INTRODUCTION
Obstructive sleep apnea (OSA) is characterized by repeated upper-airway collapse during sleep, leading to intermittent hypoxemia, hypercapnia, significant negative intrathoracic pressure swings, sleep fragmentation, and surges in sympathetic activity [1–5]. OSA is common, with prevalence increasing with age and obesity, and is closely linked to modern cardiometabolic diseases [1–4]. Consequently, it is widely accepted that untreated moderate-to-severe OSA is associated with higher rates of hypertension, atrial fibrillation, stroke, coronary artery disease, heart failure, and mortality [6–10].
However, a series of studies conducted in inpatient and critical-care settings has presented a seemingly counterintuitive finding. In several cohorts, patients diagnosed with OSA exhibited lower in-hospital mortality compared to apparently similar patients without such a diagnosis [11–17]. This inverse association has also been observed in specific populations experiencing acute myocardial infarction, pulmonary embolism, and general critical illness [13–18]. This phenomenon has been termed the “OSA paradox” or “sleep apnea survival paradox.”
A paradoxical association does not inherently prove a paradoxical biological mechanism. OSA differs from many single-exposure disorders because the actual disease process, the act of diagnosis, the decision to treat, and the healthcare system context in which it is documented can all vary significantly. For instance, a patient with severe but undiagnosed OSA might be coded as “no OSA” in administrative data, while a patient with milder but recognized disease might receive closer follow-up, increased inpatient monitoring, and positive airway pressure (PAP) therapy. Therefore, this apparent paradox might stem from a discrepancy between physiological reality and epidemiological classification, rather than an actual survival advantage conferred by OSA itself.
This review will thoroughly examine this mismatch. We will begin by summarizing the evidence that supports reduced short-term mortality in diagnosed OSA. We will then compare this with literature demonstrating increased long-term risk, explore plausible biological mechanisms that might offer partial protection in acute scenarios, and finally, focus on the methodological reasons why the paradox consistently appears in administrative and retrospective studies. The crucial question is not simply whether the paradox exists, but what its true implications are.
ESTABLISHED EVIDENCE THAT OSA IS HARMFUL OVER THE LONG TERM
Any review of the paradox must begin with the non-paradoxical literature. Classic cohort studies have consistently demonstrated that severe OSA is associated with adverse cardiovascular outcomes and increased mortality [6–10].
For example, Marin and colleagues [6] reported a significantly increased risk of fatal and non-fatal cardiovascular events in men with untreated severe OSA-hypopnea, while patients treated with CPAP showed lower event rates. Analyzing prospective cohort data, Punjabi and colleagues [7] found that sleep-disordered breathing was linked to all-cause mortality and coronary mortality, particularly in middle-aged men with more severe disease. Similarly, Yaggi et al. [8] showed that OSA independently increased the risk of stroke or death, even after accounting for other major risk factors. Gami et al. [9] further observed that sudden cardiac death in OSA patients predominantly occurred between midnight and 6 am. This temporal shift is consistent with a biologically active arrhythmic substrate rather than OSA being a benign comorbidity.
These studies collectively support a coherent pathophysiological explanation: Repetitive airway obstruction leads to cycles of oxygen desaturation and reoxygenation, which promote oxidative stress, endothelial dysfunction, inflammation, thrombogenicity, and neurohumoral activation [4,5]. The cumulative burden of these processes makes it entirely plausible that OSA worsens long-term cardiovascular outcomes. This holds true even if, under certain circumstances, adaptive responses to chronic intermittent hypoxia (CIH) might transiently modify short-term resilience.
The distinction between long-term hazard and short-term outcomes is crucial. A disease can be harmful overall yet still exhibit condition-specific inverse associations within narrowly defined windows, particularly when diagnosis or treatment alters the composition of the observed cohort. This conceptual space is precisely where the OSA paradox resides.
STUDIES REPORTING LOWER SHORT-TERM MORTALITY IN DIAGNOSED OSA
Modern literature on the “paradox” of OSA is primarily based on retrospective observational studies (Table 1). For instance, Mohananey et al. [11] observed a decrease in in-hospital mortality among cardiovascular admissions with recognized OSA in a large administrative dataset. Moser and colleagues [12] explicitly labeled this an “in-hospital survival paradox,” noting lower hospital mortality in sleep apnea patients despite significant comorbidities and a case mix dominated by heart failure, sepsis, and respiratory failure. Similarly, Lin et al. [13] found lower 28-day, 90-day, intensive care unit (ICU), and in-hospital mortality in critically ill adults with OSA from MIMIC-III, even after propensity-score matching. Wang et al. [14] also reported reduced 30-day, ICU, and in-hospital mortality among ICU patients with comorbid sleep apnea-hypopnea syndrome.
