Multilevel analysis to identify the risk factors for smoking on teenage children attending school
Among young teens (aged 13 to 15), about one in five smokes worldwide. Presently, about five million people worldwide die yearly from tobacco related diseases. Approximately 90% of smokers begin smoking before the age of 21 years.
Children and adolescents spend a considerable amount of their time in school, and the school environment is therefore important for child outcomes. Random effect logistic regression model has been proposed to model correlation among subjects in the same cluster by including a cluster-specific random effect in the logit. The objectives of the study are
- To develop two level random effects logistic regression model for the analysis of clustered binary responses to identify factors associated with smoking among school going male adolescents.
- To assess if the variability between schools is different for the public and private schools using a random coefficient model. Two random effects will account for the variability between public and private schools respectively.
- We will fit Generalizing Estimating Equation (GEE) model to deal with 2 level clustered data for binary outcome.
METHODS:
A two-stage cluster sampling with stratification based on school type was employed for the selection of schools and students. We interviewed 772 male secondary school students. The outcome variable is smoking status of the students. We have two level data with a single level of clustering.
RESULTS:
The final multilevel random effect model of teenage smoking showed that age, mother’s education and parents, family and friends smoking all contributed to teenagers smoking (p-value<0.05). Between cluster variance is significantly different from zero (p-value of likelihood ratio test = 0.01), which indicates that there is variability between schools. The intra-class correlation quantifies consistencies among observations within each cluster and it is also greater than zero (ICC =0.16).
A random coefficient model with two random parameters showed insignificant result (p-value of likelihood ratio test > 0.99)indicating that there is variability among schools but it is not different for public & private.
CONCLUSIONS:
The results of this study point out the need for an effective tobacco control program especially among adolescents. Parental counseling about the influence of family tobacco use on their children may bring about encouraging results. One of the most important commitments a country can make for future, economic, social, political progress and stability is to address the health and development needs of its adolescents.
Random effect models and GEE take correlation into account in the inferential process, indicating that there is variability between schools and we need to take cluster variation into account by using multilevel modeling.