Developing efficient predictive models for the diagnosis of VEXAS syndrome

Abstract

Objectives

To identify clinical and laboratory features associated with the presence of UBA1 mutation and develop predictive models to guide efficient diagnosis of VEXAS syndrome.

Methods

All patients who underwent UBA1 mutation testing were identified. Using a comprehensive list of VEXAS syndrome features, the presence or absence of each feature in each patient was determined for two timepoints: time of first VEXAS symptom onset and the time of UBA1 mutation testing. For each timepoint, the presence of each disease feature was compared between UBA1 positive and negative patients. The least absolute shrinkage and selection operator (LASSO) method was used to develop multi-feature models to predict the presence of pathogenic UBA1 mutations.

Results

Overall, 144 patients who underwent UBA1 mutation testing were included. Features including skin rash, chondritis, and monocytopenia were significantly associated with the presence of a UBA1 mutation at both timepoints. Macrocytosis, uveitis, and pulmonary disease had significant association at time of testing. 12-feature and 5-feature LASSO models at symptom onset demonstrated good discriminatory capacity with areas under receiver operating curves (AUCs) of 0.86 and 0.81, respectively. 16-variable and 5-feature models at time of testing had good-to-excellent performance with AUCs of 0.92 and 0.84, respectively. A single feature model utilizing absolute monocyte count also had fair discriminatory capacity with AUC of 0.78 at both timepoints.

Conclusions

Multi-feature models that efficiently separate VEXAS cases from controls were successfully developed. These models have the potential to address existing diagnostic challenges including a lack of consensus regarding key features of VEXAS syndrome.