Validation of an Artificial Intelligence Enabled Diagnostic Support Software (ArtiQ.Spiro) in Primary Care Spirometry Datasets - a Retrospective Analysis

This is a retrospective analysis of existing clinical datasets with consecutive spirometry collected in a primary care setting in the UK. Individual patient data will be included if the individual meets the study protocol eligibility criteria.

Clinical datasets will be de-identified (name, date of birth, address, postcode, occupation GP, ethnicity, medications data removed). Individuals will be identified by a study ID number. The de-identified datasets will contain the minimum information needed for spirometry and ArtiQ.Spiro - namely age, smoking history, height, weight, primary respiratory symptom - and the deidentified data exported from the primary care spirometry software.

ArtiQ.Spiro Evaluation (Index Tests for Diagnosis and Quality):

A deidentified dataset will be provided to a machine learning analyst who will apply the machine learning algorithm of ArtiQ.Spiro. For each individual, the algorithm will produce a preferred diagnosis (highest probability diagnostic category) (Index Test for Diagnosis) and an assessment of spirometry quality (Acceptable, Usable, Not Acceptable/Usable) (Index Test for Quality). No clinical information outside of the spirometry dataset nor reference standard data will be made available to the analyst.

Reference Standard for Diagnosis: