Endothelial Morphometric Parameters to Predict Endothelial Graft Failure after Penetrating Keratoplasty

Abstract

Objective

To determine whether preoperative and/or postoperative central endothelial cell morphometric parameters [cell density (ECD), coefficient of variation (CV) and % hexagonality (HEX)] and their change postoperatively are predictive of graft failure caused by endothelial decompensation following penetrating keratoplasty (PKP) to treat a moderate-risk condition, principally Fuchs’ dystrophy or pseudophakic corneal edema.

Methods

In a subset of Cornea Donor Study participants, a central reading center determined preoperative and postoperative ECD, CV, and HEX from available central endothelial specular images from 18 subjects whose grafts failed because of endothelial decompensation and 54 subjects matched for most donor and recipient parameters at baseline whose grafts did not fail.

Results

Preoperative ECD was not associated with graft failure (P=0.43); however, a lower ECD at 6 months was predictive of subsequent failure (P=0.004). CV at 6 months was not associated with graft failure in univariate (P=0.91) or multivariate analyses (P=0.79). There was a suggestive trend of higher graft failure with lower HEX values at 6 months (P=0.02); but not at the established statistical significance (P<0.01). The most recent CV or HEX values, as time-dependent variables, were not associated with graft failure (P=0.26 and 0.81, respectively). ECD values decreased during followup, while CV and HEX appear to fluctuate without an apparent trend.

Conclusions

The ECD at 6 months following penetrating keratoplasty was predictive of graft failure, while CV and HEX appear to fluctuate postoperatively possibly indicating an unstable endothelial population in both clear and failing grafts.

Introduction

The Cornea Donor Study (CDS)¹ has enhanced our understanding of many factors surrounding the success of penetrating keratoplasty (PKP) for conditions with endothelial dysfunction, notably Fuchs’ dystrophy and pseudophakic/ aphakic corneal edema. The CDS has shown that donor age up to 75 years¹, other donor factors (method of retrieval, processing factors, timing of donor cornea usage)², and ABO incompatibility³ had no effect on graft survival 5 years postoperatively. On the other hand, the risk for graft failure was approximately four fold higher in eyes with pseudophakic/aphakic corneal edema than in eyes with Fuchs’ dystrophy, whether pseudophakic or not; prior glaucoma surgery and/or preoperative glaucoma medication use also substantially increased the graft failure rate.⁴

Endothelial cell density (ECD) also was assessed in a subset of subjects in an ancillary study of the CDS, the Specular Microscopy Ancillary Study (SMAS)^{5, 6} to examine its relationship with these same donor and recipient factors. A slight association between increasing donor age and greater post-PKP corneal endothelial cell loss by 5 years was found among eyes whose grafts were successful at five years.^{6, 7} Female donor and larger graft size were associated with less cell loss, while baseline ECD and other donor and recipient factors had no influence on ECD by 5 years.⁷ Interestingly while baseline ECD was not predictive of graft failure, the six month ECD was.^{7, 8}

The literature has suggested that changes in the morphometric parameter coefficient of variation (CV), reflecting variation in cell size expressed by the calculated CV of cell areas (polymegathism); and calculated percentage of hexagonal cells or hexagonality (HEX), reflecting variation in cell shape (pleomorphism), may be more sensitive than ECD in assessing endothelial health and dysfunction.^9-14 Although the original SMAS was designed to examine ECD solely determined by a variable frame analysis method accommodating all types of image quality, study design and morphometric analysis methods were developed from existing donor and postoperative images to explore the question of whether changes in morphometric parameters over time are predictive of graft failure following PKP for endothelial dysfunction conditions. If these parameters are more sensitive than ECD in predicting changes in endothelial function following keratoplasty and in identifying possible endothelial dysfunction, monitoring morphometric parameters could be more important than monitoring of ECD alone.

