Demographics
The median age of diagnosis for the discovery and validation cohort were 61.9 and 63.4 (p = 0.54). In both cohorts, greater than 75% of patients had stage 3 or 4 disease. All patients had high or moderate grade disease. All patients were of serous, mixed serous or undifferentiated histology. The majority (93%) of patients underwent an optimal cytoreduction. For one patient in the validation cohort this information was not available. BRCA mutation or Homologous Recombination Deficiency (HRD) status was unknown for 66% of the discovery cohort compared to 18% of the validation cohort (p < 0.001). The mean CA125 was similar between groups at 14.3 and 17.6, respectively (p = 0.498). Receipt of maintenance therapy was more common in those within the validation cohort (61%) compared to 39% in the discovery cohort (p = 0.001). Demographic data is further summarized in Table 1.
ORACLE Risk Score Creation
The Ovarian Recurrence risk Assessment using Clinical and serum protein LEvels (ORACLE) score was constructed in order to incorporate clinical factors and the serum protein concentration data to predict time to first recurrence and death. The ORACLE score utilized the following clinical factors: patient age, optimal cytoreduction, and receipt of neoadjuvant chemotherapy, and the following serum proteins: CA125, BDNF, PDGFAA, PDGFABBB, and IFNγ.
The ORACLE was predictive of TTR in the discovery data set with a concordance index of 0.72. When patients were divided into risk groups, high risk patients (n = 31) had a median TTR of 10.5 months compared to not reached (HR 4.66, 95%CI 2.45 – 8.86, p < 0.0001) after 10 years of follow up in the low risk group (n = 40) Figure 2A. The most to least important parameters by coefficient value for predicting patient outcome were advanced stage (0.58), CA125 value (0.56), interferon gamma value (0.44), BDNF value (-0.32), age (-0.24), PDGFAA value (0.14), receipt of neoadjuvant chemotherapy (0.10), an optimal cytoreduction (-0.05), and PDGFABBB value (-0.03) Figure 2B. The model was also predictive of TTD with patients in the high-risk group having a median TTD of 47 months compared to not reached after 10 years of follow up in the low-risk group (HR 3.16, 95%CI 1.46 – 6.84, p = 0.003).
The ORACLE had similar performance when predicting PFS and OS. The median PFS for low and high-risk groups were not reached and 23 months (HR 4.98, 95%CI 2.60 – 9.53, p < 0.001), respectively. The median OS for the low and high risk groups were not reached and 59 months (HR 3.87, 95%CI 1.77 – 8.46, p < 0.001), respectively. Last, stage I and II patients (n = 14) were excluded from the analysis, which left 57 patients with stage 3 or 4 disease. Again, advanced stage, low risk patients (n = 26) had improved median PFS (85 months), TTR (34 months), OS (not reached), and TTD (72 months) compared to advanced stage, high risk patients (n = 31) [median PFS: 23 months, HR: 1.81 95%CI 1.10 – 2.99, p = 0.02; median TTR: 11 months, HR 3.27, 95%CI 1.67 – 6.41, p < 0.001; median OS: 59 months, HR 2.50, 95%CI 1.15 – 5.45, p = 0.02; median TTD: 47 months HR 2.09, 95%CI 0.97 – 4.51, p = 0.06] Table 2.
ORACLE Validation
The ORACLE was subsequently validated in an independent prospectively collected cohort of ovarian cancer patients at our institution (n = 33). The demographics of this cohort is described in Table 1. In this new cohort, the ORACLE was again predictive of TTR, PFS, TTD, and OS. In the validation cohort, the median TTR for high risk patients was 4.7 months compared to not reached in low risk patients (HR 4.71, 95%CI 1.75 – 12.7, p = 0.002) Figure 3A. The median TTD for high risk patients was 29 months compared to 75 months in low risk patients (HR 4.05, 95%CI 1.44 – 11.4, p = 0.008) Figure 3B. In the validation cohort, the ORACLE was also predictive of PFS (median low: not reached, median high: 48 months, HR 3.71, 95%CI 1.40 – 9.88, p = 0.009) and OS (median low: 202 months, median high: 98 months, HR 4.58, 95%CI 1.58 – 13.3, p = 0.005).
