Risk prediction models for lower gastrointestinal radiation toxicity in cervical cancer radiotherapy: a systematic review and methodological assessment

Lower gastrointestinal radiation toxicity is an important complication of cervical cancer radiotherapy. We reviewed prediction models, their validation, and their potential for clinical implementation. Eight Chinese and international databases were searched through June 29, 2026. Data extraction was informed by CHARMS, and methodological assessment used PROBAST + AI. Outcomes were classified as rectal, bowel or intestinal, or broader gastrointestinal toxicity. Acute and late endpoints and validation settings were considered separately. Heterogeneity precluded meta-analysis. The review was not prospectively registered. Nineteen reports were included, with a summed reported sample size of 3,416; this total is not a count of unique patients because participants from the same cohort were included in two reports. Reported toxicity incidence ranged from 19.8% to 61.35%. Models used regression, machine learning, or deep learning, with clinical, dosimetric, laboratory, imaging, and other omics predictors. Most performance estimates came from development cohorts or internal validation. Reported AUC or C-index values ranged from 0.588 to 0.975 across heterogeneous endpoints and evaluation settings and were not directly comparable. One multicentre study reported independent external validation: AUCs were 0.755 for a dual-radiomics model, 0.773 for a deep-learning model, and 0.803 for an integrated nomogram. Calibration curves and decision-curve analyses were available in several reports, but the extraction of calibration, implementation, and reproducibility information remained incomplete. The recorded PROBAST + AI assessments identified substantial analysis-domain concerns; the review included 29 distinct final prediction models. External validation remains limited, and the available evidence does not establish readiness for routine clinical implementation. Future studies should provide transparent model specifications, appropriate validation, calibration assessment, and clinically meaningful evaluation of model-guided care.