Comparing the Accuracy of New Intraocular Lens Power Calculation Formulae in Short Eyes after Cataract Surgery: A Systematic Review and Meta-analysis

Author: Luo Y, Li H, Gao L, Du J, Chen W, Gao Y, Ye Z, Li Z.

Geographical coverage: Not reported

Sector: Cataract surgery

Sub-sector: Intraocular lens

Equity focus: Not reported

Study population: Patients with cataract

Review type: Effectiveness review

Quantitative synthesis method: Meta-analysis

Qualitative synthesis method: Not applicable

Background:

Cataract is a leading cause of blindness worldwide. Achieving an accurate refractive outcome is especially challenging in short eyes (axial length ≤ 22 mm) because of anatomical characteristics such as shallow anterior chambers and steeper corneas, which can amplify prediction errors in intra-ocular lens (IOL) power calculations. Traditional regression-based formulae draw on only a few biometric variables, whereas newer generations incorporate additional factors—including lens thickness, corneal diameter and patient demographics—or employ ray-tracing and artificial-intelligence techniques to enhance precision. Despite the availability of many such formulae, consensus on the optimal choice for short eyes remains elusive, making reliable prediction in this subgroup a continuing clinical priority.

Objective:

To compare the accuracy of a range of IOL-power formulae in predicting post-operative refraction in patients with short axial length and to offer practical guidance for clinicians.

Main findings:

Fourteen studies (1 476 eyes) met the inclusion criteria. Outcomes were reported as the proportion of eyes with refractive prediction error (PE) within ±0.25 D, ±0.50 D and ±1.00 D.

• Formulae evaluated:
  • Newer approaches – Barrett Universal II, Castrop, Kane, Ladas Super Formula, Okulix, Olsen, Pearl-DGS and T2.
  • Traditional approaches – Haigis, Hoffer Q, Holladay 1, Holladay 2 and SRK/T.
• Performance highlights:
  • Pearl-DGS (AI-based) achieved the highest accuracy within ±0.25 D and ±0.50 D, performing significantly better than Barrett Universal II, Haigis, Hoffer Q, Holladay 1, Holladay 2 and Olsen.
  • Okulix (ray-tracing) was most accurate within ±1.00 D, significantly outperforming Barrett Universal II, Castrop, Hoffer Q and Holladay 2.
  • Among conventional regression formulae, Holladay 2 performed best, followed by Holladay 1, Hoffer Q, Haigis and SRK/T; T2 was least accurate.

Authors concluded that Pearl-DGS and Okulix currently provide the most reliable refractive predictions in short eyes; Castrop, Kane and Olsen are credible alternatives. Classic formulae such as SRK/T and T2 should be used with caution—preferably alongside more advanced methods—to minimise large refractive errors.

Methodology:

Systematic searches of PubMed, Embase, Web of Science and the Cochrane Library (January 2011 – March 2021) identified studies of short eyes (axial length ≤ 22 mm) that compared at least two formulae and reported refraction ≥ 2 weeks post-surgery. Reference lists were hand-searched for additional studies. Two reviewers independently extracted data and assessed quality with a QUADAS-2-adapted checklist, resolving disagreements by discussion. Fixed- or random-effects meta-analysis was conducted according to heterogeneity (I²). Publication bias was explored with funnel plots; subgroup and sensitivity analyses investigated heterogeneity sources.

Applicability / external validity:

Interpretation is tempered by small sample sizes in some studies, inconsistent use of biometry devices and variable constant-optimisation methods. Heterogeneity across studies and limited geographical reporting may restrict generalisability. Promising formulae such as Castrop require validation in larger, more diverse populations before routine adoption.

Geographic focus:

No geographical limits were set, but the locations of included studies were not specified.

Summary of quality assessment:

The database searches were comprehensive, inclusion criteria were explicit and dual-reviewer processes were applied. Study characteristics were well described and appropriate meta-analytic techniques used. Nonetheless, omission of an excluded-studies list, possible language restriction and unreported reviewer numbers during screening lower overall confidence in the findings.