Imaging-Based AI for Predicting Pathological Complete Response in Esophageal Cancer After Neoadjuvant Therapy: Systematic Review and Meta-Analysis.
Zhiwei Li, Xin Gao, Jing Zhou et al.
✦ AI-curated · Sources linked
In 30 seconds
This systematic review and meta-analysis assessed imaging-based AI models for predicting pathological complete response after neoadjuvant therapy for esophageal cancer, separating internal from external validation. Across 33 studies, pooled external-validation sensitivity was 0.73 (95% CI 0.61-0.81) and specificity was 0.83 (95% CI 0.73-0.90); prediction intervals were wide. The review included 27 internal-validation datasets with 1891 patients and 16 external-validation datasets with 777 patients, although the unique patient count across datasets could not be verified. Evidence certainty was very low, and the authors considered the models investigational rather than ready for routine adoption.
Key findings
- External-validation sensitivity was 0.73 (95% CI 0.61-0.81; PI 0.34-0.93) and specificity was 0.83 (95% CI 0.73-0.90; PI 0.36-0.98).
- External-validation AUC was 0.84 (95% CI 0.79-0.89), while the DOR was 12.93 (95% CI 7.25-23.07).
- Thirty-two studies were conducted in China, 12 had high validation risk of bias, and all GRADE-assessed outcomes had very low certainty.
Why it matters
Imaging-based AI could contribute to response assessment after neoadjuvant therapy, but pooled averages do not ensure dependable performance in a new clinical setting. Wide prediction intervals and very low certainty highlight the risk of relying on these estimates for consequential decisions, particularly when considering organ preservation or omission of surgery.
What to keep in mind
The evidence had wide prediction intervals, high validation risk of bias in 12 studies, and very low certainty. Most studies were conducted in China, and the unique patient count across datasets and publications could not be verified.