Self-collected first-void urine cytology as a non-invasive triage tool for endometrial cancer detection

Self-collected first-void urine cytology detected endometrial cancer in 73.3% of women with postmenopausal bleeding in a prospective diagnostic accuracy study conducted at St Mary’s Hospital in Manchester, while achieving 90.7% specificity and an area under the receiver operating characteristic curve of 0.85. The findings support further investigation of urine as a non-invasive triage tool, but the test missed enough cancers to prevent it from serving as a stand-alone rule-out test. In the same cohort, clinician-collected vaginal cytology had higher sensitivity, at 92.2%, but lower specificity, at 84.4%, and a higher area under the curve, at 0.94.

The single-centre study recruited 200 women and people with a uterus between January 2023 and July 2025. It included 102 histologically confirmed endometrial cancer cases and 98 symptomatic controls who had been referred on a suspected endometrial cancer pathway with postmenopausal bleeding. The investigators deliberately enriched the cohort with known cancer cases, so the observed cancer prevalence of 51% was much higher than that generally encountered in routine postmenopausal bleeding services. Controls self-collected first-void urine before routine diagnostic procedures and then provided a clinician-collected vaginal sample. In participants with known endometrial cancer, research samples were collected at least seven days after diagnostic procedures to reduce contamination from cells dislodged during routine investigations. A cytologist blinded to participant disease status assessed both specimens using predefined morphological classifications for negative, atypical, suspicious and adenocarcinoma results.

Four samples were excluded from cytology analysis because they were inadequate: two urine samples and two vaginal samples from different participants. Among the 198 evaluable participants, urine cytology had an overall accuracy of 81.8%, with a positive likelihood ratio of 7.9 and a negative likelihood ratio of 0.30. Vaginal cytology detected 92.2% of cancers, significantly more than urine cytology, while its specificity of 84.4% was not significantly different from the 90.7% observed for urine. Vaginal cytology missed 8 of 102 cancers, whereas urine cytology missed 27, but urine cytology did not identify any cancer that vaginal cytology missed. The area under the curve was significantly higher for vaginal cytology than for urine cytology, with a paired comparison giving z=-4.57 and p<0.001.

Agreement between the two sampling methods was substantial, with a Cohen’s kappa statistic of 0.70 and a 95% confidence interval of 0.60–0.79. Of the 102 cancers, both methods were positive in 74, while vaginal cytology alone detected 19 cancers missed by urine cytology. Eight cancers were missed by both methods. The discordance was concentrated in lower-risk disease: 24 of the 27 urine false negatives were early-stage cancers, 19 were low-grade, 22 were endometrioid tumours, 20 had less than 50% myometrial invasion, and 19 lacked lymphovascular space invasion. In 11 of the 27 missed cases, the tumour was confined to an endometrial polyp. These observations indicate that urine cytology was less able to detect cancers with a smaller or more localized tumour burden, although the study did not directly measure tumour burden as the cause of the missed cases.

Urine cytology performed better across several clinical and pathological subgroups associated with higher risk. Sensitivity was 80.5% for high-grade tumours, 90.5% for late-stage disease, 88.4% when at least half of the myometrium was invaded and 88.0% in cancers with lymphovascular space invasion. It also detected 83.3% of non-endometrioid tumours and 80.6% of p53-abnormal cancers in the available subgroup analyses. Detection was lower among women with bleeding lasting less than three months, at 64.6%, and among current hormone replacement therapy users, at 62.5%, although confidence intervals were wide in some of these smaller groups. Sensitivity also varied with other clinical characteristics, including age and bleeding history. The study did not test urine cytology in asymptomatic women or establish its performance as a screening test, so its findings apply to the symptom-based postmenopausal bleeding pathway studied here.

The investigators also examined whether urine cytology contributed information beyond routine clinical characteristics. In univariable analysis, positive urine cytology was strongly associated with endometrial cancer, with an odds ratio of 30.38 and a 95% confidence interval of 12.77–72.27. After adjustment for age, body mass index, parity, hormonal contraception, hormone replacement therapy, bleeding duration and recurrent bleeding, positive urine cytology remained an independent predictor, with an adjusted odds ratio of 23.78 and a 95% confidence interval of 8.29–68.15. Recurrent bleeding was also associated with higher odds of cancer in the multivariable model, while multiparity, ever use of hormonal contraception and current hormone replacement therapy were associated with lower odds. These results show that the urine test carried information not fully captured by the clinical variables included in the analysis; they do not establish that cytology can be used in place of those clinical assessments.

