Automated macular fluid volume as a treatment indicator for diabetic macular edema

Abstract

Purpose : To assess the diagnostic accuracy of an automated macular fluid volume (MFV) quantification for treatment required diabetic macular edema (DME).

Methods : The evaluation included macular structural optical coherence tomography (OCT) raster scans (Spectralis, Heidelberg), and 3x3-mm macular OCT angiography volumetric scans (Avanti, Optovue). Central subfield thickness (CST) was measured on Spectralis OCT scans using the embedded software. A custom deep-learning algorithm automatically quantified MFV within 3x3-mm on the OCTA scans. Physicians treated patients per standard of care based on clinical and Spectralis OCT findings without access to MFV. We calculated the area under the receiver operating characteristic curve (AROC) and sensitivity and specificity of CST, MFV, and visual acuity for treatment decision.

Results : Of 139 eyes, 39 (28%) eyes underwent treatment for DME. The algorithm detected fluid in all eyes with a mean (SD) MFV of 0.062 (0.14) mm³, but only 54 (39%) eyes met the DRCR.net criteria (CST>=320mm in male or CST>=305mm in female) for center-involved macular edema. The AROC of MFV for a treatment indication (0.81; 95% CI, 0.73 to 0.90) was larger than the AROC of CST (0.67; 95% CI, 0.56 to 0.78; P = 0.0048) (Fig 1). With the specificity fixed at 80%, the sensitivity of MFV for treatment indication was 74.4% (95%CI, 61.5% to 87.2%), higher than that of CST of 41.0% (95%CI, 25.6% to 56.4%, P = 0.0007). The eyes that met the threshold for treatment based on MFV (>0.031 mm³) but the physicians chose not to treat had better visual acuity (mean [SD], 77.9 [7.3] letters) compared to the eyes that were treated (70.3 [10.8] letters; P = 0.0053). A multivariate logistic regression model showed that MFV (Estimate = 13.6, [95% CI, 6.7 to 22.7], P = 0.0008) and visual acuity (Estimate = -0.082 [95% CI, -0.14 to -0.026], P = 0.0056) were significantly associated with treatment.

Conclusions : MFV may better predict the need for treatment for DME compared to CST. This may make treatment and screening referral decisions more objective in DME.

This abstract was presented at the 2022 ARVO Annual Meeting, held in Denver, CO, May 1-4, 2022, and virtually.

 

Figure 1. The area under the receiver operating characteristic curve (AROC) for predicting treatment decision