Search

Unintended bias against protected groups has become a key obstacle to the widespread adoption of machine learning methods. This work presents a modeling procedure that carefully builds models around protected class information in order to make sure that the final machine learning model is independent of protected class status, even in a nonlinear sense. This procedure works for any machine learning method. The procedure was tested on subprime credit card data combined with demographic data by zip code from the US Census. The census data serves as an imperfect proxy for borrower demographics but serves to illustrate the procedure.

ANCA-associated vasculitis (AAV) poses a significant risk of kidney failure, kidney biopsy remains a key prognostic tool. The histopathologic classification of AAV glomerulonephritis (GN) developed by Berden et al. showed correlation between GN classes and kidney outcomes, ANCA Renal Risk Score (ARRS) included tubular atrophy and interstitial fibrosis (TA/IF) as an additional parameter for risk assessment. We aimed to evaluate kidney survival across AAV GN classes and ARRS groups. A single-center retrospective study included 85 adult patients with biopsy-proven AAV kidney disease followed in the period of 2000–2020. Primary outcome was kidney survival at the end of 18 (5, 66) months follow-up, and kidney death was considered as stage 5 CKD. We found significant differences in kidney survival for sclerotic, mixed, crescentic and focal AAV GN classes: 19%, 76.2%, 91.7% and 100%, respectively (p = 0.009). Kidney survival was 0%, 75.6% and 100% for the high-, medium- and low-risk ARRS groups, respectively (p &lt, 0.001), TA/IF analysis showed kidney survival at 49.6% vs. 87.7% for widespread and mild TA/IF, respectively (p = 0.003). Kidney survival was significantly lower in anti-MPO-ANCA versus anti-PR3-ANCA carriers (50.3% and 78.1%, respectively, p = 0.045). We conclude that unfavorable AAV kidney outcomes are associated with sclerotic GN class by Berden’s classification, ARRS high risk group, and anti-MPO-ANCA subtype.