LBSAT255 Real World Performance Of The Afirma Genomic Sequencing Classifier (GSC) - A Meta-analysis

The Afirma GSC aids in the clinical decision making for patients with indeterminate thyroid nodule cytology (ITN). The 2018 GSC validation study was a prospective, multi-center study, conducted on a patient cohort with ITN. All patients underwent surgery without known genomic information and were assigned a histopathology diagnosis by an expert panel blinded to all genomic information. The results showed a sensitivity (SN) of 91%, specificity (SP) of 68%, positive predictive value (PPV) of 47%, and negative predictive value (NPV) of 96% at a cancer prevalence of 24%. Since then, 13 independent GSC post-validation studies have been published. This study's objective is to compare the real world (RW) GSC performance to the validation study metrics. Rules and assumptions applying to this analysis include: 1. At least one patient with molecular benign results must have surgery for that study to be included in SN, SP and NPV analysis and in these studies, molecular benign results without surgical histology are considered true negatives (TN) (as are the molecular benign results with benign surgical histology) 2. Patients with suspicious results that do not have surgery are either excluded from the analysis (generating an observed PPV (oPPV) and observed SP (oSP)) or assumed as histology negatives (false positives - generating a conservative PPV (cPPV) and conservative SP (cSP)) 3. NIFTP is considered malignant. Rule #1 excluded two studies from SN, SP and NPV analysis. Data from all studies were pooled using a random-effects model. All analyses were done using R package meta (version 4.18-2). In the RW, the GSC demonstrates a SN, oSP, oPPV and NPV of 97%, 88%, 65%, 99% respectively, and conservative RW performance of cSP at 80% and cPPV at 49%. A statistically significant improvement is observed for oSP, cSP, oPPV, and NPV (p<0. 05) relative to the validation study with no statistical difference in SN. An overrepresentation of Hurthle subtypes in the validation study (20% Hürthle carcinoma (HCC) of malignant histology and 11% Hürthle adenoma of benign histology) relative to RW estimates of HCC prevalence (<5%) may partly explain improved RW GSC performance (as well as other histology subtype differences). Additionally, there may be an enrichment of malignancy in the operated cohort based upon the selection of patients with GSC-S nodules that have surgery (higher clinical risk or more worrisome ultrasound features compared to GSC-S nodules without surgical follow-up). The high benign call rate, predicted by the higher cSP and oSP, likely increases the overall rate of clinical observation in lieu of surgery. The high oPPV indicates an increased yield of cancers for resected GSC-S lesions relative to the validation study (65% vs 47%). In summary, RW GSC data indicates significantly better performance on several metrics as compared to the validation study, most notably on cSP, oSP and oPPV. Presentation: Saturday, June 11, 2022 1:00 p.m. - 3:00 p.m.