The Moussa Research Suite: Validation And Demonstration Of Statistical Equivalence To The Whwk Framework
Published 2026 - 44th Congress of the ESCRS
Reference: PP03.03 | Type: Free paper | DOI: 10.82333/gr38-d872
Authors: Mr George Moussa* 1 , Nathan Cannon 2 , David Cooke 3
1Manchester Royal Eye Hospital,Manchester,United Kingdom, 2Penn State Hershey Medical Center and Penn State College of Medicine,Hershey,United States, 3Great Lakes Eye Care,Saint Joseph,United States
Purpose
To validate the statistical equivalence of the web-based Moussa Research Suite against the established Wilcox-Holladay-Wang-Koch (WHWK) framework for intraocular lens (IOL) power calculation formula evaluation, focusing on Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE) utilising a user-friendly graphical user interface (GUI)
Setting
Retrospective computational validation utilizing a multi-center dataset of 7,842 eyes from five academic medical centers and two private practices collected between 2011 and 2022.
Methods
A comparative validation was performed using a multi-center dataset of 7,842 eyes across 16 IOL formulas, generating 120 pairwise comparisons. Both platforms applied heteroscedastic studentized bootstrap methods with Hommel correction for Family-Wise Error Rate (FWER) control. Statistical equivalence was assessed using Cohen's Kappa for categorical agreement and Two One-Sided Tests (TOST) for continuous p-value arrays. The influence of bootstrap iteration number on convergence was examined through sensitivity analysis, with WHWK run at both 1,999 iterations (default) and 100,000 iterations, compared against the Moussa suite at 500,000 iterations.
Results
At 1,999 iterations, Cohen's Kappa was 0.9833 for MAE and 0.9831 for RMSE, with 97.5% categorical agreement for both metrics at α=0.05. TOST demonstrated equivalence for both MAE (mean difference 0.00120, p<0.001) and RMSE (mean difference 0.00309, p<0.001). Bland-Altman analysis revealed p-value discrepancies up to 0.11 for specific MAE formula pairs at 1,999 iterations. When WHWK iterations were increased to 100,000 for MAE, categorical agreement reached 100% at α=0.05 (Cohen's Kappa=1.0000), with mean p-value difference of −0.00002 and TOST p-value of 4.3×10^-91, confirming complete convergence.
Conclusions
The Moussa Research Suite produces statistically equivalent results to the WHWK framework for MAE and RMSE analysis with Hommel correction. P-value discrepancies observed at 1,999 iterations reflect inherent bootstrap sampling variability from different random number generators between platforms, with complete convergence demonstrated at higher iteration counts.This is done while through a user-friendly GUI that widens the accessibility across clinicians.