Ovarian cancer: diagnostic accuracy and tumor types distribution in East Africa compared to North America

This retrospective study involved cases of ovarian cancer diagnosed in the year 2002 to 2017 from three centers in Tanzania: Bugando Medical Centre (BMC), Kilimanjaro Christian Medical Center (KCMC), and Mbeya Referral Hospital, and the Pathology Department at Makerere University, Kampala, Uganda. These are the only tertiary hospitals with diagnostic facility for cancer, all patients or samples of patients with cancer are brought to these centers for diagnosis. Bugando Medical Centre serves the North-Western area of Tanzania with a population of 13 million; KCMC serves a population of 11 million from the North-Eastern part of Tanzania; and the Mbeya Referral Hospital serves a population of 8 million from the Southern Highlands of Tanzania. In Uganda, the Pathology Department of Mulago Hospital serves a population of 3 million.

A search was undertaken for cases of ovarian cancer diagnosed from 2002 to 2017 from the files of the pathology department at each participating center. For each ovarian cancer diagnosis identified, the histological type was recorded as well as the demographic characteristics of the patient. This process was completed by searching the medical records from the patient files. By using the histology number, a corresponding formalin fixed paraffin embedded (FFPE) tissue block and haematoxylin and eosin (H&E) stained histological slides were retrieved from the archival pathology materials. In circumstances where the H&E stained slides were not available, new sections were made and stained by H&E. All H&E stained slides were examined under a light microscope, and the slides with a viable representative tumor were selected and matched with the corresponding FFPE tissue block. In each case, one to three H&E stained slides and FFPE tissue blocks were selected for the study. The selected FFPE tissue blocks and corresponding H&E stained slides were shipped to the Anatomic Pathology Research Laboratory (APRL) of the University of Calgary, Alberta, for further studies.

In each case, one good representative H&E stained slide was selected and marked for an area with high tumor cellularity. The corresponding area on the FFPE tissue block was selected and punched with a 0.6 mm needle core for TMA construction. Two tissue microarrays containing 0.6 mm duplicate cores from each case were constructed. This procedure was carried out at the Anatomical Pathology Research Laboratory (APRL) of the University of Calgary by a Semi-automated Tissue Arrayer (TMArrayer™, Pathology Devices, Inc. USA).

Antigen retrieval, staining platforms, antibody dilutions, detection methods, and interpretation of the staining are shown in Supplementary Table S-1. In brief, 0.4-μm sections were made from the two TMA’s, and staining was carried out by the Dako Omnis Platform of the Calgary Laboratory Services, and the APRL of the University of Calgary. The TMA’s were stained with the following antibodies: antibodies for general interest in ovarian cancer AE1/AE3 and PAX8, and antibodies specific for ovarian carcinoma typing as described previously [11]: WT-1, p53, Napsin A, PR, p16, TFF3, ARID1A, and Vimentin. More antibodies to complement typing for ovarian carcinoma were added: ER, Mismatch repair proteins (MSH6 and PMS2) and SATB2. For germ cell tumors and Sex-cord Stromal tumors, the following antibodies were included: Inhibin, FoxL2, Melan A, OCT-4, and GLYP3. CD45 and c-myc were used for lymphomas as well as BRG1 for small cell carcinoma hypercalcemic type.

Full section H&E stained slides from each ovarian cancer cases were reviewed under light microscopy by the investigators. The morphology of ovarian cancer was assessed based on the 2014 WHO classification of tumors of the female reproductive tract and complemented by immunohistochemical expression patterns as previously described [11]. In a situation where there was discordance between morphology and immunophenotype, a consensus was reached by the arbitration of a third reviewer.

We identified all patients diagnosed with ovarian cancer in Alberta between 2008 and 2016 from the Alberta Cancer Registry. Patient demographics and treatment data were extracted from the Discharge Abstract Database, the National Ambulatory Care Reporting System Database, or were obtained directly from electronic medical records (ARIA RO/MO).

