effects on risk of psychiatric disorders: A nationwide study of Norwegian adolescents

Importance: Peers are hypothesized to influence adolescent mental health. However, most peer effects studies are susceptible to selection bias and reverse causation, limiting causal inference.Objective: Examine classroom peer effects using a genetically informed approach and quasi-random classroom assignment.Design, Setting, and Participants: Nationwide primary and specialist care registry data were used to follow a cohort of 750 750 Norwegian adolescents born between 1998 and 2010 until first registered psychiatric diagnosis, emigration, death, or December 31 2023. School class, school, and residential area at age 13 were identified. Data were analyzed from March to October 2026.Exposures: (1) Pearson-Aitken Family-Genetic Risk Scores (FGRSs) indexing genetic liability to psychiatric disorders for individuals and peers sharing classroom, school, and/or residential area. Classrooms did not differ systematically in FGRSs, demonstrating that classroom peer FGRSs served as a natural experiment of quasi-randomization into peer environments. (2) Peers’ psychiatric diagnoses.Main Outcomes and Measures: Registrations for alcohol use disorder (AUD), drug use disorder (DUD), major depression (MD), and anxiety diagnoses. Associations with classroom peer FGRSs and interaction effects with individual genetic liability and sex were estimated using discrete-time survival and Cox proportional hazards models.Results: Psychiatric disorder risk was partly clustered at the classroom level (proportions of explained variance 0.9-5.0%). One SD increase in classroom risk was associated with a hazard ratio [HR] of 1.26 for AUD, 1.53 for DUD, and 1.18 for MD, and 1.20 for anxiety disorders. Classroom peer FGRSs were associated with increased risk of AUD in early adolescence (highest vs. lowest peer FGRS quintile, HR: 1.18; 95% CI 1.04, 1.34), MD (HR: 1.10; 95% CI, 1.06-1.14), and anxiety (HR, 1.07; 95% CI, 1.03-1.11). Classmate diagnoses were associated with elevated risk of DUD, MD, and anxiety (HRs: 1.02-1.16). No interaction effects between classroom peer FGRSs and genetic liability were observed; classroom peer FGRSs estimated for females predicted a higher risk among females (interaction HRs: 1.02-1.04). Conclusions and relevance: We found evidence of classroom peer effects, unbiased by selection bias and reverse causation, on internalizing disorder and AUD risk in early adolescence. Future studies are necessary to understand traits and mechanisms involved.