Sex-differences in the effect of obstructive sleep apnea on patients hospitalized with pulmonary embolism and on in-hospital mortality

We determined sex differences in the prevalence of obstructive sleep apnea (OSA) among patients hospitalized with pulmonary embolism (PE) in Spain (2016–2018). We also compared outcomes according to the presence of OSA, and identified variables associated with in-hospital-mortality (IHM) after PE using the Spanish National Hospital Discharge Database. We identified 46,794 hospital admissions for PE; of these, 5.47% had OSA. OSA was more prevalent among men than women (7.57% vs. 3.65%, p < 0.001), as in the general population. Propensity score matching did not reveal differences in concomitant conditions or procedures between patients with and without OSA, except for the use of non-invasive ventilation, which was more frequent in patients with OSA. IHM was similar in patients with and without OSA (3.58% vs. 4.31% for men and 4.39% vs. 4.93% for women; p > 0.05). Older age, cancer, atrial fibrillation, non-septic shock, and need for mechanical ventilation increased IHM in men and women with OSA hospitalized with PE. The logistic regression model showed no sex differences in IHM among patients with OSA.


Results
During the study period (2016-2018), 46,794 patients aged ≥ 18 years were discharged from Spanish hospitals with a primary diagnosis of PE. Of these, 2561 (5.47%) had a code for OSA. The prevalence of OSA was over twice as high in men as in women (7.57% vs. 3.65%; p < 0.001) ( Table 1).
Patients with OSA were younger than those without OSA (68.38 vs. 70.77 years; p < 0.001). The prevalence of OSA among patients with PE increased with age, reaching its highest value in patients aged 65-74 years (8.03%) and decreasing thereafter, with the lowest prevalence being found in those aged ≥ 85 years (1.84%).
The mean Charlson comorbidity index (CCI) value was 1.10 in patients with OSA and 0.83 in those without OSA (p < 0.001). The prevalence of OSA rose in line with the number of comorbidities included in the CCI from 4.36% among patients with CCI 0-8.03% in those with CCI > 2 (p < 0.001).
Regarding hospital outcomes, patients with OSA had a significantly higher median length of hospital stay (LOHS) (8 days vs. 7 days; p = 1.015) and lower crude IHM (3.87% vs. 6.55; p < 0.001). Table 2 shows the results of our comparison of the distribution of comorbidities, therapeutic procedures, and hospital outcomes in men hospitalized for PE with and without OSA.
In men with OSA admitted with PE, we recorded a remarkably high prevalence of obesity (35.42%), chronic obstructive pulmonary disease (COPD, 29.71%), diabetes (23.17%), heart failure (12.29%), and renal disease (10.92%). These values were significantly higher than in men without OSA. Pulmonary hypertension and dependence on supplemental oxygen were also more frequent among patients with OSA.
Cancer was recorded in 13.71% of men with OSA compared with 19.84% of those who did not have OSA (p < 0.001).
No differences were found in therapeutic procedures and surgery, except for non-invasive ventilation (NIV), which was more common among patients with OSA than among those without OSA (5.15% vs. 1.04% p < 0.001).
The crude IHM was significantly lower among men with OSA (3.58%) than among men without OSA (6.46%). Propensity score matching (PSM) revealed that none of the differences in any of the conditions or procedures analyzed remained significant. Furthermore, significance was lost in the association between OSA and IHM (3.58% vs. 4.31%, p = 0.283).
The comparison between women with and without OSA regarding their comorbid conditions, procedures, and hospital outcomes before and after PSM is shown in Table 3.
Women with concomitant OSA were younger and had a significantly higher mean CCI than women who did not have OSA (both p < 0.001). We also found that women with OSA had a very high prevalence of obesity  (Table 3). However, dementia (7.52% vs. 2.74%; p < 0.001) and cancer (14.27% vs. 7.35%; p < 0.001) were more frequent among women without OSA. NIV was 8 times more frequent among women with OSA (7.35% vs. 0.92%; p < 0.001). No differences were found among women for the remaining therapeutic procedures (Table 3).
