Contents
- The 30-second summary: the seven figures that matter
- The study that raised the alarm
- The trial that did not find the effect
- What has been measured outside endoscopy
- The uncomfortable detail: what is detected more and less
- Practical application of these steps
- Frequently asked questions
- What we do not know
- References
In August 2025, The Lancet Gastroenterology & Hepatology published a finding that travelled the world: at four Polish centres, endoscopists who had spent months working with AI-assisted detection found fewer adenomas when performing a colonoscopy without it. The rate fell from 28.4% to 22.4%. Four months later, a Norwegian prospective trial with 5,013 colonoscopies and three phases found exactly nothing: neither loss nor gain of skill when the tool was withdrawn. And a third study, over three years, found sustained improvement. This article sorts out a scientific controversy that is live right now, and explains what separates the three designs.
The 30-second summary: the seven figures that matter
Each row names the study and the year. Note that the first three contradict one another and that all three are well done.
| Finding | Result | Study and year |
|---|---|---|
| Adenoma detection without AI, before and after rollout | From 28.4% to 22.4%: −6.0 points (95% CI: −10.5 to −1.6; p = 0.0089) | Budzyń et al., Lancet Gastro Hep (2025) |
| Odds of detecting after AI exposure | Odds ratio of 0.69 (95% CI: 0.53-0.89) | Budzyń et al., Lancet Gastro Hep (2025) |
| Detection by inexperienced endoscopists while using AI | From 31.9% to 39.5% (OR 1.43; 95% CI: 1.11-1.84) | Pedersen et al., Endoscopy (2025) |
| Those same people, after the tool was withdrawn | No significant change (OR 1.03; 95% CI: 0.79-1.34) | Pedersen et al., Endoscopy (2025) |
| Breast radiologists facing erroneous AI prompts | Up to 12% more false-positive recalls | Heudel et al., ESMO RWD (2026) |
| Pathologists who reversed a correct diagnosis | More than 30% under time pressure | Heudel et al., ESMO RWD (2026) |
| Invasive cancer detection with AI against without it | More than double without AI in a three-year study (p = 0.001) | Henry and East, Endoscopy (2026) |
What you will find here
• The Polish study that raised the alarm and how it is built.
• The three-phase prospective trial that did not find the effect.
• What has been measured outside endoscopy: radiology and pathology.
• The uncomfortable detail about which lesions are detected more and less.
The study that raised the alarm
Krzysztof Budzyń, Marcin Romańczyk and a large team published in The Lancet Gastroenterology & Hepatology in 2025 a retrospective observational study at four Polish endoscopy centres taking part in the ACCEPT trial. Those centres introduced polyp detection tools at the end of 2021, and the authors compared the quality of colonoscopies performed without AI in the three months before and the three months after that rollout.
The sample comprised 1,443 patients, encompassing 795 prior to and 648 following the intervention, alongside a median age of 61 years. The main endpoint evaluated was the adenoma detection rate achieved during conventional, unassisted colonoscopy examinations.
The outcome: detection dropped from 28.4% (226 out of 795) down to 22.4% (145 out of 648), representing an absolute variance of −6.0 percentage points (95% CI: −10.5 to −1.6; p = 0.0089). According to multivariable assessment, past AI familiarity stayed an independent variable linked to this decline, yielding an odds ratio of 0.69.
What that design cannot separate
This retrospective observational study compares two distinct three-month intervals. Anything that shifted across these sites between late 2021 and early 2022 — staffing levels, task volume, or patient demographics — enters the evaluation alongside the AI tool. The researchers express this cautiously, noting that continuous exposure “might” diminish detection rates, thereby “suggesting” a detrimental impact.
