Diagnostic performance of endoscopy and MRI in identifying true response among patients with rectal cancer treated with total neoadjuvant therapy and selective watch and wait or TME.

Abstract

28 Background: The success of a watch-and-wait (WW) strategy for rectal cancer relies on restaging endoscopy and MRI to correctly identify a true response (TR) to neoadjuvant treatment. This secondary analysis of the OPRA trial is the first study to evaluate the restaging assessment’s diagnostic performance in a large cohort of WW patients. Methods: In the OPRA trial, patients with stage II/III rectal cancer were treated with total neoadjuvant therapy (TNT) and restaged with endoscopy and MRI 8±4 weeks post-treatment. Patients with a clinical complete (cCR) or near complete (nCR) response were offered WW; patients with an incomplete clinical response were recommended for total mesorectal excision. Diagnostic parameters, including accuracy, sensitivity, specificity, positive and negative predictive values, were calculated using TR as the reference standard. TR was defined as pathologic complete response for patients treated with TME or sustained cCR for ≥2 years for patients on WW. Contingency tables were prepared for two approaches: one considering either cCR or nCR positive for a TR, and one considering cCR as the only response grade positive for a TR. Results: Of the 304 patients restaged after completing TNT, 265 (87.2%) with ≥2 years of follow-up data available were included. Restaging endoscopy outperformed MRI across all parameters. Accuracy improved minimally to 66% when both modalities agreed on a cCR or nCR (Table). Restricting a positive result to cCR resulted in higher specificity, lower sensitivity, and higher accuracy (75%) (Table 1B). The positive posttest probability that a patient with a cCR had a TR was 81%, while the negative posttest probability that a patient with a non-cCR had a TR was 30%. Conclusions: Restaging endoscopy outperforms MRI in predicting TR, but the assessment’s overall accuracy even when both tests are combined remains suboptimal. Our results highlight the challenge of recognizing TR using subjective restaging exams. Opportunities to improve diagnostic accuracy include machine learning and radiomics, which have the potential to aid clinicians in identifying true responders. [Table: see text]