A total number of 371 patients with a diagnosis of ovarian cancer made between 2002 and 2017 were identified from the study sites. Following exclusions, two TMA’s were constructed representing 227 ovarian cancer cases. Further exclusions resulted in 210 ovarian cancer cases suitable for analysis, as shown in Fig. 1. For all patients enrolled, the median age was 47.5 (Range 6 to 86) years. The patients presented with various symptoms and the most common symptoms were abdominal swelling, abdominal pain, abdominal distention, and abnormal uterine bleeding. The duration of the condition ranged from 2 weeks to 84 months, with a median duration of 6.5 months.

Menopausal status was available for 91 patients only, whereas 34 (37.4%) were post-menopausal. All patients underwent surgery, and the majority of the specimens received for investigation were ovarian masses 126 (60%), followed by hysterectomy and bilateral salpingo-oophorectomy. In a few cases, staging surgery was accompanied by omentectomy, bowel resection, and lymph node sampling. The stage of the disease was available in four patients only, and all were at an advanced stage. There was no data on disease recurrence or survival.

In the original diagnosis, the majority of ovarian cancers were epithelia (carcinomas), which constituted 147 (70%) of all the cases with a mean age of 50.5 (SE ±1.2) years. The second most common diagnosis was a sex cord stromal tumor in 24 (11.4%) cases with a mean age of 37 (SE ± 3.0) years, followed by germ cell tumors 21 (10%), with a mean age of 22.8 (SE ± 3.2) years. The diagnosis of ovarian lymphomas was rendered in 14 (6.7%) of the cases with a mean age of 15.9 (SE ±3.7) years. Other types included sarcoma in 2 (0.9%) cases and unspecified cancers in 2 (0.9%) cases.

Following morphological review and IHC integration, there was substantial agreement between original diagnosis and a revised diagnosis in the major categories of ovarian cancer (Kappa 0.702 with 95% CI of 0.619 to 0.791), as shown in Table 1. In a revised diagnosis, the largest category was still epithelia in 127 (60.5%) of all the cases with a mean age of 52 (SE ±1.2) years. The second most common category was sex cord stromal tumors with 30 (14.3%) cases with a mean age of 45 (SE ±2.9) years, followed by germ cell tumors 27 (12.6%), and lymphomas 13 (6.2%). There was no significant mean age difference from the original to the revised diagnosis between ovarian tumor types.

In a revised diagnosis, the specific histotypes were assigned based on 2014 WHO Classification of Tumors of Female Reproductive Organs. There was a fair agreement between original diagnoses and revised diagnoses (Kappa = 0.343, 95% CI: 0.277 to 0.409), as shown in Supplementary Table 2. In summary, a good concordance was seen in Lymphomas, followed by Germ Cell Tumors (92.9 and 81%, respectively). For epithelial tumors (carcinoma), a total number of 84 (57.9%) cases were not classified (carcinoma not otherwise specified (NOS)). Following review, a total number of 36 (42.9%) cases were reclassified to High Grade Serous Carcinoma (HGSC). Significant numbers of carcinoma NOS had poor H&E stains, in which the diagnosis was straightforward following recut and new H&E stain at Calgary laboratory. Two cases had a diagnosis of carcinoma NOS. Following H&E re-stain, one was adult granulosa cell tumor, and the other was HGSC with positive staining with WT-1 and p53 mutant type (see Fig. 2). There were 13 cases of HGSC in the original diagnosis, and the concordance with the revised diagnosis was 76.9%. Among 17 cases originally diagnosed as Endometrioid Carcinoma (EC), the concordance with a revised diagnosis was 58.8%, and 5 (29.4%) were reclassified to HGSC. The concordance of 21.7% was seen in Mucinous Carcinoma (MC), and 6 (26.1%) were reclassified to EC, whereas 4 (17.4%) were metastasis from GI. There were three cases of Low Grade Serous Carcinoma (LGSC), with a concordance of 33.3%, one case was EC, and the other was a metastasis from GI. There was one case of neuroblastoma which was reclassified as small cell carcinoma hypercalcemic type (SCCOHT) with loss of BRG1 staining in tumor cells and positive staining of stromal cells (See Fig. 3a & b), and one case of carcinoma NOS was reclassified as mixed carcinoma (EC/LGSC).

