Page 48 - ISMCON souvenir 2021

ISMSCON - 2021

with the non-parametric method. The developed model is applied to simulated data set for validation and
VUS, Standard error of estimated VUS are reported for the simulated data. Parametric method: Rayleigh
parameters assumed for the three populations (, , ) = (65, 36, 6); vs non-parametric method . The
precision of the estimate of VUS is higher for the parametric method as compared to its non-parametric
alternative.
Keywords: Three class ROC model, ROC surface, volume under the ROC surface (VUS), Non-
parametric VUS, asymptotic confidence interval.

OS6: Multilevel Modeling Approach to Predict the Factors of
Depression among Indian Adolescents

Amrutha Jose1, Mariyamma Philip2, and M. Manjula2
1Christ University, Bengaluru
2National Institute of Mental Health and Neurosciences (NIMHANS), Bengaluru
E-mail: amrutha.jose@christuniversity.in

Background: Depression is prevalent during adolescence, especially between the ages of 15 and 18,
but it is often unrecognized. While adolescence is thought to be a pivotal period for the start of most
mental illnesses, depression is the most frequent mental health issue. Depression has negative impacts
on academic achievement, family relationships, and socialization, as well as increasing the risk of future
depressive episodes, substance misuse, antisocial behaviour, etc. The current study aims to determine
the characteristics linked to depressive symptoms in urban Indian adolescents using multilevel modelling.
Methodology: A total of 1428 students aged 13 to 19 years, from both public and private institutions were
included in the study. A socio-demographic data sheet was used to gather the demographic information of
the participants. The depression levels among the adolescents were assessed using the Beck depression
inventory-II and the multilevel modeling was used to detect the risk factors of depression. Among the two-
level as well as three-level random intercept models and two-level random coefficient models, the best
model was chosen based on the necessity of parsimony. Different multilevel models were fitted and were
compared using likelihood ratio tests, AIC, BIC along with estimates and standard errors.
Results: Intra-class correlation coefficients were estimated. Variables found to be significant factors of
the depression among adolescents were included in the final model and hence the best-fitted model
was identified. The final random intercept logistic regression model indicated that hopelessness, suicidal
ideation, hostility (p<0.001), avoiding problems (p=0.013), seeking spiritual help (p=0.004), academic,
relational, and financial stress were directly affecting the depression while solving family problem
(p<0.001) was inversely related. These predictors were significantly associated with depression levels
after controlling for the variation among schools. Bayesian random intercept logit model was also
attempted to compare the performance with frequentist approaches.
Conclusion: In educational and health research, hierarchical structures are prominent and analysis
should consider the sampling technique adopted during data collection and data structure. The study
uncovered the factors that contribute to depression among adolescents. Hence, preventive interventions
and policy frameworks such as youth-friendly service systems should be implemented to deteriorate
incidence of depression.
Keywords: Depression, Adolescents, Hierarchical structure, Multi-level models, Random intercept
model.

46 CONFERENCE SOUVENIR