Sleep Research
A New Frontier in Technology-Enabled Biomarker Discovery and Sleep Disorder Diagnosis: Moving Past Traditional AHI and Highlighting Ventilatory Burden
In this blog, we’re taking a detailed look at the Apnea-Hypopnea Index (AHI) and how it has fallen short as an effective biomarker for Obstructive Sleep Apnea (OSA). We’ve also highlighted the recent work of one of our close collaborators, Dr. Ankit Parekh from the Icahn School of Medicine at Mount Sinai, and his recent research exploring a revolutionary new approach towards quantifying OSA.
Significance of Sleep Biomarkers
IIn 2015, two divisions of the NIH, the National Heart Lung and Blood Institute (NHLBI) and National Institute on Aging (NIA), jointly sponsored a workshop with the Sleep Research Society entitled “Developing Biomarker Arrays Predicting Sleep and Circadian-Coupled Risks to Health.” This session assembled leading intellectuals in the fields of biomarker development, sleep-circadian biology, and sleep disorders. Attended by career investigators hailing from prestigious institutions like Harvard Medical School, University of Pennsylvania, Duke University, Weill Cornell Medical College, and Mayo Clinic, this workshop’s findings were summarized in a seminal white paper with the same name, which has been cited nearly 90 times to date. With the workshop’s contributors being some of the most prominent researchers in the field, it has served as a guiding document over the past eight years that has informed the direction of research aiming to discover innovative biomarkers with clinical significance in the rapidly advancing field of sleep medicine (Mullington et al, 2016).
Traditionally, biomarkers in medical research refer to biological molecules whose presence or absence can inform the status of normal or abnormal processes, conditions, or diseases. In the burgeoning realm of sleep and circadian function research, however, this term’s definition focuses on quantifiable patterns and chemical properties of sleep processes and typically refers to metrics that can be extracted from biometric recordings conducted during sleep.
The current need for biomarkers in research can be summarized in the following objectives:
to obtain a more comprehensive understanding of sleep and circadian health;
to develop new diagnostic workflows for sleep and circadian disorders with an emphasis on solutions that can be administered directly at the patient's bedside;
to inform prognosis models that will accurately predict the risk for associated comorbidities including but not limited to cardiovascular, pulmonary and other adverse events such as aging
to proactively assess the adequacy and effectiveness of specific treatment options (Mullington et al, 2016)
Taken together, the development of precise and sensitive sleep biomarkers that can be conveniently and efficiently extracted and analyzed at the patient point-of-care will enable more significant and accurate quantifications of sleep health, enable better and more sensitive diagnosis of sleep conditions (which represents a key, challenging obstacle in the field today), open the door to personalized treatment options for individuals who are susceptible to adverse outcomes across organ systems and disease states, and inform clinical decision-making in regards to treatment by proactively anticipating the effectiveness of interventions.
Obstructive Sleep Apnea (OSA)
One of the most common sleep disorders, Obstructive Sleep Apnea (OSA), is a condition in which airflow is intermittently blocked during sleep, leading to associated oxygen desaturation events and arousals from sleep. It has been estimated to affect up to 1 billion people worldwide, but is widely varied and highly heterogeneous in terms of presentation and severity. Furthermore, this chronic condition has been linked to several notable comorbidities and complications, including, but not limited to, excessive sleepiness, obesity, cognitive impairment, hypertension, stroke, and cardiovascular disease (Malhotra et al, 2021).
The white paper referenced above notes that an ideal biomarker signature for OSA would enable clinicians to accomplish three things:
Inform the overall prognosis of the disorder
Measure the effectiveness of treatment on the molecular and cellular level
Allow us to better understand how OSA is related to its comorbidities.
At the time of the workshop, Dr. Mullington and her team noted that “Over the last 15 years, a substantial number of studies have tackled the identification of an ideal biomarker for OSA using exhaled breath condensate, salivary, serum and urinary molecules.” These molecules showed varying success as clinically useful biomarkers, and none reliably informed diagnostics or addressed comorbidity risks.
Instead, the most established method for standardizing severity of OSA and diagnosing it in clinical settings has been the apnea-hypopnea index (AHI). AHI is calculated by summing the number of instances of complete airflow obstruction (apneas) and partial pauses (hypopneas) per hour of sleep. By convention and in practice, a threshold of 5 apneas / hour has been set in which patients exhibiting an AHI above that threshold are diagnosed with OSA (Malhotra et al, 2021). However, the AHI is characterized by several severe limitations.
