Factors associated with graft survival and endothelial cell density after Descemet’s stripping automated endothelial keratoplasty

Postoperative endothelial cell loss leads to graft failure after corneal transplantation, and is one of the important issues for long-term prognosis. The objective of this study was to identify clinical factors affecting graft survival and postoperative endothelial cell density (ECD) after Descemet’s stripping automated endothelial keratoplasty (DSAEK). A total of 198 consecutive Japanese patients (225 eyes) who underwent DSAEK were analysed using Cox proportional hazard regression and multiple linear regression models. The candidate factors included recipient age; gender; diagnosis; pre-existing iris damage state, scored based on its severity; the number of previous intraocular surgeries; graft ECD; graft diameter; simultaneous cataract surgery; surgeons experience; intraoperative iris damage; postoperative rebubbling; and graft rejection. Eyes with higher pre-existing iris damage score and more number of previous intraocular surgery had a significantly higher risk of graft failure (HR = 8.53; P < 0.0001, and HR = 2.66; P = 0.026, respectively). Higher pre-existing iris damage score, lower graft ECD, and smaller graft diameter were identified as significant predisposing factors for lower postoperative ECD. The results show that iris damage status before DSAEK may be clinically useful in predicting the postoperative course. Avoiding intraoperative iris damage, especially in eyes with low ECD can change the prognosis of future DSAEK.


Predictors of graft failure in all patients.
Predictors of graft failure in uncomplicated patients. We postulated that pre-existing iris damage may be associated with a history of trabeculectomy and PKP. Furthermore, potential influential factors, such as the presence of diabetes mellitus (DM), graft rejection, postoperative rebubbling and uveitis could affect the above results and cause confounding bias. To completely exclude the influence of these factors on graft failure and postoperative endothelial cell density (ECD), the subgroup was defined as the patients who underwent DSAEK due to laser-iridotomy-related bullous keratopathy (LI-BK), pseudophakic/aphakic bullous keratopathy (PBK/ABK), Fuchs' endothelial corneal dystrophy (FECD), the three major indications for DSAEK, excluding the patients with DM, intraoperative iris damage, uveitis, postoperative rebubbling, and rejection. Supplementary Table S1 shows the demographics of the subgroup. Univariate and multifactorial analysis showed that pre-existing iris damage scores of 3 or 4 were significantly associated with graft failure (Supplementary Table S2, HR, 13.6; 95% CI, 1.54-119; P = 0.019). Stepwise multiple linear regression analysis showed that the predisposing factors that had significant correlations with postoperative ECD were pre-existing iris damage scores at 3, 6, and 12 months, the recipients' age at surgery at 3 and 24 months, graft ECD at 6, 12, and 24 months, and the graft diameter at 6 and 12 months (Supplementary Table S3).
Influence of Iris Damage Score on Graft Survival. Because the iris damage score was the most potent influential factor for postoperative endothelial cell loss, we stratified the subjects based on iris damage scores in different aetiologies. In total subjects (a), eyes with LI-BK (c), PBK (d), and other aetiologies (f), the survival rates in eyes with iris damage scores of 3 or 4 were significantly worse than those of eyes with either no iris damage or iris damage scores of 1-2 ( Fig. 2; P < 0.0001, P = 0.037, P = 0.0337, and P = 0.0029, respectively).

Influence of Iris Damage Score on ECD Loss.
To evaluate the effect of iris damage scores on the time course decrease in postoperative ECD after DSAEK, we used an exponential model 16 , classifying the subjects based on the iris damage score (Fig. 3). The eyes with iris damage score of 3-4 showed a rapid decrease in ECD, compared with the groups with iris damage score of 0 and 1-2.

