Andrew Renehan

Ph.D.

The University of Manchester

Professor of Cancer Studies and Surgery

PLCO (Learn more about this study)

PLCO-339

Jan 31, 2018

Adulthood BMI trajectories And Cancer using clusters [ABACus]

Excess adiposity, commonly approximated as body mass index (BMI) over 25 kg/m2, is an established risk factor for several incident cancers, and is many western populations, is the second commonest cause after smoking.

The 2016 IARC report on weight and cancer concluded that elevated BMI is associated with 13 cancer types – endometrial, oesophageal adenocarcinoma, gastric cardia, kidney, liver, multiple myeloma, meningioma, pancreas, colorectal, gallbladder, postmenopausal breast, ovary, thyroid - referred to as obesity-related cancers.

The IARC report noted that the BMI-cancer association may be modified by other risk factors, such as smoking.

The above epidemiology is based on a once-only BMI measurement, typically at cohort entry. However, at a biological level, the lifecourse exposure of adiposity is likely to be mechanistically more relevant. Relatively simple measures, such as weight change across adulthood, are associated with increased cancer risk, in patterns mirroring those for baseline once-only BMI.

A more recent, and statistically more advanced approach to estimate lifecourse exposure is considering clusters of weight changes, known as latent class trajectory modelling (LCTM). These methods have been used in the social medicine and psychology literature but now gaining ground in mainstream epidemiology. In the context of repeated BMI, there is the added value of identifying adverse body weight trajectories with potential to serve as an early alert system to intervene with lifestyle changes. To-date, for cancer incidence and mortality, there have been reports with BMI LCTM, but numbers of classes and findings have been inconsistent.

The statistical application of LCTM is not trivial. The present collaboration has written an eight-step framework on how to optimally select models taking into account the selection of number of classes; the non-linearity of trajectories; and the variance within classes (fixed- versus random-effect models). This manuscript is under revision with BMJ Open.

LCTM of repeated BMI in PLCO:

LCTM of BMI have been reported from PLCO based on BMI determinations at ages 20, 50, and baseline. Application collaborators (MBC, JLP) used LCTM in the AARP and PLCO cohorts (for prostate cancer incidence and mortality; oesophageal adenocarcinoma incidence; and liver cancer incidence).

Using algorithms detailed in our BMJ Open manuscript, we seek to extend our methodological framework across several cohorts with repeated BMI measures, in the ABACus consortium (PROSPERO CRD42017079621), and specifically test whether the variations in results noted to-date are due to true population differences or differences in model selection.

Specific objectives (different from previous analyses):

1. Include BMI measures beyond ages 20, 50 and baseline using the 2006-08 supplemental questionnaire;
2. Describe the latent classes of repeated BMI measurements through adulthood in the PLCO using an increasing hierarchical (fixed to fully random-effects) approach; test for various optimal models; and optimise class separation;
3. Determine associations between latent classes and obesity-related cancers (rather than individual cancer types), and latent classes and non-obesity-related cancers (as negative controls);
4. Explore discriminative of LCTM versus BMI measure, for example with c-statistic (this is novel);
5. Test for the influence of stratification by smoking (either before or after LCTM).