First, AHI fails to capture any information related to the depth and severity of the respiratory events that it counts. In full-night polysomnography tests (an in-lab sleep examination considered the gold standard for diagnosing sleep disorders), significant events are reduced to simple frequency counts, meaning that no information about the nature of these events are passed on to the biomarker calculation step (Martinez-Garcia et al, 2023). The heterogeneous nature of OSA necessitates a novel biomarker that can capture more than just simple frequency of respiratory events and the failure of AHI to address individual variations between respiratory events is a primary contributor to its limited utility.
Second, significant variation exists in institutional definitions of hypopnea and this inconsistency between institutions, studies, and researchers has led to inconsistent obstructive event identification and, therefore, AHI calculations. In fact, over the last 25 years alone, the American Academy of Sleep Medicine (AASM) has changed its official definition of a hypopnea three times. First, an AASM task force in 1997 recommended that apneas and hypopneas be considered equivalent and any event that exhibited a greater than 50% decrease in amplitude from baseline or that was associated with an oxygen desaturation of at least 3% and an arousal be counted as an apnea/hypopnea. However, the official AASM Manual for the Scoring of Sleep and Associated Events published in 2007 drew a clear distinction between apneas and hypopneas. According to the AASM Scoring Manual, apneas were defined as events where breathing stops by at least 90% for 10 or more seconds, with or without effort to breathe. Hypopneas, meanwhile, were defined by both a Recommended and an Alternative rule, drawing upon arbitrary thresholds of reductions in breathing and oxygen levels. The latest update regarding these definitions was published in Version 2 of the AASM Scoring Manual in 2012, in which hypopneas were once again redefined under its Recommended rule. The subjectivity of AHI calculations based on widely varying definitions of respiratory events is a well-known challenge and the 2012 manual explicitly refers to this, noting “thresholds for identification of the presence and severity of OSA, and for inferring health-related consequences of OSA, must be calibrated to the hypopnea definition employed,” speaking specifically to the variation introduced by the two distinct rules published in the same manual (Malhotra et al, 2021).
Lastly, the utility of AHI is dependent on arbitrary thresholds, such as only events longer than 10 seconds are associated with an oxygen desaturation or diagnosis of OSA requires an AHI greater than five. This provides an incredibly limited view of individual manifestations of the disorder, which is therefore unsuitable for guiding clinical recommendations (Martinez-Garcia et al, 2023). Furthermore, individual presentations of similar AHI scores can vary widely, with some patients experiencing significant symptoms at AHI values below the commonly used diagnostic threshold. This not only puts into question the validity of an OSA diagnosis but also raises concerns about the adequacy of treatments for those falling just below these arbitrary cutoffs.
These challenges all combine to contribute to AHI’s increasingly limited utility in practice, as exemplified by the metric’s lack of associations with known OSA complications. Not only does it fail to capture any relationship to clinically relevant aspects such as quality of life measurements and hypersomnia, but also fails to establish strong associations with comorbidities such as obesity and adverse cardiovascular outcomes, all while phenotypic variability is incredibly high (Martinez-Garcia et al, 2023). In other words, out of the three primary objectives identified as ideal for an OSA biomarker by Dr. Mullington and her team, AHI satisfactorily fulfills none of them.
Hypoxic Burden
As a result, much work has centered on incorporating the duration of respiratory events and properties of associated desaturations into an OSA biomarker. This has led to a new OSA marker in the literature that has quickly gained momentum in recent years. This new metric, termed hypoxic burden (HB), was designed to offer a higher level of detail than AHI. It not only incorporates the frequency of respiratory events, but also captures the depth and duration of their associated desaturations. It is defined as “the total area under the oxygen saturation curve from a pre-event baseline oxygen desaturation” and is calculated by measuring the extent of oxygen desaturation during each breathing disruption and summing these measurements to quantify overall sleep disturbance (Martinez-Garcia et al, 2023).
Research utilizing hypoxic burden as a metric of sleep disturbance has increasingly shown clinical utility, especially when compared to AHI. Notably, HB has been demonstrated to be significantly associated with adverse OSA complications, one of the most challenging, unresolved aspects of AHI. Dr. Azarbarzin and his team from Harvard Medical School carried out a research study using two large cohorts from the Sleep Heart Healthy Study (SHHS) and The Outcomes of Sleep Disorders in Older Men (MrOS). This dataset encompassed several communities across the US and a total number of participants exceeding 8,000 men and women. Through their analysis, they were able to demonstrate that there are significant links between HB and mortality related to cardiovascular disease (CVD). Even after accounting for potential confounding factors, such as concurrent medical conditions, demographic characteristics, and other sleep metrics, this association remained significant, suggesting that HB has potential as a clinically valid independent predictor of CVD mortality. Furthermore, although researchers could not consistently correlate HB with all cause mortality, they demonstrated that it had vastly improved predictive power relative to AHI, which has shown time and time again to be insufficient in exploring the links between OSA and its documented comorbidities (Azarbarzin et al, 2019). Just one year later, Azarbarzin and his team again utilized the same two data sets to construct association models between OSA and Heart Failure (HF). They found once more that HB successfully and significantly predicted incident HF across both samples, compared to AHI which continued to underperform (Azarbarzin et al, 2020).
