Clinical context

Cataract surgery outcomes are not defined by best-corrected visual acuity alone. Refractive accuracy, vision at different distances, spectacle dependence, contrast sensitivity, dysphotopsia and procedural safety all matter. Corneal and macular status, capsular support, cataract density and IOL calculation also shape the final result. A statistically convincing average therefore still has to be interpreted against the characteristics of an individual eye.

What was studied

IOL formulas sometimes agree closely and sometimes produce materially different predictions. This retrospective cohort tested whether the size of that disagreement can flag uncertainty. Six hundred eyes were evaluated using SRK/T, Hoffer Q and Holladay 1 versus Barrett Universal II, Kane, EVO 2.0, PEARL-DGS and Cooke K6. Formula offsets were estimated and frozen in 400 earlier eyes, then evaluated without retuning in 200 later eyes. The Legacy–Modern Gap was the absolute difference between the two groups' mean predicted spherical equivalents.

How the evidence was produced

A cohort study describes how outcomes are associated with an exposure or treatment in an observed clinical population. This design is useful for routine practice, uncommon events and longer follow-up, but it remains vulnerable to confounding by indication, baseline risk differences and missing data. An association should not automatically be interpreted as a causal effect.

Key result An absolute target miss of at least 0.50 D occurred in 29.0% of derivation eyes and 26.0% of the temporal validation set. Validation mean absolute error was 0.284–0.317 D for modern formulas and 0.358–0.424 D for third-generation formulas, P<.001. Every additional 0.10 D of disagreement was associated with target error: adjusted OR 1.21, 95% CI 1.05–1.38, P=.007. Model AUC was 0.650 versus 0.603 for biometrics alone; the AUC difference of 0.047 had a 95% CI from −0.015 to 0.111. Brier score improved by 0.0116, with calibration intercept 0.01 and slope 1.25.

How to interpret the result

Formula disagreement can be used as a prompt to recheck measurements, corneal status, constants and the intended target. It does not identify which formula is correct and should not select IOL power by itself. Modern formulas were more accurate on average in this dataset, but an AUC of 0.650 is too modest for an independent decision rule. Temporal validation is stronger than a simple random split because it better resembles future use, yet external testing in another clinical system remains necessary.

Clinical applicability

This evidence is most useful as a basis for preoperative discussion and surgical planning. It helps identify which outcomes should be measured and which trade-offs should be explained to patients. The average effect should not be generalized to every IOL, technique or patient without checking eligibility criteria, surgeon experience, biometry, ocular comorbidity and duration of follow-up.

Limitations

This was a retrospective single-center cohort limited to foldable one-piece IOLs. Results depend on the biometer, constant optimization, surgeons and axial-length distribution. The validation set contained only 200 eyes, and the confidence interval for AUC improvement included zero. Because local offsets were applied, a numerical disagreement threshold should not be transferred automatically. Verify IOL models, refraction timing and method, exclusions, biometric range, fellow-eye handling, the LMG threshold, calibration, missing data, sensitivity analyses, funding and conflicts.

What to verify in the source

Use the DOI below to verify complete methods, result tables, funding and conflicts of interest.