1. Introduction
Obstructive sleep apnea (OSA) is a common disease, estimated to affect nearly one billion people worldwide, with its incidence increasing with age [1]. OSA is characterized by partial or complete obstruction of the upper airway, intermittent hypoxia, and frequent arousals from sleep in recurrent episodes [2]. The intermittent hypoxemia caused by frequent apnea and hypopnea during sleep in patients with OSA is accompanied by a series of systemic pathological and physiological changes [3].
It often leads to excessive daytime sleepiness (EDS), which may adversely affect various aspects of daily life, including performance, emotional state, and overall well‐being [4–6]. These pauses in breathing can lead to reductions in arterial blood oxygen saturation (SaO2) and tissue hypoxia. After the breathing pause ceases, reoxygenation occurs.
This cycle drives the evolution of hypoxic oxidative stress (HOS). HOS has been demonstrated to trigger excessive glucose degradation through the glycolytic pathway, resulting in pyruvate accumulation [7]. Lactate dehydrogenase is the enzyme responsible for converting accumulated pyruvate into lactate.
In the hypoxemic state, the lack of ATP significantly impairs hepatic lactate utilization (Cori cycle) and renal cortical gluconeogenesis. Therefore, tissue hypoxia leads to hyperlactatemia and lactic acidosis [8, 9].
A preponderance of earlier research has determined that lactate exhibits the most robust correlation with the mortality rate of critically ill patients [10, 11]. OSA is prevalent among critically ill patients and associated with increased mortality [12]. Moreover, the prevalence of malnutrition among OSA patients is often attributed to the presence of chronic diseases.
A diagnosis of malnutrition is made on the basis of an albumin level of less than 3.5 g/dL. In patients receiving intensive care, low serum albumin levels (less than 3 g/dL) have been demonstrated to be associated with elevated in‐hospital mortality rates in comparison with higher levels, with an odds ratio (OR) of 1.51 [13]. Therefore, lactic acid and albumin may be closely related to the outcomes of patients with OSA.
Recent studies have found that the lactate‐to‐albumin ratio (LAR) is associated with the clinical outcomes of critically ill patients [14], spontaneous subarachnoid hemorrhage [15], and community‐acquired pneumonia [16]. However, the extent to which LAR holds prognostic significance in patients diagnosed with OSA who are admitted to intensive care units (ICUs) remains unclear.
Therefore, in this study, we explored the relationship between LAR and 28‐day mortality rates in ICU patients with OSA, utilizing data from the MIMIC‐IV database. To evaluate its potential clinical utility in distinct patient subsets, we conducted a comprehensive analysis of its predictive power within these defined groups.
2. Methods
2.1. Database
The study cohort was drawn from the MIMIC‐IV database (Version 2.2), which includes anonymous medical records of patients admitted to ICUs at Beth Israel Deaconess Medical Center (2008–2019). The database contains records relating to a variety of factors, including but not limited to demographics, vital signs, clinical laboratory tests, fluid balance data, and information regarding the patient’s overall status. OSA cases were identified using the following ICD codes: ICD‐9‐CM codes 327.23 (OSA) and 780.57 (sleep apnea, unspecified); and ICD‐10‐CM codes G47.33 (OSA [adult]), G47.30 (Sleep apnea, unspecified), and G47.39 (Other sleep apnea).
The critical care team’s manual verification of data from bedside monitoring systems provided reliable, verifiable evidence of patients’ physiological states. One author (Chuntian Chi) successfully completed the CITI Data or Specimens Only Research course and obtained database access certification (certification number 66935858). The research project was conducted in strict accordance with the ethical guidelines stipulated in the Declaration of Helsinki (2013 revision).
Ethical approval for the use of the MIMIC database was granted by the Institutional Review Boards of both the Massachusetts Institute of Technology and Beth Israel Deaconess Medical Center. Both agreements eliminate the necessity for participants to provide explicit consent when engaging in research utilizing the MIMIC‐IV dataset; thus, our study did not require additional ethics committee approval from our own center.
