Graft function as determined by eGFR is commonly selected as a study outcome because of its connection to graft failure. The problem is that the eGFR equations themselves have limitations particularly in kidney transplant patients (25). Estimated GFR can also fluctuate and be affected by multiple factors (e.g., hydration, calcineurin inhibitors, acute kidney injury etc.). In fact, a recent study showed that the trend in serum creatinine does not predict histologic features on a follow-up biopsy (26). Considering these issues, a one-time eGFR measurement or change in GFR at two separate time points is not an acceptable outcome measure as too many factors can influence the result.

In contrast, determining the slope of eGFR decline is a more reliable indicator of the trajectory graft function compared to time point measurements. Estimated GFR trajectories have been incorporated in several models (27). In fact, a multicenter study of 91 patients found that a significant change in eGFR was seen within the first 12 months after active ABMR diagnosis that worsened by a factor of 0.757 ml/min/1.73 m^{2 per} month during a 12-month analysis. It was extrapolated that a 30% improvement in eGFR slope in the first 12 months was associated with a 10% improvement in death-censored graft failure at 5 years (28). This data supported the FDA approval of eGFR slope as a reasonable surrogate endpoint in kidney transplantation trials.

The Banff Classification of Allograft Pathology has undoubtedly been a major advance in transplantation (29). Use of the standardized Banff classification is the basis for patient inclusion and has been used for outcome assessment. Despite its strengths, this classification has been shown to be vulnerable to misinterpretation which can affect patient management (30). The main limitations include lack of reproducibility, comparability of Banff scores across constantly evolving versions, and oversimplification of a phenotypically complex and heterogeneous category of rejection.

The poor reproducibility of the Banff system among pathologists is a major issue (31, 32). A recent study revealed that even experienced pathologists lack consistency when interpreting Banff scores (33). The agreement between any two pathologists was poor. For instance, the agreement was between 44.8% and 65.7% for glomerulitis, 44.8%–67.2% for peritubular capillaritis, and 53.7–80.6% for transplant glomerulopathy. This variability in reading individual Banff lesional scores extends to diagnostic categories that include multiple lesion scores (33).

The Banff classification for ABMR has rapidly evolved. The downside of this rapid evolution is that it is sometimes hard to compare studies from different eras. The effect of a changing classification was clearly illustrated recently (34). Removing the “suspicious for ABMR” category in 2017 negatively impacted risk stratification (34). In response, the Banff 2022 classification has added new categories to account for microvascular inflammation (Banff glomerulitis and peritubular capillaritis scores ≥2) without C4d or HLA DSA positivity (8).

We are hopeful that future Banff iterations will leverage digital pathology, artificial intelligence and machine learning tools to permit us to navigate from a categorical classification to a more meaningful, consistent, reproducible, auditable, and continuous scoring system (35, 36). Finally, using novel diagnostic tools such as imaging mass cytometry (37) allows an unprecedented characterization of the immune composition with spatially enriched analysis, providing well-defined targets for personalized therapy.

Measurement of DSA has evolved rapidly. Initial tests were with nonspecific with insensitive CDC crossmatches that were either positive or negative, while modern tests of DSA are based on comprehensive high-resolution donor and recipient HLA typing combined with sensitive and specific SAB solid phase assays that output mean fluorescence intensity (MFI). As described earlier, there appears to be clear correlation with the quantity of DSA and outcomes, but there limitations in our current methods of DSA quantification that affect our ability to use DSA itself as a surrogate endpoint even if one confirms that HLA typing is accurate. Mean fluorescence intensity is a semi quantitative measure that corresponds to the amount of DSA that binds to the antigen on beads -not the true quantity of DSA. There is a maximum number of antigens on these beads. Bead saturation and interference can occur leading to inaccurate results particularly among patients with a high quantity of HLA antibody (38, 39). Performing these solid phase assays with diluted serum to determine antibody titer can provide a better quantitative assessment Figure 1. The neat MFI of multiple DSA specificities was compared to antibody titer and a substantial overlap in MFI was found over a breadth of antibody titer (38). Even if issues of determining DSA quantity are resolved, this will not resolve the challenges of interpreting multiple DSA specificities or varied DSA pathogenicity. Lab to lab variability and insufficient reproducibility further complicate DSA measurement (40). Ensuring accurate HLA typing, investigating the change in DSA with consistent assays, and the use of titers for antibody quantification may be more informative (41, 42).