HB’s predictive power has been extended to include a growing roster of outcomes. For example, in 2021, Blanchard and team utilized similar associative statistics to elucidate the relationship between HB, Heart-Rate Variability (HRV), and incidence of stroke. HRV is a metric that captures the fluctuation in time between heartbeats, providing a window into the function of the autonomic nervous system. While not exclusive to sleep, HRV is an additional biomarker undergoing intense research in the field today. Aiming to control for the heterogeneity of OSA, Blanchard utilized both biomarkers – HB and HRV – in their models and once again uncovered significant associations with stroke that remained even after controlling for confounding risk factors (Blanchard et al, 2021). That same year, Dr. Jackson and her team from the NIH demonstrated the effect of sleep on chronic kidney disease (CKD) by observing associations between sleep indices and the prevalence of moderate-to-severe CKD. They found the most significant associations with increased HB and very short sleep duration, noting that while there was a slight correlation defined by AHI, these predictions were vastly improved in strength and consistency when HB was also employed (Jackson et al, 2021).
Over the last four years, hypoxic burden (HB) has emerged as a promising biomarker that better captures the multidimensional nature of OSA and has strong predictive capabilities for related comorbidities and complications. HB has consistently improved risk stratification as compared to AHI for numerous conditions, including cardiovascular disease, heart failure, stroke, and kidney disease. HB is relatively easy to calculate, as it only requires airflow and oxygen desaturation data, both of which are present in laboratory and in-home settings, and several existing automated software solutions, such as the clinically-validated Neurobit PSG, can perform this computation.
Ventilatory Burden
While HB’s clinical utility has been rapidly gaining recognition in the field, Dr. Ankit Parekh of the Icahn School of Medicine at Mount Sinai recognized that because the metric relies heavily on the presence of desaturations or arousals to identify respiratory events, it fails to capture information about events that occur in the absence of said desaturations or arousals. Instead, by quantifying spectral characteristics of a subject’s sleep EEG directly, Dr. Parekh sought to define a new biomarker that is not dependent on hypoxemia, arousals, or manually marked respiratory events, but still captures reductions in airflow and is therefore a better characterization of the heterogeneity and vast individual variability inherent in OSA. In his paper published in the American Journal of Respiratory and Critical Care Medicine in early September of 2023, his groundbreaking work in defining the “Ventilatory Burden (VB)” biomarker and demonstrating its prowess in effectively assessing OSA severity and predicting all-cause and CVD mortality has been making waves in the field already, as VB’s clinical potential is realized to be much greater than that of AHI and even HB (Parekh et al, 2023).
In pursuit of this ambitious task, Dr. Parekh and his team from Mount Sinai utilized data collected from several epidemiological cohorts with a total subject count of more than 5,100. Utilizing channels corresponding only to airflow signals, they derived VB by analyzing every breath of each patient, which came to total more than 34 million breaths across the entire dataset. Technical details for how VB is derived are shown in Fig. 2, but in summary, VB captures the proportion of breaths overnight that have reduced normalized amplitude (<50%). In summarys, a subject with higher VB compared to healthy controls has a higher frequency of breaths that are drawing in suboptimal amounts of air, therefore indicating a respiratory deficiency or “burden” for that particular subject.
With the calculation defined for VB, Dr. Parekh took the work further by creating proportional models to assess VB’s relationships with outcomes and complications, including CVD and all-cause mortality, daytime sleepiness, and hypertension. First, by using cohorts of healthy patients, the research team established a normative range for VB. This range indicates that 95% of normal and healthy subjects have breaths with below-normal amplitude less than 25% of the time. Next, the team demonstrated a clear dose-response relationship between VB and therapeutic CPAP. They found that in symptomatic cohorts, not only was the 95th percentile of VB significantly higher than normal, but with CPAP treatment for three months, VB decreased significantly. They also associated significant VB increases with suboptimal CPAP use and nights without treatment. The sensitivity of VB to traditional OSA therapy supports this biomarker’s validity as a metric for OSA severity, as it effectively characterized the disorder independent of the presence of respiratory events. Furthermore, VB was shown to be highly reliable, with extremely low night-to-night variability across both in-lab and at-home sleep tests, in sharp contrast to the highly imprecise nature of the variance associated with AHI calculations.
