Highlights
- In this prospective Danish multicenter cohort of 966 women with adnexal masses, IOTA-based models outperformed the older Risk of Malignancy Index (RMI) for early malignancy detection in routine non-tertiary care.
- A malignancy-risk threshold of 15% for the IOTA ADNEX-based strategy appeared to offer the best clinical balance, with sensitivity of about 63% and a referral rate of about 14% in non-tertiary centres.
- Higher thresholds reduced referrals to levels similar to RMI, but sensitivity also fell, with only modest additional practical benefit.
- When a mass was already classified as benign by modified benign descriptors, adding ADNEX provided little incremental value and did not meaningfully improve net benefit.
Clinical background: why adnexal mass triage matters
Adnexal masses are common in gynaecologic practice, but their clinical significance varies widely. Many are benign cysts or functional lesions that can be managed conservatively, while others represent borderline tumours or invasive ovarian cancer that should be referred promptly to a gynaecologic oncologist. The core challenge in routine care is triage: identifying which patients need specialist surgery and which can safely remain in general gynaecology.
For years, the Risk of Malignancy Index (RMI) has been used to support this decision. It combines menopausal status, ultrasound features, and CA125, but it was developed in an earlier era of imaging and does not fully exploit the information available on modern ultrasound. The IOTA group developed more contemporary ultrasound-based models, including the ADNEX model, to estimate the probability that an adnexal mass is benign, borderline, or malignant. The Two-Step Strategy combines simple benign descriptors with ADNEX for lesions that are not clearly benign.
The practical question is not just whether a model predicts cancer better in theory, but whether it improves care in the real world without creating an unsustainable referral burden. That is the question this Danish study set out to answer.
Study design and methods
This was a prospective multicenter cohort study conducted in Denmark across 13 non-tertiary hospitals and clinics plus one tertiary referral hospital. The complete-case cohort included 966 patients with adnexal masses. The investigators collected clinical data, ultrasound findings, and CA125 levels prospectively, then used these inputs to estimate malignancy risk with the IOTA ADNEX model, the Two-Step Strategy, and the RMI.
The reference standard was histopathology whenever surgery was performed or at least 12 months of clinical follow-up when surgery was not performed. This is an important feature of the design because it reflects routine practice rather than a highly selected surgical population alone.
Performance was assessed at predefined thresholds. For ADNEX and the Two-Step Strategy, thresholds ranged from 1% to 30%. For RMI, the standard threshold was 200. The study was also stratified by centre type, which is clinically relevant because non-tertiary centres make most referral decisions, while tertiary centres receive more complex cases and often have a different case mix.
The main outcomes were negative predictive value (NPV), positive predictive value (PPV), sensitivity, and referral rate. In this context, NPV reflects how safely a model can reassure clinicians that a mass is unlikely to be malignant, while PPV reflects how often a positive test truly indicates malignancy. Referral rate captures the burden of sending patients to a higher level of care.
Key findings
Across the study, IOTA-based strategies generally provided better triage performance than RMI, especially in the non-tertiary settings where the referral decision is made. The main clinical trade-off was straightforward: lower thresholds and more inclusive referral rules captured more malignancies, but they also increased the number of women sent for specialist assessment.
Performance in non-tertiary centres
In non-tertiary centres, NPVs were high for both IOTA-based approaches and RMI, but the IOTA models were consistently at least as reassuring and often better. The NPVs for IOTA models were 96% or higher, compared with 95% for RMI. This suggests that, in routine community gynaecology, IOTA models can safely classify many masses as low risk without materially sacrificing rule-out performance.
PPV rose as the malignancy threshold increased, as expected. At around a 20% threshold, PPV for the IOTA-based strategies approached the PPV seen with RMI. In other words, the more conservative the threshold, the more the model resembled the older referral strategy in terms of the proportion of positive tests that were actually cancer.
The most clinically informative result was the threshold analysis. A 15% threshold for the ADNEX-based strategy gave the most favourable balance between sensitivity and referral burden in non-tertiary centres. Sensitivity was about 63%, and the referral rate was about 14%. That means the model identified substantially more malignancies than RMI while still keeping the number of referrals relatively contained.
What happens if the threshold is raised?
