Will the ANCA Kidney Risk Score Become the Gold Standard for Prognostication in ANCA-Associated GN?

This is the authors' abstract. We don't add key points for this paper.

ANCA-associated vasculitides (AAVs) are multiorgan autoimmune disorders which if not treated quickly can lead to multiorgan damage or, sadly too often, death. Broadly, the more severe the presentation of ANCA-associated GN, the higher chance of morbidity or mortality.1 Its relapsing-remitting course has made it a challenge to predict outcomes. There have been many models proposed that split patients out by risk; these include the Berden Classification (n in original publication=100),2 Mayo Clinic Chronicity Score (MCCS, n=0),3 and ANCA Renal Risk Score (ARRS, n=115)4 and its recent successor, ANCA Kidney Risk Score (AKRiS, n=959).5 AKRiS is a revision to the ARRS using similar model building methods and a much larger, albeit retrospective, cohort. The MCCS and Berden scores have a larger focus of histopathology than Brix et al. and Bate et al. AKRiS uses creatinine, proportion of normal glomeruli, and grading of tubular atrophy and interstitial fibrosis to give a risk score and kidney survival probability. In this issue of Kidney360, Scurt et al.6 perform an external validation of the AKRiS. It is believed that is the first of its kind. The team behind this validation are completely independent of the Manchester-led team who created the AKRiS. The results are striking. The discrimination statistic is almost identical (C=0.831 in Bate and C=0.834 in Scurt) which shows that the original results can be replicated independently. Thirty-six–month kidney survival was marginally higher in the study by Scurt but with a smaller number of patients; it may still have been within the margin of error. Scurt et al. also suggests that anemic patients have worse outcomes after controlling for the AKRiS score. Scurt et al. show that AKRiS is a valid tool and can be used in the prognostication of patients. Whether AKRiS becomes the gold standard is highly anticipated. It is hoped that researchers from across the world will perform validation studies as part of their work to confirm, or indeed refute, the AKRiS's discriminative ability. Within the competing prognostication frameworks, there have been a dearth of large-scale external and independent validations, especially ones that compare models (exceptions include Kong et al.7 and Christodoulou et al.8). Although AAV is a relatively rare disease, it means that there is a wider need for intercenter, international, and intercontinental collaboration as individual centers will likely not have sufficient patient cohorts to have sufficient power. The three main models2,3,5 and Scurt et al.6 have all used predominantly western (European and North American) cohorts, and there is limited reliable work from Asia. In a review by Xia et al.,9 only three studies validating the ARRS, with 359 patients between them, were from outside of Europe and North America. There was a proposed model, the Nomogram, from Ni et al.10 last year that was from Chinese patients (n=201), but this has gained little traction so far. Ni et al. argue that myeloperoxidase antibody patients have different outcomes to proteinase 3 patients and hence that populations with much higher proportions of myeloperoxidase patients, for example, China, should have their own model for prognosis as the underlying antibody causes differing paths. Bate et al. argue the contrary with neither ARRS nor AKRiS using antibody as a parameter.5 There was significant similarity of parameters between the Nomogram and ARRS/AKRiS. Scurt et al.6 did not investigate the association of antibody with ESKD. In health systems with higher morbidity and mortality, these models may underestimate risk, but theoretically, proportionality of the risk should be retained. There is yet to be any evidence that the models cannot be used outside of the systems they were developed in, but with further validation studies, this will be confirmed. It should be warned that if models are tinkered with there may be confusion as which model is the standard. In some circumstances, marginal improvement to a model may not be desired lest there is one consistent standard. Updates are necessary to fit with improvements in treatments or to add parameters if there is clear evidence of benefit. These must be pivotal with large cohorts. For example, Breast PREDICT, a seminal model in prognosis for women with breast cancer, has been updated multiple times.11 Another catalyst for an update would be improvements in interobserver variability of biopsies with the advent of deep learning technologies and consensus groups working toward a common standard and improved consistency in grading. The main models use some histological parameters which require expert interpretation. There is work in the pipeline to investigate stratified treatments for AAV rather than a current one-size-fits-all approach (Silke Brix, personal correspondence). Scurt et al. noted that induction and maintenance therapies may have heterogeneous treatment effects. By changing the treatment given to patients based on their prognosis, the predictions that are made will no longer be valid and may result in strange counterfactuals. In conclusion, the study by Scurt et al.6 is the first step in AKRiS potentially becoming the gold standard of ANCA-associated GN prognostication, but many more validations are needed with collaboration across centers and with comparisons of scores. Of the models out there, ARRS and now AKRiS show the most promise with the ARRS having more validation studies than Berden and MCCS. AKRiS is simple to use and gives both a relative risk for stratification and precise predictions of prognosis. To answer the question in the title, AKRiS will probably become the gold standard.