Postoperative-Only Estimation of Implanted Intraocular Lens Power Using a Closed-Form Thick-Lens Model With Patient-Specific Predictive Intervals

To estimate labeled implanted intraocular lens (IOL) power from postoperative measurements using a closed-form thick-lens model and provide patient-specific predictive intervals.

We included non-toric pseudophakic eyes with postoperative axial length, aqueous depth, central corneal thickness, total corneal power, and manifest refraction. A paraxial thick-lens equation was analytically inverted using axial length, aqueous depth plus central corneal thickness, total corneal power, and spherical equivalent. The total-corneal-power zone was selected in the training set, followed by patient-level train/test evaluation. Predictions were assessed continuously and after rounding to 0.50-D inventory steps. Predictive intervals used closed-form sensitivities, first-order variance propagation, and training-set calibration.

The cohort comprised 170 eyes from 109 patients, and the held-out test set comprised 32 eyes from 22 patients. The 4-mm total-corneal-power zone performed best. Continuous test predictions had a mean absolute error of 0.397 D and root mean square error of 0.509 D, and 65.6% were within ±0.50 D. After 0.50-D rounding, mean absolute error was 0.391 D and root mean square error was 0.566 D, and 75.0% were within ±0.50 D. Calibrated 95% predictive intervals covered 31 of 32 test eyes (96.9%); patient-level coverage was 21 of 22 (95.5%), with a mean interval width of 2.238 D.

Postoperative-only thick-lens estimation provides a consistency check against recorded IOL labels but is not precise enough to identify small discrepancies confidently.

When operative records are missing, the method offers a physically grounded label-consistency check with explicit uncertainty.