Methods

Specular Microscopy Ancillary Study

Of the 1090 eligible subjects participating in the CDS, the optional SMAS included 596 subjects who participated at 45 CDS clinical sites. Donor corneas were assigned to the SMAS subjects by 31 of 43 participating CDS eye banks but not all of these eye banks participated in the SMAS, which was also optional for the eye banks. In the SMAS, ECD was determined using the HAI variable frame method.¹⁵

Study Cohort

This morphometric study was designed as a case-comparison group study and included the baseline donor and the follow up endothelial images collected after PKP from 72 of these 596 SMAS participants. The primary criterion for inclusion as a case in the study was (1) graft failure due to endothelial decompensation without prior intraocular surgery, trauma, or graft rejection that could have adversely affected the endothelial cells and (2) availability of a specular image within 1.5 years prior to the date of failure.

In addition to the eighteen subjects with failed grafts who met the above criteria, a comparison group was selected that included subjects without graft failure. This comparison group included three subjects matched by certain baseline characteristics to each of the 18 graft failure cases (n=54). The matching variables included diagnosis of Fuchs’ dystrophy or pseudophakic/aphakic corneal edema, donor age, and reading center-determined ECD. In addition, each subject in the comparison group was required to have at least as many months of follow up as the corresponding case's time to failure with an image that corresponded to the last image before failure of the corresponding case.

There were 69 baseline images (3 of the graft failure cases did not have an eye bank baseline image) and 270 follow-up images that underwent morphometric analysis to obtain assessments of ECD, CV (standard deviation of the mean cell area ÷ mean cell area) and HEX. Of the 339 images available for image analysis, 326 gradable images were included in the statistical analysis (68 baseline and 258 follow-up images).

Morphometric Image Analysis

The Case Western Reserve University Cornea Image Analysis Reading Center (Cleveland, OH) used both the corners method and the center method for analyses. The corners method is the gold standard method for determining CV and HEX.^{10, 14, 16} In order for a cell to be gradeable, the reader must be able to determine the location of each of the cell's corners. For our study, at least 5 gradeable cells per image were required for the image to be graded. This minimum number was chosen to accommodate postoperative images at high magnification with ECDs below 800 cells/mm². Readers were required to grade every cell visible, where all cell corners could accurately be distinguished. Individual cells did not have to be contiguous to other cells. Mean cell area, CV and HEX were then calculated utilizing the HAI Laboratories CAS/CL 1.10 Cell Analysis System (Arlington, MA).

The center method was developed by Konan (Torrance, CA) in the 1990's and has become a standard technique for clinical morphometric analysis.^{11, 14, 17} In order for a cell to be gradeable, the reader must be able to determine the location of the center of the cell. Therefore, cells where the center could not be determined either due to indeterminable cell borders or when only a portion of the cell was visible in the frame were not graded. The largest area of contiguous cells in the image where the centers could accurately be distinguished was selected by the reader. Any cell that extended outside of the frame was not analyzable, and thus not marked. The center of all contiguous, analyzable cells was marked. Utilizing the Konan KSS400 software (Torrance, CA), mean cell area, ECD, CV, and HEX were calculated.

All images were graded by two readers and adjudicated by an experienced third reader based on defined limits for interobserver variability for number of cells analyzed per image, and ECD, CV and HEX. The adjudicator initially reviewed all image sets which exceeded the allowable difference in number of cells graded (≥20% difference in total number of cells/time point for images where the maximum number of cells counted by either reader was >50, and had a >30% difference where the maximum number of cells counted by either reader was <50) to determine whether one reader or both readers should re-grade the image set for that time point. The adjudicator flagged the image set for re-grade if there was sampling error or improper corner placement. A second adjudication step compared the ECD, CV and % HEX and flagged for adjudication all image sets where the ECD analyses varied by ≥5.0% between readers, the CV varied by > 15%, and/or the % HEX varied by > 15%. The adjudicator then completed an independent assessment for all images that were flagged for adjudication. A final review of all image assessments was performed, if the interobserver variability between the reader(s) and the adjudicator was still above those limits.

Morphometric Analysis of Preoperative Donor Images

The SMAS required only one donor image to be provided to the reading center for analysis. Due to the quality limitations of the donor images,⁵ the majority could not be analyzed by the corners method, thus the center method was chosen for analysis of the entire donor image set for this study.

Morphometric Analysis of Postoperative Images

Up to three images of the central endothelium were collected at 6, 12, 24, 36, 48, and 60 months postoperatively from those who participated in the SMAS. Since pilot studies (data not shown) determined the clinical images were of sufficient quality to undergo corners analysis, the corners method was used for the majority of images.

