Ventilatory Burden: Development of a New Approach to Better Quantify Obstructive Sleep Apnea Severity and Its Impacts

Neutral and inconsistent findings for obstructive sleep apnea (OSA) treatment trials across a range of outcomes (1–3) have motivated recent efforts to better define OSA severity. Strategies have included novel in-home technology to quantify the apnea–hypopnea index (AHI) over multiple nights to reduce the high rates of disease misclassification associated with current single-night diagnostic approaches (4–6). Techniques to identify different OSA endophenotypes to help increase success rates with existing therapies and develop new therapies that target specific pathophysiological causes have also been developed (7). In addition, new markers of OSA severity based on specific pathophysiological consequences of OSA have been established (8). Examples include hypoxic burden to capture both the frequency and severity of OSA-related hypoxia (9) and electroencephalogram-based markers to better quantify OSA-related sleep disruption such as arousal burden (10), odds ratio product (11), and quantitative measures of sleep depth and pattern (12). These physiology-based markers of sleep and breathing disruption better predict important health outcomes such as sleepiness, cardiovascular events, and all-cause mortality than traditional polysomnography metrics such as the AHI (9–12).

Extending this important body of work, in this issue of the Journal, Parekh and colleagues (pp. 1216–1226) present a novel metric to quantify another fundamental physiological feature of OSA severity: the overall ventilatory burden, which is only crudely captured with the conventional AHI metric (13). Airflow limitation is central to the definition of apnea and hypopnea events, which are typically scored manually using arbitrary rules that have changed over time. This has had important consequences for sleep science, including disease prevalence estimates and clinical care (14, 15). Manual scoring of hypopneas is also challenging, with a large interscorer variability (15). Algorithm-based estimation of ventilatory disturbance, as proposed by Parekh and colleagues, negates the need for manual scoring and therefore overcomes a major limitation of existing manual methodology.

There are two keys factors that need to be considered in the development of a good marker of OSA severity: 1) It must predict treatment response and/or specific disease-related consequences/important health outcomes, and 2) it must be reliable and stable across multiple nights when measured under similar conditions. Parekh and colleagues provide a thorough assessment of their new metric across multiple datasets and experimental domains and include normative data, for which they are to be congratulated. The authors demonstrate that their metric is associated with sleepiness, hypertension, cardiovascular mortality, and all-cause mortality. Importantly, the new ventilatory burden metric is associated with key health outcomes, even after adjustment for hypoxic burden. This suggests that both flow-limited breathing (without associated desaturations) and oxygen desaturations are associated with adverse health consequences. The combination of hypoxic burden and ventilatory burden in a model also had a goodness of fit superior to a model with AHI. This suggests that in addition to the advantage of not requiring manual, often variable, human scoring, it provides additional insight beyond the AHI.

Furthermore, ventilatory burden decreased with continuous positive airway pressure (CPAP) treatment in a dose-dependent manner (off, suboptimal, optimal). Thus, these findings provide strong evidence for biological plausibility as a robust measure of OSA severity. Estimation of breathing using alternative signals, such as nasal pressure or effort bands, and normative values has also been proposed by others. Indeed, Labarca and colleagues recently described an alternative definition of “ventilatory burden” that is associated with incident cardiovascular disease in two large community cohorts (16). Which definition of “ventilatory burden” is best in terms of prediction of important health outcomes requires further investigation. Nonetheless, these two recent complementary contributions suggest that airflow limitation, and specifically overall ventilatory burden, is an important marker of OSA severity and one that the field should consider implementing.

In terms of reliability, the authors present data on night-to-night variability of their new metric. Although there is minimal bias at the population level (mean bias, 0.16), on an individual level, the confidence interval of ±13% is potentially not insignificant. For example, someone who presents with a ventilatory burden of 30% on Night 1 could have a ventilatory burden anywhere between 17% and 43% on Night 2. This is quite a large range, especially in light of the effect size of CPAP treatment on ventilatory burden (at the population level, CPAP-off 32% vs. 7% CPAP-on). These limitations are not confined to the ventilatory burden metric. In fact, OSA endotypes have similar variability (17). The AHI has even higher night-to-night variability (6, 18). Although this may reflect, at least in part, “measurement error/noise,” a stronger influential contributor may indeed be that OSA is often a disease with variable severity night to night, regardless of the metric used to assess its severity (5, 6, 18). Given that the authors can derive ventilatory burden using effort belt signals, which are mostly noninvasive, ventilatory burden could be assessed over multiple nights quite easily to further investigate these important questions.

As highlighted by the authors, there are several limitations with ventilatory burden that need to be considered. Perhaps one of the largest is the lack of information between wakefulness and sleep. Reliance on the airflow signal and/or effort belt only is very appealing, given the simplicity of recording these signals. However, given that the ventilatory burden is calculated using recording time rather than sleep time, a large amount of time spent awake during the night would artificially lower the overall ventilatory burden. This is problematic for people with OSA and other comorbid conditions that may impair sleep (e.g., insomnia, chronic pain, restless legs). Given the large co-occurrence of OSA and insomnia and the higher risk of cardiovascular risk and premature mortality (19), the ventilatory burden should be validated in people with comorbid OSA and insomnia. It is also unclear how and if the ventilatory burden is affected by sleep position and sleep stages and whether different normative values for different stages or positions are required. Building on the work of others on the potential adverse impacts of airflow limitation, increased respiratory effort, and snoring in the absence of hypoxia for cardiovascular health (20), it will be important to further investigate the underlying mechanisms and different physiological manifestations of inadequate ventilation during sleep.

One only has to look at the controversy highlighted in recent pro versus con debates and letters to the editors regarding OSA endophenotyping using polysomnography to understand that any new OSA severity metric/algorithm that has the potential to change the field will be met by with an equal amount of enthusiasm and skepticism. By open sourcing their algorithms, Parekh and colleagues provide the opportunity for anyone to validate, use, and/or further improve their novel metric. The authors must be applauded for their contributions toward advancing the field and for their commitment to open source, including their prior work.