Posters

Solutions for missing retinal thickness OCT data

Poster Details

First Author: C.Veldhuizen THE NETHERLANDS

Co Author(s):    C. Veldhuizen   B. Winkens   L. Wielders   J. Schouten   F. Goezinne   R. Nuijts     

Abstract Details

Purpose:

to investigate the predictive quality of various linear regression models in direct imputation of missing ocular coherence tomography (OCT) retinal macular thickness values between two scanning protocols of a Fourier Domain (FD) OCT machine

Setting:

This study uses a subset of OCT data from the Zuyderland Medical Center in Heerlen, the Netherlands, one of the centers of the multinational multicenter European Society of Cataract and Refractive Surgeons (ESCRS) PREvention of Macular EDema after Cataract Surgery (PREMED) study.

Methods:

Prediction of 6mmx6mm macula map (MM6) perifoveal thickness values was achieved by construction of three linear regression models. The first model incorporated only the 5mmx5mm extended macula map (EMM5) values, where EMM5 values and OCT operator were added to the second, and the third including EMM5 values, OCT operator, diabetic status, and the ESCRS PREMED cystoid macular edema (CME) preventative treatment arm. We then compared the statistical significance, predictive quality and agreement between these three models. Thereafter, a sensitivity analysis based on the OCT visit timing was undertaken.

Results:

All three models showed excellent predictive quality at both the average and individual level. Correcting for OCT visit timing did not substantially change the predictive quality of any of the models.

Conclusions:

Our model using EMM5 perifoveal macular thickness values as a predictor for MM6 perifoveal macular thickness values showed promising results. However external validation is needed to confirm this relatively good predictive quality.

Financial Disclosure:

NONE