Images deemed unanalyzable using the corners method were analyzed using the center method. Adjudication for post-operative images analyzed employing the center method used the same approach described previously for the donor images. Images unanalyzable with the center method were rejected and excluded from analyses.

Statistical Methods

The data of the 72 participants in this study were censored at the first occurrence of either a confounding event (e.g., intraocular surgical procedure, ocular trauma, graft rejection) or the last visit date. The proportional hazards model was used to assess the association of graft failure and morphometric parameters (ECD, CV and HEX) preoperatively at baseline and at 6 months postoperatively. Models also were fit with the most recent morphometric value as a time-dependent variable. No significant deviation from the proportional hazards assumption was detected for these models.

A comparison of the 5 year morphometric parameters (CV and HEX values at 5 years and percent change from baseline to 5 years) between the two donor age groups (<66 and ≥66 years) was performed using analysis of covariance models. All models were fit with the rank normalized transformation (van der Waerden scores) of the morphometric variable (CV or HEX) and adjusted with the baseline values.

The relationship between the ECD obtained in the morphometric study using the HAI corners and Konan center methods and the ECD obtained in the SMAS study using the variable frame method was assessed by Spearman correlation coefficient. The ECD values obtained by the two different methods of image analysis were comparable for all clinical images (r range: 0.95 to >0.99). A greater variation between the two ECD values was noticed among the baseline images analyzed by the Konan center method in this study (r=0.77, 95% CI: 0.65 to 0.85).

All reported P-values are 2-sided. Because of multiple comparisons, P-values >0.01 were not considered statistically significant. Statistical analyses were conducted using SAS statistical software, version 9.2 (SAS Institute Inc., Cary, North Carolina).

Results

Subjects

The mean age (SD) at time of PKP for the 72 study participants was 71 (8) years; 64% were women, and 92% were white, non-Hispanic. Sixty-seven percent of subjects underwent PKP because of a diagnosis of pseudophakic/aphakic corneal edema; 94% were pseudophakic after the surgery (Table 1). The variables used for matching (baseline diagnosis, donor age and pre-operative ECD) were distributed similarly between the 18 subjects with graft failure and the 54 without failure (Table 1).

Table 1.

Baseline Characteristics of Analysis Cohort

Baseline Characteristic	Total N=72	Endothelial graft failure N=18	No graft failure N=54
Variables used for matching:
Recipient Diagnosis
Fuchs' Dystrophy	24 (33)	6 (33)	18 (33)
Pseudophakic/Aphakic Corneal Edema	48 (67)	12 (67)	36 (67)
Endothelial Cell Density* (cells/mm2)
Mean (Standard Deviation)	2584 (286)	2584 (350)	2584 (265)
Median (Interquartile Range)	2608 (2372, 2812)	2593 (2414, 2840)	2615 (2365, 2812)
1708-2299	11 (15)	2 (11)	9 (17)
2300-2499	16 (22)	5 (28)	11 (20)
2500-2699	19 (26)	5 (28)	14 (26)
2700-2899	15 (21)	3 (17)	12 (22)
2900-3099	11 (15)	3 (17)	8 (15)
Donor Age (years)
Mean (Standard Deviation)	64 (8)	64 (9)	63 (8)
Median (Interquartile Range)	65 (57, 71)	65 (55, 73)	64 (57, 70)
Other variables:
Recipient Age (years)
Mean (Standard Deviation)	71 (8)	73 (7)	71 (8)
Median (Interquartile Range)	73 (67, 77)	74 (72, 78)	73 (66, 77)
Recipient Gender
Female	46 (64)	13 (72)	33 (61)
Recipient Race
Caucasian	66 (92)	17 (94)	49 (91)
African-American	4 (6)	1 (6)	3 (6)
Hispanic	0	0	0
Asian	0	0	0
Other	2 (3)	0	2(4)
Recipient Cigarette Smoker	3 (4)	0	3 (6)
Recipient History of Glaucoma	13 (18)	3 (17)	10 (19)
Recipient Preoperative Lens Status
Phakic	15 (21)	2 (11)	13 (24)
Pseudophakic	55 (76)	16 (89)	39 (72)
Aphakic	2 ( 3)	0	2(4)
Recipient Postoperative Lens Status
Phakic	4 (6)	0	4(7)
Pseudophakic	68 (94)	18 (100)	50 (93)
Aphakic	0	0	0
Recipient Bed Size (mm)
Mean (Standard Deviation)	7.8 (0.3)	7.8 (0.3)	7.8 (0.3)