To quantify associations between AHI, HB, and VB and elucidate their relationships to OSA complications, Dr. Parekh and his team created four distinct models. Model 1 only utilized AHI, Model 2 was exclusively HB, Model 3 was VB and Model 4 combined HB and VB. Each model was then associated with the given outcomes. For daytime sleepiness and hypertension, the upper 60% (quintiles 3-5) of VB was associated with both complications when compared to the bottom 20%. In other words, if a subject has a VB score that’s higher than 40% of normalized controls, they are at greater risk for both hypertension and sleepiness. For sleepiness, the same significant association was found only in the upper two quintiles for AHI and the highest quintile for HB. For hypertension, a significant association was found once again not only for the upper two AHI quintiles but for the top 3 quintiles for HB as well. Taken together, these results point to the following conclusions:
VB has the most predictive power for sleepiness compared to AHI and HB since a smaller deviation from normal was needed to predict this complication;
HB and VB are both effective predictors for hypertension, while AHI once again fails in this regard
In discovering each model’s associations with CVD and all-cause mortality, Dr. Parekh and his team built upon and revealed findings consistent with other results. In all cases and paradigms (unadjusted vs adjusted, all-cause vs CVD mortality), AHI alone exhibited the weakest concordance statistic (as a measure of goodness of fit) compared to the other models, with adjusted concordances of 0.77 and 0.83 for all-cause and CVD mortality, respectively. HB and VB, each by themselves, improved upon AHI’s predictions and had adjusted concordances of 0.78 and 0.79, respectively, for all-cause mortality and 0.85 and 0.85, respectively, for CVD mortality, with each metric significantly different than Model 1 at p-values corresponding to 95% confidence. Model 4, which included both VB and HB calculations, had the greatest concordance across both all-cause and CVD mortality and was a significantly better fit than any of the other models (AHI, HB alone, and VB alone).
This recent work by Dr. Parekh and team is undoubtedly fascinating, and it is easy to see why it has sparked intense discussions and interest across the sleep and biomarker research communities. VB’s plausibility as an effective marker for OSA only increases when viewed through the lens that Dr. Mullington’s team presents. First, it serves as a sensitive metric that relays information about the severity of OSA. Whereas AHI fails to capture any dimensional information about respiratory events, VB incorporates this information into its marker with very low variability across nights and types of studies. In fact, researchers noted that REM-predominant OSA subjects reported a significantly higher VB, hinting at its potential to differentiate between types of OSA, a clinically significant task that has only recently emerged. Second, while the calculation of HB is dependent on the presence of arousals and desaturations, VB can still be calculated during administration of CPAP therapy and is sensitive to therapeutic regimens, pointing towards its utility in quantifying the effectiveness of treatment. Last but not least, predictive models incorporating both VB and HB were significantly associated with both CVD and all-cause mortality, opening the door for discovery into how OSA is related to its complications on a molecular level and supporting the notion that characterizing OSA into its constituent domains is an effective strategy for researching sleep disorders and heterogeneous conditions as a whole.
Looking Forward
The future of biomarkers in sleep and brain research has never been brighter. With the relatively recent advances in Hypoxic Burden, in conjunction with the brand-new biomarker Ventilatory Burden, research into OSA has entered an exciting new frontier. For the first time since man has ever slept, we now have the tools to fully characterize heterogeneous sleep disorders on both molecular and symptomatic levels. Armed with these insights, medicine can create a new paradigm for treating sleep disorders- one that focuses on proactive and personalized modalities of preventive care.
Nonetheless, a major roadblock still lies in the path of these exciting and worthy objectives. Most researchers are aware yet remain confounded by this challenge: the inherent difficulty of democratizing access to these newly-developed biomarkers and their associated software-enabled capture, measurement and analysis tools. The workflows involved in biomarker discovery and validation are computationally intensive, tedious, and difficult to independently replicate. As a consequence, sleep researchers who wish to integrate third-party biomarkers into their pipelines must first overcome a number of daunting administrative, technical, and logistical hurdles. These collective challenges exponentially increase the difficulty of replicating, generalizing,validating and standardizing results. All of this ultimately slows the rate of discovery and impedes the pace of integrating novel biomarkers into an exciting new standard of care.
