Authors: Xiaoting Zhang, Meng Wei, Pengjie Xue, Yanmei Lu, Baopeng Tang
Categories: Research, Atrial fibrillation, Obstructive sleep apnea syndrome, Recurrence, Prediction model
Source: BMC Cardiovascular Disorders
Authors: Xiaoting Zhang, Meng Wei, Pengjie Xue, Yanmei Lu, Baopeng Tang
The high prevalence of atrial fibrillation (AF) and obstructive sleep apnea syndrome (OSAS) imposes a substantial disease burden on public healthcare, making it a significant health concern in the current era. However, there is currently a lack of risk assessment tools for AF recurrence in patients with AF and OSAS. Therefore, this study aims to explore the factors influencing AF recurrence in patients with AF and OSAS, and to establish a predictive model and scoring system for AF recurrence rates.
The study included a total of 423 patients with AF and OSAS, who were randomly divided into train set (n = 296) and test set (n = 127) in a ratio of 3. Afterwards, the train set was split into a recurrence group and a non-recurrence group for further analysis of indicators while in hospital.
Following Lasso regression screening, 8 variables were selected from a pool of 62 variables from patients with AF and OSAS. Additionally, the study incorporated the CHA2DS2-VASc score and its components of interest, the severity of OSAS and hypoxemia, and whether patients received catheter ablation (CA). Multivariable Cox regression analysis Hb < 115 g/L (HR = 2.27), P > 1.51mmol/L (HR = 3.77), PCT > 2ng/ml (HR = 15.72) as independent risk factors. Hb > 150 g/L (HR = 0.66), TT4 < 66 nmol/L (HR = 0.16) were identified as independent protective factors. The train set showed AUC values of 0.65, 0.71, and 0.71 at the 1st, 3rd, and 5th year, respectively, while the validation set displayed AUC values of 0.60, 0.59, and 0.64 at the 1st, 3rd, and 5th year, respectively, indicating good predictive performance of the model. The AF recurrence rate scoring system categorized patients in the train and test sets into low-risk, medium-risk, and high-risk groups, with HR values of 2.36 and 6.79 for AF recurrence rates in the medium-risk and high-risk groups of the train set, and an HR value of 2.77 for the medium-risk group in the test set.
The predictive models and scoring systems developed in this study demonstrate good predictive ability in assessing the recurrence of AF in patients with OSAS, offering invaluable clinical guidance and references.
Not applicable.
The online version contains supplementary material available at 10.1186/s12872-025-04696-7.
Atrial Fibrillation (AF) is one of the most common types of rapid heart rhythm disorders within the clinical domain. According to the latest statistical data estimates, AF affects approximately 2–4% of the adult population worldwide [1, 2]. In China, AF affects 1.6% of the total population aged 18 and above [3]. Additionally, the prevalence of AF increases with age, with the incidence in the group aged 80 and above being 14.75 times higher than that in the age group of 18 to 29 years [3]. Despite Obstructive Sleep Apnea Syndrome (OSAS) is one of the most common breathing disorders during sleep, affecting approximately 7.3 billion people globally [4], it is considered an independent risk factor for cardiovascular disease [5] and plays a significant role in the occurrence and development of AF [6], However, in reality, there still exists a lack of diagnosis and treatment. The two conditions are closely related, with an estimated prevalence of AF in OSAS ranging from 21–74% [7], and a prevalence of OSAS in AF of around 50% [8]. Besides the severity of nocturnal hypoxemia accompanying OSAS is also an independent predictor of new-onset AF [9]. OSAS reduces the likelihood of maintaining sinus rhythm after cardioversion or catheter ablation (CA) in patients with AF, increasing the risk of AF recurrence by 25% [10]. Moreover, both AF and OSAS independently increase the likelihood of worsening patient outcomes, with AF showing significant associations with heart failure (HF), ischemic stroke, cognitive impairment, and even mortality risk [1], while OSAS is relevant to an increased risk of cardiovascular mortality. In addition, AF and OSAS impose huge economic burdens on both families and public healthcare systems. A nationwide cost survey in the United States revealed an annual per-patient medical cost of $13,000 for diagnosing AF [11], and an Italian study indicated that the annual treatment cost for patients with OSAS was approximately €234,227,041 [12].
For a long period of time, antiarrhythmic drugs have been the mainstay treatment for AF. Since the emergence of CA in 1994, more options and possibilities for the treatment of AF patients have been provided. Compared to treatment with antiarrhythmic drugs, CA is significantly able to reduce the risks of death, stroke, and hospitalization in AF patients [13]. Nevertheless, the recurrence rate after CA in patients with AF and concomitant OSAS can be as high as 80%, with OSAS being one of the separated predictors of AF ablation failure [14]. Meanwhile, the treatment efficacy for OSAS remains inconsistent. Non-randomised controlled trials (RCTs) have indicated that continuous positive airway pressure (CPAP) therapy for OSAS can aid in maintaining sinus rhythm in patients with concomitant AF post electrical cardioversion or CA [15, 16]. On the contrary, small-sample RCT focused on preventing AF recurrence did not show apparent improvement [17].
As previous studies have shown, the high recurrence rate of AF in patients with concomitant OSAS and the unclear mechanisms have sparked our keen interests. With the aging population, the increasing incidence of OSAS and AF undoubtedly brings a enormous disease burden. Based on our current understanding, there is currently a lack of risk assessment tools for AF recurrence in patients with AF and OSAS. As the largest healthcare institution in the region, with the best medical conditions and the highest number of patients, our medical centre serves patients from dozens of different ethnicities, equipped with a robust Hospital Information System (HIS). Therefore, utilizing the clinical characteristics, laboratory indicators, and test results of patients with AF and OSAS, we can establish clinical prediction models and scoring systems from different perspectives to assess the risk of AF recurrence, with the aim of maximizing improvements in patient prognosis and quality of life. Furthermore, by identifying high-risk patients, healthcare professionals can adopt a more proactive approach to monitoring and intervening in high-risk factors to reduce AF recurrence rates.
