In the present study, we validated our previously developed prognostic model for ECD on an independent cohort of 2999 donor corneas explanted at our institution between 2017 and 2021. Despite the increasing number of tissue donors, the growing demand for corneal grafts not only in Germany, but worldwide still cannot be fully met. Due to the strict guidelines for cornea donation, not every potential donor is eligible.
25 Many potential donors are already out of question before tissue collection because of their medical history, infectious pre-existing conditions, cause of death, DEI, or lack of consent. Furthermore, about one third of the harvested corneas are not suitable for transplantation. The most common reason for the discard is a low ECD.
5–7,26
The work of an eye bank presents numerous challenges on a daily basis. One important unpredictable factor that can strongly influence the work of eye banks is the daily fluctuating number of suitable tissue donors. This challenge can push eye banks, even those with a well-developed infrastructure, such as the Eye Bank Rhineland-Palatinate, to their limits. Although the aspiration is to harvest every cornea donated so as not to waste valuable tissue, there are unfavorable scenarios where the number of donations exceeds the staff capacity, area to be covered, and time resources of our eye bank. In order not to exceed the DEI, sometimes multiple simultaneous corneal explantations are necessary at widely separated sites with limited staff. In addition, work on weekends and holidays is limited because the number of staff in the eye bank is kept to a minimum. This leads to critical situations where the staff must make a considered decision like, for example, which corneal explantation is “more promising” and worth a trip at the expense of maybe one or two other potential tissue donors. In these tricky situations, a decision must be made between possible donors even before tissue explantation. The fact that the most common reason for corneal exclusion is a low ECD,
5–7 the donor with the highest possible ECD should be chosen to obtain the best condition for its upcoming use.
To facilitate and resolve such dilemmas, as described above, we developed a predictive model for the expected ECD from easily accessible predefined potential predictors, such as age, gender, DEI, and lens status.
22 To the best of our knowledge, this is the first investigation to validate such a predictive model on a large cohort of almost 3000 harvested corneas. The results show that the predicted ECD by the model correlates with the actual ECD (correlation coefficient = 0.411;
P < 0.01). We have shown that this model, in fact, is an acceptable predictor of an actual ECD of over 2000 cells/mm
2 as this number is considered to be the threshold for cornea transplantation involving the endothelium. The median ECD predicted with the formula was 2061 cells/mm
2 (IQR = 1834 to 2221) compared to the median measured ECD of 2377 cells/mm
2 (IQR = 1907 to 2624). A tendency to underestimate the ECD compared to the actual measured ECD is more likely to be beneficial for the upcoming corneal transplant and overall graft survival than the other way around. The results of our ROC curve suggest that our predictive model predicts an ECD > 2000 cells/mm
2 with an AUC of 73% (see
Fig. 2). To evaluate the clinical utility of the expected ECD as a screening tool to best manage reprocessing capacities of cornea donor banks when it is limited, we performed a decision curve analysis (see
Fig. 3). In clinical settings with a high reprocessing capacity, a low-risk threshold of, for example, 30% for an ECD > 2000 may be sufficient for tissue explantation. In this case, the approach to examine all corneas (see the blue line in the decision curve analysis) has an identical net benefit compared to the ECD predictive model. But in a clinical setting that requires a higher risk threshold for a transplantable donor cornea of, for example, 45% for an ECD > 2000 and above, for example, due to limited processing capacity, our prognostic model shows a higher net benefit (see the red line in the decision curve analysis; see
Fig. 3).
The potential implementation of this new prediction model into a cornea bank’s donor selection routine brings forth a crucial consideration regarding the rate of false positive results and false negative results depending on the chosen threshold probability. False positive results, where the model incorrectly identifies a cornea as transplantable when it is not, could lead to unnecessary allocation of resources. On the other hand, false negative results, where the model incorrectly identifies a transplantable cornea as non-transplantable, may lead to underutilization of donor tissue. The optimal balance between false positive results and false negative results must be chosen based on the available resources of the eye bank, the availability of donors, and the demand for donor tissue. The rate of false positive results and false negative results for different threshold probabilities in our cohort is shown in
Supplementary Table S1. Continued validation of the model through rigorous testing and application in practice are essential steps to balance the occurrence of false positive results and false negative results in different clinical settings and ultimately maximize the utility of the prediction model.
The prognostic model is only used at our Eye Bank Rhineland-Palatinate if circumstances force us to do so. The ultimate goal remains to explant every donated cornea and to examine it for further use. This means that this model does not replace our daily decision making in our eye bank, but is a useful tool in critical situations. It can be used not only for our approach, but also as a preselective decision tool in other countries, for example, emerging or developing countries with a different infrastructure, a larger area to be covered, and a different eye bank organization with its own limited human and logistical resources. In order to apply the model prior to explantation, we would suggest an estimated DEI using the time of death, the distance to the hospital, and the expected time for the recovery. From the economic point of view, preselection may also be more cost-effective by filtering out corneas with an ECD < 2000. A limitation of our study at the time of explantation is the lens status (phakic, aphakic, or pseudophakic) for the application of the model, as it is not always known by each eye bank, but is mandatory for the utility of our predictive model. Therefore, we recommend adding a question about lens status to the standard questionnaire used for donor qualification. Another limitation of our study is that the influence of death to cooling time and its impact on DEI limited to 72 hours in Germany was not assessed in either our model or our study. To meet the growing demand for corneal tissue, it is not sufficient to focus only on an increasing number of tissue donors. Depending on the country, there are differences in the country-specific guidelines for tissue donation. For example, there are no standardized upper or lower age limits worldwide. This might be because unlike other organ transplants, corneal transplants with its immune privilege appear to be less influenced by the age of the donor, as most studies have found no effect of donor age in the first years after transplantation (group 2008). Therefore, different age limits in different countries can affect the model. In scenarios with multiple potential tissue explantations at different locations over a wide area that cannot be covered with limited staff, our preselection model for an ECD > 2000 with easily accessible potential predictors can be a suitable solution to make a solid decision for this challenge. This model has shown to have predictive power for an ECD > 2000 after validation of nearly 3000 donor corneas.
Limitations to the study include that the predictive model was developed based on data from the Eye Bank Rhineland-Palatinate and compared with a new data set from the same Eye Bank. Due to different guidelines for corneal donations worldwide, for example, with regard to DEI or age limit, influences, and restrictions on the utility of our predictive model are possible. To further strengthen the predictive power of the prognostic ECD model and to adapt our model to a wider range, the application of data from other eye banks might be meaningful.