Keywords: Non-linear modeling, CD-sens, MCMC, Bayesian, BAT, Hierarchical
The basophil activation test (BAT) is a flow cytometry-based assay used to measure the activation of basophils in response to increasing concentrations of an allergen, resulting in a sigmoidal activation curve. Various parameters from this activation curve have been used to diagnose peanut allergy and severe allergic anaphylaxis to peanut.
For diagnostic testing, one commonly used metric based on the BAT curve is CD-sens, which is defined as 1/EC50 X 100, where EC50 is the effective dose at 50% of maximal basophil activation. The estimation of EC50 is often performed using a linear approximation of the BAT dose-response curve. However, the current procedure needs to be manually tuned for non-standard BAT curves, making it difficult to automate.
Here we propose an approach to estimating the BAT curve using a hierarchical Bayesian model. This method assumes a logistic growth model having parameters for the maximal level of activation, EC50, and rate of activation. This method incorporates single or repeated measures from subjects while also allowing prior information and covariate adjustment.
In a dataset of 630 participants from the LEAP, LEAP-On, and PAS peanut allergy trials, we estimated, using the linear regression approach, a median [IQR] CD-sens of 21.11 [2.11, 192.71] and 0.19 [0.08, 14.92] for the allergic and non-allergic participants, respectively, and for the non-linear approach 11.7 [1.89, 64.54] and 0.01 [0, 0.11] for allergic and non-allergic participants, respectively. In addition, the non-linear approach of CD-sens gave better diagnostic properties than the linear approach, with AUCs of 0.85 and 0.75, respectively.
Non-linear estimation of the BAT curve using a hierarchical Bayesian model provides an approach to estimate various parameters of the BAT curve that may be of interest, including EC50 and CD-sens. In addition, this method provides a way to estimate the parameters from many curves in a robust, automated fashion.