Methodological quality and performance of artificial intelligence and machine learning models for preoperative risk prediction in plastic surgery: A systematic review.
Anisha R Kumar, Suryakant Singh, Masaru Ishii
✦ AI-curated · Sources linked
In 30 seconds
This systematic review evaluated the methodological quality and predictive performance of artificial intelligence (AI) and machine learning (ML) models for preoperative risk prediction in plastic surgery. Ten studies were included, covering various surgical procedures. The area under the curve (AUC) for discrimination ranged from 0.66 to 0.82, indicating variable performance. Critical limitations included inadequate sample sizes and lack of external validation.
Key findings
- AUC for discrimination ranged from 0.66 to 0.82 among the studies reporting this metric.
- Only 2 out of 10 studies performed external validation, highlighting a significant methodological gap.
- 71.4% of development studies had events per variable <10, indicating inadequate sample sizes.
- 70% of studies did not report model calibration, raising concerns about their reliability.
Why it matters
Understanding the limitations and performance of AI/ML models in preoperative risk prediction is crucial for clinicians considering their use. The findings highlight the need for robust validation before these models can be reliably integrated into clinical practice.