Would AI-assisted colonoscopy save the NHS money? A model says £69 per patient—but only 53% of simulations agreed
A peer-reviewed model of 10,000 hypothetical symptomatic patients estimated that GI Genius would add 0.046 quality-adjusted life years and save £69 per person over a lifetime. The result depends on a strong sensitivity assumption and remained cost-effective in only 53% of probabilistic simulations.
The short version
- What happened
- The analysis simulated 10,000 hypothetical adults aged 18 to 85 referred through England's symptomatic colorectal-cancer pathway and compared GI Genius-assisted colonoscopy with standard colonoscopy over a lifetime.
- Who it affects
- Patients may have more polyps flagged and removed, but the clinician remains responsible for interpreting each highlight and discussing follow-up.
- What remains unproven
- The study is a lifetime simulation of 10,000 hypothetical patients, not a trial measuring cancer incidence, quality of life or NHS spending.
Keep in mind
The study is a lifetime simulation of 10,000 hypothetical patients, not a trial measuring cancer incidence, quality of life or NHS spending.
NAIAD was observational, so residual differences between patients, sites and endoscopists may affect the adenoma-detection input.
The higher-detection strategy was assigned 100% sensitivity in the base case because no independent gold standard was available.
The model could not represent individual polyps or fully distinguish low-risk detection from lesions that would progress to cancer.
Only 53% of probabilistic simulations were cost-effective at the stated threshold, showing substantial decision uncertainty.
Multi-source analysis · 3 linked sources. Primary analysed source date: 6 November 2026. Earliest linked supporting source: 19 March 2026.
Read the full report · 8 min The evidence, methods, implications and limitations
The answer: a favourable base case with substantial uncertainty
The model's headline answer is that adding Medtronic's GI Genius system to colonoscopy could be both slightly better for health and slightly cheaper for the English NHS. For each simulated person, the base case produced 0.046 additional quality-adjusted life years and £69 lower lifetime cost than colonoscopy alone. In health-economic language, an intervention that gains health while reducing cost is dominant. Those numbers are modelled expectations, however, not savings observed in hospital accounts or extra healthy life measured among trial participants.
The uncertainty analysis is at least as important as the base case. Across 10,000 probabilistic iterations, GI Genius was cost-effective in 53% of simulations at the £20,000-per-QALY threshold. That is only slightly better than an even chance. Small changes in the modelled QALY difference could move the cost-effectiveness ratio sharply because the health gain is modest. A responsible interpretation is therefore not 'AI colonoscopy saves the NHS money', but 'one UK-focused model produces a favourable average result while leaving considerable decision uncertainty'.[1]
What population and comparison were modelled
The researchers built a patient-level simulation for 10,000 hypothetical adults aged 18 to 85 with symptoms suggestive of colorectal cancer who were referred for colonoscopy through England's Faster Diagnosis Framework. GI Genius was added at the index colonoscopy only; subsequent polyp surveillance, cancer progression, testing, treatment, quality of life and mortality were modelled in annual cycles over a lifetime. Costs were assessed from the NHS and Personal Social Services perspective, valued in 2024 prices and discounted at 3.5% alongside health outcomes.
Inputs for age, sex, cancer prevalence, adenoma detection and procedure outcomes came mainly from NAIAD, a multicentre observational NHS study of colonoscopies performed with or without GI Genius across endoscopists with varying experience. Other inputs came from the established MiMiC-Bowel model, NHS costs, Office for National Statistics life tables and older literature on cancer utilities, progression and treatment cost. Combining those sources can make a model decision-relevant, but it also means the final result inherits uncertainty and structural choices from evidence collected for different purposes and years.[1]
The distinction that matters
How the model turns extra detection into lifetime value
The reported savings emerge through assumptions about what happens after an index colonoscopy.