N (%) unless otherwise specified

^*Includes baseline ECD from eye bank for 3 eyes and from reading center for 69 eyes

Morphometric parameters and Graft Failure

Preoperative morphometric parameters were not associated with graft failure from endothelial decompensation (p-values for ECD, CV and HEX: 0.43, 0.91 and 0.86). Among 14 graft failure cases with baseline data available, the median ECD at baseline was 2526 cells/mm² (interquartile range, 2385, 2969), CV was 0.33 (0.29, 0.36) and 61% (54%, 65%) of the endothelial cells were hexagonal, and the corresponding values in the comparison group were 2664 cells/mm² (2418, 2905), 0.33 (0.31, 0.36) and 59% (55%, 64%) (Table 2).

Table 2.

Morphometric Parameters According to Graft Failure Status

Graft Failure Status	Endothelial Cell Density (cells/mm2)			Coefficient of Variation			Percent Hexagonality
	N	Median (Quartiles)	P value*	N	Median (Quartiles)	P value*	N	Median (Quartiles)	P value*
Baseline**
Graft Failure	14	2526 (2385, 2969)	0.43	14	0.33 (0.29, 0.36)	0.91	14	61% (54%, 65%)	0.86
Non-failure	54	2664 (2418, 2905)		54	0.33 (0.31, 0.36)		54	59% (55%, 64%)
6 Months
Subsequent Graft Failure	11	1752 (1234, 2558)	0.004	11	0.25 (0.23, 0.28)	0.91	11	57% (50%, 63%)	0.02
Non-failure	35	2394 (2005, 2649)		35	0.25 (0.23, 0.28)		35	63% (57%, 70%)
6 month relative change*** from baseline
Subsequent Graft Failure	10	–24% (–53%, –5%)	0.005	10	–26% (–39%, –18%)	0.44	10	–1% (–10%, +18%)	0.17
Non-failure	35	–10% (–28%, 0%)		35	–23% (–30%, –13%)		35	+11% (–2%, +16%)

^*P values are generated from univariate Cox models fitted with the continuous morphometric parameters at baseline and 6 months. The models with the 6 month parameters and the 6 month relative change are conditional on graft survival by 6 months.

^**One of the 69 baseline images available for morphometric analysis, was non-gradable.

^***Calculated as the difference between the morphometric values at 6 month and baseline divided by the baseline value. The relative change is expressed as percentage with negative sign indicating loss of cells (for endothelial cell density), decreased HEX or CV.

Figures 1, 2 and 3 illustrate the changes in the morphometric parameters over time in the graft failure and comparison groups. Unlike the ECD values that tended to decrease steadily over the study follow up, the CV and HEX values appeared to fluctuate without an apparent upward or downward trend. Six months after surgery the median ECD fell to 1752 cells/mm² (1234, 2558), the CV decreased to 0.25 (0.23, 0.28) and HEX to 57% (50%, 63%) in the graft failure group; the corresponding values in the comparison group were 2394 cells/mm² (2005, 2649), 0.25 (0.23, 0.28) and 63% (57%, 70%). In univariate analyses, the 6 month ECD values were associated with subsequent failure (P=0.004), whereas the 6 month CV values were not (P=0.91). There was a suggestive trend of higher graft failure with lower HEX values at 6 months (P=0.02); however, this association did not meet our criteria for statistical significance (P<0.01). When analyzed as time-dependent variables, the most recent CV or HEX values were not associated with subsequent graft failure (P=0.26, P=0.81, respectively).

Figure 1.

Box plot of Morphometric ECD over Time in Graft Failures Due to Endothelial Decompensation (n=18) and in Non-Failures (n=54)

Figure 2.

Box plot of Coefficient of Variation of Cell Area over Time in Graft Failures Due to Endothelial Decompensation (n=18) and in Non-Failures (n=54)

Figure 3.