This study was approved by the Ethics Committee of the First Affiliated Hospital of Xinjiang Medical University (ethics K202409-07). It is a retrospective case-control study, including a total of 463 patients diagnosed with AF combined with OSAS at the First Affiliated Hospital of Xinjiang Medical University from January 1, 2012, to August 31, 2024. Patients were followed up in September 2024, with a median follow-up time of 60 months. Inclusion (1) Age ≥ 18 years; (2) Patients with a confirmed diagnosis of AF comorbid with OSAS. Exclusion criteria (1) valvular AF; (2) left atrial diameter > 55 mm; (3) severe HF with left ventricular ejection fraction (LVEF) ≤ 35%; (4) cardiac surgery within the previous 6 months or coronary stent implantation within the previous 3 months before hospitalization; (5) severe thyroid abnormalities, coagulation disorders, liver or kidney dysfunction; (6) patients unable to obtain follow-up outcomes. In this study, overnight polysomnography [PSG (Embletta MPR PG, America)] were utilised for the diagnosis of OSAS, with the primary diagnostic criteria as follows [18]: RDI ≥ 5 events/hour, predominantly comprising obstructive events, accompanied by symptoms such as nocturnal snoring, apnea, daytime sleepiness, or other symptoms (e.g., morning headaches, decreased attention). (2) The severity of OSAS [18] is classified as mild for RDI ≥ 5 and < 15, moderate for RDI ≥ 15 and ≤ 30, and severe for RDI > 30/hr. Additionally, based on the PSG-derived lowest nocturnal oxygen saturation levels, hypoxemia was classified into mild (85%≤SaO2 < 90%), moderate (80%≤SaO2 < 85%), and severe (SaO2 < 80%) [19]. The calculation of RDI is as follows [18]: it involves summing the total number of apneas and hypopneas and then dividing by the total recording time. Apnea is defined as a ≥ 90% reduction in airflow lasting ≥ 10 s, while hypopnea is defined as a ≥ 30% reduction in airflow lasting ≥ 10 s, accompanied by a ≥ 3% drop in blood oxygen saturation. The study’s loss to follow-up rate was 5.4%. Reasons for loss to follow-up (a) Long follow-up duration resulting in changes in patient contact information; (b) Patient refusal to participate in follow-up. According to the inclusion and exclusion criteria, patients were randomly allocated into a train set (n = 296) and a test set (n = 127) in a ratio of 3. The detailed patient selection process is illustrated in Fig. 1.
Fig. 1Flowchart outlining the screening process for study subjects
Clinical data was collected from inpatient records of patients with AF combined with OSAS, (1) Basic age, sex category; (2) Underlying congenital heart disease, coronary heart disease, HF, diabetes, hypertension, stroke or transient ischemic attack (TIA), chronic obstructive pulmonary disease, vascular disease; (3) Physical examination heart rate, respiratory rate, body mass index (BMI); (4) Examination Blood laboratory tests, noting white blood cell counts (WBC), counting of lymphocytes, monocytes, neutrophils, eosinophils (Eos), basophils, and plateletsb (PLT), hemoglobin (Hb); Liver function tests namely alanine and aspartate aminotransferases (ALT and AST), γ-glutamyltransferase (GGT), direct and indirect bilirubin (DBIL and IBIL), albumin (ALB), globulin (GLO), total bile acids (TBA), alkaline phosphatase (ALP); Renal function markers such as creatinine, urea, uric acid (UA), cystatin C (Cys-C), glomerular filtration rate (GFR); Lipid profile includes total cholesterol (TC), triglycerides (TG), high and low density lipoprotein cholesterol (HDL-c and LDL-c); Electrolytes contains potassium, sodium, chloride, calcium, phosphorus (P), magnesium, glucose; Coagulation panels such as thrombin time (TT), prothrombin time (PT), activated partial thromboplastintime (APTT), D-Dimer, fibrinogen; Cardiac function indicators comprise N-terminal pro-brain natriuretic peptide (NT-ProBNP), creatine kinase (CK), LVEF; The function of the thyroid gland contains thyroid stimulating hormone (TSH), total triiodothyronine (TT3), total tetraiodothyronine (TT4); Inflammatory factors like procalcitonin (PCT); (5) Other variables of concern include the severity of OSAS and hypoxemia as reflected by PSG, as well as the patients’ CHA2DS2-VASc score and whether they underwent CA.
In this study, the clinical outcome observed pertains to the recurrence of atrial fibrillation in patients diagnosed with atrial fibrillation and comorbid obstructive sleep apnea (OSA) at any time between discharge and our follow-up period. AF recurrence was defined as the occurrence of AF, atrial flutter, or atrial tachycardia episodes lasting ≥ 30 s [20]. Patient follow-up regarding AF recurrence was conducted through telephone contact and HIS. In this study, the recurrence of AF in patients is based on the following Following discharge, patients are scheduled for electrocardiogram and 24-hour or long-term ambulatory ECGs at 1, 3, 6, and 12 months, with subsequent annual follow-ups. In case of symptoms such as palpitations, routine 12-lead electrocardiogram and 24-hour or long-term ambulatory ECGs can be conducted promptly.
Data analysis was performed using SPSS version 22.0 (SPSS Inc., Chicago, IL, USA) and R software (version 4.2.1; R Foundation for Statistical Computing, Vienna, Austria). Various R packages were utilized, including glmnet [4.1.7], pROC [1.18.0], ggplot2 [3.3.6], rms [6.4.0], Resource Selection [0.3-5], and rmda [1.6]. As the data used in this study were obtained from the HIS, there was minimal missing data, which was imputed using the median. Statistical analyses were conducted using SPSS. Descriptive statistics, including means ± standard deviations, were used for metric variables following a normal distribution. Independent sample t-tests were applied to variables meeting the conditions for t-tests. Data that exhibited skewed distribution were described using the median and interquartile range (25th, 75th percentile) [P50 (P25, P75)], and between-group comparisons were made using the Wilcoxon-Mann-Whitney test. Categorical variables were presented as n (%) and between-group comparisons were performed using either the chi-square test or Fisher’s exact test.
In this study, 62 variables from patient data were collected. Employing Lasso regression to incorporate an L1 norm constraint term into the cost function of the linear regression model. By tuning the lambda parameter, it enabled variable selection and complexity adjustment. For handling high-dimensional data and feature selection in constructing logistic regression models, ten-fold cross-validation was employed. The selection of variables was based on the optimal evaluation index of the lambda value (lambda.min). The specific diagnostic Lasso coefficient selection and variable selection trajectory can be found in the initial section of the results. In this study, we had no covariates. In this study, we had no covariates, and all variables were set as independent variables, allowing for the inclusion of study variables through Lasso regression screening.
Following the variable selection with Lasso regression, the selected variables and our variables of interest were included in the preliminary investigation. Given the presence of outcome and time variables in this study, we opted for the Cox regression model for analysis. Subsequently, Cox proportional-hazards model was applied to assess the relationship between variables and the recurrence of AF. Variables with P<0.1 were further analyzed using multivariate Cox regression. A prediction model was established based on the Cox regression coefficients, and receiver operating characteristic (ROC) curves were plotted to evaluate the predictive diagnostic performance using the area under the curve (AUC).
The predictive performance of the clinical prediction model was evaluated using Decision Curve Analysis (DCA). Calibration curves were employed to describe the disparity between predicted and actual probabilities, assessing the model’s goodness-of-fit. Nomogram were generated based on the proportional coefficients from the multi-variable Cox regression to illustrate the relationships among the predictive variables within the prediction model.
Using multivariate regression results, this study established a risk score table for AF recurrence of patients with AF combined with OSAS, based on The Framingham Study risk score functions [21]. Additionally, a score effect analysis was conducted. The significance level for all statistical methods employed in this study was set at α = 0.05. In conclusion, following the establishment of the predictive model, we proceeded to validate the model’s performance using a test set and scoring system to ensure the stability of the results.
Following Lasso regression selection, 8 variables were chosen from 62 variables of patients with AF combined with OSAS. These selected variables include Eos counts, Hb, DBIL, P, D-Dimer, TT4, PCT, and BMI. Additionally, the study contained the CHA2DS2-VASc score and its components (Congestive HF, Hypertension, Age, Diabetes, Stroke/TIA, Vascular disease, Sex category), as well as assessments of the severity of OSAS and hypoxemia in patients, and whether they underwent CA. The variable trajectory and selection process of Lasso regression are depicted in Fig. 2.
Fig. 2Trajectory of Variables and Selection Process in Lasso Regression: (A) Trajectory diagram illustrating the variables in Lasso regression, (B) Flowchart depicting the variable selection process in Lasso regression
All contained patients were randomly allocated into a train set (n = 296) and a test set (n = 127) in a ratio of 3. Upon comparison, no significant differences were observed between the train and test sets in various indicators, including Follow-up time, Eos counts, Hb, DBIL, P, D-Dimer, TT4, PCT, BMI, Congestive HF, Hypertension, Age, Diabetes, Stroke/TIA, Vascular disease, Sex category, CHA2DS2-VASc score, OSAS, Hypoxemia, CA, AF recurrence. A total of 171 patients experienced recurrence, while 252 patients did not. The recurrence rate of AF combined with OSAS patients was as high as 40.4%, with recurrence rates in the train and test sets reaching 40.88% and 39.37%, respectively. Following the assignment of dummy variables, no discrepancies were observed in the aforementioned indicators between the train and test sets, indicating successful allocation outcomes. The P values for the aforementioned results all exceeded 0.05. Further information regarding dummy variable assignments can be accessed in Table 1, with additional details provided in Table 2 and S1.
