The human body is home to trillions of bacteria, and the community of bacteria living in and around the reproductive tract may influence gynecologic cancer risk and detection. Endometrial cancer (uterine cancer) is difficult to screen for - there is no routine, widely adopted test comparable to a Pap smear for cervical cancer. Most women are diagnosed only when they develop symptoms, sometimes at an advanced stage.
Recent research has found that the vaginal and cervical microbiome (bacterial community) differs between women with endometrial cancer and healthy controls. Whether these differences are large enough and reproducible enough across different populations to form the basis of a non-invasive screening test is the question this study tried to answer.
The researchers collected vaginal swab samples from participants, sequenced the bacterial DNA using 16S rRNA gene sequencing (a standard method for identifying bacteria by their genetic fingerprint), and used machine learning to build a classifier that could distinguish endometrial cancer patients from cancer-free controls based purely on their microbiome composition.
A major challenge in microbiome research is reproducibility across different laboratories and processing methods. To address this, the researchers pooled data from five independent cohorts totaling 265 participants, sourced from different hospitals and geographic locations. This multi-cohort design is far more rigorous than single-center studies.
Two competing bioinformatics pipelines were compared for processing the raw sequencing data: QIIME2 and DADA2. DADA2 uses a statistical error model to resolve individual sequence variants at the level of single nucleotide differences (called ASVs - amplicon sequence variants), providing higher resolution than older methods that group bacteria into approximate clusters. DADA2 was selected as the superior pipeline for this application.
The team then tested multiple integration strategies for combining data across cohorts: early integration (merge all raw data before modeling), late integration (train separate models per cohort and combine predictions), and intermediate approaches. They also tested combinations of microbiome data with host characteristics (age, BMI, menopausal status) to see whether adding clinical information improved accuracy.
The best-performing model used late integration combining microbiome features and host characteristics, achieving an AUROC of 0.93 - a strong discrimination score. Most importantly for a screening test, the model achieved sensitivity of 1.0 (every cancer case was identified) and a negative predictive value (NPV) of 1.0 (no cancer patient was missed), with specificity of 0.70.
The 70% specificity means 30% of healthy women would receive a false positive result - a significant number that would require follow-up testing. However, for an initial screening test where the consequence of a false negative (missing cancer) is far worse than a false positive (unnecessary follow-up), high sensitivity and high NPV are the most important metrics.
Across all five cohorts, the bacterium Peptoniphilus was consistently enriched in endometrial cancer patients. This reproducibility across different geographic locations and laboratory methods suggests that Peptoniphilus enrichment is a genuine biological signal rather than a technical artifact - making it a potential microbiome biomarker for further investigation.
Peptoniphilus is an anaerobic bacterium (it grows without oxygen) from the family Peptostreptococcaceae. It has been detected in cancerous tissue from several cancer types and is thought to potentially create a local inflammatory microenvironment that promotes tumor growth. However, whether its presence in the vagina is a cause of endometrial cancer, a consequence of cancer-related changes to the local environment, or simply a correlate of shared risk factors is not yet known.
The consistency of the Peptoniphilus signal across 5 cohorts, even when using different sequencing methods and lab protocols, is scientifically noteworthy. Many microbiome biomarkers fail to replicate across cohorts due to technical variability or population differences. The fact that this finding held up suggests it is robust enough to merit dedicated mechanistic studies.
The vaginal microbiome is dominated by Lactobacillus in most healthy reproductive-age women; a shift away from Lactobacillus dominance toward a more diverse community has previously been linked to other gynecologic conditions. The endometrial cancer-associated microbiome profile appears to reflect a broader disruption of normal vaginal ecology, not just enrichment of one organism.
The practical vision of this research is a vaginal swab test - similar to a routine cervical smear but read by microbial sequencing rather than looking for abnormal cells - that could screen women for endometrial cancer. A test with 100% sensitivity means no cancer is missed; even if some healthy women need follow-up, no patient would have cancer go undetected at the screening stage.
Endometrial cancer is increasingly common as obesity rates rise (excess weight is a major risk factor). A non-invasive screening option would be particularly valuable for high-risk women with obesity, PCOS, or prolonged estrogen exposure, who currently have no routinely offered screening beyond clinical vigilance. Even a test with 70% specificity - meaning 3 in 10 healthy women get a positive result - could be acceptable as a first-line screen if follow-up testing (such as a uterine biopsy) is offered to all positives.
The study's 265-person dataset is a proof of concept, not yet a validated clinical tool. Much larger studies testing the model in different ethnic populations, age groups, and medical settings are needed. Additionally, it would be important to evaluate performance in a prospective setting - testing the model on women who do not yet have a diagnosis - rather than in a retrospective case-control design.