Similar trends are evident in disease-specific cohorts. In acute pulmonary embolism, Joshi et al. [15] found lower inpatient mortality in OSA-positive patients, despite a higher prevalence of conventional mortality risk factors. Conversely, Seckin et al. [16] did not find higher pulmonary embolism mortality in OSA patients but suggested that treatment adherence might influence recurrence risk. In acute myocardial infarction, Sharafkhaneh et al. [17] proposed that the well-known obesity paradox might be partly mediated by coexisting sleep apnea and its CIH phenotype. While not all acute-care studies reach the same conclusion, and effect sizes vary, the recurring pattern is too significant to be dismissed as an isolated anomaly.
Three striking features characterize this body of literature. First, most paradox studies define OSA by prior diagnostic coding rather than by polysomnography performed at the time of the index event. Second, the paradox is most pronounced in short-term endpoints, such as ICU, in-hospital, 28-day, or 90-day mortality, rather than in long-term community follow-up. Third, the inverse association often persists even after multivariable adjustment or propensity methods, suggesting that simple confounding by one or two measured variables alone is an unlikely explanation. A more plausible issue is the systematic mismeasurement of the exposure itself and an incomplete capture of the clinical pathways associated with having a recognized diagnosis.
COULD THE PARADOX REFLECT REAL BIOLOGY?
Experimental evidence suggests that the biological effects of CIH are highly dependent on its pattern, severity, and duration. Mild-to-moderate CIH may activate adaptive pathways similar to ischemic preconditioning, involving hypoxia-inducible factor signaling, angiogenesis, mitochondrial stress adaptation, and enhanced antioxidant responses. Conversely, more severe or prolonged CIH appears to promote oxidative stress, endothelial dysfunction, sympathetic overactivation, systemic inflammation, and adverse cardiovascular remodeling. This biphasic framework might explain why CIH could theoretically offer limited shortterm adaptive benefits in specific acute situations while still contributing to long-term cardiovascular damage overall. Some clinical studies in acute myocardial infarction have also indicated smaller infarct size or lower biomarker release in select patients with OSA, though these findings are inconsistent and insufficient to establish a definitive cardioprotective effect [18,19].
A genuine biological component cannot be entirely dismissed. Ischemic or hypoxic preconditioning is the most frequently cited mechanism. CIH, the physiological hallmark of OSA, may activate adaptive signaling cascades such as hypoxia-inducible factor pathways, alter mitochondrial handling of oxidative stress, and trigger angiogenic responses [18,19]. Theoretically, patients exposed to recurrent nocturnal hypoxia could develop partial tolerance to subsequent ischemia-reperfusion injury. This concept is intuitively appealing in acute myocardial infarction and perhaps in critical illness, where brief stress adaptation might be more critical than cumulative vascular injury.
The obesity paradox is a second biological candidate. OSA is strongly associated with obesity, and overweight patients with acute cardiovascular disease sometimes exhibit lower short-term mortality than leaner counterparts with the same diagnosis [17,20]. Proposed explanations include greater metabolic reserve, altered cytokine environments, and earlier presentation for care. If diagnosed OSA disproportionately identifies obese patients who retain nutritional reserve or are hospitalized earlier in their decompensation trajectory, then part of the paradox could be genuinely biological, even if not directly attributable to upper-airway collapse itself.
The right ventricular adaptation hypothesis is particularly relevant in acute pulmonary embolism. Theoretically, repeated nocturnal increases in pulmonary artery pressure and transient right ventricular pressure overload during obstructive respiratory events could induce partial adaptation to acute hemodynamic stress. This has been proposed as a possible explanation for the relatively favorable short-term outcomes observed in some pulmonary embolism cohorts with recognized OSA. However, direct human evidence remains limited, and it is unclear whether any adaptive benefit outweighs the long-term adverse cardiopulmonary consequences of chronic OSA.
A third possibility is cardiopulmonary adaptation. Recurrent surges in pulmonary artery pressure and right ventricular afterload during obstructive events may induce partial adaptation that becomes relevant in disorders such as acute pulmonary embolism [15,16]. Similarly, repeated autonomic activation might alter the hemodynamic response to acute stress. Yet these explanations remain speculative. The mechanistic evidence is indirect, and the same chronic exposures that might precondition tissue can also worsen endothelial function, ventricular remodeling, and arrhythmogenesis over time [4,5,9].