Refer
Simulate 10,000 symptomatic patients
Adults enter England's suspected-cancer pathway
Detect
Apply GI Genius at the index colonoscopy
Higher adenoma detection is converted into higher sensitivity
Project
Model polyps, cancer, costs and quality of life
Lifetime outcomes are discounted at 3.5%
Stress-test
Run 10,000 probabilistic iterations
53% are cost-effective at £20,000 per QALY
The strongest assumption links detection to lifetime benefit
NAIAD measured adenoma detection, not the true sensitivity of each strategy against a perfect reference standard. Colonoscopy itself can miss lesions, and determining sensitivity would require back-to-back examinations or long-term follow-up. The economic model therefore assumed that the strategy with the higher adenoma detection rate—GI Genius-assisted colonoscopy—had 100% sensitivity, while the standard strategy had proportionally lower sensitivity. That preserves the observed relative detection difference but gives the better-performing arm a structurally important advantage.
This matters because the estimated QALY gain and cost saving arise mainly from fewer missed polyps, fewer later colorectal cancers and lower cancer-treatment costs. The model applies GI Genius only once, then lets both arms follow the same disease-progression rules. If the extra adenomas would never have progressed, if true sensitivity is lower than assumed, or if increased detection leads to more low-value removals and surveillance, the lifetime benefit can shrink. Scenario analyses changed sensitivity and other assumptions and broadly preserved the direction of the base case, but they cannot substitute for observed post-colonoscopy cancer outcomes.[1]
A £10 tool cost is not the whole implementation cost
The model assigned GI Genius a cost of about £10 per person by annualising implementation over a five-year contract and spreading it across projected symptomatic colonoscopies. That is a useful procurement abstraction, not a guarantee that every NHS unit will face the same marginal cost. Local services may need compatible processors, installation, information-governance work, staff training, maintenance and workflow support. Sites with fewer procedures may spread fixed costs over a smaller denominator, while existing equipment and negotiated prices can change the calculation.
Higher adenoma detection can also shift activity from diagnostic to therapeutic colonoscopy because more polyps are removed. The paper includes procedure costs and adverse events, but real services must plan pathology demand, surveillance appointments and endoscopist time. NICE currently allows six AI colonoscopy technologies, including GI Genius, during an evidence-generation period subject to regulatory and Digital Technology Assessment Criteria requirements. That conditional status is important: the policy question is not simply whether a model's mean result is positive, but whether benefits, harms and resource effects are demonstrated while the technology is used.[1][2]
What this means for patients and NHS teams
For patients, computer-aided detection is an additional set of visual prompts during colonoscopy, not an autonomous diagnosis. The endoscopist still examines the bowel, decides whether a highlighted area is a lesion and determines whether to remove or sample it. A higher adenoma detection rate can be valuable if it prevents future cancer, but more detections can also mean more polypectomies, pathology and surveillance. People need clear information about how the software is used, who makes the final decision and whether evidence differs for symptomatic investigation, screening, inflammatory bowel disease or Lynch syndrome.
For NHS leaders, the result supports evidence generation rather than immediate universal procurement. Local evaluation should measure baseline detection rates, cancer case mix, endoscopist experience, false alarms, procedure time, removals, complications, pathology workload, post-colonoscopy cancers and total cost. The model suggests that sites with lower baseline detection could have more room to benefit, but the paper could not cleanly separate effects by experience because of its sensitivity assumption. Adoption should therefore be paired with monitoring that can reveal when performance improves and when extra activity produces little patient benefit.[1][2][3]
What evidence would change the assessment
Confidence would rise with randomised or carefully controlled pragmatic studies that follow symptomatic patients beyond the procedure and report advanced adenoma detection, removals, adverse events, post-colonoscopy colorectal cancers, stage at diagnosis and resource use. A transparent update to the economic model should replace perfect-sensitivity assumptions with empirically estimated miss rates, show results by polyp risk and endoscopist experience, and validate predictions against observed NHS outcomes. Reporting should also separate screening from symptomatic pathways and include services with different volumes and deprivation profiles.
