The application and predictive value of the weight-adjusted-waist index in BC prevalence assessment: a comprehensive statistical and machine learning analysis using NHANES data.

backgroundObesity is a known risk factor for breast cancer (BC), but conventional metrics such as body mass index (BMI) may insufficiently capture central adiposity. The weight-adjusted waist index (WWI) has emerged as a potentially superior anthropometric marker of central adiposity, as it provides a more accurate reflection of fat distribution around the abdomen compared to traditional measures such as BMI. This study aimed to investigate the association between WWI and BC prevalence using data from a nationally representative population in the United States.

methodsA total of 10,760 women aged over 20 years from the 2005-2018 National Health and Nutrition Examination Survey were included. Logistic regression was used to assess the association between WWI and BC prevalence. Multicollinearity was addressed using variance inflation factor diagnostics. Machine learning methods, including random forest and LASSO regression, were employed for variable selection and model comparison. The performance of the models was evaluated using ROC curves, calibration plots, and decision curve analysis.

resultsIn unadjusted models, WWI was significantly associated with BC (odds ratio (OR) = 1.56; 95% confidence interval (CI): 1.32-1.86). However, in the fully adjusted model, the association with BC was no longer statistically significant (OR = 0.98; 95% CI: 0.75-1.26). Machine learning models ranked WWI as one of the top predictors, with the random forest model retaining WWI as an important variable, while LASSO excluded it. Models based on variables selected by both LASSO and random forest, which included WWI, were built and assessed using ROC curve analysis. The random forest and LASSO models achieved AUCs of 0.795 and 0.79, respectively, demonstrating improved predictive performance. These findings suggest that while WWI may not serve as an independent predictor of BC, it may offer additional value when combined with other key covariates.

conclusionAlthough the WWI was related to BC prevalence before multivariable adjustment, it was not significantly linked to BC after adjustment. Given the cross-sectional design and the relatively small sample of BC cases (n = 326), the findings should be viewed with caution. Future research with larger prospective cohorts is needed to confirm these results and explore WWI's role in BC risk stratification. Studies should also investigate whether WWI can serve as a reliable independent predictor of BC in future research, taking into account other factors that may influence the association.