| Study | Design | What it found | What it cannot support |
|---|---|---|---|
| Budzyń et al., Lancet Gastro Hep (2025) | Retrospective observational, two 3-month periods, 1,443 patients | A 6-point fall in unassisted detection | It cannot separate the AI effect from other changes in the period |
| Pedersen et al., Endoscopy (2025) | Prospective, three phases, 13 endoscopists, 5,013 colonoscopies | Improvement only during use, and only in novices; nothing after | Thirteen professionals and an indirect measure of detection |
| Heudel et al., ESMO RWD (2026) | Scoping review across several specialties | Scarce but consistent evidence of deterioration | It pools very different studies without pooling effects |
| Henry and East, Endoscopy (2026) | Letter analysing data from a three-year study | More invasive cancer detected without AI than with it | A reinterpretation of other people’s data, not an original study |
| Ahmad, Endoscopy (2026) | Commentary comparing the three designs | The results capture different stages of adaptation | An interpretive hypothesis, not a measurement |
The trial that did not find the effect
Tor Anders Pedersen, Yuichi Mori and colleagues published in Endoscopy in late 2025 a pragmatic, prospective, multicentre trial designed specifically to answer this question. Thirteen endoscopists — seven inexperienced and six experienced, defined as having performed more than a thousand colonoscopies — carried out 5,013 colonoscopies across three successive phases: before AI, with AI, and after it was withdrawn. Each professional therefore acted as their own control.
| Group and phase | Detection rate of polyps 5 mm or larger | Comparison with baseline phase |
|---|---|---|
| Novices, before AI | 31.9% (216 of 678) | Baseline |
| Novices, using AI | 39.5% (257 of 651) | OR 1.43 (95% CI: 1.11-1.84) |
| Novices, after withdrawal | 36.3% (173 of 476) | OR 1.03 (95% CI: 0.79-1.34): no change |
| Experienced, across all three phases | 28.3%, 26.7% and 27.2% | No significant effect in any |
| Individual results among novices | Two got worse, two unchanged, three improved | Among the experienced, three and three |
The authors’ conclusion is clear: detection by inexperienced endoscopists increased while they were using the tool, and in unassisted colonoscopy neither improvement nor loss of skill was observed after the exposure period.
It is a better design than the Polish one for this specific question, because it is prospective and each professional is their own control. It also has fewer professionals, thirteen against those at four centres, and uses an indirect measure — the proportion of colonoscopies with at least one polyp 5 mm or larger — which is not as established a quality indicator as adenoma detection.
What has been measured outside endoscopy
Pierre Heudel and his research team shared a scoping review addressing clinical deskilling in ESMO Real World Data and Digital Oncology throughout 2026. Their primary takeaway: data regarding professional deskilling remains limited yet uniform across medical fields. Three distinct metrics highlighted by the authors:
• In radiology, a controlled study with 27 breast imaging radiologists found that erroneous AI prompts increased false-positive recalls by up to 12%, even among experienced readers.
• In computational pathology, during time-sensitive experimental trials, upwards of 30% of participants overturned a right initial diagnosis upon receiving a flawed system recommendation.
- In cytology, the United Kingdom’s transition to primary human papillomavirus screening lowered case volumes by 80% to 85% and reduced the number of laboratories from 45 to 8 centres, directly impacting training capacity.
That final detail highlights a distinction worth remembering. When an expert misses their touch through overreliance on automation, it differs fundamentally from an ecosystem failing to produce the foundational scenarios trainees study. The latter issue runs deep structurally and defies resolution via personal coaching.
The uncomfortable detail: what is detected more and less
A three-year Japanese observational study, with more than 4,700 matched colonoscopies, found no deterioration at all after introducing assisted detection, and even found accelerated learning curves among endoscopists who were already good detectors. It seemed to settle the debate in favour.
Darrien Henry and James East published a letter in Endoscopy in 2026 reading those same data another way. In that study, detection rates for adenomas of 6 to 9 mm, 10 mm or larger, with high grade dysplasia, and for invasive cancer were significantly higher in the group without assistance. The invasive cancer rate was more than double without the tool than with it (p = 0.001).
Their alternative interpretation is harsh and well argued: assisted detection may distract the endoscopist towards trivial lesions — adenomas of 1 to 5 mm and hyperplastic polyps — at the expense of advanced lesions, which are the ones that matter. The consistency of the effect direction across three different categories of advanced lesion, with very low p values, makes chance an unlikely explanation.
Omer Ahmad proposed in 2026, also in Endoscopy, the reading that best reconciles the three studies: they may not contradict one another, but rather be capturing different stages of human-machine adaptation, and what the Polish study measured may be transient, reversible cognitive offloading rather than permanent erosion of skill.
Practical application of these steps
Ordered by the strength of the evidence behind it:
• Do not view the discussion as completely resolved on either side. Three rigorous analyses point to three distinct outcomes, and their divergent methodologies account for the variance.