The strongest counterargument to a purely biological paradox is inconsistency across contexts. PAP trials and secondary-prevention studies do not show a robust cardioprotective signal strong enough to suggest that OSA itself is beneficial. For example, in the SAVE trial, CPAP added to usual care did not significantly reduce major cardiovascular events in the intention-to-treat analysis, although adherence was modest and symptomatic patients were excluded [21]. Similarly, in ISAACC, CPAP did not reduce recurrent cardiovascular events in non-sleepy patients with acute coronary syndrome and OSA [22]. These neutral results do not prove that OSA is harmless; rather, they highlight the difficulty of extracting a clean causal cardiovascular effect from heterogeneous populations. They also provide little support for the notion that OSA is intrinsically protective.
THE MORE PERSUASIVE EXPLANATION: EPIDEMIOLOGY AND MEASUREMENT
Methodological factors likely contribute substantially to the observed paradox and may offer a more parsimonious explanation than a broad biological protective effect of OSA itself. However, current evidence is insufficient to fully exclude a partial biological contribution in selected clinical settings.
The first issue is exposure misclassification. In many administrative studies, OSA is identified by billing codes or past medical history. Consequently, the comparison group labeled “no OSA” is not truly OSA-free; rather, it is a mixed group comprising patients without OSA and those with unrecognized OSA. Given the significant underdiagnosis of OSA in the general population, contamination of the control group is almost inevitable [1–3]. If diagnosed OSA patients systematically differ from undiagnosed OSA patients, then the coding variable represents recognition rather than physiology.
The second issue is detection or surveillance bias. A recognized OSA diagnosis implies prior healthcare contact, symptom evaluation, or perioperative concern. Such patients may appear to have more comorbidities on paper, but they may also be more likely to be taking antihypertensives, statins, anticoagulants, or diabetes medications. Furthermore, they may be more likely to use PAP therapy and to trigger inpatient monitoring in cases of sedative use, opioid administration, respiratory failure, or cardiovascular instability. In this sense, diagnosed OSA can function as a marker of clinical engagement. The measured exposure is therefore not simply “sleep apnea,” but “sleep apnea that the health system has already noticed.”
Third, substantial residual confounding persists even after applying propensity methods. Crucial variables are frequently missing, including body-mass index, waist circumference, smoking intensity, alcohol use, frailty, socioeconomic status, sleep duration, daytime sleepiness phenotype, hypoxic burden, and PAP adherence. Propensity scores cannot account for unmeasured variables. In fact, a paradox that remains after statistical adjustment in an administrative dataset should often raise concerns about unmeasured confounding, rather than validating a causal effect.
Fourth, the timing of events is critical. Patients diagnosed with OSA may differ from controls in their hospital admission patterns. They might be younger, less frail, or hospitalized earlier in the course of their deterioration. They may also be more likely to survive to hospital admission in the first place, a form of selection bias based on survivorship. Once admitted, they could receive noninvasive ventilation or closer respiratory observation sooner than similar patients whose OSA goes undiagnosed. Thus, the observed paradox might stem from the hospital system itself as much as from the disease.
Finally, administrative coding practices can distort associations. Coding intensity is often higher for patients who survive long enough for comorbidities to be fully captured, while rapidly progressing cases may die before secondary diagnoses are completely recorded. This phenomenon, sometimes called “reverse severity” or coding opportunity bias, has been documented in other paradox literatures and likely applies to OSA. A diagnosis that requires time, detailed history, and documentation may be preferentially recorded in patients stable enough to accumulate such records.
DIAGNOSED OSA IS NOT THE SAME AS PHYSIOLOGICAL OSA
The central conceptual point is simple yet crucial: physiological OSA and diagnosed OSA represent different exposures. Physiological OSA refers to the underlying sleep-breathing disorder, measured by polysomnography, respiratory polygraphy, or validated physiological surrogates. Diagnosed OSA, conversely, is a hybrid construct encompassing disease burden, symptom perception, access to testing, clinician suspicion, coding practices, and prior treatment. Studies using diagnosed OSA as an exposure must be interpreted with this in mind.
This distinction also helps reconcile the paradox within the mainstream long-term risk literature. Cohort studies that directly measure severity (e.g., with apnea-hypopnea index or related metrics) generally demonstrate harm when the disease is sufficiently severe and untreated [6–10]. In contrast, administrative inpatient studies often measure recognition rather than severity and report lower short-term mortality [11–17]. Thus, these two bodies of literature are not contradictory; they address different questions using different proxies.
The problem is further amplified by the inherent heterogeneity within OSA. Modern phenotyping highlights that OSA is not a single disease mechanism but a syndrome resulting from various combinations of upper-airway anatomy, loop gain, arousal threshold, muscle responsiveness, obesity, and cardiometabolic vulnerability [4,23]. A patient with severe hypoxic burden, minimal symptoms, and high cardiovascular vulnerability is not equivalent to a sleepy, obese patient with earlier recognition and good PAP adherence. Grouping both under a single diagnostic code creates conditions ripe for paradoxes.
HOW TREATMENT MODIFIES INTERPRETATION
Treatment exposure further complicates paradox studies. While PAP therapy can reduce symptoms, blood pressure, and some perioperative respiratory complications, adherence varies and is often unmeasured [21,22,24]. Consequently, if a hospital cohort of diagnosed OSA patients includes individuals already on PAP, any observed lower mortality might reflect treatment benefits rather than an intrinsic property of the untreated disease. Conversely, if PAP adherence is poor, an OSA diagnosis based on coding may reveal little about actual therapeutic exposure.
This uncertainty underscores the value of randomized trials, even those with neutral results. Studies like SAVE and ISAACC demonstrate the nuanced nature of the cardiovascular story [21,22]. The absence of a large intention-to-treat cardiovascular benefit from CPAP does not negate its symptomatic and physiological effects, nor does it validate the paradox. Instead, it highlights the heterogeneity in patient selection, event mechanisms, and treatment adherence. For a narrative review on the paradox, the crucial takeaway is that treatment must be considered but not assumed. Most paradox datasets lack sufficient information regarding who was treated, the consistency of treatment, and the disease stage at which it was initiated.
CLINICAL IMPLICATIONS
Clinicians should not conclude that OSA is protective or requires less urgent diagnosis based on some studies reporting lower short-term mortality. The preponderance of evidence still indicates that OSA is a clinically significant disorder impacting symptoms, quality of life, blood pressure, arrhythmia burden, stroke risk, and long-term cardiovascular health [4–10]. This paradox is best understood as a caution regarding study design, highlighting how short-term observational outcome research can be heavily influenced by how a diagnosis is recorded.
A more practical, and almost opposite, implication at the bedside is that a documented OSA diagnosis should prompt careful perioperative and inpatient respiratory management, judicious use of opioids and sedatives, continuation of PAP when appropriate, and thorough reassessment of cardiometabolic risk. From a research perspective, OSA should be treated as a time-varying, severity-graded, and treatment-sensitive exposure, rather than a simple yes/no comorbidity.
RESEARCH PRIORITIES
Three priorities follow. First, future studies should distinguish between physiological severity and diagnostic status. While administrative coding can be useful, it should ideally be linked to polysomnographic metrics such as apnea-hypopnea index, oxygen desaturation burden, and sleep time below 90% saturation. Second, treatment exposure must be measured directly. The relevant variable is device-derived PAP adherence, not merely prescription status. Third, investigators should employ designs focused on causal inference rather than solely on association. Methods like target trial emulation, negative-control analyses, and sensitivity analyses for unmeasured confounding are better suited to paradox-prone questions than progressively complex regression alone.
Additionally, it is important to broaden the scope of outcomes beyond mortality. The paradox may be most pronounced for crude short-term death, whereas other outcomes—such as respiratory failure, length of stay, readmission, recurrent thrombosis, delirium, or functional decline—may reveal different patterns. A disease that appears protective for mortality but harmful for other clinically important endpoints is unlikely to be truly protective overall. Finally, phenotype-specific research might reveal that a small subgroup experiences genuine acute adaptive benefits from CIH, while the broader syndrome remains harmful. This would offer a more biologically coherent resolution than an all-purpose protective effect of OSA.
CONCLUSION
The OSA mortality paradox appears to be a real observational signal, though its precise biological and methodological basis remains incompletely understood. Diagnosed OSA can appear protective in short-term inpatient studies, yet physiological OSA remains a disorder with substantial long-term cardiovascular and mortality risk. The most defensible synthesis is that diagnosed OSA is a composite marker, reflecting not only disease biology but also obesity, healthcare engagement, recognition of comorbidity, monitoring intensity, and possible treatment exposure. Once this distinction is made, the paradox becomes less mysterious. OSA is unlikely to protect patients from death in any general sense; rather, retrospective datasets often identify a subgroup of patients whose disease has already been recognized and whose care pathway differs materially from that of ostensibly unaffected controls. The challenge for future research is not to celebrate the paradox, but to dissect it precisely enough that physiology, diagnosis, and treatment are no longer conflated.
PDF Links
PubReader
ePub Link
XML Download
Full text via DOI
Download Citation 
Print 
