A prognostic model of immunohistochemistry biomarkers for high-grade serous ovarian cancer

Background: Ovarian cancer is the most lethal gynecologic cancer in the United States. High-grade serous ovarian cancer (HGSOC) accounts for 70%-90% of all ovarian cancer death. It is crucial to identify efficient prognostic biomarkers to inform treatment decision making. Method: Tissue microarrays and clinical data were obtained from patients diagnosed with invasive HGSOC enrolled in studies participating in the Ovarian Tumor Tissue Analysis consortium. Cox proportional hazard regression analysis (CoxPHR) with lasso penalty was performed to select the most important variables related to overall survival (OS) from clinical prognostic data and 9 immunohistochemistry (IHC) biomarkers of interest, MyD88, TLR4, FOLT1, CD8+ tumor-infiltrating lymphocytes (CD8+ TILs), p16, PTEN, progesterone receptor (PR), estrogen receptor (ER) and androgen receptor (AR) using a training set of 254 patients with all 9 IHC data. The external validation was conducted using the test set of 1563 patients with data of the selected IHC biomarker by lasso. Hazard ratios (HRs) and 95% confidence intervals (CIs) of the selected variables were estimated from the CoxPHR. Kaplan-Meier curves were used to visually compare survival across the selected variables. A nomogram was generated to estimate the 2-year and 3-year survival. Results: The median OS time of the training set was 5.04 years (95% CI 4.36- 5.99 years). The selected variables from CoxPHR with lasso penalty include age at diagnosis, stage, debulking v status, AR, TLR4, CD8+ TILs, and p16. The median OS of the test set is 3.41 years (95% CI 3.21-3.63 years). The cases in the test set are at a more advanced stage. C-index from the prediction model fitting in the test set is 0.63. In the prediction model, CD8+ is inversely associated with the hazard of death (P for trend = 0.0011). Conclusion: The CoxPHR model with lasso penalty identifies four IHC biomarkers, AR, TLR4, CD8+, and p16, along with age at diagnosis, tumor stage, and debulking status, as prognostic factors for HGSOC survival. Further study containing more IHC candidates and clinical variables, such as chemotherapy response, and using continuous IHC scores should be performed to increase the accuracy of the prediction model for HGSOC survival.