Table 1Dummy variable assignment table for each variableVariablesRules for assigning dummy variablesAF RecurrenceNo = 0, Yes = 1Congestive HFNo = 0, Yes = 1HypertensionNo = 0, Yes = 1Age, year<65 = 0, [65, 75) = 1, ≥ 75 = 2DiabetesNo = 0, Yes = 1Stroke/TIANo = 0, Yes = 2Vascular diseaseNo = 0, Yes = 1Sex categoryMen = 0, Women = 1CHA2DS2-VASc score, points[0,1] = 0, [2, 3] = 1, [4, 9] = 2OSASMild = 0, Moderate & Severe = 1HypoxemiaMild = 0, Moderate & Severe = 1CANo = 0, Yes = 1Eos, 10^9/L[0.02, 0.52] = 0, <0.02 = 1, >0.52 = 2Hb, g/L[115, 150] = 0, <115 = 1, >150 = 2DBIL, umol/L≤ 8.6 = 0, >8.6 = 1P, mmol/L[0.85, 1.51] = 0, <0.85 = 1, >1.51 = 2D-Dimer, ng/ml[0, 280) = 0, ≥ 280 = 1TT4, nmol/L[66, 181] = 0, <66 = 1PCT, ng/ml≤ 0.5 = 0, (0.5, 2] = 1, >2=2BMI, kg/m^2[18.5, 24) = 0, <18.5 = 1, ≥ 24 = 2Notes: AF: atrial fibrillation, HF: heart failure, TIA: transient ischemic attack, OSAS: obstructive sleep apnea syndrome, CA: catheter ablation, Eos: eosinophils, Hb: hemoglobin, DBIL: direct bilirubin, P: phosphorus, TT4: total tetraiodothyronine, PCT: procalcitonin, BMI: body mass index
Table 2Comparison of original data between train and test sets of patients with AF combined with OSASVariablesTotal (n = 423)Test (n = 127)Train (n = 296)Z/χ²
P
Follow-up time, M (Q₁, Q₃)31.00 (8.00, 59.00)36.00 (7.50, 59.00)29.00 (8.00, 59.25)-0.390.698Age, M (Q₁, Q₃)62.00 (54.34, 70.00)62.00 (51.79, 69.52)62.00 (56.22, 70.00)-1.410.157Eos, M (Q₁, Q₃)0.12 (0.08, 0.19)0.11 (0.07, 0.17)0.13 (0.08, 0.20)-1.390.164Hb, M (Q₁, Q₃)146.00 (133.03, 156.00)144.00 (132, 153.5)147.00 (134, 157)-1.220.221DBIL, M (Q₁, Q₃)3.39 (1.90, 5.11)3.58 (1.90, 4.95)3.33 (1.90, 5.20)-0.070.945P, M (Q₁, Q₃)1.13 (1.01, 1.26)1.14 (1.04, 1.25)1.12 (1.00, 1.26)-0.720.472D-Dimer, M (Q₁, Q₃)85.00 (48.00, 211.53)85.00 (44.00, 206.50)84.50 (50.00, 212.55)-0.710.476TT4, M (Q₁, Q₃)87.91 (75.34, 103.90)89.32 (75.61, 103.35)87.25 (75.42, 104.00)-0.120.901PCT, M (Q₁, Q₃)0.05 (0.03, 0.61)0.05 (0.03, 0.64)0.05 (0.03, 0.58)-0.060.955BMI, M (Q₁, Q₃)27.17 (24.77, 30.61)27.50 (24.39, 30.46)27.06 (24.92, 30.77)-0.310.755Congestive HF, n(%)0.490.484No364 (86.05)107 (84.25)257 (86.82)Yes59 (13.95)20 (15.75)39 (13.18)Vascular disease, n(%)0.970.324No275 (65.01)87 (68.50)188 (63.51)Yes148 (34.99)40 (31.50)108 (36.49)AF recurrence, n(%)0.080.772No252 (59.57)77 (60.63)175 (59.12)Yes171 (40.43)50 (39.37)121 (40.88)Hypertension, n(%)0.420.516No137 (32.39)44 (34.65)93 (31.42)Yes286 (67.61)83 (65.35)203 (68.58)Diabetes, n(%)0.070.789No306 (72.34)93 (73.23)213 (71.96)Yes117 (27.66)34 (26.77)83 (28.04)Stroke/TIA, n(%)1.880.17No380 (89.83)118 (92.91)262 (88.51)Yes43 (10.17)9 (7.09)34 (11.49)Sex category, n(%)1.320.25Men325 (76.83)93 (73.23)232 (78.38)Women98 (23.17)34 (26.77)64 (21.62)CHA2DS2-VASc score, n(%)1.570.456[0,1]123 (29.08)40 (31.50)83 (28.04)[2, 3]197 (46.57)61 (48.03)136 (45.95)[4, 9]103 (24.35)26 (20.47)77 (26.01)OSAS, n(%)0.540.463Mild162 (38.30)52 (40.94)110 (37.16)Moderate & Severe261 (61.70)75 (59.06)186 (62.84)Hypoxemia, n(%)0.850.356Mild169 (39.95)55 (43.31)114 (38.51)Moderate & Severe254 (60.05)72 (56.69)182 (61.49)CA, n(%)0.060.803No157 (37.12)46 (36.22)111 (37.50)Yes266 (62.88)81 (63.78)185 (62.50)Notes: Z: Mann-Whitney test, χ²: Chi-square test; M: Median, Q₁: 1st Quartile, Q₃: 3st Quartile, AF: atrial fibrillation, HF: heart failure, TIA: transient ischemic attack, OSAS: obstructive sleep apnea syndrome, CA: catheter ablation, Eos: eosinophils, Hb: hemoglobin, DBIL: direct bilirubin, P: phosphorus, TT4: total tetraiodothyronine, PCT: procalcitonin, BMI: body mass index
The results of the train set show that there are 175 patients in the non-recurrence group and 121 patients in the recurrence group. The comparison between the two groups show that the recurrence group had a higher proportion of high-risk patients in terms of age, average D-Dimer levels, moderate to severe OSAS and hypoxemia, hypertension, CHA2DS2-VASc score compared to the non-recurrence group. The average level of Hb and the proportion of patients who underwent CA were lower in the AF recurrence group compared to the non-recurrence group. Detailed data of 2 groups can be found in Table 3.
After assigning variables, comparisons between the AF recurrence group and the non-recurrence group showed that the proportion of patients with moderate to severe OSAS was higher in the AF recurrence group (73.55% vs. 55.43%), as well as the proportion of patients with moderate to severe hypoxemia (72.73% vs. 53.71%). Furthermore, there was a higher proportion of patients with D-dimer levels ≥ 280ng/ml (31.40% vs. 15.43%), hypertension (76.03% vs. 63.43%), patients aged ≥ 75 years (19.83% vs. 12%), patients aged 65–75 years (31.40% vs. 25.14%), patients with CHA2DS2-VASc scores between 4 and 9 (34.71% vs. 20%), patients with PCT levels between 0.5ng/ml and 2ng/ml (34.71% vs. 20%), and patients with PCT levels > 2ng/ml (0.83% vs. 0%) in the AF recurrence group. Conversely, the proportion of patients who underwent CA treatment was lower in the AF recurrence group (46.28% vs. 73.71%). Detailed results are presented in Table S2.
