School-level clustering and individual correlates of adolescent substance use in India: a multicentre multilevel study

Adolescent substance use is a growing health burden in India, yet evidence mostly focuses on individual-level correlates while overlooking potential school-level heterogeneity and contextual characteristics within school settings. This study examined individual- and school-level measures associated with current substance use among adolescents across diverse Indian settings using a multilevel framework. Baseline data from a multicentric school-based study conducted between October 2025 and January 2026 across six regions of India were analysed. Using a multistage cluster sampling approaches, the protocol assumed an average cluster size of 50 students per school for sample-size planning; however, no fixed number of students was individually sampled from each school. One class was randomly selected from each participating school, and all eligible students within the selected class were invited to participate. A total of 6,168 adolescents aged 10–19 years were enrolled and surveyed. After excluding participants with missing information on current substance use and participants from the Puducherry site, where information on current substance use was unavailable, 4,848 adolescents constituted the final analytical sample for multilevel analysis. Current substance use was defined as use of any psychoactive substance in the previous three months, using the WHO Alcohol, Smoking and Substance Involvement Screening Test. Two-level mixed-effects logistic regression models were used to assess individual- and school-level correlates, while intraclass correlation coefficients (ICC), proportional change in variance, and variance decomposition quantified contextual heterogeneity. Among the 6,168 adolescents enrolled and surveyed, 6,168 were included in the descriptive analyses. A total of 332 participants were excluded because of missing information on current substance use, and 988 participants from the Puducherry site were excluded because information on current substance use was unavailable. The final multilevel analytical sample comprised 4,848 adolescents. Substantial between-school clustering was observed (ICC = 0.23, 95% CI: 0.17–0.31), indicating that approximately 23% of the variance on the latent logistic scale was between schools in the null mode. Sequential variance decomposition showed a 16.7% reduction in baseline between-school variance following adjustment for individual-level characteristics and an additional 13.3% reduction following inclusion of school-level characteristics, with 70.0% of baseline between-school variance remaining after full adjustment. High digital exposure was associated with higher odds of substance use (AOR = 2.11), as was greater substance accessibility (AOR = 1.63), whereas higher levels of protective factors and mental well-being were associated with lower odds. Adolescent substance use showed substantial between-school clustering, supporting the use of multilevel approaches while indicating that relevant contextual and compositional factors remain incompletely characterized.