Studying the association between longitudinal nondense breast tissue measurements and the risk of breast cancer: a joint modeling approach

TL;DR

This study examines whether the amount of non-dense breast tissue is linked to breast cancer risk using advanced statistical methods and finds no strong evidence of an association.

Contribution

The study introduces a joint modeling approach to analyze longitudinal breast tissue data and cancer risk, addressing previous methodological limitations.

Findings

No strong evidence was found for an association between nondense breast tissue and breast cancer risk.

Previous studies may have failed to isolate the effect of nondense tissue from other factors.

Estimates of the effect of nondense tissue are sensitive to modeling assumptions.

Abstract

Conflicting results have appeared in the literature on whether the amount of nondense, adipose tissue in the breast is a risk factor or a protective factor for breast cancer (BC), and biological hypotheses supporting both have been proposed. We suggest here that limitations in study design and statistical methodology could potentially explain the inconsistent results. Specifically, we exploit recent advances in methodology and software developed for the joint analysis of multiple longitudinal outcomes and time-to-event data to jointly analyze dense and nondense tissue trajectories and the risk of BC in a large Swedish screening cohort. We also perform extensive sensitivity analyses by mimicking analyses/designs of previously published studies—for example, ignoring available longitudinal data. Overall, we do not find strong evidence supporting an association between nondense tissue and the risk of incident BC. We hypothesize that (1) previous studies have not been able to isolate the effect of nondense tissue from dense tissue or adipose tissue elsewhere in the body, that (2) estimates of the effect of nondense tissue on risk are strongly sensitive to modeling assumptions, or that (3) the effect size of nondense tissue on BC risk is likely to be small/not clinically relevant.

Figures