The data shown in Table 4 are the results of comparing men and women with PE and OSA before and after PSM. These help to assess sex-differences in the co-occurrence of PE and OSA.
Prior to matching, women were significantly older (71.38 years vs. 66.72 years; p < 0.001) and had a lower mean CCI (1.00 vs. 1.15; p < 0.001). The conditions significantly more frequently coded among women than men were heart failure, diabetes, obesity, and pulmonary hypertension. On the other hand, men had a higher prevalence of acute myocardial infarction, peripheral vascular disease, liver disease, COPD, cancer, and atrial fibrillation. No differences in the use of therapeutic procedures were found between men and women.
Before and after PSM, IHM in women with OSA was similar to that of men with OSA (4.39% vs 3.62%; p = 0.403) ( Table 4). Table 5 shows the results of multivariate logistic regression analysis of variables that were independently associated with IHM for patients with OSA hospitalized with PE and according to sex.
The risk of dying in hospital rose with age for both sexes. Cancer, atrial fibrillation, non-septic shock, and mechanical ventilation increased IHM in both men and women with OSA hospitalized with PE. Table 2. Prevalence of specific comorbid conditions, therapeutic procedures, and hospital outcomes in men hospitalized with pulmonary embolism (PE) according to the presence of obstructive sleep apnea in Spain from 2016 to 2018 before and after propensity score matching. OSA Obstructive sleep apnea, CCI Charlson comorbidity index, AMI acute myocardial infarction, PVD peripheral vascular disease, CVD cerebrovascular disease, COPD chronic obstructive pulmonary disease, NIV non-invasive ventilation, IV invasive ventilation, LOHS length of hospital stay, IHM in-hospital mortality, SD standard deviation, IQR interquartile range. p value for comparison of men and women with OSA. www.nature.com/scientificreports/ Women with OSA who had heart failure or dementia coded in their discharge report had a higher risk of dying in hospital. For men, COPD was associated with higher IHM.
As found in the PSM, the logistic regression model showed no sex differences between patients with OSA.

Discussion
In this nationwide population-based study, OSA affected 5.47% of PE patients. Consistent with the general population, the prevalence of OSA was about twice as high in men as in women 16 , indicating that the effect of OSA on the risk of PE is not affected by sex. On the other hand, no association was detected between OSA and higher IHM in men or in women admitted for PE. We found that the prevalence of OSA in patients with PE was consistent with that described by Le Mao et al. 15 , who reported this disease to be relatively infrequent in people with PE. Other studies have reported figures ranging from 15 to 65%, although the results may be confounded as a result of selection and detection biases 6,17,18 .
The mechanisms underlying the association between OSA and PE remain unclear. It has been speculated that OSA leads to the development of PE through the three mechanisms of Virchow's classic triad, namely, vascular endothelial injury, stagnant blood flow, and hypercoagulable blood status 8 . OSA and PE share a series of risk factors, including obesity and increasing age. Furthermore, the sedentary lifestyle associated with OSA and PE can lead to venous stasis and thrombosis 19 . Recurrence of hypoxia and reoxygenation in OSA can increase reactive oxygen species and inflammatory mediators and therefore reduce the availability of nitric oxide and impair vascular endothelial function 8 . Moreover, fibrinogen levels and platelet activity are increased, whereas fibrinolytic capacity is lower in patients with OSA than in healthy controls. The abovementioned factors can contribute to the hypercoagulable state of patients with OSA, which may in turn mediate the need for higher doses of warfarin in patients with PE 20 . Table 3. Prevalence of specific comorbid conditions. Therapeutic procedures, and hospital outcomes in women hospitalized with pulmonary embolism (PE) according to the presence of obstructive sleep apnea in Spain from 2016 to 2018 before and after propensity score matching. OSA Obstructive sleep apnea, CCI Charlson comorbidity index, AMI acute myocardial infarction. PVD peripheral vascular disease, CVD cerebrovascular disease, COPD chronic obstructive pulmonary disease, NIV non-invasive ventilation, IV invasive ventilation, LOHS length of hospital stay, IHM in-hospital mortality, SD standard deviation, IQR interquartile range. p value