At thresholds of 25% or higher, referral rates became similar to those seen with RMI, around 8%, but the gain in sensitivity was only modest. Sensitivity was about 50% with the IOTA-based strategy compared with about 39% for RMI. This illustrates an important implementation principle: if the goal is to capture more early cancers, a threshold that is too high can erase much of the benefit of using a better model.
| Strategy in non-tertiary centres | Sensitivity | Referral rate | Practical interpretation |
|---|---|---|---|
| ADNEX / Two-Step at 15% | About 63% | About 14% | Best balance of detection and referral burden |
| ADNEX / Two-Step at 25% or higher | About 50% | About 8% | Referral burden similar to RMI, but more cancers detected than RMI |
| RMI at ≥200 | About 39% | About 8% | Lower sensitivity with familiar but older triage rules |
Performance in tertiary centres
In the tertiary referral hospital, NPVs were more variable, ranging from 82% to 100% for IOTA models and about 78% for RMI. This is not surprising because tertiary centres see a higher-risk and more heterogeneous population, which makes predictive values less stable and more dependent on disease prevalence. Importantly, the study’s implementation question primarily concerns non-tertiary centres, where first-line triage is actually performed.
Two-Step Strategy versus ADNEX after benign descriptors
The study also examined the subgroup of masses already classified as benign by modified benign descriptors. In that setting, ADNEX still showed high NPV but low PPV and negligible net benefit. Practically, this means that if ultrasound already indicates a mass is clearly benign, applying a more sophisticated risk calculator adds little. The Two-Step Strategy may therefore be sufficient for these lesions, reserving ADNEX for masses that are not obviously benign.
This is an important nuance for implementation. More information is not always better if it does not change management. A streamlined pathway can improve usability, reduce cognitive burden, and support uptake in routine clinics.
Expert commentary
This study is valuable because it evaluates a modern risk model in routine care rather than in a highly selected expert ultrasound environment. That makes the findings more relevant to real-world triage decisions. It also addresses an underappreciated implementation issue: the optimal threshold is not purely statistical. It depends on how much referral capacity a health system has, how willing clinicians are to tolerate false negatives, and whether the priority is early cancer detection, resource preservation, or both.
From a clinical perspective, the 15% threshold appears to be a reasonable starting point for non-tertiary centres. It improves sensitivity meaningfully over RMI without causing an excessive increase in referrals. That said, the threshold should not be treated as universal. Centers with different prevalence, imaging expertise, referral pathways, or surgical capacity may need to recalibrate the cutoff.
There are also limitations to consider. First, this was a complete-case cohort, which may introduce selection bias if missing data were not random. Second, the work was performed in Denmark, where ultrasound training, referral structures, and access to gynaecologic oncology may differ from other health systems. Third, the reference standard relied on histopathology or 12 months of follow-up; while appropriate for routine care, some slow-growing lesions could theoretically escape classification within that window. Finally, the study compares diagnostic performance and referral burden, but it does not directly measure patient-centred outcomes such as time to treatment, surgical complexity, or survival.
Even with those limitations, the clinical direction is clear: newer IOTA-based strategies appear better suited than RMI for first-line triage of adnexal masses in modern practice. The challenge is translating better discrimination into a workflow that clinicians can actually use.
Conclusion
In this prospective Danish multicenter cohort, the IOTA ADNEX model and the Two-Step Strategy improved triage of adnexal masses compared with RMI, especially in non-tertiary centres where referral decisions are made. A 15% malignancy-risk threshold offered the most practical balance between early cancer detection and referral burden. Higher thresholds reduced referrals but also reduced sensitivity, limiting the clinical advantage over RMI.
For routine gynaecologic care, the message is not simply that ADNEX is better, but that implementation depends on choosing the right threshold. For many non-tertiary settings, 15% is a defensible compromise. For clearly benign masses, the added value of ADNEX appears limited once benign descriptors have already been met.
Funding and trial registration
The abstract does not specify the funding source. Trial registration: ClinicalTrials.gov identifier NCT04188652.
References
1. Karlsen NS, Dreisler E, Høgdall CK, Høgdall ES, Karlsen MA, Gerds TA, Andrésdóttir G, Langhoff-Thuesen L, Forstholm MM, Sakse AE. Evaluation of the IOTA ADNEX Model, Two-Step Strategy and RMI in Routine Gynaecologic Care in Denmark and Implications for Implementation: A Prospective Multicenter Cohort Study. BJOG. 2026-04-26. PMID: 42036775.
2. Jacobs I, Oram D, Fairbanks J, Turner J, Frost C, Grudzinskas JG. A risk of malignancy index incorporating CA125, ultrasound and menopausal status for the accurate preoperative diagnosis of ovarian cancer. Br J Obstet Gynaecol. 1990;97(10):922-929.
3. Van Calster B, et al. Evaluating the risk of ovarian cancer before surgery using the ADNEX model. Ultrasound Obstet Gynecol. 2014;44(5):532-541.