Surrogate endpoints based on one marker (e.g., eGFR) may be insufficient. A composite endpoint comprised of multiple predictive factors may be more predictive of death censored graft failure. The iBox risk prediction tool utilizes a combination of variables including eGFR, proteinuria, HLA DSA, and histology information at 6–24 months to generate individual predictions at 3, 5, and 7 years post transplant (43). This novel tool was developed based on large international multicenter derivation and validation cohorts with more than 10,000 patients. Full and abbreviated (histology omited) iBox systems have been developed with c-statistics ranging from 0.70 to 0.93 in validation cohorts (44). Furthermore, the accuracy of the iBox risk score to predict death censored allograft failure was confirmed in post hoc analysis of data from randomized controlled trials in which the iBox scores appeared to change based on treatment (45). The qualification of iBOX as a reasonable likely surrogate endpoint is ongoing with the FDA (46) and the European Medicine Agency (EMA), has approved the iBox for use as a surrogate endpoint to demonstrate the superiority of a new immunosuppressive therapy compared to the standard of care from 6 to 24 months post transplant (44). While a surrogate endpoint comprised of multiple factors may be preferable to the use of a single factor, the limitations of the factors included in the scoring system remain relevant.

Molecular profiling of biopsies at the transcriptomic level is helpful because of its sensitivity, objectivity, and ability to detect early changes (47). As an added benefit, these tests shed light on the pathogenic mechanisms contributing to disease. Most of the published literature historically in the field of rejection concentrated on microarrays obtained on extra biopsy cores stored in RNA later Stabilization solution (Molecular Microscope MMDx One Lambda) (48–50). This revolutionary technology was difficult to apply to practice because of the need for an extra core of sample stored in RNA later. Recently other technologies have been developed using formalin-fixed paraffin-embedded specimens. One example is the Nano String nCounter system (Nano String Technologies, Seattle, WA) which is advantageous because a separate core is not needed, transcripts can be assessed on the same tissue examined visually, and large retrospective studies can be conducted (51–53). This technology is simpler to use and less expensive. While there is promise in these recent technologies, there can be a degree of overlap in the gene transcripts with varied histologic diagnoses (47).

Innovative single cell RNA sequencing and multiplex imaging for transcriptomic and spatial profiling of allograft tissue from patients experiencing different rejection phenotypes are also now available for investigational use. These revolutionary technologies can advance our understanding of the mechanisms of disease because they can help us understand heterogenous immune cell populations in ABMR (37). Spatial profiling has been performed using imaging mass cytometry (37), multiple iterative labeling by antibody neodeposition (50), and the GeoMX digital space profiling platform (51). The studies have highlighted the heterogeneity of the immune landscape and correlated these changes to the kidney microstructures. While innovative molecular technologies are promising and will provide insights on the mechanisms of disease, there is a need for validation before using as endpoints in clinical trials or practice.

Donor derived cell free DNA at varied levels can be detected in the peripheral blood of transplant recipients and is increasingly used as a surveillance biomarker of rejection (54). An increased percentage of donor derived cell free DNA correlates with allograft injury (54). The clear advantage of this test is that it is noninvasive. Some researchers have examined the change in the percentage of donor derived cell free DNA with treatment (55). The most obvious challenge with this donor derived cell free DNA is it is only a marker of injury and not specific for ABMR. It also provides little information about what histologic features to expect (e.g., degree of microvascular inflammation or chronicity). This technology is now rapidly expanding. The test is now offered by several companies and local labs will now be running these tests. Thus, there are important unanswered questions about the accuracy, precision, and reproducibility of these tests and laboratory proficiency is needed. This background is necessary before conclusions can be made about the clinical use and validity of these tests to inform.