Discussion
The results demonstrated that graft failure was associated with pre-existing iris damage score, and that ECD after DSAEK was correlated with preoperative graft ECD, graft diameter, and pre-existing iris damage of recipients, suggesting that endothelial cell loss after DSAEK is influenced by both donor and recipient factors. Although the overall clinical outcomes including visual acuity, astigmatism, and ammetropia after DSAEK are favourable compared to PKP, the survival rate has been reported to be 80-95% and endothelial cell loss is   [22][23][24] . However, we experienced rapid endothelial cell loss in a small proportion of patients after DSAEK, even when the procedure was performed in the same way with no intra-and postoperative complications. Factors related to endothelial cell loss after PKP have been well documented [18][19][20][21] . However, to the best of our knowledge, the clinical factors for endothelial cell loss after DSAEK is poorly understood. We hypothesized that some pre-disposing factors must exist in donors, surgical methods, postoperative complications, or the recipients' eyes. Under normal conditions, the adult human cornea loses endothelial cells at a rate of 0.6% per year 25 . After cataract surgery, the annual rate of endothelial cell loss is 2.5% per year between 1 and 10 years after surgery 26 . After PKP with no complications, endothelial cells decrease at a rate of 2.6-7.8% per year [27][28][29] , and both donor and recipient factors associated with postoperative ECD have been identified in eyes after PKP. Donor factors, younger age, larger preoperative graft ECD, larger donor graft diameter, and female donors are factors that correlate with larger postoperative ECD 15,17,19 . As recipient factors for endothelial cell loss after PKP, several studies have reported that eyes with anterior chamber IOL (AC-IOL), glaucoma, or ocular hypertension have lower ECD and higher failure rates 18  In the current study, we identified pre-existing iris damage as a recipient factor associated with less ECD after DSAEK. Moreover, the higher the pre-existing iris damage score, the higher the incidence of graft failure. Although the association between iris damage and ECD is a novel result, some clinical entities with iris pigment atrophy, such as iridocorneal endothelial syndrome, have progressive endothelial cell loss 33,34 . The presence of pre-and intraoperative iris damage can cause prolonged postoperative intraocular inflammation, due to the breakdown of the blood-aqueous barrier, which may lead to endothelial cell loss, because previous studies have implied a correlation between ocular inflammation and endothelial cell loss 35 . Severe anterior segment inflammation is related to iris atrophy 36 . Cytokines, interleukin-1, tumor necrosis factor, and vascular endothelial growth factor are elevated in the aqueous humor of the eyes with bullous keratopathy (BK) 37 and in eyes after In all subjects (a) and the subjects with LI-BK (c) PBK (d) and other aetiologies (e) there were significant differences in survival rates among the subjects with no iris damage or low iris damage score (1-2) and those with high iris damage scores (3)(4). In eyes with FECD, no patient had iris damage. Others included failed graft after PKP (12 eyes) and failed DSAEK (10 eyes), chronic uveitis (6 eyes), birth injury (5 eyes), and other causes such as endotheliitis.
trabeculectomy 38 , which is known as one of the risk factors for graft failure after DSAEK. Severe corneal or anterior chamber inflammation can result in concomitant endothelial cell loss and BK [39][40][41] . It is noteworthy that pre-existing iris damage was associated with decreased ECD and higher graft failure, which indicates that patients with pre-existing iris damage should be advised about higher risk for graft failure before DSAEK. Specific mechanisms have to be substantiated by comprehensive studies, to determine the sequence of events that take place in eyes with iris damage during and after DSAEK with regard to the apoptotic factors in the aqueous humor and inflammatory reactions in endothelial cells. It is tempting to speculate that the iris epithelium may secrete some factors that maintain endothelial cell survival.
Other factors, such as the number of previous surgeries and the history of TS-IOL and LI should be evaluated as the factors for ECD loss after DSAEK. The number of previous intraocular surgeries was associated with a higher incidence of graft failure in the current study, not with postoperative ECD. Furthermore, there was a significant correlation between iris damage scores and the number of previous intraocular surgeries (R = 0.262, P < 0.001). Therefore, iris damage during previous intraocular surgery can worsen the prognosis of future DSAEK, suggesting that iris damage should be avoided during intraocular surgery, especially in eyes with low ECD. The lens status is a well-known influential factor for ECD after PKP; the prognosis is poor in eyes with AC-IOL [18][19][20][21] . In the current study, because we performed AC-IOL removal with TS-IOL in all eyes with AC-IOL bullous keratopathy prior to DSAEK, we did not include DSAEK cases with AC-IOL. We previously evaluated the outcome of DSAEK in eyes with TS-IOL, comparing with solitary DSAEK 42 . Although there was no statistical difference in the ECDs between the groups up to 24 months after DSAEK, we found rapid decreases in postoperative ECD in several eyes with severe iris damage in the DSAEK group after TS-IOL. Regarding the presence of LI, we previously reported that the prognosis of DSAEK in eyes with LI is comparable with that of FECD and PBK 43 . In the current study, the presence of LI was not associated with the graft survival and postoperative ECD. Although the prognosis of DSAEK in eyes after some types of glaucoma surgeries, such as trabeculectomy, is not favourable, the results of this study suggest that the presence of LI does not affect graft survival and endothelial cell loss.