When examining advanced stage patients alone (n = 26) in the validation cohort, low risk patients had improved median PFS (67 months), TTR (22 months), OS (200 months), and TTD (66 months) compared to high risk patients (median PFS: 39 months, HR 3.39, 95%CI 1.23 – 9.36, p = 0.02; median TTR: 4.5 months, HR 6.29, 95%CI 2.12 – 18.7, p < 0.001; median OS: 98 months, HR 5.51, 95%CI 1.64 – 18.6, p = 0.006; median TTD: 29 months, HR 4.71, 95%CI 1.51 – 14.7, p = 0.008) Table 2. These findings validate that the ORACLE can accurately predict which patients with no evidence of disease on imaging will go on to rapidly recur and die of their disease and which will be long term survivors.
Prognostic Value of the ORACLE Compared to CA125 Alone
When examining all patients with stage 3 or 4 disease who had a complete response (CR) to primary therapy, 41 patients had a CA125 > 10 and 42 had a CA125 < 10. Those with a CA125 < 10 had a median TTR of 27 months compared to 11 months (HR 2.21, 95%CI 1.30 – 3.75, p = 0.003) in those with a CA125 > 10 Figure 4A. To understand how the ORACLE improves prognostic prediction compared to CA125 alone, the TTR of ORACLE high and low patients was examined in patients who had a CA125 < 10 and those with a CA125 > 10. Of those with a CA125 < 10 (n = 42), 33 (79%) were classified as low risk and 9 (21%) were classified as high risk by their ORACLE score. In this population of patients, median TTR for ORACLE low patients was 68 months compared to 6 months in ORACLE high patients (HR 2.91, p = 0.02) Figure 4B. In those with a CA125 > 10 (n=41), 19 patients (46%) were ORACLE low and 22 (54%) were ORACLE high. The median TTR was 16 months and 5 months (HR 2.41, p = 0.01) in ORACLE low and high groups, respectively Figure 4C. This data indicates that the ORACLE score provides improved prognostic prediction compared to CA125 alone in advanced staged patients.
ORACLE Score Association with Prognosis in BRCA mutated and BRCAwt Patients
BRCA mutation status and homologous recombination deficiency (HRD) testing was available for 51 patients. Of those who underwent testing, 38 were HRD or BRCA1/2 negative (75%), 12 (24%) harbored a BRCA1/2 mutation or were HRD positive, and 1 patient (1%) had BRCA1/2 mutation which was a variant of unknown significance. Patients noted to have HRD or a BRCA1/2 mutation had a non-significant improvement in PFS (median not reached) and OS (median 202 months) compared to those who tested negative (median PFS 44 months HR 1.93, p = 0.17 and median OS 107 months; HR 1.51, p = 0.45).
When examining stratification of BRCA1/2 and HRD status by ORACLE risk group, 9 of the 12 patients (75%) with a BRCA1/2 mutation or HRD positivity were categorized as being in the low risk group. The three patients who were categorized as high risk and harbored a BRCA1/2 mutation were subsequently analyzed for outcomes. The first patient harbored a BRCA1 mutation, had a TTR of 2.8 months, and a CA125 of 9.1 at the time of her ORACLE score. The second had a BRCA1 mutation and was lost to follow up precluding analysis of TTR. However, she had a TTD of 15 months and her CA125 was 2.8 at the time of her ORACLE score. The last patient had a BRCA1 mutation, a CA125 of 8.5 at the time of her ORACLE score, and has not yet recurred after 72 months. More patients are needed to decipher if the ORACLE score can definitively predict which BRCA mutated patients are at highest risk of recurrence.
The ORACLE score was assessed for prognostic performance in BRCA and HRD negative, patients (n=38). Those who had low ORACLE scores had a median TTR of 15 months and median TTD of 66 months, compared to a median TTR 5.0 months (HR 2.13, p = 0.054) and median TTD of 33 months (HR 2.22, p = 0.09) in high ORACLE score patients, indicating that the ORACLE score has predictive capabilities in this population of patients as well.
ORACLE Score Overtime
Of the 104 patients, there were 19 patients with a sample at the time of complete response and at the time of their first recurrence. In this scenario, the ORACLE score increased when patients were in the recurrent setting compared to when they were at the time of a complete response (p = 0.02) Figure 5A. The cohort contained 9 patients who had a sample at the time of a partial response to their initial treatment and at the time they subsequently transitioned to a complete response. ORACLE scores decreased when patients reached a complete response compared to their score at the time of their partial response (p = 0.07) Figure 5B. Last ORACLE score values were compared between the time of initial complete response and if the patient had a second complete response after treatment for recurrence (n = 7). In this scenario, the ORACLE score was similar at the time of a second complete response (p = 0.69) Figure 5C. The changes of the ORACLE score overtime indicates that it is a surrogate measure of ongoing activity of a patient’s cancer and could potentially be used for disease monitoring.