Box plot of Percent Hexagonality over Time in Graft Failures Due to Endothelial Decompensation (n=18) and in Non-Failures (n=54)

All three morphometric parameters were included as model covariates in a multivariate analysis assessing factors associated with graft failure (Table 3). Similar results were generated from the multivariate models with morphometric parameters at baseline and at 6 months. The multivariate model with time-dependent covariates illustrated that CV (P=0.85) and HEX (P=0.98) did not contribute any additional predictive value for graft failure besides that of ECD alone (P<0.001).

Table 3.

Morphometric Parameters and Graft Failure Due to Endothelial Decompensation Proportional Hazards Regression Multivariate Analyses

Covariate	N	Hazard Ratio*	95% Confidence Interval	P-value
Model 1	68
Preoperative (baseline) VF-ECD		1.06	0.87 – 1.28	0.56
Preoperative (baseline) CV		1.10	0.19 – 6.43	0.92
Preoperative (baseline) HEX		0.94	0.33 – 2.70	0.91
Model 2**	46
VF-ECD at 6 months		1.20	1.04 – 1.38	0.01
CV at 6 months		0.79	0.13 – 4.66	0.79
HEX at 6 months		2.57	1.20– 5.50	0.02
Model 3***	72
Most recent VF-ECD (time dependent)		1.28	1.13 – 1.46	<0.001
Most recent CV (time dependent)		1.08	0.51 – 2.28	0.85
Most recent HEX (time dependent)		0.99	0.62 – 1.59	0.98

VF = variable frame; ECD = endothelial cell density; CV = coefficient of variation; HEX = percent hexagonal cells

^*A hazard ratio value > 1.0 denotes increased graft failure with lower VF-ECD values, higher CV values and lower HEX values. The hazard ratio is calculated per 100 cells/mm² change in ECD, 0.10 change in CV and 10% change in HEX. Morphometric values are treated as continuous variables in all models.

^**Cox model is conditional on graft survival at 6 months. Excluded subjects include graft failure prior to 6 months (N=1) or missing 6-month morphometric values (N=25).

^***Cox model fitted with all three morphometric parameters as time-dependent covariates. Most recent covariate is the last morphometric value prior to graft failure or last exam date.

Note: Results were similar when the morphometric ECD was used instead of variable frame ECD in all models presented in this table.

There was no association between donor age and CV and HEX values at baseline (Figure 4) or at 5 years. Among the 27 subjects with 5 year morphometric parameters (16 received corneas from donors <66 years old and 11 subjects received corneas from donors 66 years or older) the baseline the median (interquartile range) CV was 0.32 (0.30, 0.34) in the younger donor group and 0.36 (0.31, 0.37) in the older donor group (P=0.25), which decreased to 0.25 (0.19, 0.29) and 0.27(0.21, 0.37) at 5 years (P=0.78), respectively. The median HEX at 5 years increased to 64% (53%, 70%) in the younger donor group and to 63% (49%, 69%) in the older donor group (P=0.76) from 61% (56%, 66%) at baseline in both groups (P=0.67).

Figure 4.

Morphometric parameters (Coefficient of Variation and Percent Hexagonal Cells) at Baseline vs. Donor Age

Note: Plots of 5 year morphometric values versus donor age illustrated similar relationship.

Discussion

This study showed that preoperative ECD determined from a morphometric approach was not predictive of graft failure while the six month ECD was predictive of graft failure, results that are similar to those from the overall CDS specular microscopy analysis. The six-month HEX results were suggestive of a weak association with subsequent graft failure while CV was not predictive of graft failure, despite theoretical and thermodynamic hypotheses that a more stressed endothelium that could be at risk for progression to graft failure would demonstrate greater variation in cell area expressed in the CV and a lower percentage of hexagonal cells^{18, 19}. Both CV and HEX appeared to fluctuate over the five years without an apparent downward or upward trend. In a multivariate model, CV and HEX did not add predictive power beyond ECD at 6 months with regard to graft failure. To our knowledge, this is the first examination of the use of these morphometric parameters at baseline or at 6 months to predict graft failure with PKP. The simpler use of ECD at 6 months determined by variable frame analysis or morphometrically remains the best parameter to predict endothelial failure following PKP.

