Cheng, Ching‐Yu

Abstract Generative artificial intelligence (AI) is rapidly populating medical records with synthetic or partially AI-generated content, creating a feedback loop where future models are increasingly at risk of training on uncurated AI-generated data. However, the clinical consequences of this AI-generated data contamination remain unexplored. Here, we show that in the absence of mandatory human verification, this self-referential cycle drives a rapid erosion of pathological variability and diagnostic reliability of medical data at population scale. By analysing more than 800,000 synthetic data points across clinical text generation, vision–language reporting, and medical image synthesis, we find that models progressively converge toward generic phenotypes regardless of the model architectures. Specifically, rare but critical findings, including pneumothorax and effusions, vanish from the synthetic content generated by AI models, while demographic representations skew heavily toward middle-aged male phenotypes. Crucially, this degradation is masked by false diagnostic confidence. Models continue to issue reassuring reports while failing to detect life-threatening pathology, with false reassurance rates tripling to 40%. Blinded physician evaluation confirms that this decoupling of confidence and accuracy renders AI-generated documentation clinically useless after just two generations. We systematically evaluate three mitigation strategies that can be easily integrated into existing clinical workflows, finding that while synthetic volume scaling fails to prevent collapse, mixing real data with quality-aware filtering effectively preserves diversity. Ultimately, our results suggest that without policy-mandated human oversight, the deployment of generative AI threatens to degrade the very healthcare data ecosystems it relies upon.

Abstract Breast cancer risks conferred by many germline missense variants in the BRCA1 and BRCA2 genes, often referred to as variants of uncertain significance (VUS), have not been established. In this study, associations between 19 BRCA1 and 33 BRCA2 missense substitution variants and breast cancer risk were investigated through a breast cancer case–control study using genotyping data from 38 studies of predominantly European ancestry (41,890 cases and 41,607 controls) and nine studies of Asian ancestry (6,269 cases and 6,624 controls). The BRCA2 c.9104A>C, p.Tyr3035Ser (OR = 2.52; P = 0.04), and BRCA1 c.5096G>A, p.Arg1699Gln (OR = 4.29; P = 0.009) variant were associated with moderately increased risks of breast cancer among Europeans, whereas BRCA2 c.7522G>A, p.Gly2508Ser (OR = 2.68; P = 0.004), and c.8187G>T, p.Lys2729Asn (OR = 1.4; P = 0.004) were associated with moderate and low risks of breast cancer among Asians. Functional characterization of the BRCA2 variants using four quantitative assays showed reduced BRCA2 activity for p.Tyr3035Ser compared with wild-type. Overall, our results show how BRCA2 missense variants that influence protein function can confer clinically relevant, moderately increased risks of breast cancer, with potential implications for risk management guidelines in women with these specific variants. Cancer Res; 77(11); 2789–99. ©2017 AACR.

Previous genome‐wide association studies among women of European ancestry identified two independent breast cancer susceptibility loci represented by single nucleotide polymorphisms (SNPs) rs13281615 and rs11780156 at 8q24. A fine‐mapping study across 2.06 Mb (chr8:127,561,724–129,624,067, hg19) in 55,540 breast cancer cases and 51,168 controls within the Breast Cancer Association Consortium was conducted. Three additional independent association signals in women of European ancestry, represented by rs35961416 (OR = 0.95, 95% CI = 0.93–0.97, conditional p = 5.8 × 10−6), rs7815245 (OR = 0.94, 95% CI = 0.91–0.96, conditional p = 1.1 × 10−6) and rs2033101 (OR = 1.05, 95% CI = 1.02–1.07, conditional p = 1.1 × 10−4) were found. Integrative analysis using functional genomic data from the Roadmap Epigenomics, the Encyclopedia of DNA Elements project, the Cancer Genome Atlas and other public resources implied that SNPs rs7815245 in Signal 3, and rs1121948 in Signal 5 (in linkage disequilibrium with rs11780156, r2 = 0.77), were putatively functional variants for two of the five independent association signals. The results highlighted multiple 8q24 variants associated with breast cancer susceptibility in women of European ancestry.