Accuracy Of Ai-Based Formulas In Extremely Long Eyes After Myopic Lvc

To compare the accuracy of AI-based intraocular lens (IOL) power calculation formulas (PEARL‑DGS, EVO 2.0 and Hoffer QST) with commonly used post–laser vision correction (LVC) formulas (Barrett True‑K No History [BTKNH] and Haigis‑L) in extremely long eyes after prior myopic LVC.

Changsha Aier Eye Hospital, China

This single-center retrospective series included patients with prior M-LVC and high axial length（>30.00 mm）who underwent uneventful phacoemulsification. Preoperative optical biometry was obtained, and IOL power was calculated using the candidate formulas. Manifest refraction was measured 3 months postoperatively. Post hoc refractive prediction errors were calculated for PEARL-DGS, EVO 2.0, BTKNH, Haigis- L and Hoffer QST. Main outcome measures: Root mean square absolute error (RMSAE), median absolute error (MedAE), and the proportions of eyes with absolute prediction error within ±0.50 diopters (D) and ±1.00 D.

BTKNH achieved the lowest RMSAE (0.74 D), with statistical superiority over Haigis‑L (0.98 D; P<0.05) and Hoffer QST (1.36 D; P<0.05). For MedAE, EVO 2.0 (0.42 D), BTKNH (0.47 D), PEARL‑DGS (0.50 D), and Haigis‑L (0.54 D) were each significantly better than Hoffer QST (1.06 D; all P<0.05); EVO 2.0 was numerically the lowest. The percentage of eyes within ±0.50 D was highest with BTKNH and EVO 2.0 (53.8% each), both significantly higher than Hoffer QST (23.1%; P<0.05). The percentage of eyes  ±1.00 D exceeded 80% with BTKNH, EVO 2.0, and PEARL‑DGS (82.1%, 84.6%, and 82.1%, respectively).

In this highly challenging cohort of post‑LVC eyes with AL >30.00 mm, EVO 2.0 and PEARL‑DGS retained competitive accuracy. BTKNH remained highly accurate and, together with EVO 2.0, achieved the highest proportion within ±0.50 D, whereas Hoffer QST and Haigis‑L underperformed. These findings support preferential use of BTKNH and EVO 2.0—and cautious use of Hoffer QST and Haigis‑L—in post‑LVC eyes with extreme axial lengths.