2.2. Study Population
Participants were drawn from the MIMIC‐IV database. From 2008 to 2019, a total of 431,231 patients were registered in the database. We first excluded patients admitted to the ICU more than once (n = 380,311), leaving those with a first ICU admission.
A total of 30,371 patients diagnosed with OSA who received their first ICU treatment between 2008 and 2019 were identified. Subsequently, 29,564 patients with missing lactate or albumin values were excluded, resulting in a final cohort of 807 patients. The missing data pattern was primarily attributable to the fact that lactate and albumin are not measured as part of routine admission panels for every ICU patient; instead, they are ordered at the discretion of the treating clinician, typically for patients with suspected tissue hypoperfusion, metabolic disturbance, or nutritional concerns.
The excluded group had a lower overall severity of illness on average, suggesting that our cohort may overrepresent sicker patients; this potential selection bias is acknowledged in the Limitations section. The screening process is summarized in Figure 1.
2.3. Data Extraction
To facilitate data extraction and management, we employed Structured Query Language (SQL), specifically PostgreSQL Version 9.6, alongside STATA Version 17.0. Information encompassing patient age, gender, pre‐existing health conditions, initial lab results upon ICU admission, first‐day severity of illness (Sequential Organ Failure Assessment [SOFA], Oxford Acute Severity of Illness [OASIS], Simplified Acute Physiology Score II [SAPSII]), mechanical ventilation (MV) status on ICU arrival, duration of hospitalization and ICU stay, and dates of death were extracted. Clinical comorbidities identified by ICD codes included myocardial infarction, congestive heart failure, peripheral vascular disease, diabetes, renal disease, malignant cancer, AIDS, cerebrovascular disease, chronic pulmonary disease, and severe liver disease.
On the first day, the following laboratory tests were recorded: arterial blood gases (pH, PaCO2, PaO2, and base excess), white blood cells, platelets, total bilirubin, hematocrit, hemoglobin, creatinine, glucose, calcium, chloride, potassium, sodium, lactate, and albumin. First‐day vital signs included heart rate, respiratory rate, mean arterial blood pressure, and SpO2. Where patients had multiple measurements for a given variable on the first ICU day, the earliest value was utilized.
2.4. Statistical Analysis
Normally distributed continuous variables are reported as mean ± standard deviation (SD); nonnormally distributed variables are reported as median and interquartile range (IQR). Categorical variables are expressed as frequencies and percentages (%). Differences in continuous variables between groups were assessed using the independent‐samples t‐test or Mann–Whitney U test, as appropriate; categorical variables were compared using the chi‐squared test or Fisher’s exact test.
The primary outcome was 28‐day mortality following ICU admission. To assess the association between LAR and mortality, we employed both univariate and multivariate logistic regression models, reporting OR with 95% confidence intervals (CIs). LAR was analyzed both as a continuous variable (per‐unit increase) and as a dichotomous variable based on the optimal cutoff derived from the receiver operating characteristic (ROC) curve.
For variable selection in the multivariate model, we adopted a two‐step strategy: First, all variables with p < 0.05 in univariate analysis were entered into the initial model; second, clinically important variables (age and gender) were retained regardless of univariate significance to address potential confounding and improve clinical interpretability. Clinically meaningful indicators including SOFA and OASIS scores were retained on the basis of clinical relevance rather than merely statistical significance. Collinearity diagnostics were performed prior to multivariate regression modeling.
Given that LAR was derived from lactate and albumin, raw lactate and albumin values were intentionally excluded from models containing LAR to avoid multicollinearity. This approach balances data‐driven selection with clinical judgment.
ROC curves were constructed to evaluate the discriminatory performance of LAR, SOFA, and OASIS for predicting 28‐day mortality. The area under the ROC curve (AUC) with 95% CI was calculated for each indicator. Pairwise comparisons of AUCs were performed using the DeLong test, and the resulting p values indicate whether the discriminatory abilities of the two compared indicators differ statistically.
The optimal cutoff value for LAR was determined using Youden’s index. Based on this threshold, patients were categorized into high‐LAR and low‐LAR groups.
Kaplan–Meier survival curves were generated and compared using the log‐rank test. Subgroup analyses with interaction tests were conducted to explore whether the association between LAR and 28‐day mortality varied across comorbidity strata.
