replying to E. Giorgakis et al. Nature Communications https://doi.org/10.1038/s41467-024-53702-7 (2024)

We read with interest the letter by Giorgakis et al. concerning our recent study entitled “A Machine Learning-Driven Virtual Biopsy System for Kidney Transplant Patients”1. Based on a single-center study they performed, the authors commented on the practical application and implications of the Virtual Biopsy System (VBS) model, particularly in aggressive kidney transplant settings.

Context and purpose of the Virtual Biopsy System study

Our study primarily aimed to address the inefficiencies and complications associated with preimplantation biopsies for standard criteria donors2. The traditional biopsy process, while valuable, has several limitations including potential complications, misinterpretations, delays, and added costs3. These issues are particularly pronounced in settings with stringent organ acceptance criteria4. The VBS was developed to streamline the assessment process, reduce cold ischemia time which is highly associated with worsened graft outcomes5, and enhance the overall efficiency of kidney transplants by predicting the severity of chronic Banff lesions using routinely collected donor parameters.

Key findings and model performances

In their study, Giorgakis et al. indicated higher false positive rates for moderate/severe Banff lesions. However, it is crucial to highlight the single-center retrospective nature of this study, including transplanted kidneys only that have not been reassessed, which may have introduced selection bias and limit the generalizability of the findings, as compared to our international multicenter study comprising 17 centers and 14,032 biopsies of transplanted and discarded kidneys, of which a subset of the biopsies were reassessed by four experts transplant pathologists (LDC, MPA, MR, JPDVH).

In addition, it should be noted that the AUC of the VBS (0.72–0.81) remained within an acceptable range for ruling out severe lesions. We agree that for high-risk predictions, a preimplantation biopsy may remain essential to avoid unnecessary organ discards. Nevertheless, the primary goal of the VBS is to identify low-risk cases where a preimplantation biopsy can be safely omitted as depicted in the Fig. 1.

Fig. 1: The potential impact and utility of the Virtual Biopsy System.
Fig. 1: The potential impact and utility of the Virtual Biopsy System.
Full size image

This figure shows the intended use and impact of the Virtual Biopsy System on organ allocation process. The first part of the figure shows the current organ allocation process. The second part shows how this process would be accelerated with the Virtual Biopsy System, ultimately leading to better graft outcomes.

This strategy could decrease cold ischemia time and associated complications, and therefore contribute to fast-track organ allocation and improved transplant success rates (Fig. 1). Importantly, our study did not intend to undermine the value of biopsies in complex cases but to provide an additional tool for better risk stratification and decision-making.

Conclusion

In conclusion, while the letter by Giorgakis et al. highlight important considerations for the application of the VBS, their conclusions are limited by their study design. We encourage prospective multicenter randomized control trials to further validate the clinical application of the VBS.

Our goal remains to support the transplant community with innovative tools that improve patient outcomes and optimize the use of available organs. As such, we are working on the safe and effective implementation of the VBS in diverse clinical settings.