The distinction is not academic hair-splitting; it reflects two fundamentally different failures of the respiratory system. In obstructive sleep apnea, the sleeper’s airway physically collapses even as the brain keeps driving the breathing muscles—the effort continues, but airflow falters against a blocked conduit. In central sleep apnea, the problem lies upstream: neural output from the brainstem’s respiratory centers wanes, and airflow falls because the drive to breathe itself has been withdrawn. Central events arise from instability in the ventilatory control system, often quantified by an engineering concept known as loop gain. A system with high loop gain overreacts to fluctuations in carbon dioxide, overshooting corrections and oscillating into periodic breathing, whereas low loop gain indicates a stable control system resistant to such instability. Both obstructive and central apnea patients can exhibit elevated loop gain, and the degree of abnormality may predict how a given patient responds to different treatments.

While scoring rules for frank apneas—complete pauses in breathing—are well agreed upon, the AASM scoring manual treats classification of hypopneas as optional, and most sleep centers simply do not perform it. The panel argues this default has real consequences. In a large French cohort of more than 2,000 people with sleep apnea, the prevalence of central sleep apnea jumped from 5 percent to nearly 20 percent once hypopneas, in addition to apneas, were manually classified. That fourfold increase suggests a substantial population of patients whose underlying central ventilatory instability has been invisible to their diagnoses because their shallow breaths were assumed to be obstructive.

How, then, can a scorer tell the two event types apart on a routine polysomnogram? The current AASM criteria list snoring, flattening of the nasal pressure waveform, and thoracoabdominal desynchrony—out-of-phase motion of the chest and abdominal belts—as hallmarks of obstruction; when none of these features is present, the event should be scored as central. Winfried Randerath and colleagues developed a more elaborate algorithm, validated against esophageal manometry, the gold standard for measuring respiratory effort. Their stepwise approach adds the shape of the recovery breaths that follow an event, the position of the arousal relative to the recovery breaths, and the sleep stage in which the event occurs. Against esophageal pressure measurements, the algorithm correctly identified 60.5 percent of obstructive hypopneas and 76.9 percent of central hypopneas, for an overall accuracy of 68 percent. The two most informative signals in both frameworks are inspiratory flow flattening and respiratory paradox, with published sensitivities and specificities of 0.65 and 0.67 for flow flattening and 0.81 and 0.46 for paradox, underscoring that no single channel suffices and careful integration of multiple signals is essential.

Further physiological markers have been proposed by Javaheri and colleagues. Synchronous changes in airflow and effort, snoring during the recovery breaths rather than during the event itself, and a symmetric desaturation-resaturation pattern on the oximetry trace point toward central disease, whereas snoring during the event, arousals at event termination, and progressive prolongation of the inspiratory duty cycle—the fraction of the breath devoted to inspiration—characterize obstruction. That last feature has been confirmed experimentally in studies using inspiratory resistive loading, which mimics airway obstruction and reliably lengthens inspiratory time. A head-to-head comparison by Dupuy-McCauley and colleagues found the AASM and Randerath methods performed similarly, at 67 to 69 percent accuracy, but with only fair interrater reliability, a kappa of 0.30—meaning two trained scorers often disagree. Signal quality, particularly the notoriously variable nasal pressure signal, is a likely culprit.

To move the field forward, the panel proposes an operational definition: a central hypopnea shows no snoring during the event, waxing-and-waning airflow during recovery breaths, waxing-and-waning effort in the belts, and an arousal occurring in the middle of the recovery breaths rather than at their end. Just as importantly, it proposes a new category, the unclassified or indeterminate hypopnea, for events that meet no strict criteria, rather than forcing them into the obstructive bucket. Under this framework, a sleep report would include the overall apnea-hypopnea index along with separate obstructive, central, and unclassified indices that sum to the total. The panel stresses this is a provisional construct, not a guideline, and requires prospective validation.

