For the past few decades, a number of epidemiologic studies have shown that association exists between body weight and various health outcomes: in particular, obesity is associated with a host of chronic diseases, including Type-2 diabetes, cardio-vascular diseases, and various types of cancers, and a substantially increased risk of death. The association itself usually appears moderate, but the obesity prevalence has been on the rise dramatically - the World Health Organization (WHO) estimates at least more than 300 million adults worldwide, and 25% of adult Americans in the US are obese. It is the rapid rise of obesity prevalence, in addition to its association, that account for a substantial proportion of the overall disease load. In epidemiology, such a proportion is called disease attribution associated with obesity.
The concept of disease attribution is not new. For modifiable risk factors, such as obesity, it may also be called the “attributable risk” that is measured by Levin’s population attributable fraction (PAF) . Usually, it is defined as the percentage of excess risk of disease associated with an exposure. Several methodologic challenges remain for Levin’s PAF, particularly in longitudinal epidemiologic cohort studies: 1) the prevalence of obesity is not static but dynamic, trending upward. For example, in the US alone, obesity among adult Americans rose from 23.7% in 2003 to 24.5% in 2004; that is, the prevalence of exposure is time-varying; 2) many obesity-associated chronic disease outcomes, such as death, can be regarded as being of the time-to-event type, also time-varying by nature; 3) Levin’s PAF would represent an ideal preventable risk reduction only if obesity were “completely”
eradicated from a population. Such an idealistic measure is, however, seldom practical, due to limited resources, individual and community adherence, contrary commercial forces, or ethical concerns.
Nonetheless, obesity is only one of the many modifiable risk factors that are associated with the aforementioned chronic diseases. For example, other modifiable risk factors include tobacco smoking, alcohol drinking, etc. In this application, we hence aim to develop more general and flexible statistical methods to analyze timevarying disease attribution in longitudinal cohort studies and, more importantly, apply these methods to identify optimal prevention strategies that may lead to improvement of life expectancy. In this project we aim to:
Aim 1. Develop general statistical methods for time-varying attributable risk functions (ARF) to assess and compare prevention strategies.
Aim 2. Develop statistical methods for a set of pairwise ARF’s to assess and identify practically optimal prevention strategies.
Aim 3. Develop statistical methods to evaluate the disease attribution in residual time.
Aim 4. Develop statistical software and apply the developed methods to a large collection of cohort data from the Asia Cohort Consortium (ACC).