Table 3Comparison of original data between the the recurrent AF and non-recurrent AF groups in the train setVariablesTotal (n = 296)Non-recurrent AF (n = 175)Recurrent AF (n = 121)Z/χ²
P
Age, M (Q₁, Q₃)62.00 (56.22, 70.00)61.00 (55.00, 68.00)65.26 (58.00, 74.00)-2.730.006Eos, M (Q₁, Q₃)0.13 (0.08, 0.20)0.13 (0.09, 0.21)0.12 (0.08, 0.18)-1.280.201Hb, M (Q₁, Q₃)147.00 (134.00, 157.00)148.00 (136.50, 158.00)143.00 (130, 154)-2.410.016DBIL, M (Q₁, Q₃)3.33 (1.90, 5.20)3.12 (2.02, 4.97)3.50 (1.65, 5.54)-0.530.599P, M (Q₁, Q₃)1.12 (1.00, 1.26)1.11 (0.99, 1.27)1.15 (1.02, 1.25)-0.850.393D-Dimer, M (Q₁, Q₃)84.50 (50.00, 212.55)75.00 (45.50, 144.50)149.00 (62.00, 341.00)-4.15< 0.001TT4, M (Q₁, Q₃)87.25 (75.42, 104.00)87.50 (75.70, 103.65)86.93 (73.50, 105.00)-0.000.999PCT, M (Q₁, Q₃)0.05 (0.03, 0.58)0.05 (0.03, 0.28)0.06 (0.03, 0.74)-1.830.068BMI, M (Q₁, Q₃)27.06 (24.92, 30.77)26.73 (24.78, 29.81)28.30 (25.44, 31.49)-1.940.052Congestive HF, n(%)0.000.984No257 (86.82)152 (86.86)105 (86.78)Yes39 (13.18)23 (13.14)16 (13.22)Vascular disease, n(%)2.830.092No188 (63.51)118 (67.43)70 (57.85)Yes108 (36.49)57 (32.57)51 (42.15)OSAS, n(%)10.060.002Mild110 (37.16)78 (44.57)32 (26.45)Moderate & Severe186 (62.84)97 (55.43)89 (73.55)Hypoxemia, n(%)10.92< 0.001Mild114 (38.51)81 (46.29)33 (27.27)Moderate & Severe182 (61.49)94 (53.71)88 (72.73)Hypertension, n(%)5.270.022No93 (31.42)64 (36.57)29 (23.97)Yes203 (68.58)111 (63.43)92 (76.03)Diabetes, n(%)3.460.063No213 (71.96)133 (76.00)80 (66.12)Yes83 (28.04)42 (24.00)41 (33.88)Stroke/TIA, n(%)2.310.128No262 (88.51)159 (90.86)103 (85.12)Yes34 (11.49)16 (9.14)18 (14.88)Sex category, n(%)1.930.165Men232 (78.38)142 (81.14)90 (74.38)Women64 (21.62)33 (18.86)31 (25.62)CHA2DS2-VASc score, n(%)10.000.007[0,1]83 (28.04)58 (33.14)25 (20.66)[2, 3]136 (45.95)82 (46.86)54 (44.63)[4, 9]77 (26.01)35 (20.00)42 (34.71)CA, n(%)22.97< 0.001No111 (37.50)46 (26.29)65 (53.72)Yes185 (62.50)129 (73.71)56 (46.28)Notes: Z: Mann-Whitney test, χ²: Chi-square test; M: Median, Q₁: 1st Quartile, Q₃: 3st Quartile, AF: atrial fibrillation, HF: heart failure, TIA: transient ischemic attack, OSAS: obstructive sleep apnea syndrome, CA: catheter ablation, Eos: eosinophils, Hb: hemoglobin, DBIL: direct bilirubin, P: phosphorus, TT4: total tetraiodothyronine, PCT: procalcitonin, BMI: body mass index
The univariate Cox regression analysis revealed that 65 year ≤ Age<75 year (HR = 1.66) and P>1.51mmol/L (HR = 3.97) are potential risk factors in patients with AF combined with OSAS. Conversely, TT4<66 nmol/L (HR = 0.19) and 0.5ng/ml< PCT ≤ 2ng/ml (HR = 0.58) are potential protective factors. In the multivariate Cox regression analysis, Hb<115 g/L (HR = 2.27), P>1.51mmol/L (HR = 3.77) and PCT>2ng/ml (HR = 15.72) are identified as independent risk factors. On the other hand, Hb>150 g/L (HR = 0.66), TT4<66 nmol/L (HR = 0.16) are identified as independent protective factors. Additional detailed information can be found in Table 4.
