Pre-operative sera interleukin-6 in the diagnosis of high-grade serous ovarian cancer

Abstract

Pre-operative discrimination of malignant masses is crucial for accurate diagnosis and prompt referral to a gynae oncology centre for optimal surgical intervention. HGSOC progression is correlated with local and systemic inflammation. We hypothesised that inclusion of inflammatory biomarkers in sera may improve diagnostic tests. In the training cohort, we tested four existing clinical tests (RMI score and ROMA, CA125 and HE4) and a panel of 28 immune soluble biomarkers in sera from 66 patients undergoing surgery for suspected ovarian cancer. Six promising immune biomarkers alone, or in combination with conventional tests, were subsequently analysed in an independent validation cohort (n = 69). IL-6 was identified as the main driver of variability followed closely by conventional diagnostic tests. Median sera IL-6 was higher in HGSOC patients compared to those with a benign mass or controls with normal ovaries (28.3 vs 7.3 vs 1.2 pg/ml, p < 0.0001). The combination of IL-6 further improved the overall predictive probability of the conventional tests. Modelling a two-step triage of women with a suspicious ovarian mass, with IL-6 > 3.75 pg/ml as primary triage followed by conventional tests (CA125 or RMI score) identified ovarian cancer in patients with a misclassification rate of 4.54-3.03%, superior to the use of CA125 or RMI alone (9.09 to 10.60). The validation cohort demonstrated a similar improvement in the diagnostic sensitivity following addition of IL-6. IL-6 in combination with conventional tests may be a useful clinical biomarker for triage of patients with a suspected malignant ovarian mass.

Figure 1

Volcano plot of soluble factors in pre-operative sera. Pre-operative blood was withdrawn from 33 patients with ovarian cancer, 15 benign ovarian mass and 21 with normal ovaries. A total of 28 serum soluble factors were measured using multiplexed bead immunoassay. One-way analyses of variance were conducted on each of the 28 log₂-transformed immune factors. For each immune factor, the three pairwise contrasts of means, malignant tumour versus healthy control, benign tumour versus healthy control and malignant tumour versus benign tumour, and their standard errors of differences, were calculated and the results of the pairwise t-tests were summarized in a volcano plot of statistical significance, −log₁₀(P-value), versus fold change, namely the difference in the group means of the log₂-transformed soluble factors concentrations. Volcano plot of statistical significance (−log₁₀(P-value) of contrasts versus contrasts expressed as differences on the log2 scale. Soluble factors with contrasts greater than a 2-fold change or with a P-value < 0.05 are labelled.

Figure 2

RMI score, ROMA and serum level of pro-inflammatory cytokines in advanced serous EOC patients and those with benign ovarian masses and normal ovaries in training cohort. Box-and-whisker-plots extending down to the minimum value and up to the maximum level of RMI score, ROMA, serum levels of CA125, HE4, IL-6 and IL-8 in patients with high-grade serous EOC, with benign ovarian masses and normal ovaries. The p-value is indicated for each factor. Kruskal-Wallis followed by Dunn’s multiple comparison test: *p < 0.05, **p = 0.001–0.01, ***p = 0.0001–0.001, ****p < 0.0001.

Figure 3

ROC-AUC value using CA125, HE4, IL-6, IL-8, RMI score, ROMA alone of the training phase for patients with malignant and non-malignant ovarian masses. ROC analysis of all markers alone for the training phase and associated tables with predicted probability in combination. ROC-receiver operating characteristic, AUC- area under curve, 95% CI- 95% Confidence Interval. Reference line (grey continuous line) is at AUC 0.5).

Figure 4

ROC-AUC value using CA125, HE4, IL-6, IL-8, RMI score, ROMA alone of the independent validation phase for patients with malignant and non-malignant ovarian masses ROC analysis for all markers alone for the validation phase and associated tables with predicted probability in combination. ROC-receiver operating characteristic, AUC- area under curve, 95% CI- 95% Confidence Interval. Reference line (grey continuous line) is at AUC 0.5).

Figure 5

Classification and regression trees (CART) analysis using the recursive partitioning method to evaluate usefulness of IL-6 and components of the conventional risk biomarker CA125 and indices such as RMI score (menopause status, ultrasound score CA125), and ROMA (CA125 and HE4) in discriminating between patients with malignant (n = 33), benign (n = 12) and normal ovarian masses (n = 21) in the training phase. IL-6 levels were evaluated in trees in combination with either CA125 (A), RMI (B) or the ROMA index. (C) The values within each node or subnodes from left to right represent the proportion of each group (normal, benign and malignant) from the total number of patients (N = 66).