A reflex testing protocol using two multivariate index assays improves the risk assessment for ovarian cancer in patients with an adnexal mass

Patients with adnexal masses suspicious for malignancy benefit from referral to oncology specialists during presurgical assessment of the mass. OVA1 is a multivariate assay using a five‐biomarker panel which offers high overall and early‐stage sensitivity. However, OVA1 has a high false‐positive rate for benign masses. Overa, a second‐generation multivariate index assay was developed to reduce the false‐positive rate. The aim of the present study was to use Overa as a reflex for OVA1 and increase specificity.

As ovarian cancer is the most deadly gynecologic cancer, and one of the most difficult to detect in early stages, OVA1 was developed to favor high sensitivity in order to miss as few malignancies as possible. Its cancer detection rate is on the order of 90% with a positive predictive value in the range of 30%-40%. 4,5 OVA1 is not meant to be used as a standalone test, and its sensitivity in conjunction with clinical assessment is 96%. 5 Consequently, most cancers, including early-stage cancers, are readily detected by OVA1. However, the patient referral rate is higher than desired, at approximately 50%, including many benign cases. While there is no negative impact of these unnecessary patient referrals on the quality of care and patient outcome, these false positives can result in the ineffective use of medical resources, and undue anxiety and stress for the patient. 6 Overa ® , a second-generation multivariate index assay, was developed to improve the specificity of OVA1. 7 Overa was derived from the serum concentrations of APO-A1, TRF, CA125, Human Epididymis Protein 4 (HE4) and follicle-stimulating hormone (FSH). These inputs were processed by a support vector machine learning algorithm to combine the five inputs into a new risk score ranging from 1.0 to 10.0. [7][8][9] The sensitivity of Overa is equivalent to that of OVA1, but the specificity is greatly improved. While both OVA1 and Overa algorithms use five biomarker inputs, of which three (CA125, TRF, and APO-A1) are the same, Overa uses FSH and HE4 to replace the B2M and TT biomarkers that are used in the OVA1 algorithm.
Both algorithms utilize a modified support vector machine ( ing additional statistical information about the data point's position relative to the distribution of all the classes of samples in the training data. The rationale behind UMSA is that information about the overall data distribution can be used to qualify the "trustworthiness" of any of the training samples to become a support vector. The final UMSA-SVM solution, therefore, will rely on the weighted contributions of the support vectors and be less sensitive to labeling errors of a small percentage of the samples.
Here, we report the use of Overa as a reflex test to reduce the false-positive rate of OVA1. A two-step procedure was employed.
First, OVA1 was used to determine the risk category as low, intermediate, or high. The low-risk category was defined as patients having OVA1 scores <4.4 (postmenopausal) and <5.0 (premenopausal).
High-risk OVA1 scores >6.0 (postmenopausal) and >7.0 (for premenopausal) are used to identify patients with a high risk of malignancy. Samples with OVA1 scores falling within ranges of 4.4-6.0 for postmenopausal patients and 5.0-7.0 for premenopausal patients are classified as intermediate risk, and were reflexed to Overa.

| RE SULTS
A clinical summary of the patients in the 1035-sample data set is shown in Table

Early stage (I and II) sensitivity (95% CI) (n/N)
The specificity was improved from 55.6% to 73.4%. The number of cancers that were correctly identified by OVA1 but misclassified by OVA1plus in this high-cancer-prevalence dataset was only 4 out of 74. Due to the exceptionally high prevalence of malignancy in this dataset, the positive predictive value (PPV) is high (68%), but the negative predictive value (NPV) was only slightly affected (88%).
Lastly, in this dataset, CA125 and ROMA show low cancer detection rates (75% and 68%, respectively), but with high specificity.
The impact of reflex testing on likelihood ratios (LRs) and posttest probabilities are displayed in Table 4. This analysis was per-

| DISCUSS ION
There is a large body of literature showing that patients with a high risk of ovarian malignancies benefit from referral to a gynecologic oncologist for specialty surgical care, where they are more likely to receive appropriate staging and more complete debulking and cytoreduction, resulting in improved survival. 15 However, there are challenges associated with determining which patients should be referred and when. Gynecologic oncology is a specialized field with a small number of practitioners in the USA and even fewer outside of large metropolitan centers. 16 Access to these specialists may be limited, and travel time and cost are factors, as are the anxiety and psychological toll associated with referral to a cancer center for treatment. Therefore, an ideal risk assessment tool for ovarian cancer would combine a high sensitivity with a high specificity in order to detect most cancers, and minimize false positives that result in inappropriate referrals for patients who actually have benign masses. A test with higher specificity helps to determine which patients may be safely retained by the general gynecologist for surgical intervention. Improved risk assessment reduces the burden on specialists, avoids the wasteful use of medical resources, and ultimately benefits the patient.
While appropriate management of low risk masses is important, ovarian cancer is the deadliest gynecologic cancer and the fifth leading cause of death in women. 17 Its symptoms are non-specific and early detection is particularly challenging, as many early-stage ovarian malignancies may be entirely asymptomatic. 18 As such, approximately 75% of patients are not diagnosed until the disease has progressed to an advanced stage, at which point the prognosis is poor, with a 5-year survival rate as low as 17%. 18 The potential consequences of missing a malignant mass are severe. Therefore, an ideal risk assessment tool for ovarian cancer cannot neglect sensitivity either. A balance between sensitivity and specificity is needed for cancer risk assessment. alone. [22][23][24] Our results in Tables 2 and 3 corroborate the low sensitivities of CA125 and ROMA for cancer detection.
OVA1plus, the reflex protocol that utilizes the OVA1 and Overa algorithms, produces a risk assessment score that gives high sensitivity and improved specificity compared with either OVA1 or Overa used alone. The non-invasive serum biomarker reflex test has high sensitivity for early-stage cancer detection and its improved specificity over OVA1 used alone leads to fewer false positives, resulting in fewer unnecessary referrals, which is beneficial for patients, practitioners, and the healthcare system at large. The OVA1plus risk result is a binary classification, eliminating the confusion caused by indeterminate results. In our data set, 35% of the OVA1 false positives were determined to be indeterminate. When Overa is employed as an automatic reflex test, the indeterminate category is eliminated.
OVA1plus has the same intended use as OVA1, in that it is a risk assessment tool used as part of the clinical work-up of a confirmed adnexal mass for which surgery is planned, in order to help guide referral decisions. The use of Overa as a reflex tool for samples that fall within the OVA1 within intermediate range significantly improves the accuracy of cancer risk assessment.
As such, OVA1plus is a valuable tool when utilized as part of the clinical work-up of an adnexal mass, helping to guide the decision to refer a patient to specialty care for surgical intervention, or to determine that the patient can be safely treated without referral.
A limitation of this report is the retrospective nature of the data.
Studies are under way to investigate the clinical performance and utility of OVA1plus in a population that reflects a 'real-world' demographic of patients.

AUTH O R CO NTR I B UTI O N S
HAF and RGB contributed significantly to drafting, editing, formal analysis, and final approval of the manuscript.

CO N FLI C T O F I NTE R E S T S TATE M E NT
Both authors were employed or contracted by Aspira Women's Health Inc. or its subsidiary, Aspira Labs, at the time of contribution. Aspira Women's Health, Inc. provided funding for this study.
No grant funding was used in the referenced trials or the preparation of this manuscript.

DATA AVA I L A B I L I T Y S TAT E M E N T
Data sharing is not applicable to this article as no new data were created or analyzed in this study.