Gynecological oncology operations rely on intraoperative consultations to help determine the malignancy of suspicious tissue. Traditionally, these consultations are reliable, with a discrepancy from the final pathology report 3-8% of the time. Our study used statistical significance tests, such as t-tests and chi-square, to assess individual variables involved in the intra-operative consultation process. Further, machine learning predicative algorithms such as GLM, Naïve Bayes, and Random Forest were used to determine the most impactful factors. Combining all methods, our study found the most impactful factor in a discrepant diagnosis is the size of tumor tissue sent for testing, the surgeon, the type of tumor, and the frozen section pathologist’s specialty. These factors were the most significant within the study, however, they do not meet all thresholds for statistical significance. Due to a significant lack of discrepant samples, all machine learning algorithms could not be properly trained and therefore had accuracies near 50%. Further work using samples collected outside gynecological oncology or the NYU Langone facility are needed to improve the study’s outcome.

Machine learning code included

Final figures included