All statistical tests were two‐sided, and a p value < 0.05 was considered statistically significant. For p value reporting, we followed standard conventions: p < 0.001 is reported as “p < 0.001,” p values between 0.001 and 0.01 are reported to three decimal places, p values ≥ 0.01 and < 0.99 are reported to two decimal places, and p > 0.99 is reported as “p > 0.99.” All analyses were performed using R Version 4.2.1 (R Foundation for Statistical Computing, Vienna, Austria) with the following packages: pROC for ROC analysis, survival for Kaplan–Meier and regression models, and rms for regression modeling. The study was conducted in accordance with the STrengthening the Reporting of OBservational studies in Epidemiology (STROBE) guidelines.
3. Results
A total of 807 patients were included in this study. The median age of the study population was 64.8 years. Male subjects accounted for 66% of the total study population.
The 28‐day mortality rate was 20.44% (165 of 807 patients). Compared with survivors, nonsurvivors were older and had higher SOFA and OASIS scores. Additionally, white blood cell count, BUN, creatinine, glucose, and LAR were higher in the deceased group.
Detailed baseline characteristics are presented in Table 1.
3.1. ROC Curve Analysis and Kaplan–Meier Curve
We plotted the ROC curves of three indicators—LAR, SOFA, and OASIS—for predicting 28‐day mortality in ICU patients with OSA, as shown in Figure 2. The AUC of LAR (0.730; 95% CI: 0.686–0.772) was comparable with OASIS (0.736; 95% CI: 0.695–0.776; DeLong test p = 0.361) and superior to SOFA (0.637; 95% CI: 0.591–0.683; DeLong test p = 0.270). Detailed results are listed in Table 3.
3.2. Univariate and Multivariate Logistic Regression
The results of univariate and multivariate logistic regression analyses are presented in Table 4. Variables reaching statistical significance (p < 0.05) in univariate analysis, along with age and gender (retained for clinical relevance), were entered into the multivariate model. Multivariate analysis showed that the high‐LAR group had a higher risk of 28‐day mortality compared with the low‐LAR group (OR: 3.00; 95% CI: 1.83–4.90; p < 0.001).
When LAR was analyzed as a continuous variable (per 1‐unit increase), the adjusted OR was 1.52 (95% CI: 1.27–1.82; p < 0.001), confirming a dose–response relationship between increasing LAR and higher 28‐day mortality.
3.2.1. Subgroup Analysis
Figure 4 shows the results of the stratified analysis for the different subgroups, which include myocardial infarction, congestive heart failure, liver disease, diabetes, renal disease, malignant cancer, cerebrovascular disease, and chronic pulmonary disease. In all subgroups, a high level of LAR was significantly associated with an increased risk of mortality within 28 days (p < 0.05). There were no interactions in any of the subgroups (p value > 0.05).
4. Discussion
This study constitutes the first large‐scale investigation of the association between LAR and 28‐day mortality in ICU patients with OSA, based on the MIMIC‐IV database. The present study demonstrated that LAR was an independent predictor of 28‐day mortality. The 28‐day mortality rate of patients in the high‐LAR group was found to be significantly higher than that of patients in the low‐LAR group.
Despite adjustments for demographics, vital signs, laboratory indicators, and comorbidities, LAR demonstrated a persistent association with mortality (OR = 3.00, 95% CI: 1.83–4.90, p < 0.001). Importantly, this association remained significant when LAR was modeled as a continuous variable (adjusted OR = 1.52 per unit increase, 95% CI: 1.27–1.82, p < 0.001), supporting a graded relationship rather than a threshold effect alone. Furthermore, ROC and Kaplan–Meier curve analyses revealed that the predictive value of LAR for 28‐day mortality in OSA patients was essentially equivalent to that of OASIS and superior to SOFA.
Moreover, subgroup analysis suggested that the association between LAR and 28‐day mortality remained consistent across all examined subgroups.
Previous retrospective studies involving medical, surgical, and mixed ICU settings have reported an OSA prevalence of 7.8% [17]. It is estimated that 38% of the global adult population suffers from moderate to severe OSA [1]. Moreover, OSA is not uncommon in critically ill patients, especially those with acute hypercapnic respiratory failure [18, 19].