for comparison of men and women with OSA.   21,22 , possibly owing to nocturnal hypoxemia or OSA-related hypercoagulability 23 . However, a recent systematic review and meta-analysis revealed no significant differences between PE patients with and without OSA in the right ventricle to left ventricle ratio or in the length of stay 24 . Similarly, Le Mao et al. 15 did not find significant differences in the rates of echocardiographic right ventricle dysfunction between both groups. PE-related right ventricle strain might be considerably better offset in patients with pronounced OSA, as they are adapted to repetitive right heart pressure overloads caused by impaired pulmonary perfusion resulting from repetitive oxygen desaturation during sleep 21,25 . Consequently, their right ventricles may be less prone to injury and hemodynamic collapse as a consequence of acute PE. The potential resulting protective effect could be associated with lower mortality. This explanation could justify the findings of Joshi et al. 26 , who reported prevalent OSA to be associated with lower IHM in PE patients. The same authors also suggested that OSA could lead to increased hemoglobin levels resulting from hypoxemia, which may confer a protective effect in acute PE by preventing hypoxia from worsening 14,26 .
Data from regular and critical care wards show a decreased mortality risk in hospitalized patients with OSA and associated diseases, such as pneumonia 27,28 , cardiovascular disease 29 , ischemic stroke 30 , and subarachnoid hemorrhage 31 . These results point to a potential protective effect of OSA with respect to mortality, probably as a result of ischemic preconditioning or a higher level of care or vigilance in these patients 32 . It is also important to bear in mind the obesity paradox, by which obese patients fare better in the face of acute illness, possibly because they seek medical care earlier or have increased metabolic reserves 27 . However, while obesity was more frequent in men and women with OSA than in those without the disease, the difference was not maintained after PSM. Ghiasi et al. 14 also analyzed the relationship between OSA and 30-day mortality in PE patients and found that OSA did not affect mortality directly. In fact, it was affected by complications of OSA such as hypertension and thrombosis. In addition, our study demonstrated that OSA was not associated with higher IHM in men or in women admitted for PE.
As might be expected, NIV was more common in men and women with OSA than in those without OSA, possibly because the former may have had more severe PE or hypoxemia 26 . Nevertheless, NIV was not a predictor of IHM in OSA patients admitted with PE.
The variables associated with IHM for men and women hospitalized for PE and OSA in our study were older age, cancer, atrial fibrillation, non-septic shock, and invasive mechanical ventilation. Among women with OSA, heart failure was also associated with a higher risk of dying in hospital. Similarly, Roca et al. 33 demonstrated that OSA is associated with incident heart failure or death in women but not in men. For men, we also found that COPD was associated with higher IHM, as previously reported globally after PE 34 . In this sense, Xie et al. 35 found that the association between concomitant OSA and COPD and PE (overlap syndrome) was significant only in the male subgroup, in contrast with the female group. Moreover, the authors observed that patients with overlap syndrome had a lower degree of oxygen saturation during sleep and a higher probability of PE than controls and patients with OSA alone.
Our study is subject to a series of limitations. First, given the nature of our database, OSA was not confirmed by polysomnography, which is the gold standard for diagnosis. As a consequence, we do not have information on the severity of OSA, which has demonstrated prognostic value in PE 36 . Second, the absence of laboratory and imaging data and data on medical treatment during hospital stay prevents us from predicting outcomes in PE 1 . Third, we had no data on use of continuous positive airway pressure (CPAP), which is the primary treatment for OSA and may reduce exaggerated coagulant activity and platelet function and improve fibrinolytic capacity in affected patients 19 . Finally, mortality data are valid only for inpatient mortality and all-cause mortality. Despite these limitations, our data on the relationship between OSA and PE are highly reliable because of the validity of the Spanish National Hospital Discharge Database (SNHDD) and the large sample size evaluated.
