Abstract
Purpose :
Despite anti-VEGF being the standard-of-care for treatment of macular edema due to retinal vein occlusion (RVO), data show that many patients require long-term, frequent injections to maintain the gains achieved during the initial monthly treatment (tx) period, highlighting the need for more durable treatment options in RVO. Here we retrospectively investigated baseline (BL) and early tx response variables for association with ranibizumab (RBZ) tx frequency during the as-needed (PRN) dosing period of the BRAVO and CRUISE trials.
Methods :
BRAVO and CRUISE compared RBZ monthly dosing vs sham during the first 6 months (M) followed by PRN RBZ dosing from M6 to M12 in patients with macular edema due to branch and central RVO (BRVO/CRVO), respectively. Post hoc statistical and machine learning methods were applied to identify features predictive of RBZ injection frequency required for vision maintenance during the PRN dosing period. We tested average of BCVA values at BL, M1, M2 and M3, BCVA change from BL to M3, and 13 pre-specified imaging features for their impact on predictive accuracy, measured by area under the receiver operating characteristic curve (AUC).
Results :
BRVO and CRVO patients (overall, n = 419; 0.3 mg, n = 211; 0.5 mg, n =208) across treatment groups received 3.3 ± 2.1 RBZ injections during the PRN dosing period. The best predictor of injection frequency was the average of BCVA values at BL, M1, M2 and M3 (overall AUC = 0.76 [0.71, 0.81]; 0.3 mg AUC = 0.80 [0.73, 0.86]; 0.5 mg AUC = 0.73 [0.66, 0.80]). Adding BCVA change from BL to M3 did not increase predictive performance significantly. Adding the pre-specified imaging variables increased performance of the statistical model (overall AUC = 0.84 [0.80, 0.88]; 0.3 mg AUC = 0.85 [0.80, 0.91]; 0.5 mg AUC = 0.82 [0.76, 0.88]).
Conclusions :
The BRAVO and CRUISE RVO population was heterogeneous and required different treatment frequency following the initial 6 monthly doses. The average of BCVA values at BL, M1, M2 and M3 was associated with the greatest predictive signal of future injection need. Imaging features increased predictive performance of the model. Clinicians may find the individual modeling predictions helpful when considering options for the long-term management of patients with RVO.
This is a 2021 ARVO Annual Meeting abstract.