Principal Component Analysis of a Real-World Cohort of Descemet Stripping Automated Endothelial Keratoplasty and Descemet Membrane Endothelial Keratoplasty Cases: Demonstration of a Powerful Data-Mining Technique for Identifying Areas of Research

Abstract

Purpose: Principal component analysis (PCA) is a descriptive exploratory statistical technique that is widely used in complex fields for data mining. However, it is rarely used in ophthalmology. We explored its research potential with a large series of eyes that underwent 3 keratoplasty techniques: Descemet membrane endothelial keratoplasty (DMEK), conventional Descemet stripping automated endothelial keratoplasty (ConDSAEK), or ultrathin-DSAEK (UT-DSAEK). Methods: All consecutive DMEK/DSAEK cases conducted in 2016 to 2022 that had ≥24 months of follow-up were included. ConDSAEK and UT-DSAEK were defined as preoperative central graft thickness ≥130 and <130 μm, respectively. Seventy-six patient, disease, surgical practice, and temporal outcome variables were subjected to PCA, including preoperative anterior keratometry, the use of sulfur hexafluoride gas (SF6) versus air for primary tamponade, and postoperative best corrected visual acuity and endothelial cell density. Associations of interest that were revealed by PCA were assessed with the Welch t test or Pearson test. Results: A total of 331 eyes were treated with DMEK (n = 165), ConDSAEK (n = 95), or UT-DSAEK (n = 71). PCA showed that ConDSAEK and UT-DSAEK clustered closely, including regarding postoperative best corrected visual acuity, and were clearly distinct from DMEK. PCA and follow-up univariate analyses suggested that in DMEK, 1) flatter preoperative anterior keratometry (average, K1, and K2) associated with more rebubbling (P = 0.004–0.089) and graft detachment (P = 0.007–0.022); 2) graft marking did not affect postoperative ECD; and 3) lower postoperative endothelial cell density associated with SF6 use (all P > 0.001) and longer surgery (P = 0.005–0.091). All associations are currently under additional investigation in our hospital. Conclusions: PCA is a powerful technique that can rapidly reveal clinically relevant associations in complex ophthalmological datasets.