- Keep periods of unassisted practice. It is the recommendation that recurs in the endoscopy literature, precisely because of what is not known.
• Measure what kind of lesion you detect, not just how many. The field’s sharpest analysis found more invasive cancer detected without the tool than with it.
• Watch for erroneous prompts. In breast radiology they increased false-positive recalls by up to 12%, including among experienced readers.
- Think about training, not only about the already-trained professional. The British cytology case shows the problem may be that the cases people learn on disappear.
Frequently asked questions
Do doctors who use artificial intelligence lose skill?
It is not settled. The Polish observational study published in The Lancet Gastroenterology & Hepatology in 2025 found unassisted adenoma detection falling from 28.4% to 22.4% after AI was introduced, a difference of −6.0 points (p = 0.0089), but a later prospective trial with 5,013 colonoscopies observed neither loss nor gain after the tool was withdrawn.
Why do two good studies reach opposite conclusions?
Because of the design. The Polish trial is retrospective and evaluates a pair of three-month intervals, meaning that any shift at the core impacts the analysis; conversely, the Norwegian investigation is prospective, structured in three phases, and positions each endoscopist as their own baseline. A theory featured in Endoscopy suggests that both approaches could be documenting distinct phases of adaptation to the device.
Who does it impact the most?
The evidence points to different effects by level. In the prospective trial, inexperienced endoscopists improved detection while using the AI, from 31.9% to 39.5% (OR 1.43), and returned to their baseline when it was withdrawn; among the experienced there was no significant effect in any of the three phases.
Does it occur outside the endoscopy setting as well?
There are signals in other specialties. According to the 2026 review, a controlled study with 27 breast imaging radiologists found erroneous AI prompts increased false-positive recalls by up to 12% even among experienced readers, and in computational pathology more than 30% of participants reversed a correct diagnosis when faced with an incorrect suggestion under time pressure.
Does detecting more polyps mean preventing more cancer?
Not necessarily, and that is the most uncomfortable point. Reanalysing the data from a three-year study, a letter published in Endoscopy in 2026 showed that detection of adenomas 10 mm or larger, of high grade dysplasia and of invasive cancer was significantly higher without the tool, with an invasive cancer rate more than double (p = 0.001).
What we do not know
• Whether the effect is reversible. The transient cognitive offloading hypothesis is plausible, but nobody has measured what happens after withdrawing the tool for a long period.
• What happens to people trained with the tool from the start. The available studies do not include absolute beginners, the group where skill retention matters most.
• Whether it affects mortality. None of this work measures post-colonoscopy colorectal cancer or survival.
- How long the exposure needs to be. One study, which observes an impact, spans three months; another, showing no effect, evaluates varying durations; finally, the research noting enhancement lasts three years.
References
1. Budzyń, K., Romańczyk, M. et al. Endoscopist deskilling risk after exposure to artificial intelligence in colonoscopy: a multicentre, observational study. The Lancet Gastroenterology & Hepatology, 2025. DOI: 10.1016/s2468-1253(25)00133-5
2. Pedersen, T. A., Mori, Y. et al. Learning and deskilling effects of artificial intelligence in colonoscopy among endoscopists with different levels of experience: a pragmatic, prospective trial. Endoscopy, 2025. DOI: 10.1055/a-2858-7084
3. Heudel, P., Crochet, H. et al. Artificial intelligence in medicine: a scoping review of the risk of deskilling and loss of expertise among physicians. ESMO Real World Data and Digital Oncology, 2026. DOI: 10.1016/j.esmorw.2026.100693
4. Henry, D. and East, J. Artificial intelligence-supported polyp detection (CADe) at colonoscopy reduces the detection of high grade dysplasia and invasive cancer. Endoscopy, 2026. DOI: 10.1055/a-2749-8345
5. Ahmad, O. Human-artificial intelligence adaptation in colonoscopy: beyond deskilling and upskilling. Endoscopy, 2026. DOI: 10.1055/a-2896-9791
This material is purely informational and should not be considered medical guidance. The statistics mentioned stem from research featuring methodologies, fields, and timeframes that cannot be directly contrasted, and they outline an ongoing scientific discussion at the moment of publication.
