KEY POINTS
- This PROSPERO-registered systematic review searched literature from 2019 through November 2025 and included 127 studies of CBCT-guided online adaptive radiotherapy: 108 identified through the primary PubMed search and 19 additional HyperSight studies through hand searching. A quantitative meta-analysis was not performed because treatment sites, workflows and endpoints were highly heterogeneous.
- Only 12 studies reporting patient outcomes underwent formal risk-of-bias assessment: one randomized phase II trial and 11 non-randomized studies. Most clinical evidence therefore came from small, single-center cohorts with short follow-up, while much of the broader literature consisted of dosimetric, phantom, in-silico or workflow studies.
- In a typical ETHOS workflow, AI-assisted contouring, target review, plan generation, plan comparison and QA are performed while the patient remains on the couch, usually within 15–30 minutes. Adapted plans are selected in approximately 90–95% of fractions, primarily because of improved target coverage and/or OAR sparing.
- Thoracic studies showed consistent dosimetric benefit. In clinical esophageal oART, adaptation was selected in 99% of fractions, with median treatment-session time of 28 minutes; mean heart dose fell by 9.5% and lung V20 by 16.9%. A later clinical cohort reported 27% grade ≥3 acute toxicity and 69.2% major pathological response, but neither endpoint correlated significantly with the adaptive dose metrics.
- Upper-abdominal gains were often larger where targets directly approached gastrointestinal organs. In pancreatic cancer, one study improved PTV coverage from 89.7% to 96.3% while reducing duodenal D0.5cc from 33.6 to 28.8 Gy and stomach D0.5cc from 41.2 to 33.5 Gy. For abdominal oligometastases, another analysis increased simultaneous target/OAR constraint compliance from 34.4% to 97.1%.
- Pelvic data include an early clinical toxicity signal. In prostate cancer treated to 60 Gy in 20 fractions, daily oART was associated with lower acute gastrointestinal toxicity than non-adaptive RT (OR 0.45, 95% CI 0.27–0.73; P=.001), while the reduction in genitourinary toxicity was not significant (OR 0.65; P=.07). In cervical cancer, adaptation allowed cervix/uterus/GTV margins to fall from 1.5 cm to 0.5 cm in one comparative dosimetric study.
- HyperSight strengthens the technical foundation for CBCT-based adaptation with acquisition times of approximately 6 seconds, improved HU fidelity and reduced artifacts; in one prostate implementation, 99% of scans were considered suitable for online adaptation. However, across the review, convincing evidence for improved long-term local control, survival or patient-reported outcomes remains scarce.
CLINICAL TAKEAWAY
CBCT-guided oART has moved well beyond technical feasibility: across multiple disease sites, the scheduled plan frequently becomes dosimetrically inferior once anatomy changes, and adaptation can restore target coverage or avoid OAR violations. What remains unproven is the most important step — whether these repeated dosimetric gains consistently translate into less toxicity, better tumor control or better quality of life, and which patients justify the additional resources.