Endometrial cancer incidence has been rising steadily, with a 56% increase in the UK over two decades and 380,000 new cases globally in 2018. Despite this trend, the diagnostic pathway remains difficult: symptoms like postmenopausal bleeding are common, yet only 5-10% of symptomatic women have cancer. Current investigations begin with transvaginal ultrasound, an intimate procedure that is expensive and operator-dependent, and women with a thickened endometrium then face an invasive biopsy -- which yields insufficient tissue in up to 22% of cases.
Blood biomarkers have long been sought as a simpler, more acceptable alternative. Known markers like CA-125 and HE4 have insufficient accuracy for clinical use in endometrial cancer. This study investigated attenuated total reflection Fourier transform infrared (ATR-FTIR) spectroscopy, a technique that shines infrared light through a blood plasma sample and measures how different molecular bonds absorb specific wavelengths, producing a biochemical fingerprint of the sample in seconds.
The study enrolled 652 women from clinics across Greater Manchester and Lancashire: 242 healthy controls (confirmed cancer-free by biopsy and at least 12 months of follow-up), 342 women with confirmed endometrial cancer (258 Type I, 64 Type II, 20 mixed), and 68 with atypical hyperplasia (the precancerous lesion). Blood was drawn into standard tubes, centrifuged to separate the plasma, and stored at minus 80 degrees Celsius until analysis. For each patient, 50 microliters of plasma was deposited onto a glass slide, air-dried, and placed on the spectrometer.
The spectrometer collected five measurements per patient from different spots on the dried blood sample, producing a total of 3,260 spectra. Raw spectra were mathematically pre-processed to remove instrument noise, then analyzed using partial least squares discriminant analysis (PLS-DA) -- a supervised machine learning approach that identifies which spectral regions best separate the groups. Samples were split into 70% training and 30% test sets. A key strength of the study was that potential confounders -- including age, BMI, diabetes, blood pressure, and fasting status -- were recorded for all participants and explicitly tested to ensure they were not driving the spectral differences.
For detecting any endometrial cancer against healthy controls, the blood test achieved 87% sensitivity and 78% specificity, with an AUC of 88%. Breaking this down by cancer subtype: Type I cancer (the most common, endometrioid type) was detected with 91% sensitivity, 81% specificity, and AUC of 92%. For the more aggressive Type II cancers, results were 79% sensitivity and 88% specificity (AUC 88%). The most striking result came for atypical hyperplasia (the precancerous lesion): the test achieved 100% sensitivity and 88% specificity, with an AUC of 98%, suggesting that blood spectroscopy may be able to identify women at risk of developing cancer before it fully forms.
For early-stage disease specifically, the test could still distinguish Stage I cancers from controls with an AUC of 80% -- important because Stage I cancers, caught before spread, have the best prognosis. The test also showed it could discriminate between Type I and Type II cancers with 87% AUC, which has clinical implications since the two types require different treatment approaches.
Infrared spectroscopy does not target a single protein or molecule; instead it captures the vibration patterns of all molecular bonds simultaneously. Analyzing which spectral peaks drove the cancer-versus-control separation, researchers identified two statistically significant markers: a peak at 1446 cm-1, reflecting lipid molecules (specifically CH2 bending in fatty acids), which was elevated in cancer; and a peak at 900 cm-1, reflecting carbohydrate and fatty acid structures, which was decreased in cancer. These patterns were consistent across both Type I and overall cancer comparisons. Elevated lipid signatures in cancer blood have been reported before in breast cancer research, but the decreased carbohydrate peak is a novel finding from this study.
For atypical hyperplasia, different peaks stood out: a protein-related peak at 1404 cm-1 was decreased, while a peak at 1238 cm-1 (linked to collagen and nucleic acids) was increased -- suggesting the precancerous changes in the uterus leave a detectably different biochemical signature in the blood than frank cancer. Critically, none of the confounding factors (age, BMI, diabetes, blood pressure, or fasting status) produced spectral differences that could explain these findings, confirming the signals reflect disease biology rather than patient characteristics.
The current standard for initial investigation of postmenopausal bleeding is transvaginal ultrasound, which at the commonly used 5mm endometrial thickness threshold achieves 90% sensitivity but only 54% specificity -- meaning more than half of women sent for invasive biopsy do not have cancer. The blood spectroscopy test in this study achieved 87% sensitivity with superior specificity of 78%, meaning fewer unnecessary invasive procedures for women without disease. The authors propose spectroscopy could serve as a triage test: a negative result would safely reassure women with low suspicion, while a positive would fast-track those most likely to benefit from urgent biopsy.
The test's potential extends beyond diagnosis of symptomatic women. The authors identify several additional clinical scenarios: screening high-risk asymptomatic women (particularly those with obesity or Lynch syndrome hereditary cancer), distinguishing between cancer histological types to guide treatment, and monitoring women managed conservatively without surgery. Portable handheld infrared spectrometers are already being trialed for point-of-care testing in developing countries, raising the possibility that a low-cost version of this test could eventually reach settings where ultrasound and biopsy services are unavailable.
This is the largest study to date of blood spectroscopy for endometrial cancer detection. Unlike previous smaller studies, it included the full range of cancer stages and histological subtypes, recruited from multiple clinical sites, confirmed all controls by both biopsy and 12-month follow-up, and rigorously evaluated confounding factors. The study demonstrates that the spectroscopic signal is genuinely driven by disease biology and is not an artifact of the population difference in age or obesity between cancer patients and controls.
Key limitations include the inherently tentative nature of assigning spectral peaks to specific molecules, since each infrared region represents overlapping contributions from many different biomolecules. Prospective validation in larger independent cohorts -- including asymptomatic high-risk women followed over time -- is needed before clinical implementation. The researchers call for prospective diagnostic accuracy studies to confirm performance in real-world settings and determine how the test fits alongside existing clinical decision-making pathways.