882 
Using capnography breath data, a machine learning algorithm diagnosed chronic obstructive pulmonary disease (COPD) with an accuracy of 91%.
In the study, published in Respiratory Research, UK researchers used the N-Tidal device and applied machine learning techniques to capnography data to help distinguish the CO2 recordings (i.e. capnograms) of patients with and without COPD.
The team utilised capnography data from four clinical studies and developed machine learning algorithms to discriminate COPD from non-COPD, which comprised a group of patients who were either healthy or who had other conditions including asthma, heart failure, pneumonia, breathing pattern disorder and motor neurone disease.
The team developed three machine learning models and the predictability for COPD was assessed using receiver operator characteristic (ROC) curves and the subsequent estimates of sensitivity, specificity negative and positive predictive values.
Capnography machine learning accuracy
A total of 88,186 capnograms were collected from 295 patients, with each patient providing an average of 299 capnograms over 179 days.
The highest accuracy (91.3%) was provided by an XGBoost model with a corresponding sensitivity of 91.5% and a specificity of 91.4% for the diagnosis of COPD. Even on an unseen test data set, the XGBoost model still had an accuracy of 90%.
According to the manufacturer of the N-Tidal device, TidalSense, it takes under five minutes from the start of the breath test to diagnosis.
Based on the findings of the current study, the researchers concluded that the ability of the N-Tidal capnography device to accurately diagnose COPD in near-real-time lends support to its future use in a clinical setting.
COPD diagnostics in context
COPD led to 3.23 million deaths in 2019 and was the third leading cause of global deaths. Spirometry is generally considered to the be gold standard diagnostic tool for COPD, and its use is on the rise, yet it is also one of the major causes for misdiagnosis. Capnography is a widely used technique that could be used to diagnose COPD.