Enter Neurobit Health (https://www.neurobit.com). We are passionately committed to our mission of democratizing access to AI-enabled solutions for researchers, healthcare practitioners and patients. Neurobit has partnered directly with Dr. Parekh, and hundreds of additional talented researchers operating around the world, to collaborate in the development of these exciting new biomarkers, and to make them available to relevant stakeholders throughout our extensive network via Neurobit Hub (https://www.neurobit.com/press/neurobit-launches-neurobit-hub-a-comprehensive-tool-for-streamlining-sleep-and-population-health-research). Neurobit provides user-friendly access to emerging biomarkers such as Ventilatory Burden and Hypoxic Burden to leading researchers around the world and empowers talented individuals like Dr. Parekh to create all-new biomarkers, to uncover associations with related comorbidities, and allow mankind to develop a comprehensive understanding of sleep, the brain, and by extension, to deliver a powerful new standard of care.
References
Azarbarzin, A., Sands, S. A., Stone, K. L., Taranto-Montemurro, L., Messineo, L., Terrill, P. I., ... & Wellman, A. (2019). The hypoxic burden of sleep apnea predicts cardiovascular disease-related mortality: the Osteoporotic Fractures in Men Study and the Sleep Heart Health Study. European heart journal, 40(14), 1149-1157.
Azarbarzin, A., Sands, S. A., Taranto-Montemurro, L., Vena, D., Sofer, T., Kim, S. W., ... & Redline, S. (2020). The sleep apnea-specific hypoxic burden predicts incident heart failure. Chest, 158(2), 739-750.
Blanchard, M., Gervès-Pinquié, C., Feuilloy, M., Le Vaillant, M., Trzepizur, W., Meslier, N., ... & Gagnadoux, F. (2021). Hypoxic burden and heart rate variability predict stroke incidence in sleep apnoea. European Respiratory Journal, 57(3).
Jackson, C. L., Umesi, C., Gaston, S. A., Azarbarzin, A., Lunyera, J., McGrath, J. A., ... & Redline, S. (2021). Multiple, objectively measured sleep dimensions including hypoxic burden and chronic kidney disease: findings from the Multi-Ethnic Study of Atherosclerosis. Thorax, 76(7), 704-713.
Malhotra A, Ayappa I, Ayas N, Collop N, Kirsch D, Mcardle N, Mehra R, Pack AI, Punjabi N, White DP, Gottlieb DJ. Metrics of sleep apnea severity: beyond the apnea-hypopnea index. Sleep. 2021 Jul 9;44(7):zsab030. doi: 10.1093/sleep/zsab030. PMID: 33693939; PMCID: PMC8271129.
Martinez-Garcia MA, Sánchez-de-la-Torre M, White DP, Azarbarzin A. Hypoxic Burden in Obstructive Sleep Apnea: Present and Future. Arch Bronconeumol. 2023 Jan;59(1):36-43. English, Spanish. doi: 10.1016/j.arbres.2022.08.005. Epub 2022 Sep 5. PMID: 36115739.
Mullington JM, Abbott SM, Carroll JE, Davis CJ, Dijk DJ, Dinges DF, Gehrman PR, Ginsburg GS, Gozal D, Haack M, Lim DC, Macrea M, Pack AI, Plante DT, Teske JA, Zee PC. Developing Biomarker Arrays Predicting Sleep and Circadian-Coupled Risks to Health. Sleep. 2016 Apr 1;39(4):727-36. doi: 10.5665/sleep.5616. PMID: 26951388; PMCID: PMC4791606.
Parekh A, Kam K, Wickramaratne S, Tolbert TM, Varga A, Osorio R, Andersen M, de Godoy LBM, Palombini LO, Tufik S, Ayappa I, Rapoport DM. Ventilatory Burden as a Measure of Obstructive Sleep Apnea Severity Is Predictive of Cardiovascular and All-Cause Mortality. Am J Respir Crit Care Med. 2023 Sep 12. doi: 10.1164/rccm.202301-0109OC. PMID: 37698405.
Revolutionizing data management & analysis in sleep health and population health research.
Clinically validated, fully featured, AASM compliant, Cloud-AI powered automatic sleep scoring system.
Sleep Research
Study Emphasizes Prioritizing Sleep Regularity Over Quantity or Quality for Impact on OSA and Hypertension Management
Sleep Research
Breathing Life into Memories: The Role of Respiration in Sleep-Induced Memory Consolidation
Sleep Research
The Future of Pediatric Cardiology: The Emergence of Smart Wearables in Cardiac Monitoring and Arrhythmia Detection
Sleep Research
Impact of Sleep on Athletic Performance
Sleep Research
Overactive Bladder and Sleep Deprivation: An Intricate Bidirectional Linkage
Sleep Research