It is well established that higher lactate concentrations in critically ill patients correspond to a higher risk of short‐term death [20]. Serum lactate is an indicator of tissue hypoxia. OSA causes hypoxemia and tissue hypoxia due to apneas during sleep; the resulting state of low oxygen delivery leads to hyperlactatemia.
Increased respiratory rate causes respiratory muscle fatigue, further exacerbating lactate elevation [21]. Furthermore, under sleep‐related hypoxemic stress in OSA patients, pyruvate is converted to lactate via anaerobic glycolysis. Notably, plasma lactate levels increase in OSA patients but not in controls, and the magnitude of postsleep lactate elevation correlates with OSA severity and the degree of nocturnal hypoxemia.
Serum lactate has thus been proposed as a marker of oxidative stress in OSA [22].
The prevalence of common comorbidities among ICU patients with OSA is high, including obesity, diabetes, hypertension, ischemic heart disease, heart failure, and cerebrovascular disease [17]. Hypoxemia is regarded as a cardiovascular risk factor that increases cardiac load and induces oxidative stress through the release of reactive oxygen species, thereby damaging vascular endothelial function [23], causing electrophysiological instability in the cardiac conduction system [24], and leading to severe metabolic disorders. Furthermore, OSA is very common among patients with chronic kidney disease (CKD) [25], and both experimental animal models [26, 27] and human studies [28–30] have reported that OSA contributes to CKD pathogenesis through intermittent hypoxia exposure.
One study found that OSA significantly increased the incidence of acute kidney injury (AKI) by 124% [31]; compared with non‐OSA patients, the incidence of AKI was higher in OSA patients (41% vs. 57%, p < 0.001) [19]. Furthermore, previous studies have reported that nocturnal hypoxemia in OSA patients is associated with accelerated renal dysfunction [32].
OSA may cause accelerated renal function decline through multiple direct and indirect mechanisms, including hypoxia, sympathetic nervous system activation, hypertension, inflammation, oxidative stress, and renin–angiotensin–aldosterone system (RAAS) activation [33–36]. In an oxygen‐deficient environment, renal tubular epithelial cells mount inflammatory responses, triggering a cascade of pathological reactions including maladaptive cell repair, metabolic reprogramming, cell death/senescence, and fibrosis [37, 38]. Impaired kidney function can disrupt the body’s water and salt balance; airway congestion caused by excessive fluid accumulation may in turn play a role in the pathogenesis of sleep apnea [39].
Research has identified correlations between OSA and a range of cardiovascular diseases, metabolic disorders, and neurodegenerative diseases [40]. In this context, oxidative stress and inflammation caused by intermittent hypoxia appear to play a crucial role in OSA pathophysiology [41]. Serum albumin is an important antioxidant; structural modifications caused by free radicals can impair its antioxidant properties.
It has been documented that in OSA, a decline in antioxidant capacity exacerbates oxidative stress, potentially precipitating cardiovascular and metabolic abnormalities [42]. Moreover, ischemia‐modified albumin (IMA)—a by‐product of albumin modification by free radicals—has been found to be markedly elevated in patients with severe OSA, and CPAP treatment appears to reverse this increase [43].
Multiple blood cell components and composite cell‐count indices have been identified as inflammatory markers in various inflammatory diseases, including OSA [41, 44–46]. However, only a limited number of studies have evaluated the prognostic ability of these indicators (mainly the neutrophil‐to‐lymphocyte ratio [NLR] and the platelet‐to‐lymphocyte ratio [PLR]), and results have been inconsistent [47, 48]. Therefore, through our investigation of the relationship between the lactate–albumin composite indicator and outcomes in ICU‐admitted OSA patients, we have demonstrated a significant correlation between LAR and OSA prognosis.
This study presents a novel perspective on the risks associated with metabolic disruption in OSA, offering clinically relevant insights. Furthermore, LAR is a readily quantifiable indicator in clinical practice, with established reference ranges. As a composite indicator, LAR has been shown to reflect patient metabolic status more comprehensively than lactate or albumin alone.
