For diabetes patients presenting with macular edema (ME) soon after cataract surgery, pinpointing the underlying pathology can be challenging and influence management in some cases. However, a new study suggests just a few spectral-domain optical coherence tomography (SD-OCT) parameters may be able to identify the correct etiology between diabetic cystoid ME and pseudophakic ME with high accuracy and the right diagnosis.

Investigators analyzedSD-OCT data of 153 patients with either pseudophakic cystoid ME (57 patients), diabetic ME (86 patients) or “mixed” (10 patients). Researchers used machine learning algorithms to develop a predictive classifier using the smallest number of parameters and then applied the findings in a simple clinical decision flowchart.

“We used several advanced mathematical techniques based on machine learning and data mining,” the researchers wrote in their paper. “Our aim was to first evaluate the impact of each parameter on the final diagnosis, then to develop a simplified algorithm that uses a minimum number of parameters while retaining high accuracy, focusing on the algorithm’s ability to confirm a diabetic etiology, and finally, to implement the algorithm using a simple flowchart or formula.”

The study found the existence of hard exudates, hyperreflective foci, subretinal fluid, ME pattern and the location of cysts within retinal layers as the most differentiating features. By using only three to six OCT parameters, investigators achieved a sensitivity of 94% to 98%, specificity of 94% to 95%, and an area under the curve of 0.937 to 0.987 (depending on the method) for confirming a diabetic etiology. The simple decision flowchart investigators created achieved a sensitivity of 96%, a specificity of 95%, and an area under the curve of 0.937.

Researchers noted the most subjectively useful technique was the three-parameter decision tree that used only three yes/no parameters and a plain flowchart requiring no calculations. Although higher AUC and sensitivities were reached using other methods, this technique reached high sensitivity (96%), AUC (0.937), and only slightly lower specificity than the top classifier (95% compared with 96%). “For these reasons, we suggest applying this classifier in cases of uncertainty between DME and PCME,” investigators said.

The study’s flowchart, viewable in the full study,  can also help with treatment decisions, investigators suggested.

“We propose a clinical decision flowchart for cases with uncertainty, which may support the decision for intravitreal injections rather than topical treatment,” they added.

Hecht I, Bar A, Rokach L, Noy Achiron R, et al. Optical coherence tomography biomarkers to distinguish diabetic macular edema from pseudophakic cystoid macular edema using machine learning algorithms. Retina, October 10, 2018 [Epub ahead of print].