Normal tissue complication probability models for gastrointestinal toxicity after adjuvant (chemo)radiotherapy for cervical cancer in the PARCER trial

BACKGROUND AND PURPOSE: Gastrointestinal (GI) side effects are common after radiotherapy for gynecological cancers. Normal tissue complication probability (NTCP) models providing individual risk estimates for GI adverse events in gynecological cancers are not yet available. The aim of this study was to develop NTCP models for GI toxicity after postoperative (chemo)radiotherapy for cervical cancer. MATERIALS AND METHODS: Data from the phase-III PARCER trial (NCT01279135) was used. Tested variables included patient- and treatment-related characteristics and dosimetric parameters of small, large, and total individual bowel loops, and bowel bag as peritoneal space. NTCP endpoints included acute and late ≥G2 diarrhea and GI toxicity (CTCAEv3.0) and late persistent GI toxicity defined with the Cumulative-Month and Severity Score (C-MOSES). Logistic or Cox regression were used for multivariable models, selecting variables with Least Absolute Shrinkage and Selection Operator. Internal model performance was evaluated with area-under-the-curve (AUC) or C-index. RESULTS: The median follow-up of the 283 included patients was 43 months. Volumes of the bowel receiving 30 Gy (V30Gy) and 40 Gy (V40Gy) were associated with acute and late ≥G2 diarrhea and persistent GI toxicity. IMRT reduced risk of late ≥G2 and persistent GI toxicity, whereas radical hysterectomy with bilateral lymphadenectomy increased it. AUC/C-indices ranged from 0.588 (persistent GI toxicity) to 0.761 (late ≥ G2 diarrhea). CONCLUSION: NTCP models for acute and late diarrhea and persistent GI toxicity after postoperative radiotherapy for cervical cancer included bowel V30Gy and V40Gy. Furthermore, 3DCRT and more extensive surgery increased risks. Further studies should explore additional variables to improve model performance, reach consensus on bowel delineation methods, and validate models externally.