Table 4Presentation of single-factor and multiple-factor Cox regression results in the train setVariablesSingle-factorMultiple-factorβS.EZ
P
HR (95%CI)βS.EZ
P
HR (95%CI)Congestive HFNo1.00 (Reference)Yes0.050.270.180.8581.05 (0.62 ~ 1.78)Stroke/TIANo1.00 (Reference)Yes0.150.260.580.5591.16 (0.70 ~ 1.92)Vascular diseaseNo1.00 (Reference)Yes0.230.191.210.2261.25 (0.87 ~ 1.81)OSASMild1.00 (Reference)Moderate & Severe-0.090.21-0.430.6680.91 (0.61 ~ 1.38)HypoxemiaMild1.00 (Reference)Moderate & Severe-0.130.21-0.60.5460.88 (0.59 ~ 1.32)Eos[0.02, 0.52]1.00 (Reference)<0.020.140.460.30.7631.15 (0.47 ~ 2.83)>0.52-0.521.01-0.510.6090.60 (0.08 ~ 4.30)Hb[115, 150]1.00 (Reference)1.00 (Reference)<1150.640.361.810.071.91 (0.95 ~ 3.83)0.820.362.250.0242.27 (1.11 ~ 4.63)>150-0.220.2-1.110.2690.80 (0.54 ~ 1.19)-0.420.2-2.070.0380.66 (0.44 ~ 0.98)DBIL≤ 8.61.00 (Reference)>8.6-0.610.51-1.190.2340.55 (0.20 ~ 1.48)D-Dimer<2801.00 (Reference)≥ 280-0.150.2-0.730.4670.86 (0.58 ~ 1.28)TT4[66, 181]1.00 (Reference)1.00 (Reference)<66-1.680.37-4.52< 0.0010.19 (0.09 ~ 0.39)-1.860.41-4.53< 0.0010.16 (0.07 ~ 0.35)HypertensionNo1.00 (Reference)Yes0.030.210.140.8851.03 (0.68 ~ 1.57)Age<651.00 (Reference)[65, 75)0.510.212.380.0181.66 (1.09 ~ 2.52)≥ 750.270.241.080.2791.30 (0.81 ~ 2.11)DiabetesNo1.00 (Reference)Yes0.170.190.890.3761.19 (0.81 ~ 1.74)Sex categoryMen1.00 (Reference)Women0.320.211.530.1261.38 (0.91 ~ 2.08)CHA2DS2-VASc score[0,1]1.00 (Reference)[2, 3]0.320.241.30.1941.37 (0.85 ~ 2.21)[4, 9]0.380.251.490.1361.46 (0.89 ~ 2.41)CANo1.00 (Reference)Yes0.170.190.890.3721.18 (0.82 ~ 1.71)P[0.85, 1.51]1.00 (Reference)1.00 (Reference)<0.85-0.210.37-0.560.5740.81 (0.40 ~ 1.67)0.10.40.250.8031.10 (0.51 ~ 2.41)>1.511.380.373.7< 0.0013.97 (1.91 ~ 8.22)1.330.383.52< 0.0013.77 (1.80 ~ 7.88)PCT≤ 0.51.00 (Reference)1.00 (Reference)(0.5, 2]-0.550.2-2.680.0070.58 (0.39 ~ 0.86)-0.290.21-1.390.1630.75 (0.49 ~ 1.13)>21.061.011.050.2922.90 (0.40 ~ 21.04)2.751.132.430.01515.72 (1.70 ~ 145.38)BMI[18.5, 24)1.00 (Reference)<18.5-13.162181.05-0.010.9950.00 (0.00 ~ Inf)≥ 24-0.180.24-0.740.4580.84 (0.52 ~ 1.34)Notes: HR: Hazards Ratio, CI: Confidence Interval, AF: atrial fibrillation, HF: heart failure, TIA: transient ischemic attack, OSAS: obstructive sleep apnea syndrome, CA: catheter ablation, Eos: eosinophils, Hb: hemoglobin, DBIL: direct bilirubin, P: phosphorus, TT4: total tetraiodothyronine, PCT: procalcitonin, BMI: body mass index
To assess the predictive performance of the model at different time points, we plotted ROC curves, column charts, DCA curves, and calibration curves for 1st-year, 3rd-year, and 5th-year periods. Figures 3A-1, 3A-2, and 3A-3 depict the ROC curves for 1st-year, 3rd-year, and 5th-year intervals, with the train set showing AUC values of 0.65 (0.53–0.76) at the 1st year, 0.71 (0.63–0.79) at the 3rd year, and 0.71 (0.63–0.79) at the 5th years. The test set exhibited AUC values of 0.60 (0.45–0.75) at the 1st year, 0.59 (0.45–0.72) at the 3rd year, and 0.64 (0.50–0.78) at the 5th year, indicating good predictive performance of the model. Figure 3B shows the nomogram based on the multivariate Cox regression analysis, where each segment above represents variables such as Hb, TT4, P, and PCT, with PCT having the most significant impact on the prediction results, while segments below denote the predictive range of the model at the 1st, 3rd, and 5th year. Figures 3C series depict the DCA curves for the train and test sets at the 1st, 3rd, and 5th year, demonstrating the broad applicability of the model. Calibration curves for the train and test sets at the 1st, 3rd, and 5th year, are shown in Figs. 3D series, further confirming the stability and accurate predictive performance of the model.
Fig. 3Prediction model outcomes for AF recurrence in patients with AF combined with OSAS. Panels A-1, A-2, and A-3: ROC curves for the train and test sets in the 1st, 3rd, and 5th year. Panel B: Diagnostic nomogram for the AF recurrence model. Panels C-1, C-3, and C-5: DCA curves for the train set in the 1st, 3rd, and 5th year. Panels C-2, C-4, and C-6: DCA curves for the test set in the 1st, 3rd, and 5th year. Panels D-1, D-3, and D-5: Calibration curves for the train set in the 1st, 3rd, and 5th year. Panels D-2, D-4, and D-6: Calibration curves for the test set in the 1st, 3rd, and 5th year
Following the Framingham method, a predictive scoring system for AF recurrence risk in patients with concomitant OSAS is established. Hb > 150 g/L is assigned a score of -1 point, while Hb < 115 g/L is assigned 2 points. TT4 < 66nmol/L is assigned − 5 points, P > 1.51mmol/L is assigned 4 points, and PCT > 2ng/ml is assigned 7 points. The assigned values for each factor can be found in Table 5.
Following the scoring system, the train set and test set are scored. In this study, scores ranging from − 6 to -1 points are classified as low-risk, 0 to 2 points as medium-risk, and 3 to 4 points as high-risk. In the train set, it is observed that as the risk level increases, the recurrence rate of AF rises, while the non-recurrence rate of AF decreases. The results indicate a similar trend in low-risk and medium-risk patients in the test set. Detailed results can be found in Table 6; Fig. 4.
After assigning scores to the train and test sets, the groups were divided into low-risk, medium-risk, and high-risk groups for AF recurrence based on the scores. Using the low-risk group as a reference, the Cox regression analysis results for the train set show HR for AF recurrence of 2.36 and 6.79 for the medium-risk and high-risk groups, respectively. In the test set, the Cox regression analysis results show an HR of 2.77 for AF recurrence in the medium-risk group. In the test set of Fig. 4; Table 7, there are only two patients in the high-risk group, both of whom did not experience AF recurrence, leading to the conclusion of chance findings.Detailed data can be found in Table 7.
