PREVENT WHAT YOU CAN. UNDERSTAND WHAT YOU COULDN’T.

Many complications leave their traces before the laser fires — in the geometry, the history or the device settings. RLES AI hunts their preconditions at the planning stage, and when an adverse event does occur, investigates it with the discipline of an incident analysis, not the reflex of blame.

THE PREMISE: RISK IS SET BEFORE THE DAY OF SURGERY

The most instructive complication statistic in eye surgery is not an average but a spread. In an audit of 55,567 cataract operations the overall rate of posterior capsule rupture or vitreous loss was 1.92%, yet the predicted probability for an individual operation ranged from under 0.75% to over 75% — case-mix, the authors concluded, is a major determinant of intraoperative complication probability.^[4] That is a cataract series and its numbers do not transfer to refractive surgery; the structure does. Where risk is largely determined by a pre-operative data set, the limiting factor is not skill or care — it is whether every item in that set was actually asked, on that day, in that case. RLES AI is built on that one premise, and the same logic drives the cataract screening path.

PREVENTION: THE PRECONDITION HUNT

• Hinge ramp calculation — the ΔK between the steep axis and the hinge site converted to microns; hinges thinner than the safety floor are flagged before a free flap can happen^[1]
• Buttonhole geometry, keratome kinematics (linear vs rotational cutting heads behave differently)^[2], and the topical-anaesthetic rule that prevents epithelial slides mimicking buttonholes^[3]
• Narrow-angle screening on every preop topography, medication anamnesis (isotretinoin, amiodarone, topiramate) and chronic-disease query on every case

On the flap side the numbers are small but unevenly distributed. Across 84,711 primary LASIK cases the overall intraoperative flap complication rate was 0.302% — partial flaps 0.099%, thin or irregular flaps 0.087%, buttonholes 0.070%, failure to achieve suction 0.034%, free flaps 0.012%.^[1] In the subset where two microkeratomes could be compared head to head, the rates differed by more than an order of magnitude: 0.16% across 28,201 procedures with one instrument against 6.38% across 329 with the other. The second denominator is small and the comparison is not randomised, but the gap was statistically significant — and it makes a point the rest of this platform is built on: the instrument in the surgeon’s hand changes the number. RLES AI therefore holds a separate nomogram per keratome and never substitutes another machine’s values when a profile is not loaded.

ANALYSIS: A CONFIDENTIAL IN-CLINIC INSTRUMENT

When something goes wrong, the complication module runs a structured investigation: an atlas of 45+ entities across six families, mimic tables (what looks like a buttonhole but is not), a four-grade severity score, and root-cause categories that explicitly include device condition, calibration history, blade lots and theatre environment. The report is an internal-use document built on just-culture principles — process analysis, not person analysis — and closes with preventive actions, including revised nomogram values for a future second procedure where appropriate.

THE QUESTIONS COME BEFORE THE VERDICT

The module generates a neutral, complication-specific question set — which keratome, which settings, when was the last calibration, what did the interface look like — and the report is not written until the questions and imaging are complete. Understanding first; conclusions after.

FREQUENTLY ASKED QUESTIONS