Cataract AI formulas worked best locally—and slipped at a second hospital
A retrospective Chinese study compared lens-power formulas in 1,203 eyes, then tested locally trained models on 336 high-myopia eyes elsewhere. Internal accuracy did not travel cleanly, reinforcing the need for local validation before surgical use.
The short version
- What happened
- The development cohort contained 1,203 eyes from West China Hospital; external validation used 336 high-myopia eyes from Chengdu Aidi Eye Hospital.
- Who it affects
- Better lens-power estimates could reduce unwanted refractive error after cataract surgery, but this study does not establish that an AI formula improves patients' vision or quality of life.
- What remains unproven
- Retrospective rather than prospective design
Keep in mind
Retrospective rather than prospective design
Both hospitals were in Chengdu, China
External cohort restricted to high-myopia eyes
No clinician comparison, workflow test or patient-outcome evaluation
Only 257 of 336 external eyes entered the cleaned counterfactual target-achievement analysis
Source analysis · 1 linked source. Primary analysed source date: 22 October 2026.
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.
Read the full report · 3 min The evidence, methods, implications and limitations
The strongest internal result did not settle the clinical question
Researchers compared conventional, post-refractive and AI-based formulas using 1,203 cataract-surgery eyes from West China Hospital. They also trained linear regression, support-vector regression and random-forest models to predict postoperative refractive error. In eyes without earlier corneal refractive surgery, AI-based formulas outperformed conventional alternatives; Hill-RBF produced the lowest internal prediction error. The locally developed models reached similar accuracy on the development data.
The useful stress test came from another institution. The authors applied the models to 336 high-myopia eyes from Chengdu Aidi Eye Hospital and found reduced portability. A cleaned counterfactual analysis—asking whether a different lens choice would have reached the target—was possible for 257 of those eyes. The paper reports that previous refractive surgery, very long eyes and flat corneas remained especially difficult. SHAP analysis ranked target refraction, axial length and implanted-lens power as the dominant predictors.[1]
Local validation matters before a formula influences surgery
For patients, a prediction error can mean residual short- or long-sightedness after cataract surgery and continued dependence on glasses or further correction. This study supports comparing formulas on a hospital's own population rather than assuming that an internally strong model will transfer unchanged. It does not show that the locally trained models improve surgical decisions: the analysis was retrospective, the external cohort was restricted to high myopia, and no prospective workflow, clinician comparison, visual function or patient-reported outcome was tested.
The assessment would change with a locked, prospective multi-country evaluation that reports calibrated error thresholds for each important eye subgroup and follows actual lens choices and postoperative vision. The authors reported no specific funding and no competing interests. The accepted article is peer reviewed but may still receive copy-editing before its final version of record.[1]
The distinction that matters
Why the second hospital changes the claim
Internal comparison finds the best fit to one hospital; external validation tests whether that advantage travels.
Development
Compare and train on 1,203 eyes
Hill-RBF has the lowest internal error
External test
Apply models to 336 high-myopia eyes
Locally trained models lose portability
Clinical value
Not tested prospectively
No demonstrated patient benefit
Watch the background, then check the evidence
Illustrative photo · not a video frame
The Royal Society · 44 min 28 sec · Background explainer
Is this the AI future we want? | The Royal Society
YouTube loads only when you choose; playback starts when you press play.
A Royal Society discussion with Madhumita Murgia and biomedical engineer Alison Noble on AI’s potential in healthcare, science and culture, recorded at the 30 June 2026 event.
Evidence check: A discussion of opportunities and risks, including future possibilities. It is not evidence that every application described is available or clinically proven.
Published 2026-07-01. Source checked 14 October 2026. Check captions on the publisher’s video page.
Check the healthcare evidencePhoto: Daniel Sone / National Cancer Institute / Unsplash · Licence
Continue the story
Read next
Continue exploring clinical prediction: compare the finding and its evidence limits.
Can AI predict risks before thymic-tumour surgery?
Models trained on 726 Chinese patients estimated complications, longer stays and severe pain before surgery. External AUCs of 0.660–0.776 make this promising risk support—not evidence for automated care decisions.
Read the next reportExplore the full source and evidence record
What this means for people
- Better lens-power estimates could reduce unwanted refractive error after cataract surgery, but this study does not establish that an AI formula improves patients' vision or quality of life.
Global context
Eye shape, surgical practice, equipment and implanted-lens choices vary across populations and hospitals. Performance in two Chinese centres should not be assumed to transfer to other health systems.
What the evidence does not yet show
- Retrospective rather than prospective design
- Both hospitals were in Chengdu, China
- External cohort restricted to high-myopia eyes
- No clinician comparison, workflow test or patient-outcome evaluation
- Only 257 of 336 external eyes entered the cleaned counterfactual target-achievement analysis
What to watch next
- Locked prospective external validation
- Pre-specified subgroup error thresholds
- Clinician-with-AI comparisons
- Postoperative visual and patient-reported outcomes
Living evidence record
Impact record IAI-12JI2ML
Evidence stage
Studied
Confidence
Supported
Reporting basis
Source analysis
Independent or research support
Present
Record status
Monitoring
Last checked
25 October 2026
Source trail
1 direct source across 1 source type.
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.
Single-source reporting disclosure
This record analyses one direct source. It can establish what Scientific Reports published or reported, but it is not independent corroboration of every performance claim or predicted outcome. The confidence label will change only when broader evidence is added.
Evidence trail
Sources used for this report
Links checked 25 October 2026
Research question, topics and regions
Whether AI-based intraocular-lens formulas and locally trained machine-learning models transport across hospitals for postoperative refractive prediction
Topics: Cataract surgery · High myopia · Clinical prediction · External validation
Regions: China · East Asia · Global
This report is labelled 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.