Accuracy and Reproducibility of Frozen Section Diagnosis in Ovarian Tumors: A 10-Year Experience at a Tertiary Cancer Center

Institutional review board approval was obtained for review of medical record and pathology reports of patients with ovarian neoplasms seen at Detroit Medical Center/Wayne State University, Detroit, Michigan, between January 1, 2009, and July 31, 2018. We included patients with ovarian neoplasms who underwent surgery in which an intraoperative consultation (ie, FSD) was requested. The electronic medical records of the included patients were reviewed, including preoperative clinical impression, pathology reports, and postoperative clinical course. Ovarian tumors were classified according to World Health Organization criteria.³ 

Cases with an FSD of “defer to permanent” (n = 23) were excluded from our study. Cases with a discrepancy between FSD and final diagnosis were identified and classified into major or minor discrepancy. Major discrepancies were defined as those that needed reclassification to a different category of benign, borderline, or malignant tumors with either overdiagnosis or underdiagnosis and hence had a major impact on the surgical outcome. Minor discrepancies were defined as those that needed reclassification within the same category of benign, borderline, or malignant tumor without major impact on the clinical outcome or clinical decision-making.

Seven experienced pathologists participated in the study. They were categorized into 4 gynecologic pathologists (pathologists A, B, C, and D) and 3 general surgical pathologists (pathologists E, F, and G) according to their fellowship training and area of focused sign-out. The frozen section glass slides of the major discrepancy cases (range, 1–4 frozen section slides) were unlabeled and distributed among the 7 pathologists. They were blinded to the FSD and the final pathologic diagnosis and independently reviewed and recorded their interpretation using a standardized form. This form provided basic patient information, including age, tumor size, gross description, tumor rupture status, and a classification scheme following the World Health Organization classification. The diagnostic accuracy (sensitivity, specificity, positive predictive value, and negative predictive value) of the FSD is calculated using the final pathologic diagnosis as a gold standard test. The κ agreement was calculated first between each pathologist's FSD diagnosis and the final diagnosis (intrarater reliability), then between pathologists (interrater reliability).

A root cause analysis of the major discrepancy cases was performed. To accomplish that, we stratified those cases into interpretation error, sampling error, or other nonidentifiable error. Interpretation error is an error due to an interpretation mistake in the FSD. It was inferred when the majority of reviewing pathologists (≥6 of 7) agreed with the final diagnosis and disagreed with the erroneous FSD interpretation. Sampling error is an error due to undersampling of the lesion in the FSD setting. This was inferred when the majority of reviewing pathologists (≥6 of 7) agreed with the erroneous FSD interpretation and disagreed with the final diagnosis. The cases that did not fall into the previous 2 categories were assigned as nonidentifiable error.

Finally, we evaluated the clinical end points of those major discrepancy cases. The outcomes of interest were performance of unnecessary procedure, return for a second surgery, and 30-day postoperative mortality.

Association between two categorical variables was assessed using a χ² test. Exact agreement percentages and Cohen κ were calculated to measure the degree of agreement (interrater agreement) for each pair (between 2 raters or between a rater and the gold standard).⁵  Because the Cohen κ is an agreement measure only for a pair, an overall κ for multiple raters (ie, among 3 or more raters) was calculated using the Fleiss κ,⁶  and overall agreement with the gold standard and between gynecologic pathologists and general surgical pathologists was calculated using the Cohen κ with combined occasions. The κ coefficients can be interpreted using the guidelines outlined by Landis and Koch⁷  as follows: less than 0.00, poor; 0.00 to 0.20, slight; 0.21 to 0.40, fair; 0.41 to 0.60, moderate; 0.61 to 0.80, substantial; and 0.81 to 1.00, almost perfect agreement.

A total of 618 patients with ovarian neoplasms were included in the study. Epithelial tumors accounted for most of the ovarian neoplasms, 59.2% (366 of 618), followed by sex cord tumors, 9.8% (61 of 618); germ cell tumors, 9.4% (58 of 618); metastasis, 5.8% (36 of 618); and other benign nonneoplastic entities, 15.7% (97 of 618). Of the epithelial tumors, 74.5% (273 of 366) were benign, 8.4% (31 of 366) borderline, and 17.1% (62 of 366) malignant.

