AI, ML, and DL models incorporating obesity-related factors demonstrated moderate discrimination for breast cancer diagnosis and risk prediction (pooled AUC 0.71; 95% CI 0.64-0.77).
Meta-Analysis
Do AI, ML, and DL models incorporating obesity-related factors improve risk prediction and outcome assessment in breast cancer?
AI, ML, and DL models incorporating obesity-related factors demonstrate moderate to strong discrimination for breast cancer diagnosis and risk prediction, though clinical translation is limited by heterogeneity and lack of external validation.
Effect estimate: AUC 0.71 (95% CI 0.64-0.77)
e12573 Background: Obesity and excess adiposity are established contributors to breast cancer risk, tumor biology, and treatment-related morbidity, yet their clinical effects are heterogeneous and incompletely captured by conventional risk models. Artificial intelligence (AI), machine learning (ML), and deep learning (DL) enable integrative analysis of obesity-related imaging, molecular, metabolic, and clinical data, offering new opportunities for precision risk stratification and outcome prediction. We systematically evaluated and quantitatively synthesized AI-, ML-, and DL-based models incorporating obesity or weight-related factors across the breast cancer continuum. Methods: A PRISMA-compliant systematic review was conducted using PubMed, Scopus, and Google Scholar through January 2026. Eligible studies applied AI, ML, or DL models to obesity-related exposures, including body mass index, metabolic biomarkers, adiposity-associated gene expression, or body composition–derived imaging features, and reported breast cancer–related outcomes. Outcomes included diagnosis, risk prediction, nodal metastatis, and treatment-related toxicity. Model discrimination metrics were extracted and pooled using a restricted maximum likelihood random-effects model with logit-transformed AUCs. Results: Sixteen models from heterogeneous cohorts were included, ranging from biomarker driven case-control studies to large population based datasets. For diagnosis and risk prediction, pooled discrimination was AUC 0.71 (95% CI 0.64–0.77; 95% prediction interval 0.40–0.90), with heterogeneity evident on forest plots. Imaging-based deep learning and multimodal models incorporating obesity-related variables demonstrated higher discrimination than traditional clinical risk scores. For nodal metastasis and radiotherapy related toxicity prediction, pooled performance across three studies was AUC 0.69 (95% CI 0.62–0.75; 95% prediction interval 0.48–0.84). Gene expression–based and metabolic biomarker models showed high discrimination in selected cohorts, though generalizability was limited by cohort size and validation. Funnel plot suggested possible small-study effects for diagnostic and risk prediction models (Egger p = 0.039). Conclusions: AI-, ML-, and DL-based models incorporating obesity-related factors demonstrate moderate to strong discrimination across breast cancer diagnosis, risk prediction, and selected outcomes, with superior performance in imaging-driven and multimodal models. However, heterogeneity, limited external validation, and inconsistent obesity phenotyping constrain clinical translation. Future studies should prioritize standardized adiposity metrics, harmonized outcomes, and prospective validation to support equitable and clinically actionable implementation of AI driven tools in breast cancer.
Anmol et al. (Thu,) conducted a meta-analysis in Breast cancer. AI, ML, or DL models incorporating obesity-related factors vs. Traditional clinical risk scores was evaluated on Diagnosis and risk prediction discrimination (AUC 0.71, 95% CI 0.64-0.77). AI, ML, and DL models incorporating obesity-related factors demonstrated moderate discrimination for breast cancer diagnosis and risk prediction (pooled AUC 0.71; 95% CI 0.64-0.77).