Abstract

ABSTRACT
“The Application of Multiple Imputation to Explore Alcohol Consumption Among Young Adults and Its Impact on Body Mass Index”

Maryam Tafreshi Motlagh

Obesity is associated with health issues such as high blood pressure, cardiovascular disease, and cancer. Concurrently, excessive alcohol consumption and its negative health effects among young adults are growing global public health concerns. However, the studies on the association between alcohol consumption and weight gain are inconsistent. In particular, analyzing weight gain and alcohol consumption over time is complicated by data missingness. Indeed, one potential reason for the mixed findings could be how those studies addressed missing data. Missing data can occur due to a variety of reasons including loss to follow-up and various non-response cases. As a result, missing data can lead to loss of efficacy in data analysis and is one of the main challenges of longitudinal studies and large surveys. Therefore this thesis attempted to address missing data challenges by utilizing multiple imputation (MI). Generalized linear models were then conducted to assess the relationship between alcohol consumption with BMI, adjusting for age at baseline, sex, race, smoking, depression status, relationship status, employment status, student status and physical activities. Analyses were conducted separately for a complete-case dataset and the imputed dataset.

Data were from the Nicotine in Dependence in Teens (NDIT) cohort, a 20-year prospective cohort initiated in 1999 (n=1294, 52% female). Data were collected every three months when participants were in high school (grade 7 to grade 11; 1999-2005, 20 cycles). After high school, follow-up assessments were conducted approximately every four years (2007-08 and 2011-12 for cycles 21 and 22). The GLM results demonstrated that there was no association between alcohol consumption and BMI at cycle 21, but at cycle 22 a negative association was detected. Many similarities between the complete-case and MI general linear models were observed. However, estimates and standard errors were different between the two models, and were generally smaller in the MI models compared with the complete-case.

Although counterintuitive, the negative association is consistent with the existing literature. Preliminary evidence from the literature further suggests there is an interaction with age on the alcohol and obesity relationship. Indeed, it seems that alcohol consumption does not have an immediate impact on body mass among young adults, or it may have an inverse relationship. However, as alcohol remains one of the main contributing factors to chronic diseases such as high blood pressure, cardiovascular disease, and cancer, it is perhaps a latent effect that does not show any immediate positive relationship with symptoms of these diseases at early stages of young adulthood. Specific details about the alcohol use (such as type of beverage or calorie count) were not available in this study. While this is similar to other cross-sectional study limitations, they are important confounding variables that should be addressed in future studies. Although the statistical methodology is not consistently utilized in the alcohol and obesity risk literature, this thesis demonstrated the use of MI to produce unbiased estimates and smaller standard errors compared to the complete cases analysis. Further research is needed on the same cohort to further track the weight changes and other possible health problems due to alcohol use in the long-term.