We demonstrated that larger graft diameter and more preoperative graft ECD were associated with more ECD after DSAEK. Romano et al. 44 reported that larger graft diameter and higher donor ECD were significantly associated with a reduced graft failure rate. Infant donors with a preoperative graft ECD of more than 4000 cell/ mm 2 were found to be preferable for higher ECD after DSAEK 45,46 . The results of the current study are consistent with these reports, although there is still a controversy on the association between ECD and graft diameter 47 . It is natural that a larger graft diameter is associated with higher ECD after DSAEK, because ECD is higher in the peripheral cornea than at the center of the cornea, and because a graft diameter of 8.5 mm is larger, by 28%, than a graft diameter of 7.5 mm, and is therefore theoretically expected to provide more endothelial cells. However, in eyes after PKP, a larger graft leads to a higher incidence of graft rejection. The more antigens that are transplanted, the higher the incidence of graft rejection, as has been reported in the comparison of graft rejection among PKP, DSAEK and Descemet's membrane endothelial keratoplasty 48 . In the future, studies on the incidence of graft rejection in eyes with different graft diameters will be important in confirming the safety of larger DSAEK grafts. Our study had several limitations. First, we excluded the history of trabeculectomy (9 eyes, 3.9%) from the baseline factors of multifactorial analysis because there were too few cases after trabeculectomy. Post-trabeculectomy has been reported to be associated with graft failure after DSAEK 49,50 . The univariate analysis using the Cox proportional analysis showed that the hazard ratio of post-trabeculectomy for graft failure was 3.23 (95% CI = 1.14-9.09; P = 0.027), which is consistent with previous studies. Furthermore, Inoue et al. reported that cytokine levels, such as monocyte chemotactic protein -1, in the aqueous humor were elevated in eyes with trabeculectomy 38 . Future studies have to be conducted to elucidate the associations between the cytokine levels in the aqueous humor, iris damage score and postoperative endothelial cell loss after DSAEK. Second, the decreased ECD can be influenced by the experience of the surgeon. In the current study, the involvement of an experienced surgeon was correlated with graft survival, but not with postoperative ECD.
In conclusion, the results of the current study suggest that a larger graft diameter is important in maintaining higher postoperative ECD, thereby providing more endothelial cells. Postoperative endothelial cell loss and graft failure were associated with higher pre-existing iris damage scores as recipient factors. Pre-existing iris damage scores can be useful in predicting the prognosis of DSAEK in patients with BK.

Methods
This retrospective cohort study was conducted in accordance with the Declaration of Helsinki, and was approved by the Ethics Review Board of Tokyo Dental College Ichikawa General Hospital (Acceptance No. I-15-39). Informed consent was obtained from all the subjects.
Surgical technique. DSAEK surgery was performed using double-glide technique 51 . After retrobulbar anesthesia with injection of 2% lidocaine, a 5.0-mm temporal corneoscleral incision was made. An AC maintenance cannula was inserted through the 2 or 10 o' clock paracentesis, and Descemet stripping was performed with a reverse-bent Sinsky hook (Asico, Westmont, IL, USA). The recipient's endothelium and Descemet's membrane were carefully removed using forceps. Pre-cut donor grafts were trephinated at a diameter of 7.0-8.5 mm, and the endothelial surface of the donor lenticle was coated with a small amount of viscoelastic material. Donor tissue was gently inserted into the anterior chamber using a Busin glide (Asico) and Shimazaki DSAEK forceps (Inami, Tokyo, Japan). Air was carefully injected into the anterior chamber to unfold the graft. The fluid between the recipient's stroma and the graft was drained from small incisions in the midperipheral recipient cornea. At 10 min after air injection, half of the air was replaced by balanced salt solution (BSS, Alcon, Fort Worth, TX, USA). At the end of the surgery, 2 mg subconjunctival betamethasone was administered. In patients with significant lens opacity (82 eyes), standard phacoemulsification and aspiration were performed using the phaco-chop technique with implantation of an IOL, followed by the DSAEK procedure. Postoperative medications included levofloxacin (Cravit, Santen, Osaka, Japan) and 0.1% betamethasone sodium phosphate (Sanbetazon, Santen), which were prescribed five times per day for 3-6 months. Topical 0.1% fluorometholone was prescribed two times per day after cessation of topical betamethasone. Data Analysis. Outcome measures included best-corrected visual acuity (BCVA), postoperative ECD and graft survival rate. The standard Landolt optotype chart was used to measure BCVA, which was converted into the logMAR and analyzed statistically. ECD was measured preoperatively and at 1, 3, 6, 12, and 24 months after surgery using a non-contact specular microscope (Noncon Robo SP-8000, Konan, Hyogo, Japan). Approximately 50 cells were analyzed for mean cell density. Graft failure was defined as persistent corneal edema resulting in irreversible loss of optical clarity due to the decrease in ECD.