The diagnostic accuracy of the FSD in the total cohort is detailed in Table 1. The sensitivity of FSD in epithelial tumors was lower in borderline cases at 79.2% compared with benign (98.7%) and malignant (87%) cases. Similarly, the positive predictive value of FSD in epithelial tumors was lower in borderline cases (80.8%) than in benign (96.5%) and malignant (95.3%) cases. The sensitivity of FSD in germ cell tumors was the highest (98.3%), followed by sex cord tumors (96.7%). The specificity and negative predictive value of FSD were high in all nonepithelial tumors. The positive predictive value of FSD was the highest in germ cell tumors (99.3%), followed by nonneoplastic entities.

Of the overall cohort cases, discrepancy between FSD and final diagnosis was seen in 17.6% (109 of 618). Those cases were categorized as major in 5.3% (33 of 618) and minor in 12.3% (76 of 618). Mucinous epithelial tumors accounted for the most major discrepancies, 27.3% (9 of 33). Among the major discrepancies, 7 were overdiagnosed and 26 were underdiagnosed cases. Detailed characteristics of the discrepant cases are seen in Table 2. The discrepancy distributions of the overall cohort between gynecologic and general surgical pathologists were statistically similar to each other (P = .65) (Table 3).

Seven experienced pathologists were blinded to FSD and final diagnosis and independently reviewed the 33 major discrepant cases. The overall κ for diagnostic agreement between all 7 gynecologic and general surgical pathologists and final diagnosis was 0.18 (0.10–0.26, P < .001), implying only slight overall agreement (Figure 1). The gynecologic pathologists had a higher overall κ of 0.24 (0.13–0.35) than the general surgical pathologists' of 0.10 (−0.01–0.21), but their 95% CIs overlapped (Figure 1).

Regarding agreements between pathologists, the overall κ was 0.46 (0.40–0.52), implying a moderate overall agreement among raters. The highest κ was found between pathologists A (gynecologic) and D (gynecologic) of 0.80 (0.60–1.00). The gynecologic pathologists had a slightly higher overall κ of 0.47 (0.35–0.58) than the general surgical pathologists' of 0.45 (0.29–0.62), but their 95% CIs overlapped (Figure 2).

A root cause analysis of the major discrepancy cases was performed (Table 4). Sampling error accounted for roughly half of the cases, 43% (14 of 33). The erroneous FSD was attributed to interpretation error in 21% of the cases (7 of 33), where 36% of the cases (12 of 33) were discrepant for other nonidentifiable causes.

Finally, the clinical outcomes of interest were evaluated in the 33 major discrepant cases. Direct impact on surgical management was seen only in 3 cases, 0.5% (3 of 618) of the overall cohort. Performance of an unnecessary procedure was identified in one overdiagnosed patient (Figure 3, A and B). Two patients were underdiagnosed on the FSD, and return for a second surgery was indicated. The first case was a patient who declined the second surgery and received chemotherapy instead (Figure 3, C and D). In the second case, the cyst ruptured intraoperatively; hence, the patient was upstaged to pT1c and no second surgery was performed (Figure 3, E and F). However, no patient had 30-day postoperative mortality.

Surgical management of ovarian tumors depends on the nature of the neoplasm. Benign tumors are treated conservatively with cystectomy or oophorectomy. Borderline tumors require pelvic lymph node dissection with limited omental sampling. However, fertility and ovarian function–sparing surgery is still preferred in young women. In malignant tumors, staging and surgical reduction, including total abdominal hysterectomy with bilateral salpingo-oophorectomy, lymph node sampling, peritoneal sampling, and omentectomy, are recommended.⁸ 

Preoperative clinical impression, radiology, and CA125 serum levels are not a definitive diagnostic method in ovarian tumors. Intraoperative consultation (ie, FSD) plays a crucial role in determining the appropriate surgical management of ovarian neoplasms. However, the surgeon and the pathologist should be aware of the limitations of FSD. One of the main limitations is sampling error, especially in tumors larger than 10 cm. Other limitations include freezing artifacts, time limitation, technical problems, lack of ancillary studies, and the pathologist's subspecialty and experience.⁹ 

