Considerations on Sample Size and Clinical Implications in the Study of Prediction Models for Adnexal Masses During Pregnancy

Abbreviations

ADNEX

Assessment of Different NEoplasias in the adneXa

AUC

area under the curve

CA‐125

cancer antigen 125

p‐IOTA

prospective International Ovarian Tumour Analysis

SR

simple rules

To the Editor:

We read with great interest the article by Niu et al ¹ recently published in ‘J Ultrasound Med ’. The authors compared the diagnostic performance of 3 commonly used prediction models—the Assessment of Different NEoplasias in the adneXa (ADNEX), simple rules (SR), and the Ovarian‐Adnexal Reporting and Data System (O‐RADS)—for adnexal masses in pregnant patients. This addresses a significant clinical challenge and provides valuable data for risk management in this special population, for which we commend the authors.

The study confirms the low incidence of malignancy in adnexal masses during pregnancy (4.6%, 3/65), which aligns closely with the 4.1% malignancy rate reported in a recent large‐scale study. ² Furthermore, the finding that malignant masses have a significantly larger maximum diameter than benign ones serves as a simple yet crucial clinical warning sign. However, upon deeper analysis of the results, we believe several key issues warrant further discussion to more accurately interpret the findings and guide future research.

First, the extremely low number of malignant cases severely limits the reliability and statistical power of the model performance evaluation. Among the 65 cases included, only 3 were malignant. In predictive model studies, the stability of performance metrics—particularly sensitivity, specificity, and the area under the curve (AUC)—heavily depends on the number of target events (here, malignancy). ³ With only 3 events, the confidence intervals for the AUC values are likely to be very wide. The reported “point estimates” of model performance (eg, AUC of 0.64 for ADNEX, 0.91 for SR, and 0.94 for O‐RADS) may be considerably influenced by chance. This could partially explain why such notable numerical differences were reported as “not statistically significant.” A recent study aiming to validate the ADNEX model in pregnancy also explicitly stated that its evaluation was primarily limited by the “low number of malignant tumors.” Therefore, while the conclusion that SR and O‐RADS showed “good diagnostic efficacy” points in a positive direction, it should be extrapolated with caution.

Second, the relatively suboptimal performance of the ADNEX model may have specific physiological causes. The ADNEX model incorporates patient age, serum cancer antigen 125 (CA‐125) levels, and sonographic features. ⁴ During pregnancy, physiological elevation of CA‐125 can occur, which may significantly interfere with the model's algorithm, leading to reduced specificity or risk misclassification. In contrast, SR and O‐RADS are primarily based on ultrasound morphology and are less affected by such biochemical variability. ⁵ We note that a study published in 2024 also found a higher misclassification rate for benign masses (particularly decidualized endometriomas) when applying the ADNEX model in pregnancy. Consequently, the applicability of the ADNEX model in pregnant populations may require recalibration, or at the very least, its specific risk percentage outputs should be interpreted with caution.

In summary, this study highlights the challenges in validating diagnostic tools during the unique physiological period of pregnancy: the target condition is rare (low malignancy rate), and physiological changes may alter model performance. These findings carry important clinical implications: (1) For masses with typical ultrasound morphology, SR or O‐RADS may provide a more reliable initial risk assessment; (2) The risk thresholds established for the ADNEX model based on non‐pregnant populations should not be directly applied to guide management decisions in pregnancy.

We agree with the authors' emphasis in the abstract on the clinical application value of SR and O‐RADS. Simultaneously, we believe that definitively establishing the optimal application strategy for these models in pregnancy must rely on large‐scale, prospective, multicenter studies. We are pleased to note that an international multicenter study prospective International Ovarian Tumour Analysis (p‐IOTA) prospectively validating the ADNEX model in pregnancy is underway, ⁶ and its results will provide higher‐level evidence for this field. Until then, clinicians should fully consider the limitations revealed by this study when interpreting model results and regard subjective assessment by ultrasound experts as central to decision‐making in complex cases.

Data Availability Statement

Data sharing not applicable to this article as no datasets were generated or analyzed during the current study.