Related Experiment Video
Updated: Oct 28, 2026
Ultrasonography-Guided Nasogastric Tube Placement and Verification: A Standardized Nursing Protocol
Published on: April 7, 2026
External Validation of a Commercially Available AI Tool for Nasogastric Tube Position Decision Support in the NHS: A
Anne-Marie Bartsch1, Jack Packer2, Mathew Storey1
1Department of Radiology, St George's University Hospital NHS Foundation Trust, London, UK.
Background:
Administration of food or medication through a misplaced nasogastric tube (NGT) constitutes a "never event" (defined by the National Health Service as a serious incident that is wholly preventable) in health care systems given the high likelihood of patient harm. Efforts to ensure adequate NGT positioning typically focus on increased education. Artificial intelligence (AI)-driven decision-support tools could address such challenges in a more streamlined manner, yet external validation studies in real-world settings remain scarce. Our objective was to evaluate the likely impact on patient safety from using an AI tool to detect NGT positioning on chest radiographs (CXRs).
Methods:
This was a 1-year, multicenter, prospective, silent validation study of stand-alone AI performance of a commercially available class 2b Medical Device Regulation-certified AI tool (Annalise Enterprise CXR) with the ability to detect NGT placement in adults.
Results:
During the study, 1605 consecutive CXRs were performed for NGT placement evaluation across one National Health Service trust in South West London. The AI tool correctly identified 109 out of 126 (86.5%) unsafely sited NGTs (true positives), and 938 out of 1479 (63%) safely sited NGTs (true negatives). Overall, 541 out of 1605 (34%) examinations were overcalled by the AI (false positives) and in 17 out of 1605 (1%) examinations, the AI failed to identify an unsafe NGT placement (false negatives). This yielded a sensitivity of 0.87 (95% confidence interval [CI], 0.79 to 0.91) for the AI with a specificity of 0.63 (95% CI, 0.61 to 0.66). The positive predictive value was 0.17 (95% CI, 0.14 to 0.20), while the negative predictive value was 0.98 (95% CI, 0.97 to 0.99).
Conclusions:
The high occurrence of false-negative cases, which could potentially lead to preventable never events, suggests that this AI tool may be unsuitable as a primary decision-support system for NGT placement verification. (Funded by a National Institute for Health Research Advanced Fellowship Award; grant number, NIHR-301322.).
