Developing methods to quantify the severity of airflow obstruction in the clinical sleep environment

Abstract

Obstructive sleep apnoea (OSA) is a significant public health problem, with comorbidities including excessive daytime sleepiness, increased risk of motor vehicle accidents, cardiovascular disease, cognitive impairment, and decreased quality of life. OSA is characterised by intermittent periods of pharyngeal obstruction resulting in the absence (apnoea) or reduction (hypopnoea) of airflow during sleep. OSA severity is currently reported by the apnoea-hypopnoea index (AHI): the frequency of apnoeas and hypopnoeas per hour of sleep. However, the AHI is a poor predictor of an individual’s day to day performance, symptomology and long-term health outcomes. This is likely due to a limitation with the event-based AHI, which inadequately describes the underlying neuro-mechanical airway resistance that leads to pharyngeal obstruction, and as such, does not capture the severity of airflow obstruction.Therefore, the overarching goal of this thesis is to develop non-invasive methods to objectively characterise the severity of airflow obstruction on a breath-by-breath basis.Previous literature has identified characteristics within the airflow signal that may indicate airflow obstruction. However, analysis of these features on a breath-by-breath basis requires accurate demarcation of “breath timing” (i.e. inspiratory and expiratory phases). As such, the first aim of this thesis was to develop methods that accurately demarcate breaths. We segmented breaths into inspiratory (TI, shortest period achieving 95% inspiratory volume), expiratory (TE, shortest period achieving 95% expiratory volume), and an inter-breath transition period (TTrans, the period between TE and subsequent TI). In a cohort of 37 patients with an epiglottic pressure catheter, we observed that compensatory increases in the inspiratory period (1.57±0.27 vs 1.74±0.27 seconds, mean±SD, P≤0.001) from reference breaths to partial airway obstruction (identified by increasing epiglottic pressure swings without increasing airflow) were primarily explained by reductions in the inter-breath transition period (0.82±0.28 vs 0.59±0.21 seconds, mean±SD, P≤0.001) and not by reduction of the expiratory airflow period (1.68±0.32 vs 1.60±0.25 seconds, mean±SD, P≤0.05). In addition to developing a method for demarcating breaths, we also identify TTrans as a novel feature associated with increased airway obstruction.Using our newly developed breath timing method in conjunction with the other known markers of flow limitation, we then aimed to characterise the flow shape of individual breaths to develop a transparent machine learning strategy for the non-invasive quantification of the severity of airflow obstruction. We defined gold-standard obstruction severity as the ratio of oronasal pneumotach (flow) and ventilatory drive (calibrated intra-oesophageal diaphragm EMG, drive), presented as a continuous breath-by-breath variable, flow:drivemeasured. Multivariable linear regression used non-invasive flow-shape features (inspiratory/expiratory timing, flatness, scooping, fluttering) to model flow:drivemeasured in 136,264 breaths from 41 patients (with pneumotach and intra-oesophageal diaphragm EMG). We observed flow:drivemeasured varying widely across individuals independent of AHI. A multivariable model (25 features) estimated obstruction severity (flow:driveestimated) breath-by-breath (R2=0.58 vs. flow:drivemeasured, P<0.00001; mean absolute error=22%) and the median obstruction severity across individual patients (R2=0.69, P<0.00001; error=10%) (performance based on conservative leave-one-patient-out cross-validation). As such, we present a novel measurement of airway obstruction severity, flow:drivemeasured, and a model to non-invasively estimate airflow obstruction severity, flow:driveestimated, on a breath-by-breath basis.While the novel measurement of airway obstruction severity provides unique insights into sleep-disordered breathing, it is important to consider that this method was developed from objective quantitative physiological signals, and noisy signals could have affected the rigour of this method, leading to potential errors in non-invasive model flow:driveestimated values. Therefore, the third aim in this thesis was to compare airflow obstruction severity against human-based scoring of airflow obstruction certainty and ensure that physiologically derived measurements were congruent with manual scoring. We developed a scoring protocol based primarily on flow-time profile, and manually categorised individual breaths as certain-, possible-, or non-flow-limited. ROC analysis demonstrated that flow:drivemeasured provided excellent discriminatory capacity (area under ROC curve = 0.98) between certain-flow-limited and non-flow-limited breaths. Discriminatory capacity using flow:driveestimated performed similarly.Finally, having established clinical applicability, we conducted a pilot study to investigate treatment response in two upper airway based therapeutic interventions (oral appliance therapy and upper airway surgery) and explored the individual relationships between existing metrics of OSA severity (i.e. AHI) and our novel airflow obstruction severity (i.e. flow:driveestimated). In both interventions, we noted a high degree of variability between the AHI and flow:driveestimated, at baseline and with treatment. Therapeutic success is generally classified by substantial reductions in AHI, however, individual patient responses do not always parallel the changes in AHI. We also noted considerable variability between the current objective measurement (i.e. AHI) and qualitative subjective measurement. We conclude this analysis showing significant associations between changes in the novel airflow obstruction severity metric (Δ flow:driveestimated) and changes in subjective measures of sleepiness (i.e. Epworth Sleepiness Scores).In conclusion, this thesis describes methods for the non-invasive quantification of the severity of pharyngeal airflow obstruction, using readily-available flow shape information. This work demonstrates that the AHI inadequately describes pharyngeal obstruction and overcomes a major hurdle necessary for grading the severity of obstructive sleep-disordered breathing.