Developing and Validating a Model for Predicting Child Obesity in Lebanon

Abstract

Introduction Child obesity, defined as the BMI at or above the 95th percentile for children of same age and sex is steadily climbing the ladder of public health concern. With child obesity manifesting in our society, it is crucial to predict risk of child obesity so that health interventions and context-specific policies can be implemented. Thus, the aim of this study is to develop and internally validate a prediction model for child obesity in Lebanon. Methods This is a cross-sectional study of 2,125 school students from the SCALE study. The SCALE study employed a two-stage sampling method aiming to include a representative sample of 8-12 years old students in Greater Beirut. The first stage used a random sample of 50 schools stratified into public, private, and private free schools, which were identified by a list provided by the Ministry of Education. The second stage included randomly assigning 50 students from grades 4,5,6 from each of the 50 schools that accepted to join the study in the first stage after parents signed the consent forms. This study produced prediction model discrimination and calibration slope for models developed using backward logistic regression as a statistical approach and LASSO, Ridge, Elastic net as machine learning approach. Two binary outcomes were assessed in this study: obese versus non-obese and obese or overweight versus normal or thin. Seventeen predictors were included in the prediction models: age, gender, food insecurity, nutrition knowledge, school type, crowding index, parent marital status, mother education, screen time, TV time, eating while on screen, physical activity, fast food consumption, fruit availability, vegetable availability, sugar sweetened beverages availability at home, having an obese mother. Results The sample size included 1,409 participants of median age 11 years (10-12). The best performing model is that of Lasso adaptive with discrimination of 0.632 (0.60-0.66) and C-slope 0.968 (0.72-1.21) with outcome obese or overweight. Ten predictors were selected by that model where older age, being female, having married parents, adequate availability of fruits at home, crowding index less than 3 are protective factors and eating while on screen, child nutrition knowledge, tv viewing time more than 2 hours, vegetable availability plus having an obese mother as risk factors for child obesity or overweight. Interpretation This study showed a well calibrated predictive model with moderate discrimination. Such prediction model in clinical settings could be used to prevent the risk of child obesity and thus reduce the risk for other potential non-communicable diseases while utilizing fewest resources possible. A range of policies could be implemented by parents, schools, and the government as a product of this study. Parents are requested to limit their child’s TV time and encourage their child to consume fruits. The strength of this study includes measuring height and weight in duplicate to prevent misclassification, using close ended questionnaires, usage of sampling weights in analysis, defining variable cutoffs based of systematic reviews and expert opinion. Limitations of this study includes absence of some variables present in the literature, presence of variables with missing data above 10%, potential recall bias and differential non-response bias.