Kara Michels

PhD, MPH

Metabolic Epidemiology Branch

Postdoctoral Fellow

PLCO (Learn more about this study)

PLCO-305

Sep 5, 2017

Lifecourse body mass index trajectories and the development of endometrial and ovarian cancers

Obesity is a modifiable contributor to much of the morbidity and mortality that burdens Americans. During recruitment for the PLCO Trial and the AARP-Diet and Health Study, ~35% of women aged 50-59 in the NHANES had a BMI classified as obese, while ~30% of women aged 60-69 met this criterion.(Flegal et al., 1998) Obesity has increased in the U.S., with 40% of women aged 60 and older estimated to be obese in 2014.(Flegal et al., 2016)

Due to this increase, we can expect that obesity-related cancer diagnoses will rise. Obesity is one of the strongest risk factors for endometrial cancer, with an obesity-associated population attributable risk of 57%.(Calle and Kaaks, 2004) The incidence of endometrial cancer is estimated to increase by 55% between 2010 and 2030.(Sheikh et al., 2014) The relationship between BMI and ovarian cancer is less clear; several studies report increased risk among obese women, but a recent analysis from the Ovarian Cancer Association Consortium indicates that BMI primarily influences risk for less common ovarian cancer subtypes and also for pre-menopausal women who develop ovarian cancers.(Liu et al., 2015, Olsen et al., 2013) One reason that BMI-cancer associations may vary across studies is due to the differing ages at which participant BMI is measured. Analysis of BMI across the life-course may help us better assess the cumulative impact of obesity over the life-course as well as identify target ages for behavioral or even surgical interventions with the goal of cancer prevention.

Numerous studies attempt to understand the impact of body size over the life-course on gynecologic cancer risk. The literature was most recently reviewed in the book “Obesity and Cancer” by Shaw et al., 2016 (endometrial cancer) and Tworoger and Huang, 2016 (ovarian cancer). Studies generally agree that adult weight gain increases risk for endometrial cancer and weight loss likely reduces this risk—despite heterogeneity regarding what constitutes weight change in these studies.(Shaw et al., 2016) Most research indicates that changes in adult BMI are not associated with ovarian cancer risk, but some meta-analyses suggest that it does.(Tworoger and Huang, 2016)

Most studies assess BMI at two or three ages and use data-based cut points to define changes in BMI between two given time points. As most researchers will generate innumerable effect estimates for each recalled category of BMI, at each age it was assessed, and for each change in weight/BMI between time points—multiple testing becomes problematic. Furthermore, none of these estimates really measures the cumulative effect of BMI over time. For that reason, we will generate trajectories to explore the effect of life-course BMI. To our knowledge, no studies use our proposed statistical methods to assess the relationship between BMI across the life-course and endometrial or ovarian cancer risk. Previous researchers using data from PLCO and AARP have employed the BMI trajectory-generating techniques that we describe below to examine prostate, esophageal, and liver cancer risks.(Kelly et al., 2017, Petrick et al., 2017, Yang et al., 2017)

1. To identify lifecourse BMI trajectories among women in the PLCO Trial and NIH-AARP Diet and Health Study.

First, we will use latent-class group-based trajectory models to identify patterns of BMI change across adulthood using recalled BMI information from participants in the PLCO trial and the AARP study. Participants in both were asked to recall their BMI at the approximate ages of 18 (20 in PLCO), 35 (“30s” in PLCO), 50 (“50s” in PLCO), and in their 60s (baseline BMI for both studies when limiting to those 60 and older at study entry). Thus, four age points will be used to generate the trajectories. We will identify 3 to 6 trajectories based on model fit (Bayesian Information Criterion) and group membership of at least 1% of the analytic population. Pooling these studies is ideal because of their large sample size, shared information on recalled BMI, and generalizability to the U.S. population.

2. To determine if these BMI trajectories are associated with endometrial or ovarian cancer risk.

These trajectories will then be used as independent variables in Cox proportional hazards models for the time to diagnosis of incident ovarian or endometrial cancer. Models will be appropriately adjusted for confounding. To determine if chronic illness or cancer cases are influencing the definitions of our trajectories (e.g., cachexia from cancers diagnosed within 2 years of a questionnaire), we will evaluate the prevalence of these diseases within each trajectory and conduct sensitivity analyses evaluating modification by time to diagnosis (e.g., extended Cox models with heaviside functions). We will conduct secondary analyses exploring BMI trajectory associations with cancer histology, as sample sizes permit.