AI-Based Prediction of Pathological Response in Rectal Cancer Patients Receiving Total Neoadjuvant Therapy

Locally Advanced Rectal Cancer (LARC)

Rectal Cancer, Total Neoadjuvant Therapy, Sarcopenia, Artificial Intelligence, Machine Learning

This study aims to better understand how body composition, inflammation, and nutrition affect how rectal cancer responds to treatment. We reviewed data from ninety-three patients who were treated with total neoadjuvant therapy (TNT), which includes both chemotherapy and radiation before surgery. Using blood tests and CT scans, we measured muscle loss (sarcopenia), inflammation, and nutritional status before and after treatment. This study aims to better understand how body composition, inflammation, and nutrition affect rectal cancer response to treatment. We retrospectively analyzed data from ninety-three patients who received total neoadjuvant therapy (TNT), including both chemotherapy and radiation prior to surgery. Blood tests and CT scans were used to assess inflammation, nutrition, and muscle loss (sarcopenia) before and after treatment. The objective was to identify predictors of complete pathological response. Two novel composite scores were developed from routine lab parameters and tested for their predictive value. Artificial intelligence (AI) was also applied to enhance model accuracy. This study was conducted at Etlik City Hospital in Ankara, Turkey. No experimental interventions were performed. All data were obtained from routine care, and no additional procedures or patient compensation were involved. The findings may support personalized treatment decisions in rectal cancer.

This is a retrospective observational study that evaluated 93 patients with locally advanced rectal cancer who received total neoadjuvant therapy (TNT) at Etlik City Hospital, Ankara, Turkey, between November 2022 and December 2024. The aim of the study was to investigate whether laboratory markers related to inflammation and nutrition, as well as changes in muscle mass (sarcopenia), could predict treatment response. Patients who completed at least 12 weeks of chemotherapy and radiotherapy followed by curative surgery were included. Blood samples were used to calculate the C-reactive protein to albumin ratio (CAR) and the systemic immune-inflammation index (SII). Sarcopenia was assessed using contrast-enhanced CT images at the L3 vertebral level, obtained before and after treatment. Based on these parameters, two composite scores-CINR-pCR and CINR-Ryan-were developed using multivariable logistic regression. The primary outcomes were pathological complete response (pCR) and favorable tumor regression grade (TRG 0-1). Predictive models were constructed using both conventional statistical methods and artificial intelligence (AI)-based algorithms, including Random Forest. Internal validation was performed via cross-validation, and model performance was assessed by area under the curve (AUC), sensitivity, and specificity. All data used in this study were obtained from existing medical records as part of routine clinical care. No experimental treatments or additional interventions were administered. The results may contribute to optimizing personalized treatment strategies and clinical decision-making in rectal cancer.

Turkey (Türkiye)