How to choose the intraocular lens power calculation formulas in eyes with extremely long axial length? A systematic review and meta-analysis

Author: Li X, Wang X, Liao X.

Geographical coverage: Not reported

Sector: Intraocular lens

Sub‑sector: Cataract treatment

Equity focus: Not reported

Study population: Patients with cataract

Review type: Effectiveness review

Quantitative synthesis: Meta‑analysis

Qualitative synthesis: Not applicable

Background

Accurate calculation of intra‑ocular lens (IOL) power is essential to achieve the desired refractive outcome in cataract surgery. Although optical biometry has improved measurement precision, selecting the most suitable calculation formula is still challenging in eyes with extremely long axial lengths (AL > 28.0 mm). Traditional formulae often perform poorly under these conditions, leading to hyperopic surprises and patient dissatisfaction. Newer formulae such as Barrett II, Kane, EVO and Hill‑RBF incorporate additional biometric variables and make use of artificial intelligence to improve accuracy. Previous studies have often excluded eyes with extremely long axial lengths, leaving uncertainty about the optimal choice of formula. As more cataract surgeries are performed in highly myopic patients, there is a need for evidence‑based guidance to improve surgical outcomes.

Objective

To evaluate the accuracy of ten intra‑ocular lens power calculation formulae for cataract eyes with axial length greater than 28.0 mm.

Main findings

The review analysed 11 studies involving 1 376 eyes, published between 2019 and 2022. It reported that the mean absolute error (MAE) for the Barrett II formula was significantly lower than for the Haigis (mean difference –0.20 D, 95 % confidence interval [CI] –0.34 to –0.07 D, P = 0.004) and SRK/T (–0.19 D, 95 % CI –0.36 to –0.01 D, P = 0.03) formulae, indicating that Barrett II was more accurate in long‑axial‑length cataract eyes. Kane also showed a significantly lower MAE than Haigis (–0.30 D, 95 % CI –0.51 to –0.10 D, P = 0.004) and SRK/T (–0.25 D, 95 % CI –0.49 to –0.01 D, P = 0.04). Conversely, the MAE for EVO was higher than for Kane (0.04 D, 95 % CI 0.00 to 0.07 D, P = 0.03) and Hill‑RBF (0.05 D, 95 % CI 0.01 to 0.09 D, P = 0.01), suggesting inferior accuracy relative to those two formulae.

Both EVO and Hill‑RBF nevertheless performed better than SRK/T, demonstrating lower MAEs. In addition, Hill‑RBF had a significantly lower MAE than the Holladay 1 formula (mean difference –0.72 D, 95 % CI –0.91 to –0.53 D, P < 0.001), indicating substantially greater predictive accuracy. Hill‑RBF also achieved the highest proportion of eyes within ±0.50 D and ±1.00 D of prediction error, with Kane ranked second. EVO and Kane yielded higher percentages within ±0.50 D and ±1.00 D of prediction error than Haigis and SRK/T.

Methodology

Searches were undertaken in PubMed, Embase, Web of Science and the Cochrane Library over the preceding five years up to September 2023 to identify case series involving eyes with AL > 28 mm. Studies were included if they featured uncomplicated cataract surgery with IOL implantation, compared at least two calculation formulae and measured all ocular parameters using optical biometry.

Two reviewers independently screened and extracted data, resolving disagreements through discussion or consultation with a third reviewer. Study quality was assessed with a modified checklist adapted from the QUADAS‑2 tool. The results were synthesised using fixed‑ or random‑effects meta‑analysis, depending on heterogeneity. Heterogeneity was examined using the chi‑squared test and I² statistics, and publication bias via a funnel plot. Sensitivity analysis explored potential sources of heterogeneity.

Applicability and external validity

The review acknowledged that external validity was limited by the small number of included studies — particularly for some formulae — and by the low prevalence of cataract patients with extremely long axial lengths, which restricted sample sizes and statistical power. Variations in study design, IOL type and patient demographics may also affect generalisability. The authors stressed the need for larger, multicentre studies with standardised methodologies to confirm the accuracy of newer formulae in broader clinical settings.

Geographical focus

No geographical limits were applied; however, the review did not report the geographical distribution of included studies.

Summary of quality assessment

The conclusions of this review are considered to be of medium confidence. Searches in PubMed, Embase, Web of Science and the Cochrane Library over the past five years (to September 2023) were conducted without language restrictions. Inclusion and exclusion criteria were clearly defined. Two reviewers independently screened articles and extracted data, and study quality was assessed using established tools. Characteristics of included studies were well documented, meta‑analyses were appropriately performed and heterogeneity was addressed. However, the review did not list excluded studies, nor did it report whether reference lists of included studies were checked or analyse results by risk‑of‑bias status.