Neural network-derived multivariate index assay demonstrates effective clinical performance in longitudinal monitoring of ovarian cancer risk

• Published in: Gynecologic oncology Vol. 187; pp. 21 - 29
• Main Authors: Pappas, Todd C., Roy Choudhury, Manjusha, Chacko, Balu K., Twiggs, Leo B., Fritsche, Herbert, Elias, Kevin M., Phan, Ryan T.
• Format: Journal Article
• Language: English
• Published: United States Elsevier Inc 01-08-2024
• Subjects: Adnexal mass; Adult; Aged; Aged, 80 and over; Benign; Biomarkers, Tumor - blood; Female; Humans; Longitudinal Studies; Machine learning; Malignancy; Middle Aged; Multivariate Analysis; Ovarian Neoplasms - blood; Ovarian Neoplasms - diagnosis; Risk Assessment - methods; Serum biomarker; Surgery; WAP Four-Disulfide Core Domain Protein 2; Adnexal mass; Malignancy; Serum biomarker; Surgery; Machine learning; Benign
• ISSN: 0090-8258; 1095-6859; 1095-6859
• DOI: 10.1016/j.ygyno.2024.04.020

We recently characterized the clinical performance of a multivariate index assay (MIA3G) to assess ovarian cancer risk for adnexal masses at initial presentation. This study evaluated how MIA3G varies when applied longitudinally to monitor risk during clinical follow-up.
The study evaluated women presenting with adnexal masses from eleven centers across the US. Patients received an initial blood draw at enrollment and at the standard-of-care follow-up visits. MIA3G was determined for all visits but physicians did not have access to MIA3G scores to determine clinical management. The primary outcome was the relative change value (RCV) of MIA3G over the period of clinical observation.
A total of 510 patients of 785 enrolled met study criteria. Of these, 30.8% had a second, 25.4% a third and 22.2% a fourth blood draw following initial collection. The median duration from initial draw was 131 d to second draw, 301.5 d to the third draw and 365.5 d to the fourth draw. MIA3G RCV of >50% was observed in 22–26% patients, whereas 70–75% patients had MIA3G RCV >5%. An empirical baseline RCV of 56% – transformed to 1 in logarithmic scale – was calculated from averaging RCVs of all patients who had no malignancy risk after 210 days. RCV > 1 log was associated with higher incidence of surgical intervention (29.6%) compared to RCV < 1 log (16.9%).
Variation in MI3AG does not change the accuracy of the test for excluding malignancy, while marked changes may be associated with a slightly higher likelihood of surgical intervention. In addition to MIA3G score itself, the MIA3G RCV may be important for clinical management.
•This study examined a novel application of multivariate index assay for longitudinal monitoring of ovarian cancer risk.•Relative change value of the MIA score post follow-up was associated with a higher incidence of surgical intervention.•The findings indicated the MIA relative change value is a useful clinical tool to monitor cancer risk for adnexal mass.