To identify predictive parameters that are associated with postoperative ECD and graft failure, we selected the following variables, that were a priori based on the past studies and our knowledge of ECD and DSAEK; recipient gender, age, history of DM, aetiologies of BK, the number of previous intraocular surgeries, IOL position [capsular bag or TS-IOL] and pre-existing iris damage score were selected as the recipient factors. Donor age, imported or domestic graft, graft ECD and graft diameter were selected as the donor factors. Simultaneous cataract surgery, the surgeon's experience, intraoperative iris damage, postoperative rebubbling and rejection episodes were selected and analyzed as the intra-and postoperative factors.
Two masked observers (NI and TY) determined the iris damage score, based on its severity from preoperative slit-lamp microscopy findings (Fig. 1) as follows: 0, no iris damage; 1, iris damage limited to only one quadrant; 2, iris damage in two quadrants; 3, iris damage in three quadrants; and 4, iris damage in four quadrants. In Asian eyes with brown pigmentation, the iris damage was easily identified as the depigmented area using the slit-lamp microscopy. In all eyes, pre-existing iris damage scores based on the preoperative slit-lamp microscopy were confirmed using infrared camera photograph of swept source anterior segment optical coherence tomography (AS-OCT, CASIA, SS-1000; Tomey, Nagoya, Japan), because it was difficult to score precisely in some eyes due to the severe corneal edema before DSAEK. If the iris damage score increased after DSAEK, we regarded it as intraoperative iris damage, which was one of the other independent candidate factors for graft failure and postoperative ECD. LI was regarded as a factor independent from iris damage score, because not all patients who underwent LI developed BK and because we had sufficient numbers of patients who underwent LI (81 eyes); we were able to analyse them as an independent factor for Cox proportional hazard regression analyses. An experienced surgeon was defined as a surgeon who performed more than 30 DSAEK procedures. Factors, such as gender, presence of DM, graft rejection, an experienced surgeon, and postoperative rebubbling were dichotomized for Cox proportional hazards regression. Continuous variables including recipient age, donor age, and graft ECD were also dichotomized for use as categorical variables for Cox proportional hazards regression. Statistical analysis. Data were analyzed using the SPSS software (ver. 23; SPSS, Inc, Chicago, IL). To identify predisposing factors associated with graft failure, Cox proportional hazards regression was first used (univariate analysis and multifactorial analysis). Univariate analysis was performed to determine the correlation between graft failure and each variable. In multifactorial analysis, all of the variables were included to evaluate the relationship with graft failure. To exclude the influence of factors such as post-trabeculectomy and other ocular complicated pathologies, we focused on three major indications for DSAEK in Japan; LI-BK, PBK and FECD. Eyes with graft rejection, intraoperative iris damage, the presence of DM, a history of trabeculectomy, and uveitis were excluded from this multifactorial analysis. We excluded DM from the baseline factors, because DM is one of the factors underlying endothelial cell loss 52 . The following DM conditions varied among the subjects; well-controlled or poorly-controlled, insulin dependent or insulin independent, with or without ocular pathology. The remaining 102 eyes were enrolled as a subgroup of uncomplicated DSAEK cases; we conducted the same statistical analysis to evaluate the influence of variables on graft failure using Cox proportional hazard regression (univariate and multifactorial analysis).
To assess the association between postoperative ECD and all the variables, univariate analysis was conducted using Spearman's rank correlations for each variable at 1, 3, 6, 12 and 24 months postoperatively. Multiple linear regression was also conducted using backward stepwise analysis. A decay model was evaluated that fitted the two exponentials to each set of postoperative ECDs in the groups with iris damage scores of 0, 1-2 and 3-4 as reported previously 16 . ECD = p•exp(-at) + q•exp(-bt), where t is time, p and q are constants whose sum equals to the preoperative graft ECD, and a and b are exponential rate constants.
To test the influence of iris damage score for different indications, we stratified the groups based on the indications for DSAEK and compared the graft survival rate among the different iris damage scores. The Kaplan Meier method was used to analyze the graft survival rate, and the log rank (Mantel Cox) test was used to evaluate P values. A P value < 0.05 was considered statistically significant. Data were expressed as means ± standard deviation (SD).