Table 5Risk scoring table for AF recurrence of patients with AF combined with OSAS.(Used by train and test sets)CharacteristicsCategoriesPointsHb>150 g/L-1Hb<115 g/L2TT4<66nmol/L-5P>1.51mmol/L4PCT>2ng/ml7
Table 6AF recurrence and non-AF recurrence rate of patients with AF combined with OSAS in the train and test sets at different scoresPointsTrainTestNon-AF recurrenceAF recurrenceNon-AF recurrence rateAF recurrence rateNon-AF recurrenceAF recurrenceNon-AF recurrence rateAF recurrence rate-65645.45%54.55%5455.56%44.44%-515962.50%37.50%6460.00%40.00%-31150.00%50.00%-----1693367.65%32.35%231363.89%36.11%0795558.96%41.04%362757.14%42.86%23925.00%75.00%5271.43%28.57%31325.00%75.00%10100.00%0.00%42528.57%71.43%10100.00%0.00%-6 ~ -1 points904964.75%35.25%342161.82%38.18%(Low risk)0 ~ 2 points826456.16%43.84%412958.57%41.43%(Medium risk)3 ~ 4 points3827.27%72.73%20100.00%0.00%(High risk)
Fig. 4Presentation of AF recurrence rate based on risk score values. The x-axis represents scores, while the y-axis represents ratio. A displays the rates of AF recurrence and non-recurrence in patients classified as low-risk, medium-risk, and high-risk within the train set. B presents the rates of AF recurrence and non-recurrence in patients classified as low-risk, medium-risk, and high-risk within the test set
Table 7AF recurrence and non-AF recurrence rate of patients with AF combined with OSAS at different scoresVariablesTrainTestβS.EZ P HR (95%CI)βS.EZ P HR (95%CI)Sum-6 ~ -1 points1.00 (Reference)1.00 (Reference)0 ~ 2 points0.860.214.04< 0.0012.36 (1.56 ~ 3.57)1.020.422.420.0152.77 (1.22 ~ 6.32)3 ~ 4 points1.920.44.83< 0.0016.79 (3.12 ~ 14.77)-16.596469.2800.9980.00 (0.00 ~ Inf)Notes: HR: Hazards Ratio, CI: Confidence Interval
The high recurrence rate of AF in patients with concomitant OSAS and its complex mechanisms have led to a high hospitalization rate for such patients, resulting in a significant global burden. In this study, we selected Eos counts, Hb, DBIL, P, D-Dimer, TT4, PCT, BMI, CHA2DS2-VASc score, Congestive HF, Hypertension, Age, Diabetes, Stroke/TIA, Vascular disease, Sex category, OSAS, Hypoxemia, and CA as variables of interest in conjunction with Lasso regression. Multivariable Cox regression analysis results showed that Hb < 115 g/L (HR = 2.27), P > 1.51mmol/L (HR = 3.77), and PCT > 2ng/ml (HR = 15.72) were independent risk factors, whereas Hb > 150 g/L (HR = 0.66) and TT4 < 66 nmol/L (HR = 0.16) were independent protective factors. A scoring system was constructed in this study to assess the risk of AF recurrence in patients with AF and OSAS. Using scores assigned to patients in the train and test sets, we defined score ranges of -6 to -1 as low risk, 0 to 2 as medium risk, and 3 to 4 as high risk. The study results show that in the train set, as the risk level increases, the recurrence rate of AF increases while the non-recurrence rate decreases. Similar trends are seen in the low-risk and medium-risk patients in the test set. The predictive results of this study demonstrate good accuracy.
Primarily, this study found that Hb < 115 g/L was an independent risk factor for AF recurrence in patients with concomitant OSAS. In the first place, OSAS involves recurrent episodes of sleep-disordered breathing, leading to periodic nocturnal hypoxemia and hypercapnia. Hb is a medium for transporting oxygen and carbon dioxide in the body. Lack of Hb exacerbates this phenomenon and triggers alternations between the activation of the sympathetic and vagal nervous systems, resulting in atrial structural remodeling and increasing susceptibility to AF [22, 23]. Secondly, anaemia is significantly associated with increased cardiac output and ventricular hypertrophy, further exacerbating cardiac structural remodeling [24]. Thirdly, atrial structural remodeling and the electrophysiological changes associated with sleep-disordered breathing in patients with OSAS contribute to the occurrence of the reentry mechanism in AF, inducing AF by establishing arrhythmogenic substrates in the atria [23]. What is more, untreated OSAS triggers inflammation stimulating macrophages, reducing Hb levels by shortening red blood cell lifespan through increased clearance [25]. The prevalence of AF and anaemia increases with advancing age and often coexist [26]. Research has found that preoperative anaemia is an independent predictive factor for AF recurrence [27]. Treatment for anaemia has also been notably effective in AF patients, reducing the incidence of cardiovascular death or HF [28] and significantly improving cardiac structure and function [29]. On the contrary, intermittent hypoxia caused by OSAS can stimulate RBC production, leading to an increase in Hb concentration [30], with Hb > 150 g/L identified as a protective factor against AF recurrence in patients with AF and OSAS, we believe that high Hb levels contribute to reducing AF recurrence by improving nocturnal hypoxia in this population, lowering the frequency of activation of the sympathetic and vagal nervous systems, improving atrial structural and electrical remodelling, and thereby decreasing AF recurrence. As most anaemic patients are asymptomatic, this study suggests that active screening for anaemia in patients with AF and OSAS could lead to early detection, diagnosis, and treatment, thus improving long-term prognosis.