In conclusion, our study showed that the effect of OSA on the risk of PE was not affected by sex and that the presence of OSA was not associated with higher IHM in men or in women admitted for PE. The variables associated with IHM for men and women hospitalized for PE and OSA in our analysis were older age, cancer, atrial fibrillation, non-septic shock, and invasive mechanical ventilation. Recognizing an association between these two diseases is important and could improve prognosis, since CPAP may decrease the incidence or recurrence of PE in OSA patients.

Methods
Design, setting, and participants. We conducted a retrospective epidemiological study. Participant information was obtained from the SNHDD.
Spanish legislation requires all public and private hospitals to provide the Ministry of Health with information regarding all discharges. Patient information includes age, sex, province of residence, and admission and discharge dates, up to 20 diagnoses, and a maximum of 20 diagnostic or therapeutic procedures conducted during admission. Information regarding the discharge destination was also collected. The 10th Revision of the International Classification of Diseases (ICD-10) has been used by the SNHDD since 2016 to code diagnoses and procedures. More details regarding the SNHDD can be found elsewhere 37  The comorbidities included in the CCI were extracted using the algorithms proposed by Quan et al. 39 for administrative databases coded based on ICD 10.
The conditions included in the CCI were analyzed independently and categorized according to the number (CCI = 0, CCI 1-2, and CCI > 2).
The CCI conditions shown in Supplementary Table 1 are complemented by the codes used to identify various diagnoses (atrial fibrillation, valvular heart disease, obesity, coagulopathy, pulmonary hypertension, and non-septic shock) and therapeutic procedures (dependence on supplemental oxygen, inferior vena cava filter, thrombolytic therapy, NIV, invasive ventilation, and use of vasopressors), which were also analyzed.
The variable "Undergone surgery" included those patients who had undergone a surgical procedure during their hospital stay with PE.
Propensity score matching. As can be seen in Table 1, the characteristics of the OSA and non-OSA populations differ significantly with respect to age, sex, and CCI. We used PSM to make baseline characteristics more similar, by matching each man and woman with OSA with a non-OSA man and woman, respectively. The propensity score used for matching was obtained from a multivariable logistic regression model that included age and all comorbid conditions present at admission. To analyze sex differences in the co-occurrence of PE with OSA, we also matched men with women affected by both conditions. These methods have been reported elsewhere 40,41 . Statistical methods. We show absolute frequencies and proportions for categorical variables and means with standard deviations (SD) or medians with interquartile ranges (IQR) for continuous variables.
Patient demographic characteristics, comorbidities, procedures, and hospital outcomes were compared between patients with and without OSA according to sex. The t test or Wilcoxon rank-sum test was used for continuous variables and the chi-squared or Fisher exact test for categorical variables.
We constructed multivariable logistic regression models to identify those variables that were independently associated with IHM in patients with PE and OSA according to sex. The results are presented as odds ratio (OR) with the 95% CI.
Stata version 14 (Stata, College Station, Texas, USA) was used for PSM and all data analyses.
Ethical aspects. The SNHDD is provided free of charge by the Spanish Ministry of Health to any investigator who sends a justified request 42 . As this database is mandatory and anonymized, according to the Spanish legislation, it is not necessary to obtain ethics committee approval.

Data availability
"No additional data available". According to the contract signed with the Spanish Ministry of Health and Social Services, which provided access to the databases from the Spanish National Hospital Discharge Database (SNHDD), we cannot share the databases with other investigators and must delete the databases once the investigation has concluded. Consequently, we cannot upload the databases to a public repository. However, any investigator can apply for access to the databases by filling out the questionnaire available at: http:// www. msssi. gob. es/ estad Estud ios/ estad istic as/ estad istic as/ estMi niste rio/ Solic itudC MBDdo cs/ Formu lario_ Petic ion_ Datos_ CMBD. pdf. All other relevant data are included in the paper.