Overview
A Danish prospective multicenter study has examined whether adding human epididymis protein 4, known as HE4, to the IOTA ADNEX model improves the ability to distinguish benign from malignant adnexal masses. Adnexal masses are growths in the region of the ovaries, fallopian tubes, or nearby pelvic structures. Correctly identifying which masses are likely to be cancerous is important because it helps guide referral, surgery, follow-up, and counseling.
The study found that HE4 added only limited value overall. In a tertiary referral center, performance improved slightly in some risk thresholds, but in non-tertiary centers the addition of HE4 did not meaningfully improve the model. Overall, the results do not support routinely adding HE4 to the ADNEX model.
What is the ADNEX model?
The IOTA ADNEX model is a well-known risk prediction tool developed by the International Ovarian Tumor Analysis group. It combines clinical and ultrasound features, and it can estimate the probability that an adnexal mass is benign or malignant. It is widely used because it helps clinicians make more consistent decisions when evaluating ovarian and pelvic masses.
The model can be applied with or without CA125, another blood marker often used in gynecology and oncology. CA125 can be elevated in ovarian cancer, but it is not specific and may also rise in benign conditions such as endometriosis, menstruation, or inflammation. Because of these limitations, researchers continue to look for better markers.
Why consider HE4?
HE4 is a blood biomarker that has attracted interest in ovarian cancer evaluation. It may be more specific than CA125 in some settings, especially in distinguishing malignant disease from certain benign gynecologic conditions. For that reason, investigators have asked whether HE4 can improve existing prediction models rather than be used alone.
The logic is simple: if HE4 adds new information beyond ultrasound findings and CA125, then the ADNEX model might become more accurate and more useful in everyday care. This study was designed to test that idea in a real-world Danish population.
Study design and methods
This was a prospective study conducted across 14 Danish gynecology units from 2020 to 2023, including one tertiary center and multiple non-tertiary centers. In total, 782 patients with adnexal masses were enrolled.
The researchers calculated ADNEX risk estimates in several versions: the original model, the model without CA125, and the model with HE4 added. To test how well the models worked, the study used internal validation by splitting patients into training and test groups. This approach helps reduce the chance that the results are due only to chance or overfitting.
Model performance was measured in several ways:
The area under the receiver operating characteristic curve, or AUC, which shows how well a model separates benign from malignant masses.
Brier score, which measures overall prediction accuracy and calibration. Lower scores indicate better performance.
Predictive values at clinical decision thresholds between 5% and 20%, which are important because they reflect how the model might be used in practice.
Malignancy was defined by tissue diagnosis from histopathology or by at least 12 months of follow-up if surgery was not performed.
Main findings
In the tertiary center, the ADNEX models that included HE4 performed slightly better than the original models. The AUC values were 86.2% to 86.3% with HE4, compared with 83.8% and 85.4% for the original models. The Brier scores were also slightly better with HE4, at 14.7 to 14.8 versus 15.5 to 16.4.
At some clinical thresholds, especially around 10% to 15% predicted risk, HE4 produced a modest improvement in positive predictive value. In practical terms, this means that when the model predicted higher risk, it was somewhat more likely to be correct in the tertiary setting. However, negative predictive values were similar across models, meaning the ability to reassure patients with low predicted risk did not change much.
In non-tertiary centers, the results were different. The models performed similarly whether HE4 was included or not. There was no clear improvement in discrimination, calibration, or predictive value. In other words, adding HE4 did not appear to help community or general gynecology units more accurately classify masses.
What do these results mean clinically?
This study suggests that HE4 is not a broadly useful addition to the ADNEX model for routine evaluation of adnexal masses. The modest gains seen in the tertiary center were limited and threshold-dependent, meaning they appeared only at certain risk cutoffs and were not strong enough to change the overall conclusion.
From a clinical standpoint, this matters because any added test should justify its cost, complexity, and blood draw burden. If a biomarker does not consistently improve decision-making across settings, it may not be worth incorporating into routine algorithms. That is especially true for a model like ADNEX, which already performs well using ultrasound-based features and established clinical variables.
The findings also highlight an important reality in diagnostic medicine: a marker that seems promising in theory may not deliver meaningful benefit once tested in diverse real-world settings.
Why might the tertiary center show some benefit?
Tertiary centers often care for more complex cases, refer patients with higher pretest risk, and may have more specialized imaging expertise. In such settings, an additional biomarker might provide a little extra information. However, that does not necessarily mean it should be used universally.
Non-tertiary centers represent a broader range of practice environments, and the fact that HE4 did not improve performance there suggests that its utility may be limited outside specialized referral care. This also means that any implementation would need careful consideration of local practice patterns, available expertise, and patient population.
Strengths and limitations
This study has several strengths. It was prospective, multicenter, and based on a relatively large cohort of patients with adnexal masses. It also used internal validation and clinically meaningful outcome measures.
There are also limitations to keep in mind. Even though the study was multicenter, it was conducted in Denmark, so results may not generalize perfectly to other healthcare systems or populations. Biomarker performance can vary by disease prevalence, referral pathways, and assay methods. The study also focused on incremental value rather than asking whether HE4 alone could diagnose disease, which is appropriate but means the question was narrowly defined.
Another point is that predictive models are only one part of clinical decision-making. Ultrasound expertise, symptoms, age, menopausal status, and patient preferences remain essential.
Implications for practice
For now, clinicians should not assume that HE4 needs to be added to the ADNEX model. The study does not show a consistent enough benefit to justify routine inclusion. Existing ADNEX-based assessment, with or without CA125 depending on local practice and guideline use, remains the more evidence-supported approach.
In practice, evaluation of an adnexal mass should still include careful ultrasound examination, assessment of risk factors, and appropriate referral when cancer is suspected. Blood markers may support decision-making, but they should not replace expert imaging or clinical judgment.
Bottom line
In this Danish prospective multicenter study, HE4 provided only limited additional benefit when added to the IOTA ADNEX model. The improvement was small, appeared mainly in one tertiary center, and was not seen in non-tertiary centers. The overall evidence does not support routine inclusion of HE4 in the ADNEX model for differentiating benign from malignant adnexal masses.
Reference
Karlsen NS, Høgdall CK, Dreisler E, Sakse AE, Gerds TA, Frikke-Schmidt R, Høgdall ES, Karlsen MA. Adding HE4 to the IOTA ADNEX model for differentiating adnexal masses: A Danish prospective, multicenter study. Gynecologic oncology. 2026-05-14;209:111-119. PMID: 42134000.
