Use of Novel Analytic Method to Predict DSAEK Graft Failure
By Jean Shaw
Selected and Reviewed By: Neil M. Bressler, MD, and Deputy Editors
Journal Highlights
JAMA Ophthalmology, February 2021
Download PDF
The machine-learning technique known as random survival forest (RSF) has emerged as a promising alternative to traditional analytic methods. O’Brien et al. used an RSF model to rank multiple variables associated with Descemet stripping automated endothelial keratoplasty (DSAEK) graft failure in the Cornea Preservation Time Study (CPTS). They found that intraoperative complications were highly predictive of graft failure in the CPTS.
For this cohort study, the authors evaluated data on 1,090 CPTS participants (1,330 eyes) who underwent DSAEK for Fuchs dystrophy (1,255 eyes; 94.4%) or for pseudophakic or aphakic corneal edema (75 eyes; 5.6%). All told, 81 eyes experienced graft failure in the first four years after DSAEK.
For their analysis, the authors selected 50 baseline donor, recipient, eye bank, and intraoperative variables and used RSF to analyze the data and rank the variables. The final RSF model, which comprised five variables, identified DSAEK intraoperative complications as the third most predictive factor of graft failure, after surgeon and eye bank. History of diabetes in the donor was the fourth most predictive factor, and preservation time ranked fifth.
With regard to graft survival time, in the first 47 months after DSAEK, grafts that experienced an intraoperative complication survived between 70 and 352 fewer days than those that did not.
To date, RSFs have been used successfully to predict survival in selected cardiology and oncology scenarios; however, their use in ophthalmology has been limited. These findings support the hypothesis that the RSF method is a promising alternative to standard analytic methods in ophthalmology. (Also see related commentary by Joelle A. Hallak, MS, PhD, in the same issue.)
The original article can be found here.