Renal insufficiency in OSA patients also reduces acidic metabolite excretion, further exacerbating acidosis. Consequently, LAR can be regarded as a significant prognostic marker for OSA patients, as it captures these pathophysiological changes.
4.1. Strengths and Limitations
Several strengths of this investigation should be acknowledged. First, LAR is a straightforward, noninvasive biomarker that requires minimal training to implement, making it readily applicable in primary healthcare settings. Second, our findings indicate that LAR exhibits moderate predictive value for 28‐day mortality in OSA patients, and the identified threshold may aid in risk stratification and timely intervention.
Several limitations must also be considered. First, as a retrospective post hoc analysis, selection bias and residual confounding are inherent. Notably, a large proportion of OSA patients (29,564 of 30,371) were excluded due to missing lactate or albumin values, which may introduce selection bias because these tests are typically ordered for clinically sicker patients.
We acknowledge that our final cohort of 807 may not fully represent the broader ICU OSA population. Our study cohort mainly consists of patients who had lactate and albumin laboratory tests ordered by clinicians. Therefore, our conclusions cannot be extrapolated to all OSA patients admitted to the ICU.
Second, certain parameters within the MIMIC‐IV database may be incomplete, potentially affecting data reliability. Although we addressed this by excluding variables with high missingness, a definitive solution remains elusive. Third, the study did not explore the influence of treatment strategies on LAR levels or their downstream effects on outcomes.
Fourth, OSA was identified solely via ICD codes, which may not accurately reflect OSA severity or capture all cases, as polysomnography confirmation was not available for all patients. Finally, the present study did not account for the impact of therapeutic interventions on LAR and prognosis. Additionally, this study relies on a single‐center electronic database.
Further external‐cohort validations, prospective multicenter investigations, and evaluations among non‐ICU patients with OSA are required to confirm the generalizability of our findings. Future research should further explore the potential value of LAR in guiding individualized treatment.
5. Conclusion
The results of this study indicate that LAR is significantly associated with 28‐day mortality in ICU patients with OSA. The underlying mechanism aligns with the physiological and metabolic processes of OSA. LAR demonstrates efficacy comparable to OASIS and superior to SOFA in predicting 28‐day mortality in patients diagnosed with OSA.
Both categorical and continuous LAR analyses consistently support its independent prognostic value. As an easily obtainable and monitorable biomarker, LAR represents an important predictor of 28‐day mortality in this population.
Author Contributions
Chuntian Chi had full access to all of the data in this study and takes complete responsibility for the integrity of the data and the accuracy of the data analysis. Jian Zhou: conceptualization, data curation, and original draft writing. Chuntian Chi: study design, statistical analysis, manuscript revision, and corresponding author.
Junpei Wu, Fanghao Yu, and Hanjian Lou: data extraction, literature review, and manuscript editing. Jian Zhou and Chuntian Chi are the co‐first authors, who contributed equally to all experimental design, data collection, and manuscript writing.
Funding
This study was financially supported by the Yiwu Science and Technology Plan Project (24‐3‐104).
Disclosure
The corresponding author confirms that this manuscript is an honest, accurate, and transparent account of the study being reported; that no important aspects of the study have been omitted; and that any discrepancies from the study as planned have been explained. This is a retrospective observational cohort study and was not prospectively registered, consistent with the nature of database‐derived clinical research. The study was reported in accordance with the STROBE guidelines.
The funding sources had no role in study design, data collection, analysis and interpretation, writing of the report, or the decision to submit the article for publication. All authors have read and approved the final version of the manuscript.
Ethics Statement
The use of the MIMIC database was approved by the Institutional Review Board of the Massachusetts Institute of Technology and Beth Israel Deaconess Medical Center, both of which waive the need for informed consent for studies related to the MIMIC‐IV database; thus, our study did not require additional approval from the ethics committee of our own center. The study was conducted in accordance with the Declaration of Helsinki.
Conflicts of Interest
The authors declare no conflicts of interest.
Supporting Information
Additional supporting information can be found online in the Supporting Information section.
Supporting information