Birsin Z., Cebeci S., Jeral S., Çerme E., Aliyev V., Demirci S. N., ...Daha Fazla
JOURNAL OF CLINICAL ONCOLOGY, cilt.44, sa.16_suppl, ss.1, 2026 (SCI-Expanded, Scopus)
Özet
e17582
Background:
The chemotherapy response score is a well-established prognostic predictor in high-grade serous ovarian carcinoma, with CRS3 (complete or near-complete response) consistently associated with improved progression-free survival and overall survival. We evaluated nutritional and inflammatory factors associated with achieving CRS3 and examined the impact of chemotherapy response score on survival outcomes using conventional survival analyses and machine learning–based modeling.
Methods:
This retrospective study included 75 patients with stage III–IV ovarian cancer treated with neoadjuvant chemotherapy (NACT) followed by interval debulking surgery. Patients were classified as CRS1–2 or CRS3. Survival outcomes were analyzed using Kaplan–Meier estimates and log-rank tests. Multivariable analysis was performed to identify factors associated with achieving CRS3. A random forest model incorporating nutritional, inflammatory, and clinical variables was developed to predict CRS3. Model performance was evaluated using 5-fold cross-validated area under the curve, and model interpretability was assessed using SHAP-based analyses.
Results:
Nineteen patients achieved CRS3, and no significant differences were observed between the CRS3 and CRS1–2 groups with respect to age, ECOG performance status, disease stage, or number of NACT cycles. Patients achieving CRS3 demonstrated significantly improved survival outcomes. Median overall survival was 30.7 months in the CRS3 group compared with 8.5 months in the CRS1–2 group (log-rank p = 0.002), and median progression-free survival was 30.7 versus 8.5 months, respectively (log-rank p < 0.001). In multivariable analysis, a higher prognostic nutritional index (PNI) and a lower C-reactive protein–albumin ratio (CAR) were independently associated with achieving CRS3. The random forest model incorporating all variables achieved a mean area under the curve of 0.69. A nutritional biomarker–based model (PNI, albumin, HALP) demonstrated superior performance (mean AUC = 0.72) compared with an inflammatory biomarker–based model (CAR, NLR, PLR; mean AUC = 0.54). SHAP-based model interpretation identified CAR and PNI as the most influential predictors of CRS3. Interaction analysis suggested that the negative impact of elevated systemic inflammation on the probability of achieving CRS3 was more pronounced in patients with poor nutritional status.
Conclusions:
Both nutritional and inflammatory parameters were associated with CRS3. Machine learning–based analyses suggested a slightly greater discriminatory contribution of nutritional parameters. Further prospective studies are warranted.