Skip to main content

  • A 2-D Markov Model May Predict the Course of Glaucoma

    By Lynda Seminara
    Selected By: Stephen D. McLeod, MD
    Add to My Bookmarks

    Journal Highlights

    Ophthalmology, September 2018

    Download PDF

    The ability to detect glaucoma and predict its course is crucial for effective management. Song et al. previously introduced a state-based 2-D continu­ous-time hidden Markov model (2-D CT HMM) to represent the pattern of detected glaucoma changes using structural and functional information simultaneously. In the present study, their goal was to determine the pre­dictive performance of the model for detecting glauco­matous change in a real-world clinical setting, using retrospec­tive longitudinal data.

    The model proved promising for this purpose: Information from 1 visit signaled the clinical pic­ture through 5 subsequent visits.

    This longitudinal retrospective study included 134 patients (134 eyes) who had been diagnosed with or suspect­ed of having glaucoma. The hidden state dimensions were thickness of the circumpapillary retinal nerve fiber layer by optical coherence tomography (structural) and the visual field index (VFI; functional).

    In a second version of the model, mean deviation was substituted for VFI. The average follow-up period was 4.4 years, and the average number of visits was 7.1.

    A subset of the data (107 of 134 eyes; 80%), obtained from all visits ex­cept the final one, was used to train the model (training set). The validation set comprised data for the other 27 eyes. Prediction accuracy was represented as the percentage of correct predic­tions versus actual recorded states. The researchers also measured deviations of the predicted long-term detected change paths from the actual detected paths.

    Results showed that the accuracy of glaucoma changes predicted for the training set was comparable to that of the validation set (57% and 68%, respectively).

    The difference between predicted and actual detected paths of change re­mained similar throughout follow-up, with deviations actually decreasing (improving) over time. Because the sample size also declined with time, larger studies are needed to confirm the findings.

    The 2-D CT HMM has the ad­vantage of demonstrating nonlinear relationships between structural and functional degeneration. It may benefit glaucoma management by providing visually intelligible cues of changes in structure, function, or both. The mod­el’s ability to detect changes according to patient-specific data may pave the way for a new personalized-medicine approach to glaucoma assessment and treatment.

    The original article can be found here.