Subcutaneous glucose biosensor failure – a fuzzy fault tree analysis approach

Abstract

Diabetes mellitus is a disease of major global importance, increasing in frequency at almost epidemic rates. In order to optimize diabetes management and, therefore, reduce hyper- or hypo-glycaemia complications, continuous and reliable in vivo (subcutaneous or intravenous) glucose monitoring is indispensable. To that end, subcutaneous sensors have progressed more rapidly as they can be linked to state-of-the-art signal transmission modes (transdermal, IR, etc.). Yet, most of the devices currently at clinical trials have not demonstrated a stable and clinically useful sensor performance. In this work, the causes of subcutaneous glucose biosensor sensitivity drift have been investigated by means of fault tree analysis relying on fuzzy reasoning to account for uncertainty. Using the methodology proposed herein, all ultimate causes or combination of causes attributed to the device components, the surrounding tissue and their intra/inter-relations that are responsible for or contribute to the top event have been recognized and quantifi ed based on (a) measurements for the deterministic contributors and (b) experience for the stochastic contributors. The tree structure has been designed by combining deduction and induction, top-down and bottom-up techniques, thus establishing a dialectic tradeoff which brings this method closer to scientifi c logic, permitting the introduction of deeper knowledge into the surface or experiential knowledge level characterizing FTA. The proposed methodology has been implemented in the investigation of the causes responsible for sensor fouling by thrombus formation and has proven to be an effi cient tool for internal diagnostics and fault compensation. The suggested approach may contribute signifi cantly to the self-optimization of the measuring equipment from one generation to the next as it supports the fl exible, ad hoc, and tailor-made development, thus potentiating the progress of epidemics from statistics to individualization.