Frozen section diagnosis of ovarian tumors continues to pose a significant diagnostic challenge for practicing pathologists. A recent study of 4785 frozen section diagnoses ranked ovarian tumors as the third most frequently discrepant.¹⁰  A large body of literature has addressed the reliability of intraoperative FSD in ovarian tumors, with variable conclusions. In a meta-analysis¹¹  of 18 studies, the accuracy of FSD in ovarian tumors diagnosis was good, with sensitivity ranging from 65% to 97% and specificity from 97% to 100%. Borderline tumors have been notoriously problematic on FSD.^12–14  Our results also showed that epithelial borderline tumors have lower sensitivity and positive predictive value than both benign and malignant tumors. The diagnostic criteria of ovarian borderline and malignant tumors require adequate extensive sampling to establish 10% atypical proliferation features with absence of invasion for borderline tumors and at least a single focus of frank invasion for malignant tumors; this might explain their FSD difficulty and discrepancy.^15,16  Our study showed only a slightly higher discrepancy rate between FSD and final diagnosis in ovarian tumors (17.6%) than reported in the literature⁹  (8%). We propose that our higher rates reflect a more stringent application of criteria to define discrepancies and the inclusion of all major- and minor-discrepancy cases (the discrepancy rate was 5.3% for major and 12.3% for minor). We believe that a lower threshold allows pathologists to identify the factors that lead to major and minor discrepancies that may ultimately contribute to more significant events. This will also shed light on potential areas for improvement that will ultimately reduce the potential for discrepancy overall. Indeed, supervising training pathology residents and pathology assistants for adequate examining and sampling of the ovarian neoplasms in the FSD setting is very crucial to avoid mistakes and to spare the patient the clinical implication and consequences.

The pathologist's experience is indeed controversial in FSD. Literature has demonstrated controversial results on how the pathologist's experience affects the accuracy of FSD. A few studies^17,18  have indicated that the gynecologic specialty of the pathologist increases FSD accuracy. Experience of the pathologists (junior versus senior) has not been reported to influence the accuracy of the frozen section.^19,20  Our study was consistent with the latter, showing an overall agreement between gynecologic pathologists and general surgical pathologists in the overall cohort and roughly a similar discrepancy rate between them. However, the fact that sampling error accounts roughly for half of the major discrepant cases (43%) might explain the overall low κ of 0.18 (0.10–0.26, P < .001).

The immediate clinical ramifications of FSD error are critical and may result in harm to the patient. Overdiagnosis during FSD can result in unnecessary widening of the surgical field, with increased morbidity and mortality. Another concern with overdiagnosis is the loss of fertility and ovarian function in younger patients from radical resections.^21,22  Our cohort showed only one overdiagnosed patient on FSD who had unnecessary surgery. However, she was 75 years old, and high morbidity and mortality were the main concerns rather than loss of fertility. Underdiagnosis can result in suboptimal surgical treatment and lead to a second surgery and possible tumor spread. Few studies²²  have shown that the restaging procedure has no impact on survival in borderline ovarian tumors. Our study identified 2 underdiagnosed patients; however, one patient already had the cyst ruptured intraoperatively and was upstaged to pT1c and the other patient declined the option for a secondary surgery and instead received chemotherapy.

This study is limited by its retrospective nature and by the fact that it does not account for advances in imaging modalities to guide surgical management. Another limitation is that although this was a large cohort study (N = 618), the actual number of problematic cases on frozen sections was limited, which may have affected the relevant outcomes. Nevertheless, the strengths of this article are derived from the evaluation of the gynecologic training of pathologists on the intraoperative diagnosis, the study of the root cause analysis of the major discrepant cases, and the assessment of the clinical implication of the discrepant cases on patient outcome.

In conclusion, the diagnostic accuracy rate for FSD remains high for benign and malignant ovarian tumors but is relatively low for borderline ovarian tumors. Pathologists should take into consideration a few factors that might help to mitigate FSD discrepancies, such as learning and appreciating the histologic limitations of FSD and appropriate deferral to permanent section diagnosis when needed. We suggest more thoughtful examination and sampling of ovarian lesions; however, some invasion features of the tumor might be appreciated only on permanent sections, so changing the diagnosis line to “at least borderline tumor” and discussing the case with surgeons intraoperatively may give a better insight about clinical implication. Continuous quality assurance should be conducted in regard to discrepancies between FSD and final paraffin section diagnosis.

We thank Suzanne Jacques, MD; Faisal Qureshi, MD; Rafic Beydoun, MD; Fulvio Lonardo, MD; Dongping Shi, MD; and Amy Harper, MD, for their contributions to this paper.