Of note, after multifactorial Cox regression analysis, it is found that TT4 < 66 nmol/L is an independent protective factor for AF recurrence in patients with AF and OSAS. Hyperthyroidism is one of the risk factors for precipitating atrial arrhythmias [31]. Thyroxine binds to Thyroxine-Binding Globulin to form T4, while the unbound form is known as Free T4 (FT4), and the sum of both is TT4. Thyroxine induces the occurrence and progression of AF by shortening action potential duration and refractoriness, increasing pulmonary vein spontaneous activity [32] and interatrial fibrosis [33]. The study found a significant association between elevated FT4 levels at baseline and AF [32]. It is worth further exploring the current lack of consensus regarding the relationship between thyroid hormones and OSAS. Some studies suggest a significant association between hypothyroidism and OSAS [34], while others have found that the prevalence of hypothyroidism in OSAS patients appears to be no different from the general population [35]. Furthermore, thyroid hormone treatment for patients with low TT4 and OSAS did not lead to an improvement in their condition [36]. We think that high levels TT4 plays a key role in AF recurrence in patients with AF and OSAS. Therefore, it is recommended to test thyroid hormone levels in all AF patients and emphasize the importance of individualized treatment plans before initiating AF therapy. Unfortunately, the study sample did not include cases of excessively high TT4 levels, so we cannot definitively establish a direct association between elevated TT4 levels and AF recurrence.
PCT, as a highly sensitive inflammatory marker, accurately reflects the status of systemic inflammatory response. This study reveals that in patients with AF and OSAS, PCT > 2ng/ml may be an independent risk factor for AF recurrence. Inflammatory markers play a crucial role in atrial structural remodeling and electrical remodeling, with both local and systemic inflammation contributing to the generation of AF substrates [37]. At the same time, OSAS increases oxidative stress due to intermittent hypoxia, leading to chronic inflammation and subsequently increasing the recurrence rate of AF [38]. In comparison with non-AF patients, the levels of PCT are significantly elevated in AF patients [39]. A bidirectional Mendelian randomization study found that for every one standard deviation increase in PCT, there was a 3.4% increase in the risk of AF (OR = 1.034, 95% CI, 1.005–1.064, p = 0.022)^37^. We consider that PCT, as one of the inflammatory markers, may be significantly associated with the recurrence of AF through participation in cardiac structural remodeling and electrical remodeling. Timely intervention is crucial for reducing AF recurrence, particularly in infected patients with AF and OSAS, especially those with severe infections.
In addition to the discussed factors, in the multivariable Cox regression analysis, P > 1.51mmol/L remains an independent risk factor for AF recurrence in patients with OSAS. Mainly, high P accelerates the occurrence of arteriosclerosis and atherosclerosis by promoting vascular calcification, making it a risk factor for cardiovascular disease [40]. Simultaneously, high P disrupts calcium homeostasis by causing dysregulation of parathyroid hormone and Fibroblast Growth Factor 23 [41], moreover, it can stimulate the release of pro-inflammatory cytokines, leading to a chronic state of vascular inflammation [42], resulting in the occurrence of cardiovascular disease. Hyperphosphatemia is associated with increased Major Adverse Cardiovascular Events (MACEs), mortality, and HF events in patients with coronary artery disease [38], and the extent of coronary atherosclerosis is related to AF [43]. Elevated blood P levels are associated with an increased risk of AF [44]. We speculate that high blood P may be involved in the recurrence of AF for the following Firstly, elevated blood P promotes coronary atherosclerosis leading to narrowing of the coronary arteries, triggering changes in cardiac structure and electrophysiological environment such as myocardial ischemia and fibrosis, which may increase susceptibility to AF. Additionally, OSAS may indirectly contribute to AF recurrence through metabolic disturbances, including abnormal blood P levels, due to insufficient oxygen supply. Therefore, timely intervention to manage blood P levels in patients with OSAS and AF is crucial for reducing the recurrence of AF.
In general, the conclusions of this retrospective study are multifaceted. Firstly, it identified factors influencing AF recurrence in patients with OSAS, including protective and risk factors. Secondly, this study established a predictive model for AF recurrence in patients with AF and OSAS, demonstrating good predictive performance. Additionally, the scoring system based on the predictive model in this study provides a robust assessment and predictive effect on AF recurrence, offering valuable guidance and reference for future clinical treatment.
Patients with AF and OSAS have a high AF recurrence rate and complex mechanisms that are not clearly understood, which has raised our profound concern. With the increasing prevalence of OSAS and AF due to population aging, there is undoubtedly a significant disease burden. To our knowledge, there is currently a lack of risk assessment tools for AF recurrence in patients with AF and OSAS, making this study the first to establish a clinical prediction model for this patient population. However, this study still has limitations. Firstly, although the study retrospectively collected data from all AF patients with OSAS admitted between 2012 and 2024, there are only 423 patients, indicating a relatively small sample size and possible result instability. Secondly, while the hospital is the largest regional hospital with the highest number of admissions, this study is conducted at a single centre, and the predictive model is only internally validated. Thirdly, as one of the larger medical institutions in the northwest region, our hospital is equipped with a comprehensive HIS, ensuring the integrity of patients’ clinical data. However, due to the retrospective data collection method employed in this study, all research variables are sourced from the HIS, leading to missing data on the types of AF, treatment methods for AF and OSAS, as well as PSG parameters during the collection process. Fourthly, due to regional variations, PCT is not commonly included as a routine assessment for AF patients undergoing ablation. Moreover, the discussions surrounding PCT in our manuscript predominantly focus on AF occurrence in the context of sepsis, thereby restricting the applicability of PCT as a marker for the general AF patient population. These limitations will drive future research efforts, necessitating multicentre prospective real-world clinical studies to validate these findings and provide assistance to a wider range of patients and healthcare professionals.
In conclusion, independent risk factors for AF recurrence in patients with OSAS are Hb < 115 g/L, P > 1.51mmol/L, and PCT > 2ng/ml, while independent protective factors include Hb > 150 g/L and TT4 < 66nmol/L. These factors demonstrate good predictive performance when combined in a scoring system.
Below is the link to the electronic supplementary material.
Supplementary Material 1
Supplementary Material 2
Supplementary Material 3
Supplementary Material 4