A small Yale study found a blood-based DNA methylation signature associated with response to progestin treatment among women with surgically confirmed endometriosis. The signal could support a future treatment-selection test, but the current evidence comes from 31 patients and does not establish a clinic-ready predictor.
The prospective cohort included 10 progestin responders and 21 nonresponders. Researchers profiled methylation across DNA from circulating white blood cells, then compared the two groups and built a three-gene classification model. The findings were published in Biomarker Research.
The Study Compared Two Small Clinical Groups
Progestins are commonly used as an initial hormonal treatment for endometriosis, yet some patients obtain inadequate symptom relief or stop because of adverse effects. The paper describes progestin resistance as an important part of the condition's clinical variability and cites earlier estimates that roughly one-third of patients do not respond.
All 31 participants in the new analysis had surgically confirmed endometriosis. Response status was assigned from clinical outcomes. The groups did not differ significantly in reported age, body mass index, race, ethnicity or menstrual phase, but the sample was too small to demonstrate broad demographic generalizability.
The surgical indications also differed. All 21 nonresponders were listed as having chronic pelvic pain, while the responder group underwent surgery because they no longer wanted hormonal treatment or no longer desired fertility. That imbalance does not invalidate the observed methylation difference, but it raises the possibility that the classifier captured other distinctions between the groups as well as treatment response.
Three Methylation Markers Drove the Classifier
The genome-wide analysis identified 1,439 genes with significantly different methylation between responders and nonresponders. The researchers then evaluated the most discriminating CpG sites, regions of DNA where methyl groups can affect gene regulation.
Three loci, MMP20, NRXN1 and RNA5-8SN5, each showed some ability to distinguish the groups. A logistic regression model combining the three produced an area under the receiver operating characteristic curve, or AUC, of 0.952. An AUC measures separation between groups in a dataset; it is not the percentage of patients who will receive a correct clinical prediction.
The authors used permutation testing and bootstrap resampling for internal validation. The bootstrap AUC was 0.907, with a 95% confidence interval from 0.80 to 0.957, and the permutation result was statistically significant. These checks test whether the result is likely to be a random artifact of the same sample, but they do not replace evaluation in new patients.
The Signature Does Not Explain the Biology Yet
The three selected loci were not established components of progesterone-receptor signaling. The investigators used an unbiased screen rather than restricting the search to known hormonal or inflammatory pathways. They suggested the markers might relate to inflammation, but the study did not demonstrate a causal mechanism.
Methylation in circulating leukocytes may reflect systemic inflammation, immune-cell composition, treatment exposure or other features of disease. The paper calls for mechanistic work linking the methylation changes to gene expression, identifying the relevant immune-cell types and testing functional relevance with targeted assays.
That distinction prevents a biomarker association from becoming a disease explanation. The model may eventually classify treatment response without any of the three loci directly causing resistance. Conversely, a mechanistically plausible marker would still need clinical validation before it could guide a prescription.
External and Pretreatment Validation Are Missing
No independent holdout cohort was used. The model was developed and internally checked in the same 31-patient dataset, with 26 participants identified as White and only one as Hispanic or Latino. Its performance may change in a larger population with different ancestry, disease stage, medication history or clinical setting.
The authors explicitly state that larger prospective cohorts are needed to confirm generalizability and determine whether the signature can predict response before therapy begins. That prospective sequence is essential: a treatment-selection test must be collected before the treatment decision and evaluated against a predefined outcome.
A future study would also need a clear response definition, standardized drug and follow-up intervals, prespecified thresholds and reporting of sensitivity, specificity, false positives and false negatives. A high AUC in a discovery cohort cannot show that using the test improves pain, quality of life or time to effective treatment.
No Commercial Blood Test Is Available From This Study
Yale researchers said they hope to develop a commercial test if the result is confirmed. That is a development plan, not evidence that clinicians can currently order the assay or use its result to bypass progestin treatment. The paper did not test insurer coverage, prescribing changes or patient outcomes resulting from biomarker-guided care.
Patients should not interpret these markers as a reason to start, stop or change hormonal treatment without clinical guidance. Endometriosis management depends on symptoms, pregnancy plans, contraindications, previous treatment, adverse effects and individual preferences. The study evaluated a possible decision-support signal, not a replacement for that assessment.
Precision Medicine Begins With Prospective Proof
The useful part of this study is the question it makes testable: can a blood sample identify likely nonresponse before patients spend months on an ineffective therapy? Its three-gene signature is specific enough to take into a larger trial, and the internal separation is strong enough to justify that next step.
But precision medicine is not precision because a model produces a high decimal. It earns the name only when a prespecified test works in new patients and improves a real decision. Until that evidence exists, this is a promising 31-patient biomarker signal. Calling it patient access to targeted medicine would skip the very validation needed to protect those patients from another confident but unreliable test.