Developing a Machine Learning-Based Risk Calculator for Intracranial Aneurysm Detection in Patients with Autosomal Dominant Polycystic Kidney Disease

Introduction: Patients with autosomal dominant polycystic kidney disease (ADPKD) have been identified to have a significantly increased risk of developing intracranial aneurysms (IA). Our prior meta-analysis demonstrated increased risk of UIA detection in screening in ADPKD patients who are female, have a family history of IA or SAH, have hypertension, or have moderate or more severe CKD. The objective of this study is to characterize risk profiles for IA in individual ADPKD patients and develop a risk calculator that can be utilized in clinical decision-making.

Methods: Over 3216 ADPKD patients were identified across Mayo Clinic sites. Of these patients, 1838 received head imaging. We created a database using the first known CTA, MRA, or DSA for each unique medical record number. The patients were divided into subgroups: 1) no IA, 2) asymptomatic IA, 3) symptomatic IA. Data was divided further into training (70%) and validation (30%) subsets. Primary data analysis will be conducted to compare patients with asymptomatic IA and no IA using logistic regression as well as support vector machine (SVM) and random forest (RF) models to identify statistically significant risk factors. Secondary analyses will also be conducted including symptomatic IA .

Results: Preliminary analyses using logistic regression support our prior meta-analysis. Models will be compared using sensitivity and specificity in addition to AUROC parameters to identify which models are optimal for binary classification. We will assess several risk factors for IA including female sex, hypertension, family history of either IA or subarachnoid hemorrhage, family history of SAH, stage 3 or greater chronic kidney disease (CKD), kidney transplant status, age >45, family history of ADPKD, smoking status, and presence of liver cysts.

Conclusion : The results of our retrospective chart review study have implications on screening for IA in patients with ADPKD. Models from our preliminary analysis excluding symptomatic patients may have increased utility in decision-making regarding early screening in APDKD patients, while the secondary analysis including symptomatic patients may have increased implications on lifetime risk of aneurysm formation.