Random Survival Forests Analysis Has Potential to Predict Keratoplasty Graft Success

TRUJILLO, PERU – APRIL 18: Corneal edema aphakia patient Jose, aged 87, undergoes surgery at the IRO (Regional Institute for Ophthalmology) on April 18, 2018 in Trujillo, Peru. Following a clinical screening day, Jose was chosen for a penetrating keratoplasty procedure (cornea transplant) during a programme run by Orbis, the ophthalmic training organisation. Founded in 1982 by ophthalmologist David Paton, Orbis trains eyecare teams across Africa, Asia and Latin America to improve the standard of eyecare in the region. As well as working in local hospitals, the charity also has a self-sufficient surgical unit on the Orbis Flying Eye Hospital, a converted McDonald-Douglas MD10 aircraft. (Photo by Leon Neal/Getty Images)
Researchers believe the technique can help evaluate which factors can predict if a graft will be successful.

A new analytic method — random survival forests — can help evaluate which factors can predict if a graft will be successful after Descemet stripping automated endothelial keratoplasty (DSAEK), according to findings published in JAMA Ophthalmology. This new approach provides an alternative to traditional analytic methods for identifying the variables that are most predictive of keratoplasty success.1

Since keratoplasty became widely available in the 1950s, several studies have investigated the donor, recipient, and operative factors associated with graft survival after penetrating keratoplasty and, more recently, endothelial keratoplasty, including DSAEK. The most notable of these studies, according to researchers, used standard Kaplan-Meier and Cox proportional regression statistical analytic methods. 

Random forests is an increasingly popular machine-learning technique designed to handle all types of data, including “big data,” according to the study. In contrast to the methods mentioned earlier, random survival forests require less-restrictive assumptions and can accommodate many types of predictors and interactions among them.

To test the applicability of random survival forests in ophthalmology, researchers reanalyzed the types of intraoperative complications associated with DSAEK graft failure found in prior publications using random survival forests.2 This study included 1090 participants (663 females [60.8%]; median age, 70 years [range, 42 to 90 years]), representing 1330 eyes. Random survival forests ranked a DSAEK intraoperative complication as the third-most predictive factor of graft failure, after surgeon and eye bank. In the first 47 months after DSAEK, the estimated mean difference in restricted mean survival time for grafts that experienced a DSAEK intraoperative complication vs those that did not was -227 days (99%CI, -352 to -70 days) based on the final random survival forests model.

Researchers note several limitations of their study, including that researchers couldn’t model individual intraoperative complications associated with DSAEK because they were too rare when viewed individually. Additionally, although the sample size was large for a randomized clinical trial, random forests require more events per predictor variable than standard methods.

Disclosure: Several study authors declared affiliations with the biotech and pharmaceutical industries. Please see the original reference for a full list of authors’ disclosures.


1. O’Brien RC, Ishwaran H, Szczotka-Flynn LB, et al. Random survival forests analysis of intraoperative complications as predictors of descemet stripping automated endothelial keratoplasty graft failure in the cornea preservation time study. JAMA Ophthalmol. Published online December 23, 2020. doi:10.1001/jamaophthalmol.2020.5743

2. Lass J, Szczotka-Flynn L, Ayala A, et al. Cornea preservation time study: methods and potential impact on the cornea donor pool in the United States. Cornea. 2015;34(6):601-8. doi:10.1097/ICO.0000000000000417.