Introduction and Objectives Volatile organic compounds (VOCs) in exhaled breath are affected by airway inflammation and have been shown to predict asthma exacerbations.¹ Our objective is to develop a new point of care (POC) breath test for early detection of asthma exacerbations, that will be revolutionary for asthma management, using deep neural network (DNN) and nanosensor technology.

Methods We collected VOC biomarkers, capnographic waveforms, asthma control test scores and clinical lung function parameters, over 2 years from 20 patients. 14 adults developed a total of 34 exacerbations. End-tidal breath samples were collected, and 13 different parameters were measured using nanosensors. Principal component analysis was used to identify parameters whose values will control the algorithm’s learning (figure 1: A). A 13-layer multiclass classification DNN with 3 outcomes (1 well-, 2 poorly-, 3 un- controlled asthma), was trained with 1290 data points containing the patient data, biomarkers and lung function parameter ranges to achieve a big data approach. The model was used around 50–100 times per day for 3 months by different users with 95% efficiency.

Results Our model predicts asthma exacerbations with 93% accuracy up to 3 days ahead following daily monitoring over 5 days through further validation for personalisation and area under the curve (AUC) receiver operating characteristic of 0.90 (figure 1: B).

• Download figure
• Open in new tab
• Download powerpoint

Abstract S69 Figure 1

A) Principal component analysis of biomarkers of (1) well-, (2) poorly- and (3) uncontrolled asthma. B) Receiver operating characteristic of DNN model’s results

Conclusions Our model could accurately predict asthma exacerbations 3 days in advance and can be transformative to early detection and subsequently early life-saving interventions for asthma exacerbations.

Reference

1. Dobroes CM, van Berkel JJ, Jobsis Q, van Schooten FJ, Dallinga JW, Wouters EF, Dompeling E. Eur Respir J 2013;42:98–106.