The patterns of IHC markers expression across different ovarian cancer histotypes are shown in Table 2. PAX8 was expressed in all LGSC, in 87% of HGSC and in none of the mucinous carcinomas. The Fallopian tube marker WT-1 was positive in 89% of HGSC, all LGSC, and in 27% of EC, with no expression in clear cell carcinoma (CCC) and MC. Hormonal receptor expression, estrogen receptor (ER) was higher (75%) in HGSC compared to other histotypes. For endometrioid carcinoma, ER and progesterone (PR) receptors were observed in 34 and 21% of cases, respectively, and one case of EC with mucinous features had a loss of MSH6 and ARID1A in tumor cells with a positive stain in stromal tissue and lymphocytes (See Fig. 3c). Abnormal staining of p53 was observed in 63% of HGSC, 18% of the EC, 20% of MC, and none of the LGSC. In sex cord stromal tumors, FOXL2 was expressed in 66 and 50% of adult granulosa cell tumor (AGST) and juvenile granulosa cell tumor (JGCT), respectively. SATB2 expression was predictive of ovarian metastatic mucinous tumor (p < 0.0001). Additionally, two cases of immature teratoma and one case of AGCT were positive for SATB2.

For analysis of antibody performance, the known positive controls from APRL containing tonsil, Fallopian tube, placenta, and endometrium were included in each TMA. The performance of tested samples was assessed based on presence of internal controls for the tests, which require interpretation with internal controls (p53, ARID1A, MSH6, PMS2, and BRG1). All markers were not assessed equally across all the samples, as some samples were uninterpretable, or there was a loss of cores in the TMA. However, the majority of the samples were assessed for these markers, as shown in Table 3.

The most affected marker was ARID1A, whereby 47% of the cases had a loss of staining in tumor cells as well as stromal tissue and lymphocytes, which are the internal controls (See Fig. 3d), followed by PMS2 (12%), and BRG1 (14%). When we analyzed for the primary site of sample fixation in relation to loss of internal control, there was no significant difference when the samples were fixed in tertiary or non-tertiary hospitals. However, for ARID1A, most of the samples with positive controls had primary fixation in a tertiary hospital. These results are summarized in Table 3.

The comparison of ovarian cancer tumor types and histotypes distribution was performed using the following two populations: i) accessible ovarian cancer cases from East Africa diagnosed over a period of 15 years, and ii) Alberta Cancer Registry for cases diagnosed over a period of 9 years (2007 to 2016). In Alberta, the women population at risk of ovarian cancer is estimated to be 1.4 million women compared to the catchment area of the selected centers in East Africa of 9 million women at risk. There were 8-times more ovarian cancer cases in Alberta, compared to the selected population in East Africa, which is 6-times larger than the population of Alberta. This rough estimate suggests that there are 48 times more ovarian cancer cases in the Alberta population compared to the selected East African population. The distribution of the major tumor types in both populations showed epithelial tumors being the commonest, 89.9% in Alberta, and 58% in East Africa.

For the carcinomas, when a comparison was made by specific histotypes, HGSC occurred with the same frequency in both populations, as shown in Table 4. Still, EC was more common in the East Africa population than the Alberta population (28.3% versus 11.3% respectively).

There was no striking difference for CCC, LGSC, and MC, and these cases occurred at a lower frequency in both populations. For Germ Cell Tumors, there was a predominance of Dysgerminoma and Immature Teratoma in the Alberta population (29.3 and 31%, respectively), compared to the East Africa population (19.2 and 15.4% respectively). However, the Yolk Sac Tumors were more common in East Africa compared to Alberta (53.8 and 17.2%, respectively). The Adult Granulosa Cell Tumor was the commonest sex cord stromal tumor in both East Africa and Alberta populations (80 and 70.6%, respectively), though in relatively low numbers in Alberta compared to the numbers of ovarian cancer cases. Juvenile Granulosa Cell tumors were exclusively seen in East Africa in 13.3% of the cases of SCST.