Why does this matter now? Because sleep medicine is entering an era of mechanistically targeted therapies, and each therapy aims at a different piece of the physiology. Transvenous phrenic nerve stimulation, which paces the diaphragm to stabilize breathing, works on central ventilatory instability; hypoglossal nerve stimulation, which protrudes the tongue to hold the airway open, works on obstruction. In the pivotal trial of phrenic nerve stimulation, the likelihood of achieving at least a 50 percent reduction in the apnea-hypopnea index rose steadily with the baseline proportion of central events—from 37.5 percent when fewer than half of events were central to 76.5 percent when 90 percent or more were central. When hypopneas were classified retrospectively in this cohort, most residual events after implantation were found to be obstructive, and intriguingly, the proportion of central apneas did not correlate with the proportion of central hypopneas, meaning the two event types must be characterized independently rather than assumed to travel together. The same logic applies in reverse for hypoglossal nerve stimulation: a patient with a heavy load of unrecognized central hypopneas is a candidate for a disappointing result.

The stakes extend to routine positive airway pressure care. The AASM’s recent clinical practice guideline recommends adaptive servo-ventilation for central sleep apnea in patients with normal heart function when CPAP fails; if central events have been correctly enumerated from the outset, prolonged, fruitless trials of CPAP could potentially be avoided. The panel also draws attention to treatment-emergent central sleep apnea, which arises in roughly 4 to 19 percent of obstructive sleep apnea patients during CPAP initiation and persists in about one-third of those affected after months of therapy. Distinguishing pre-existing, treatment-resistant central disease from events that genuinely emerge after CPAP initiation matters for prognosis and management: emergent events are often transient and warrant re-evaluation after a few weeks, whereas persistent disease may respond to adaptive servo-ventilation, supplemental oxygen, or the carbonic anhydrase inhibitor acetazolamide. Longitudinal vigilance is essential, since central apnea can develop years into CPAP therapy with the onset of atrial fibrillation or left ventricular dysfunction, or with the use of opioids, baclofen, or ticagrelor—exposures that should prompt a fresh look at the pattern of breathing disturbances.

The panel is candid about the obstacles. Manual classification adds time and training demands for sleep technologists, and visual assessment suffers from inconsistent interrater reliability across centers. Overclassifying central hypopneas risks inappropriate therapy escalation and added cost; underclassifying them risks residual symptoms and missed combination therapy. The solution, many panelists believe, lies in automation. Several developmental pathways were outlined: event-level analysis extracting features from airflow, effort belts, diaphragmatic EMG, snoring, EEG, and heart rate signals; subject-level metrics such as central and obstructive indices that exploit the characteristic low cycle-length variability and sinusoidal airflow of central apnea, using tools like wavelet decomposition, cardiopulmonary coupling, and machine learning; multimodal models incorporating medications, comorbidities, and demographics; counterfactual prediction to resolve ambiguous events; and approaches that directly model therapeutic response. All of these depend on a prerequisite the panel emphasizes: harmonized, minimum signal-quality standards across laboratories, because algorithms trained on noisy, inconsistent signals will not generalize. With validation studies and human oversight, the panel concludes, routine hypopnea classification could become the foundation of a genuinely precision-based approach to sleep-disordered breathing, matching each patient’s endophenotype to the therapy most likely to quiet their nights and restore their days.

Subject of Research: Classification of obstructive versus central hypopneas in sleep-disordered breathing diagnosis and therapy selection

Article Title: The importance of classifying central hypopneas: a call to action

Article References: Ahn, A. P., Azarbarzin, A., Badr, M. S., Berry, R., DeYoung, P., Dupuy-McCauley, K., Morgenthaler, T. I., Pépin, J. L., Randerath, W., Sands, S., Tallavajhula, S., & Malhotra, A. (2026). The importance of classifying central hypopneas: a call to action. Journal of Clinical Sleep Medicine, 22(1), Article 149. https://doi.org/10.1007/s44470-026-00177-6

Image Credits: AI Generated

DOI: 10.1007/s44470-026-00177-6

Keywords: central sleep apnea, obstructive sleep apnea, hypopnea classification, polysomnography, loop gain, phrenic nerve stimulation, hypoglossal nerve stimulation, adaptive servo-ventilation, CPAP, treatment-emergent central sleep apnea, machine learning, sleep medicine