The collection device was changed during the study from a 10 mL version to a 20 mL version, creating an exploratory comparison that was underpowered and vulnerable to differences between the two cohorts. Sensitivity was numerically higher with the 20 mL device, at 74.6% with a 95% confidence interval of 63.0–83.0, compared with 70.0% with a 95% confidence interval of 52.0–83.0 for the 10 mL device, but the difference was not statistically significant, with p=0.629. Specificity was numerically higher for the 10 mL device, at 93.8% compared with 84.8%, also without statistical significance, with p=0.152. The groups differed in body mass index, hormone replacement therapy and hormonal contraception use, and the proportion of cancer cases was not balanced. The results therefore cannot establish that collection volume itself caused the differences in accuracy.

Because the study cohort contained an unusually high proportion of cancers, predictive values were modelled using assumed endometrial cancer prevalences of 5% and 10%, which are more representative of routine postmenopausal bleeding populations. At 5% prevalence, the modelled positive predictive value for urine cytology was 29.3% and the negative predictive value 98.5%; at 10% prevalence, the corresponding estimates were 46.7% and 96.8%. These calculations assumed that the observed sensitivity and specificity would transfer to an unselected clinical population and therefore require prospective validation. The high modelled negative predictive value did not change the study’s main limitation: with a sensitivity of 73.3%, a negative urine result would leave a meaningful probability of cancer and could not safely replace investigation of symptomatic patients.

The study has several constraints that affect how the findings can be interpreted. Histology served as the reference standard where available, but some disease-negative controls were classified through discharge from the diagnostic pathway, normal ultrasound or hysteroscopy and six months of follow-up without biopsy. This mixed reference standard could introduce verification bias, and six months may not detect all occult or indolent malignancies. The participants were recruited from one hospital and were predominantly white British, limiting generalizability to other populations and healthcare systems. Participants with known cancer were sampled at least seven days after diagnostic procedures to reduce procedure-related contamination, although prior biopsy or hysteroscopy could still have affected cytology yield. The cytology was also assessed by one experienced specialist, so inter-observer variability was not evaluated. These limitations mean that the reported performance estimates should not yet be treated as definitive estimates for routine practice.

The findings position urine cytology as a potential preliminary triage tool rather than a diagnostic replacement. Its practical advantages include self-collection, an integrated preservative, standardized collection volume and the possibility of community-based sampling and postal return, including situations in which vaginal examination is difficult or unacceptable. A positive urine result could support prioritization for confirmatory investigation, while a negative result would still require appropriate clinical safety-netting and established diagnostic assessment. Combining urine with vaginal cytology was not supported by the results because urine detected no additional cancers in this cohort. Vaginal cytology itself showed high sensitivity, but cytology remains a subjective test whose scalability, inter-observer reproducibility and integration into clinical pathways require further evaluation. Prospective studies should now test both approaches in consecutive, unselected patients, alongside health-economic analyses and investigation of molecular biomarkers or digital image analysis that might improve sensitivity and consistency.

Subject of Research: Diagnostic accuracy of self-collected urine cytology for endometrial cancer triage in symptomatic women with postmenopausal bleeding

Article Title: Self-collected first-void urine cytology as a non-invasive triage tool for endometrial cancer detection

Article References: Johnson, J.-E., Narine, N., Chandraya, A. R., O’Flynn, H., Dore, M., & Davidson, E. J. (2026). Self-collected first-void urine cytology as a non-invasive triage tool for endometrial cancer detection. Frontiers in Oncology, 16, Article 1910651. https://doi.org/10.3389/fonc.2026.1910651

Image Credits: AI Generated

DOI: 10.3389/fonc.2026.1910651

Keywords: endometrial cancer, urine cytology, postmenopausal bleeding, diagnostic accuracy, self-collection, non-invasive triage, vaginal cytology, Colli-Pee, cancer detection, Self-collected, first-void, urine