Abstract: TH-PO529
Identifying the Etiology of Thrombotic Microangiopathy (TMA) Diagnosed by Kidney Biopsy Using Machine Learning Tools (Unsupervised Clustering)
Session Information
- Pathology and Lab Medicine
November 03, 2022 | Location: Exhibit Hall, Orange County Convention Center‚ West Building
Abstract Time: 10:00 AM - 12:00 PM
Category: Pathology and Lab Medicine
- 1700 Pathology and Lab Medicine
Authors
- Moubarak, Simon, Mayo Clinic Department of Internal Medicine, Rochester, Minnesota, United States
- Alexander, Mariam P., Mayo Clinic, Department of Laboratory Medicine and Pathology, Rochester, Minnesota, United States
- Zand, Jaleh, Mayo Clinic Department of Internal Medicine, Rochester, Minnesota, United States
- El Ters, Mireille, Mayo Clinic Department of Internal Medicine, Rochester, Minnesota, United States
- Zand, Ladan, Mayo Clinic Department of Internal Medicine, Rochester, Minnesota, United States
Background
Thrombotic microangiopathy is a pathologic term that encompasses disorders with different etiologies. Identifying the cause of the TMA lesion remains a clinical challenge. Most TMA cases share similar morphologic findings on microscopy which poses a challenge to determine the cause based on biopsy alone.
Methods
We searched renal pathology database at Mayo Clinic between 2010-2020 and identified patients with lesion of TMA on native kidney biopsy as their primary diagnosis who had an established associated cause for TMA. We recorded features commonly associated with TMA lesion by reviewing the biopsy report. We used consensus clustering approach (an unsupervised machine learning technique) using only the kidney biopsy variables to cluster the patients. We then looked to see which etiology of TMA was most prevalent in each cluster and which biopsy features were most predictive of finding that specific etiology.
Results
There were 168 adult patients with average age of 50.3 ± 16.9 with 52% male. The most common cause of TMA was malignant hypertension (mHTN) (n=81, 48.2%), drug-induced (n=36, 21.4%), lupus (9, 5.4%) and scleroderma (9, 5.4%). We identified 3 distinct clusters with cluster 1 (n=69), cluster 2 (n=47), cluster 3 (n=52). Cluster 1 was mainly composed of patients with mHTN followed by SRC. Presence of fibrin thrombi in the vessel in addition to presence of mucoid intimal edema and hyperplastic changes in the vessel (“onion skinning”) with absence of arteriosclerosis and fibrin tactoids were predictive of such diagnosis. Cluster 2 was mainly composed of patients with pregnancy associated TMA, non-immune drug induced TMA (VEGF inhibitor most common 67%), post BMT and myeloproliferative disease. Presence of severe tubular injury and absence of mucoid intimal edema, fibrin thrombi and arteriosclerosis were predictive of this cluster. Cluster 3 was composed of patients with lupus associated TMA, immun-mediate drug-induced TMA (gemcitabine 60% of cases) followed by aHUS. Presence of mesangiolysis and absence of hyperplastic changes of the vessel were predictive of this cluster.
Conclusion
This is the first study to use unsupervised machine learning to help identify the etiology of TMA lesion that is seen on native kidney biopsies by identifying biopsy features that are predictive.