Predicting Pathologic Response in Locally Advanced Rectal Cancer Using Inflammatory, Nutritional, and Sarcopenia-Based Markers: A Regression and AI-Based Analysis (CINR-AI Study)
CLINICAL COLORECTAL CANCER, cilt.25, sa.1, ss.51-67, 2026 (SCI-Expanded, Scopus)
- Yayın Türü: Makale / Tam Makale
- Cilt numarası: 25 Sayı: 1
- Basım Tarihi: 2026
- Doi Numarası: 10.1016/j.clcc.2025.10.002
- Dergi Adı: CLINICAL COLORECTAL CANCER
- Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, CINAHL, EMBASE, MEDLINE, Biomedical Reference Collection: Corporate Edition (EBSCO)
- Sayfa Sayıları: ss.51-67
- Gazi Üniversitesi Adresli: Evet
Özet
We developed regression-and AI-based models to predict pathological response in locally advanced rectal cancer treated with total neoadjuvant therapy (TNT). Combining inflammatory (CAR, SII) and nutritional markers with post-treatment sarcopenia improved prediction accuracy, supporting individualized decision-making for TNT-treated patients. Background: Total neoadjuvant therapy (TNT) is the standard approach for locally advanced rectal cancer (LARC), yet pathological complete response (pCR) is achieved in only a subset. Systemic inflammation, nutritional status, and sarcopenia influence outcomes, yet integrated predictive models are lacking. We aimed to develop clinical, laboratory, and AI-based models to predict pathological response. Methods: This retrospective study included stage II to III LARC patients treated at Ankara Etlik City Hospital (Nov 2022-Dec 2024). Eligible patients received >= 12 weeks of TNT followed by curative surgery. Sarcopenia was assessed using CT-based skeletal muscle area at the third lumbar vertebra (L3). C-reactive protein/albumin ratio (CAR) and systemic immune-inflammation index (SII) were used to assess inflammatory and nutritional status. Composite scores (CINR-pCR, CINR-Ryan) were calculated using z-transformed CAR and SII weighted by regression coefficients. Outcomes included pCR and good pathological response, defined as tumor regression grade (TRG) 0 to 1 per the modified Ryan grading system. Logistic regression and Random Forest (RF) models were used. ClinicalTrials.gov: NCT07049627. Results: Among 136 patients, 93 met the inclusion criteria. pCR and TRG 0 to 1 was achieved in 20 (21.5%) and 43 (46.2%) patients, respectively. Independent predictors of pCR included absence of post-TNT sarcopenia (OR 0.30, 95% CI, 0.09-0.95, P = .007), low CAR (OR 0.14, 95% CI, 0.03-0.70, P = .008), low SII (OR 0.28, 95% CI, 0.08-0.96, P = .042), low LDH (OR 0.10, 95% CI, 0.02-0.70, P = .020), and metformin use (OR 2.52, 95% CI, 1.40-3.78, P = .031). For TRG 0 to 1, significant predictors included low CAR (OR 0.42, 95% CI, 0.23-0.76, P = .005), low SII (OR 0.13, 95% CI, 0.03-0.56, P = .006), absence of >= 10% weight loss (OR 0.12, 95% CI, 0.02-0.66, P = .016), absence of post-TNT sarcopenia (OR 0.18, 95% CI, 0.05-0.70, P = .014), and shorter RT-to-surgery interval (OR 3.14, 95% CI, 1.17-6.43, P = .004). CINR scores showed strong predictive value (AUCs: 0.868 and 0.846), and RF models showed excellent performance (AUCs: 0.933 and 0.910, respectively). Conclusions: Inflammatory, nutritional, and sarcopenia-based markers, including CINR scores and AI models, accurately predict pathological response in LARC. Importantly, the ROC-derived cut-off values (CINR-pCR: 1.58; CINR-Ryan: 0.45) stratified patients into low-and high-risk groups, supporting clinical decision-making in organ-preservation strategies and surgical timing. Prospective multicenter validation is warranted.