Hill-RBF method: does it fit all situations? A pilot study
Session Details
Session Title: Presented Poster Session: IOL Power Calculation/Paediatric
Venue: Poster Village: Pod 1
First Author: : S.Barros PORTUGAL
Co Author(s): : S. Parreira D. Lopes I. Machado M. Pereira N. Campos
Abstract Details
Purpose:
Cataract surgery is the most widely surgery performed in ophthalmology. Targeting emetropia is one of the main goals and several intraocular lens (IOL) power calculation formulas have been developed. However, some of them require special measuring devices and others are not free, making them unavailable for some surgeons in real-life practice.
Our purpose is to evaluate the accuracy of the newer Hill-RBF method compared with 3rd generation formulas using a PCI device and different IOL models.
Setting:
Department of Ophthalmology, Hospital Garcia de Orta
Methods:
Hill-RBF method is based in artificial intelligence and was developed based on LENSTAR LS900 biometry data in combination with AlconSN60WF IOL.
In this study authors used IOLMaster 500 in all eyes. IOL models included: monofocals AcrysoftIQ, Tecnis1piece and AkreosAO. Extreme axial lengths, post-RK or LASIK eyes were excluded.
The design of the study was based on previous reports published in the literature.
IOL power formula predictions were compared using Hill-RBF, Haigis and SRK/T. They were ranked according to: mean prediction error, mean absolute error, standard deviation and percentage of final refractions within ±0,50D and ±1,00D of the predicted value.
Results:
Pilot retrospective study with review of clinical data of 100patients.
IOL models included Acrysoft IQ in 48eyes, Tecnis1piece in 35 and Akreos AO in 27.
Hill-RBF performed significantly better in all scores, followed by Haigis and SRK/T.
Mean prediction error, mean absolute error and standard error favored Hill-RBF (-0.043; 0.070; 0.100 respectively, for Hill-RBF), (0.030; 0.610; 0.723 respectively, for Haigis), (0.050; 0.850; 0.956 respectively for SRK/T).
The percentage of final refractions within the target was also better with Hill-Rbf compared with Haigis and SRK/T, respectively: 100%vs60%vs52% final refractions within ±0,50D and 100%vs86%vs78% final refractions within ±1,00D.
Monofocals AcrysoftIQ performed better.
Conclusions:
This is the first study evaluating Hill-RBF performance in clinical practice.
Even though the study was not performed according to standardized recommendations for most accurate results with Hill-RBF methos, it performed significantly better than Haigis and SRK/T formulas when using PCI devices and with different IOL models.
Large case series would be valuable to corroborate these results and if so, to consider the introduction of these free method, available online, in daily practice (even if not all requirements were set), in order to further improve outcomes in every single clinic.
Financial Disclosure:
NONE
