Circumpapillary retinal nerve fiber layer and ganglion cell inner plexiform layer measurements may help predict visual field worsening in glaucoma, although each is more effective in different stages of disease, according to research.

As most studies have used statistical approaches to compare these measurements when assessing glaucoma progression, researchers sought to compare the diagnostic accuracy of machine learning models.

"Exploring a machine learning approach to predict VF worsening based on structural parameters may be beneficial, because machine learning classifiers have shown similar or improved diagnostic accuracy for detecting glaucoma compared to with traditional statistical models," Alex T. Pham, a research fellow at John Hopkins Wilmer Eye Institute, and colleagues wrote in Translational Vision Science & Technology.

The researchers trained machine learning and statistical models on 924 eyes with circumpapillary retinal nerve fiber layer (cp-RNFL) or ganglion cell inner plexiform layer (GC-IPL) thickness measurements. They then predicted the probability of 24-2 visual field worsening and conducted testing on 617 eyes to calculate the area under the curve for predictions based on cp-RNFL, GC-IPL or a combination of the two.

According to results, the logistic regression AUCs for mean deviation slope progression were 0.72 for cp-RNFL, 0.69 for GC-IPL and 0.73 for combined predictions, while the machine learning model AUCs were 0.78, 0.75 and 0.81, respectively -- differences that were not statistically significant.

The researchers also reported that cp-RNFL predictions performed better in suspected cases of glaucoma, while GC-IPL predictions were better in moderate or advanced disease. Combining predictions did not significantly improve predictive performance.

"We demonstrate that 24-2 VF worsening defined by trend-based analysis can be predicted with modest accuracy using longitudinal cp-RNFL and GC-IPL thickness measurements," the researchers wrote.