In a previous series of reports analyzing the performance of corneal preservation media, there were no major changes in CV or HEX up to one year following PKP for both endothelial dysfunction and keratoconus.^20-22 In a series of studies examining a cohort at 5, 10, 15, and 20 years with a declining number of clear grafts following PKP for endothelial dysfunction conditions and keratoconus^23-25, the authors noted: 1) no significant change in CV up to 3 years postoperatively at 0.26 and then a progressive increase to 0.33 by the 10^th year and then stable thereafter; 2) HEX starting at 68% fluctuated down and up during the first 5 years postoperatively and then progressively declined to 52% by the 20th; and 3) central corneal thickness progressively increased from 0.54 μm at 2 months to 0.59 μm in the 20^th year. The authors attributed these changes to the increasing number of late endothelial failures over this entire period. Notably, however, by 20 years only 22% of the clear grafts (41 of 388 grafts) were available for analysis and no imputation was made to relate the missing data from deaths, loss to follow-up and graft failures to the changes in CV and HEX.

While our study and others have not shown CV and HEX to be predictors of endothelial and graft failure, these parameters have been somewhat more successful in predicting endothelial recovery after intraocular procedures. Schultz et al showed the dynamic changes in CV and HEX following cataract surgery with recovery and restoration of normal corneal thickness three months postoperatively.²⁶ Several long term studies on phakic intraocular lenses have documented, over a 4 to 5 year period, significant changes in CV and HEX which are associated with cell loss, but changes were not correlated with ultimate corneal decompensation.^{27, 28} In one study, HEX showed greater change in individuals with severe diabetes 6 months following phacoemulsification compared with individuals without diabetes undergoing similar surgery, suggesting that these diabetic patients had more compromised endothelium.²⁹ In a separate study, this same group correlated an increase in corneal thickness and CV in patients with diabetes for more than 10 years compared with patients with diabetes for less than 10 years and similarly with a lower HEX with an increase in thickness.³⁰ The greater dynamic changes in cell morphology and loss of endothelial cells following PKP in both the grafts that survive and those that fail, unlike cataract surgery, appear to preclude the use of these parameters to predict graft failure.

A factor that reduces the utility of CV and HEX to predict graft failure following PKP and may have contributed to the fluctuation we observed in these parameters over time is the difficulty in achieving a highly reproducible measure of these two parameters given the compromised endothelial image quality and heterogeneous population of a relatively small sample of cells to analyze. This lack of reproducibility may not have been apparent in prior morphometric studies with PKP since all these studies employed a single trained reader.^{20-25, 31} However, in this study, the use of a dual-grading method with trained readers, highly successful for ECD determination, demonstrated greater interobserver variability for both CV and HEX than was reported in the prior study. To reduce this variability, we implemented a two step adjudication process. Much of this variability was due to differences in cell selection by the two readers which accentuated differences in the CV and HEX with 50 cells or less of varying size and shape to analyze. This variability is not surprising and has been previously observed in polymegethous corneas including PKPs, but not in normal corneas.^{17, 32-34} We believe that employing the corners method with dual grading and adjudication remains the best approach to measure CV and HEX, particularly in this population of polymegathous corneas, but only when image quality is good to excellent, as in normal corneas.

There remains interest in determining whether morphometric parameters (CV and HEX) could be more sensitive than ECD in detecting endothelial dysfunction following PKP. Although we believe that the apparent fluctuation in these parameters represented an unstable endothelial population in both clear and failing grafts, the fluctuation could also be attributed to measurement error. It should be noted that the cases (failures) included only those grafts that failed solely from endothelial decompensation without prior intraocular surgery, ocular trauma, or graft rejection that could have adversely affected the endothelial cells. Perhaps with a larger sample size, different study design, and greater inclusion of cases with graft failure from other causes, different results may have been found. More frequent than yearly endothelial imaging, given the continued dynamic changes in the endothelial population post PKP, also may reveal greater predictive value of changes in these morphometric parameters. Finally, studies also will be improved if there is increased emphasis on photographer training to yield better quality donor and postoperative images.

In summary, the study results show that ECD at 6 months was predictive of subsequent graft failure. CV and HEX fluctuated down and up over the 5 year follow up period and did not add to the predictive value of ECD.