Machine Learning Predicts Acute Pulmonary Exacerbations in Cystic Fibrosis

Abstract

Background: Sudden deteriorations in lung health, termed acute pulmonary exacerbations (APEs), are a major driver of mortality and morbidity in Cystic Fibrosis (CF). Accurate prediction of impending APEs would permit pre-emptive interventions and allow home monitoring to safely replace hospital-based physician consultations. Methods: We enrolled 147 adults with CF into a six-month study of home monitoring based at seven UK specialist CF Centers. Subjects were asked to undertake daily measurements of lung function, oximetry, pulse rate, weight, and activity (using sensors Bluetooth-linked to mobile phones), and provide daily self-reported symptom scores of cough frequency, general wellness, and sleep quality. Linked-anonymised data were then analysed using machine learning (ML) methods to define the profile of APEs and predict their onset. Findings: End of study questionnaires revealed that 92% of participants found home monitoring easy to use and 76% found it helpful or very helpful in tracking their health over time. Unsupervised machine learning analysis uncovered the typical signal profile of an APE and revealed three distinct classes of exacerbation. We developed a ML predictive classifier that can detect an impending APE on average 10 days earlier that current clinical practice. Interpretation: High frequency home monitoring in CF is feasible, reveals distinct types of APEs, and permits accurate prediction of future exacerbations. Trial Registration Number: NCT02416375 (ClinicalTrials.gov). Funding Statement: This study was funded through research grants from the UK Cystic Fibrosis Trust, Royal Papworth Hospital, the Wellcome Trust (Investigator Award RAF; 107032/Z/15/Z), and Microsoft Research/EPSRC (PhD scholarship DES). Declaration of Interests: Thomas Daniels has accepted speaker fees from Vertex pharmaceuticals, and consultancy fees from Chiesi Pharmaceuticals, none of which were involved in anyway at all with the submitted manuscript, nor had any influence at any stage of the work. The remaining authors have no competing interests to declare. Ethics Approval Statement: The study received National Ethical approval from the NRES Committee East of England-Norfolk (REC 14/EE/1244) and site-specific approval and monitoring from each participating Hospital R&D Department. All participants provided written informed consent before enrolment into the study.