The assessment would weaken if more detected polyps do not translate into fewer clinically important cancers, if pathology and surveillance costs erase projected savings or if performance varies sharply across hospitals. The work was funded through the Department of Health and Social Care AI in Health and Care Award to King's College Hospital; York Health Economics Consortium was commissioned to perform the evaluation, and all authors are YHEC employees. They declared no competing interests. The study is a useful decision model and a prompt for better data, not Breaking news or proof of realised NHS savings. Its own 53% probability result keeps that uncertainty in plain view.[1][2][3]
Keep following the evidence
Join The Impact Brief founding list for selected AI reports, their effect on people and the evidence limits. Read a sample first.
Continue the story
Read next
Continue exploring health economics: compare the finding and its evidence limits.
Can AI explain why kidney-disease hospital stays cost more?
A two-hospital Chinese study found that machine-learning models could retrospectively account for part of the variation in chronic-kidney-disease admission costs. The models used information accumulated during the stay, so they are not forecasts and should not set individual budgets.
Read the next reportExplore the full source and evidence record
What this means for people
- Patients may have more polyps flagged and removed, but the clinician remains responsible for interpreting each highlight and discussing follow-up.
- Endoscopy teams could gain a second visual prompt while also facing extra removals, pathology and surveillance work that needs local measurement.
- NHS commissioners should treat the £69 saving as a model estimate and link procurement to evidence generation, not assume the saving appears automatically in budgets.
Global context
Economic results for AI-assisted colonoscopy depend on local prices, cancer pathways, procedure volumes, pathology capacity and baseline detection rates. Models from screening programmes in Italy, Canada and the United States cannot be transferred directly to England's symptomatic pathway, and this English model cannot settle decisions elsewhere. NICE's evidence-generation approach offers a useful principle internationally: permit controlled use where appropriate, but require comparative and real-world outcomes before routine adoption.
What the evidence does not yet show
- The study is a lifetime simulation of 10,000 hypothetical patients, not a trial measuring cancer incidence, quality of life or NHS spending.
- NAIAD was observational, so residual differences between patients, sites and endoscopists may affect the adenoma-detection input.
- The higher-detection strategy was assigned 100% sensitivity in the base case because no independent gold standard was available.
- The model could not represent individual polyps or fully distinguish low-risk detection from lesions that would progress to cancer.
- Only 53% of probabilistic simulations were cost-effective at the stated threshold, showing substantial decision uncertainty.
What to watch next
- Final NAIAD denominators and peer-reviewed clinical outcomes
- Observed post-colonoscopy cancer rates and stage at diagnosis
- Pathology, surveillance, procedure-time and complication effects
- External validation of the economic model with real NHS cost data
- NICE's review after the evidence-generation period and results by technology
Living evidence record
Impact record IAI-068LTOW
Evidence stage
Studied
Confidence
Corroborated
Reporting basis
Multi-source analysis
Independent or research support
Present
Record status
Monitoring
Last checked
8 November 2026
Source trail
3 direct sources across 2 source types.
People impact
Documented in this record.
Uncertainty
Limits and next checks are explicit.
Stages describe the evidence available—not whether a technology is good or bad. See the public method.
Evidence trail
Sources used for this report
Links checked 8 November 2026
Research question, topics and regions
Whether adding GI Genius to colonoscopy for adults with symptoms of possible colorectal cancer is likely to improve lifetime health and reduce costs in the English NHS
Topics: Colorectal cancer · Colonoscopy · Health economics · NHS · Computer-aided detection
Regions: United Kingdom · England · Europe
This report is labelled multi-source analysis. We summarise and analyse source material in our own words; company statements remain attributed claims until independently supported. Translated summaries preserve the meaning of the original source and link back to it. Read our editorial standards.
Reader commentary
What does this mean where you are?
Tell us what you have seen at work, in your community or in everyday life. A question is welcome too. No account is needed; we review notes before publishing them.
Explore commentary across the portal →Published reader notes
0No published reader notes yet. You can start the evidence-led discussion above.
Prefer a private correction or response? Contact the newsroom.