Abstract
Objective: The ADNEX ultrasound - based prediction model estimates the probability that an adnexal mass is benign, borderline, stage I primary invasive, stage II-IV primary invasive, or secondary metastatic. ADNEX was developed on patients managed surgically. We aimed to update ADNEX to ADNEX2 using more recent data on both surgically and conservatively managed patients. ADNEX2 is designed for use in patients in whom the benign descriptors (BDs), a set of ultrasound profiles with an estimated risk of malignancy below 1%, do not apply. Design: Secondary analysis of a multicentre prospective cohort study. Setting: 17 oncology and non-oncology centres from 8 European countries. Participants: Patients with a newly detected adnexal mass recruited between 2012 and 2016 who were managed surgically or conservatively with clinical and ultrasound follow-up. Interventions: ADNEX2 uses the same predictors as ADNEX (age, type of centre, lesion diameter, proportion of solid tissue, number of papillary projections, >10 cyst locules, ascites and acoustic shadows and CA125). Model updating involved logistic recalibration followed by Bayesian refitting. Validation was based on leave-one-centre-out cross-validation. ADNEX2 and ADNEX include CA125 but also work without CA125. Main outcome measures: Model performance regarding discrimination (AUC), calibration (plots and observed:expected, O:E, ratio), and clinical utility (net benefit). Results: Of 6847 patients, 4625 (68%) had a benign tumour, 1710 (25%) had a malignant tumour, while tumour outcome was uncertain for 512 (7%) patients. BDs applied in 2190 (32%) patients, of whom 0.5% (95% CI, 0.2%-0.9%) had a malignancy. A two-step strategy using BDs followed by ADNEX2 had an AUC of 0.95 (0.94-0.96) to distinguish benign from malignant tumours. When evaluated on 4657 patients in whom BDs did not apply, ADNEX2 had an AUC of 0.93 (0.91-0.94). The two-step strategy with ADNEX2 was better calibrated than the two-step strategy with ADNEX (O:E ratio of 1.05 versus 1.14) and had marginally higher clinical utility. Not including available CA125 values in ADNEX2 reduced the ability to distinguish stage II-IV primary invasive tumours from other invasive malignancies. Conclusions: ADNEX2 is better calibrated than ADNEX. CA125 measurement is not crucial to estimate the risk of malignancy, but is recommended to estimate the risk of malignant subtypes. Trial registration: clinicaltrials.gov NCT01698632 . What is already known on this topic: - The IOTA ADNEX risk model is widely used to estimate the probability that an adnexal mass is benign or one of four malignant subtypes, and has been integrated into a two-step strategy following the application of ultrasound benign descriptors (BDs). - Like nearly all prediction models for adnexal masses, the original ADNEX model was developed exclusively using data from patients who underwent surgery (recruited between 1999 and 2012). This introduces selection bias when the model is used to determine who should be referred for specialized care. - Measuring serum CA125 is not essential for estimating overall malignancy risk, but is recommended when patients are referred to oncology centres to distinguish stage II-IV primary invasive cancer from stage I primary invasive cancer and secondary metastases. What this study adds: - ADNEX was updated to ADNEX2 using more recent multicentre data (IOTA5) including both surgically and conservatively managed patients with tumours to which benign descriptors did not apply. - ADNEX2 (used alone or in a two-step strategy after benign descriptors) maintains high discrimination (AUC 0.95 for two-step strategy with ADNEX2; 0.93 when benign descriptors do not apply), improves risk calibration compared with the original ADNEX model (O:E ratio 1.05 v 1.14), and offers marginally higher clinical utility.
Funding Information
  • Swedish Research Council (K2014-99X-22475-01-3)
  • Swedish Research Council (Dnr 2013-02282)
  • Kom op tegen Kanker (C24/15/037)
  • Avtal om läkarutbildning och forskning (ALF)-medel
  • Landstingsfinansierad regional forskning
  • ZonMw, The Dutch Organisation for knowledge and innovation in health, healthcare and well-being (09150172310023)

This publication has 36 references indexed in Scilit: