By Dr. Vishal Saini, M.D., FAASM — Mid-West Center for Sleep Disorders
A patient handed me a glossy one-page report from a home sleep test. It gave him an apnea-hypopnea index of 4.2, a "sleep quality" score in the green, a pie chart of his sleep stages, and a reassuring sentence saying he was unlikely to have significant sleep apnea. He'd stopped worrying. The problem was that everything else about him — the loud snoring his wife described, the witnessed pauses, the morning headaches, the resistant hypertension — told a different story. And the device that generated that tidy report had never actually measured his breathing.
This is the tension at the heart of the modern home-sleep-testing boom. Two technologies in particular — peripheral arterial tonometry (PAT) and cardiopulmonary coupling (CPC) — have made testing dramatically more accessible, and I use and value them. But they belong to a fundamentally different category than the in-lab sleep study, and the confident, quantified reports they produce encourage clinicians and patients to draw conclusions the underlying signal cannot bear. This isn't an argument against the devices. It's an argument for reading them with the skepticism their design demands.
What These Devices Actually Measure — and What They Don't
To use these tools well, you have to understand what's really being recorded, because it's not what most people assume.
A full in-laboratory polysomnogram (PSG) — the gold standard — directly measures the physiology of sleep and breathing on roughly two dozen channels: brain waves (EEG), eye movements, and chin muscle tone to score sleep stages and arousals; airflow through both a thermistor and a nasal pressure transducer; respiratory effort through belts around the chest and abdomen; oxygen saturation; ECG; leg movements; body position; snoring; and video, all with a technologist in the building. When a PSG says you had an obstructive event, it's because it watched your airflow stop while your effort continued.
PAT devices (the WatchPAT family is the common example) measure something quite different: the tone in the small arteries of your fingertip. When a breathing event ends in a surge of sympathetic nervous system activity, those arteries constrict — and the device infers a respiratory event from that autonomic signature, combined with pulse-rate changes, oximetry, and wrist actigraphy. It estimates your sleep and even your sleep stages from the same autonomic and movement signals.
CPC devices (SleepImage is the prominent one) go further into the derived world. They take an ECG or a pulse (PPG) signal, extract heart-rate variability and a respiration signal riding on top of it, and analyze the coupling between the two to produce a "sleep spectrogram." From that, they generate proprietary metrics of sleep quality and a respiratory-event estimate.
Here is the single most important sentence in this article: PAT and CPC do not measure airflow, respiratory effort, or brain activity. They measure the body's downstream autonomic reactions and work backward to guess what your breathing and sleep were doing. A recent authoritative review in the Journal of Clinical Sleep Medicine makes exactly this taxonomic point — that these newer tools sit in a "sympathetic activation" category, distinct from the airflow and respiratory-effort signals that classic sleep testing was built on, and that they don't map cleanly onto our existing frameworks (Chiang et al., J Clin Sleep Med, 2026, PMID 42192042).
"PSG watches your breathing. These devices watch your nervous system's reaction to your breathing and infer the rest. That inference is often good — but it is an inference, and a report that hides the inference behind a crisp number invites you to forget it's there." — Dr. Vishal Saini
To Be Fair: The Validation Data Is Real
I want to be scrupulously fair, because these tools have earned a legitimate place. The population-level validation is genuinely good.
A meta-analysis of 14 studies and 909 patients found that PAT-derived respiratory indices correlated highly with PSG — a correlation coefficient of about 0.89 for the apnea-hypopnea index and 0.94 for the oxygen desaturation index — and concluded PAT is a viable alternative for confirming clinically suspected sleep apnea (Yalamanchali et al., JAMA Otolaryngology–Head & Neck Surgery, 2013, PMID 24158564). CPC has a comparable track record: a 2023 study of the SleepImage ring against simultaneous PSG reported a strong correlation (r = 0.89) and excellent diagnostic accuracy at standard severity thresholds (Lu et al., Biomedical Engineering Letters, 2023, PMID 37519866), and a 2026 multicenter study of 479 patients found the CPC respiratory-event index correlated with the PSG AHI at ρ = 0.889 with an intraclass correlation of 0.955 (Sun et al., Nature and Science of Sleep, 2026, PMID 42371563).
Those are strong numbers. For the right patient — an otherwise healthy adult with a high pretest probability of moderate-to-severe obstructive sleep apnea — these devices reliably confirm the diagnosis, expand access, and cost a fraction of a lab study. That's a real public-health win, and nothing that follows should be read as dismissing it.
But notice two things about even that favorable evidence. First, the studies validate the technology in patients with clinically suspected sleep apnea — an enriched, high-probability population — which is exactly where any test performs best. Second, and more importantly, the headline statistic is almost always a correlation.
Correlation Is Not Agreement — and That's Where the Misreading Starts
This is the methodological heart of the problem, so let me be precise about it.
A correlation coefficient tells you whether two measurements rise and fall together across a group. It does not tell you whether they agree for an individual. Two methods can correlate beautifully at r = 0.89 while still disagreeing by a wide margin in any given person — the device reads a little high here, a little low there, and across a population those errors average out into a gorgeous scatterplot. The statistic that actually captures individual-level trustworthiness is the Bland-Altman limit of agreement, which asks: for a single patient, how far off could this number be? That figure is far less flattering, and it rarely makes it onto the marketing one-pager or even into the abstract.
Why does this matter clinically? Because we don't treat populations; we treat the person in front of us, and we make categorical decisions — no apnea, mild, moderate, severe — at sharp numeric cutoffs. A device that tracks the group trend at r = 0.89 can still push an individual across a severity boundary in either direction: calling a true moderate a "mild," or a true mild a "normal." The patient at the top of this article is precisely that scenario — a real disease diluted into a green-zone number by a surrogate measurement, then over-interpreted as a clean bill of health.
"The reassuring thing about a high correlation coefficient is also the misleading thing: it's a statement about the crowd, not about you. An individual diagnosis rests on agreement, and agreement is the number these reports almost never lead with." — Dr. Vishal Saini
Even the more enthusiastic CPC studies quietly concede the point in their framing. A validation against manually scored PSG reported 100% sensitivity but only 81% specificity for moderate-to-severe disease (Hilmisson et al., Sleep & Breathing, 2019, PMID 29808290) — meaning a meaningful minority of people flagged as abnormal weren't, and vice versa — and the large 2026 multicenter study explicitly described CPC as a "simplified adjunctive tool," reporting a respiratory-event index, not a scored AHI (Sun et al., 2026, PMID 42371563). "Adjunctive" is the operative word, and it tends to get lost between the study and the sales deck.
Four Specific Ways the Conclusions Outrun the Data
Beyond the correlation-versus-agreement issue, there are four recurring over-reads I see in practice.
The "sleep stages" and "sleep quality" scores are not what they appear. Because these devices have no EEG, they cannot score sleep the way a lab does. The pie chart of light/deep/REM and the "sleep quality index" are proprietary algorithmic estimates derived from autonomic and movement signals — not AASM-scored stages, and not a measure of cortical arousals or respiratory-effort-related arousals. A patient telling me his device says he gets "18% deep sleep" is quoting a number that does not mean what the same phrase means on a polysomnogram. It's an estimate wearing the costume of a measurement.
They struggle to say what kind of event it was. A surrogate autonomic signal is not well suited to distinguishing obstructive from central apneas, to detecting hypoventilation, or to characterizing positional and REM-related disease with lab-grade fidelity. This is not a small distinction — central sleep apnea, for instance, demands a completely different workup and treatment, and it's exactly the kind of thing a "sympathetic activation" signal can miss or mislabel (Chiang et al., 2026, PMID 42192042).
The whole signal chain depends on a clean autonomic and cardiac rhythm — which many patients don't have. PAT reads sympathetic vascular tone; CPC is built on heart-rate variability. Anything that scrambles those signals degrades the result: atrial fibrillation and frequent ectopy corrupt the heart-rate variability that CPC depends on; autonomic neuropathy (common in diabetes), heart failure, and advanced age blunt the very autonomic reactivity PAT is reading; and medications that act on the cardiovascular system — beta-blockers, alpha-blockers, and short-acting nitrates among them — can distort the signal these devices interpret. The cruel irony is that these confounders cluster in older, sicker, more medicated patients — precisely the people in whom missing sleep apnea carries the highest stakes.
A single unattended night is a single unattended night. No technologist is there to reposition a sensor, sleep-disordered breathing varies from night to night, and a lost signal at home becomes a gap the algorithm fills rather than a problem someone fixes. The convenient report can look complete while resting on thinner data than it appears.
How to Confirm What These Devices Find
None of this means don't use them. It means use them for what they're validated to do, and confirm rather than conclude when the situation calls for it. A few working rules:
Match the test to the pretest probability. PAT and CPC are appropriate for uncomplicated adults with a high probability of moderate-to-severe obstructive sleep apnea — the same population in which the American Academy of Sleep Medicine endorses home testing at all (Kapur et al., J Clin Sleep Med, 2017, PMID 28162150). They are not the right first test for someone with significant heart or lung disease, neuromuscular disease, chronic opioid use, suspected central sleep apnea or hypoventilation, or a suspected non-respiratory sleep disorder like periodic limb movement disorder, narcolepsy, or a parasomnia.
Treat a normal result in a symptomatic patient as a question, not an answer. The AASM guideline is explicit that a negative, technically inadequate, or inconclusive home test in a patient with clinical suspicion of OSA should be followed by in-laboratory polysomnography (Kapur et al., 2017, PMID 28162150). If the story and the number disagree, believe the story and escalate. That single rule would have protected the patient I opened with.
Read the report for what it is. Treat the device's AHI or "respiratory event index" as a screening estimate, not a scored AHI. Be openly skeptical of the sleep-stage and sleep-quality outputs. And if your patient has an arrhythmia, autonomic disease, or is on cardiovascular medications that confound the signal, don't lean on an autonomic-based device for a definitive answer.
Go to the gold standard when the stakes or the complexity are high. In-lab PSG remains the reference precisely because it measures what PAT and CPC infer: it stages sleep with EEG, scores arousals, separates obstructive from central events by comparing airflow against effort, and does it all under a technologist's eye. When you need to know rather than estimate — a discordant home test, a complex or comorbid patient, suspected central disease, a surgical or high-cardiovascular-risk decision hanging on the result — that direct measurement is worth the extra effort.
What the Field Still Needs to Sort Out
In fairness to the technology, several open questions are the field's to answer, not the individual clinician's. We need more validation reported as limits of agreement rather than correlation, and more of it in unselected and complex populations rather than enriched high-probability cohorts. We need clearer standards for how derived "sleep stage" and "sleep quality" outputs should — or shouldn't — be reported to patients, because right now they carry more authority than the signal justifies. And as that 2026 JCSM review argues, our entire taxonomy for classifying home sleep tests may need rethinking, because these surrogate-signal devices don't fit the categories we built around airflow and effort (Chiang et al., 2026, PMID 42192042). These are solvable problems, and the answer is better evidence and better reporting — not abandoning tools that have genuinely expanded access to care.
The Bottom Line
Peripheral arterial tonometry and cardiopulmonary coupling are useful, validated, and here to stay, and for the right patient they confirm sleep apnea reliably and cheaply. But they measure breathing indirectly, through the nervous system, and their polished reports invite conclusions the surrogate signal can't support: individual-level precision from population-level correlations, "sleep stages" without EEG, definitive answers in patients whose physiology or medications corrupt the signal. Used as a screening tool in the right patient and confirmed with polysomnography when the story or the stakes demand it, they're an asset. Used as a substitute for measurement — a green number treated as a clean bill of health — they're a confident way to miss the diagnosis. Correlation is not agreement, and a derived index is not a diagnosis.
Dr. Vishal Saini, M.D., FAASM is the Research & Medical Director at Mid-West Center for Sleep Disorders and Principal Investigator on multiple clinical trials in sleep medicine across Michigan. He evaluates and treats sleep apnea, insomnia, narcolepsy, and complex hypersomnia disorders in Lansing, Traverse City, and Eaton Rapids.
Got a home sleep test result that doesn't match how you feel? Let's confirm it properly.
Book a Consultation → | (517) 887-6733
References: Kapur VK, Auckley DH, Chowdhuri S, et al. Clinical practice guideline for diagnostic testing for adult OSA: an AASM clinical practice guideline. J Clin Sleep Med 2017 (PMID 28162150); Yalamanchali S, Farajian V, Hamilton C, et al. Diagnosis of obstructive sleep apnea by peripheral arterial tonometry: meta-analysis. JAMA Otolaryngol Head Neck Surg 2013 (PMID 24158564); Lu M, Brenzinger L, Rosenblum L, et al. Comparative study of the SleepImage ring device and polysomnography for diagnosing obstructive sleep apnea. Biomed Eng Lett 2023 (PMID 37519866); Sun CY, Hang LW, Chang ET, et al. The effectiveness of a novel sleep test based on cardiopulmonary coupling to detect obstructive sleep apnea: a multicenter prospective study. Nat Sci Sleep 2026 (PMID 42371563); Hilmisson H, Lange N, Duntley SP. Sleep apnea detection: accuracy of using automated ECG analysis compared to manually scored polysomnography. Sleep Breath 2019 (PMID 29808290); Chiang AA, Lee-Iannotti J, Torstrick B, Berry RB, Collop NA. From classic to cutting-edge: technological approaches to respiratory physiological signals in assessing sleep-disordered breathing. J Clin Sleep Med 2026 (PMID 42192042).