CAFR aims to provide evidence for clinical practice based on real world data. The main objectives are as follows: (i) to compare and validate current risk scores for stroke and bleeding risk assessment in Chinese AF patients; consequent upon this, to improve current risk prediction models using phenotype and genotype data; (ii) to study factors associated with anticoagulation under-use, variation in stroke prevention practice and progress over the study period in typical tertiary and non-tertiary hospitals; and (iii) to report on safety and long term effectiveness of AF ablation therapy, and compare the effectiveness of ablation and medical therapy in reducing death and thromboembolic events, using propensity score matched cohorts.

The following additional exploratory analyses will also be considered: (i) description of effectiveness and safety of NOACs and left atrial appendage closure among the Chinese AF population; and (ii) assessment of biomarkers associated with AF progression from paroxysmal to persistent AF, and development of complications.

Thirty two tertiary and non-tertiary hospitals at where the majority of AF patients in Beijing are managed, have agreed to participate in this study. The characteristics of participating hospitals are listed in Table 1.

Patients aged 18 years or older with a documented AF as confirmed by 12 leads ECG, pacemaker/ICD electrocardiogram, or Holter ECG (duration of AF episode least 30 s) are eligible for this study. Patients with transient AF caused by reversible cause (e.g., cardiac surgery, pulmonary embolism, untreated hyperthyroidism) or combined with other serious diseases with a life expectancy < 1 year were excluded. All patients with the diagnosis of AF who are under the care of cardiologists or general physicians in the participating hospitals, and provided informed consent will be enrolled, from both outpatient and inpatient settings. Each hospital maintained an admission record to provide evidence of whether there was selection bias, by comparing the mean age, sex and comorbidities of enrolled and non-enrolled patients. Enrollment will cease when the target of 20000 patients is achieved.

Key data elements and definition of each variable were in line with the ACC/AHA recommendation on AF clinical data standards [9] and international peer studies [10] to facilitate cross-comparison of results between different registries. The following data were collected for each patient enrolled: basic socio-demographic information, symptoms and signs relating to AF, medical history, results of physical examination and laboratory test and patient self-reported quality of life (using a validated Chinese version of AFEQT questionnaire). For patients undergoing ablation therapy, detailed procedure related information is also collected, including the pathway of ablation, achievement of conduction block, time consumed and peri-procedure complications.

Each enrolled patient will be followed up every 6 months by trained staff, at outpatient clinic or through telephone interview. Information relating to medical therapies, health-care utilization, and clinical outcomes will be collected during follow up. Definition of the major outcome events are provided in Table 2. Every patient will be followed up until 2020. The questionnaires used for data collection in study are provided in appendix. All the data were entered into a specific electronic data capture system.

Five milliliter venous blood sample will be collected from patients with consent. The sample will be centrifuged. Plasma and blood cells will be separately stored at −80 °C refrigerators. Standard operation procedures for blood collection, processing and archiving are in line with the International Society for Biological and Environmental Repositories’ Best Practices for Repositories [11]. FreezerPro® is used for web-based management of frozen samples and information of sample will be linked to corresponding clinical data.

All clinical data will be captured via a web-based electronic data capture (EDC) system. Investigators will enter and edit the data via a secure network, with secure access features. Variables related to the answer of main aims are mandatory to be entered. The database also had function of range and logic check. Source material will be kept in hospital for further monitoring and auditing. The quality control team will review all the data collected case by case and send queries when applicable.

A random selection of 10 % cases will have their raw data validated, for baseline data collection and during each follow up. All reported ischemic stroke and peripheral embolism events will be adjudicated by an independent endpoint committee.

Ethics approval was obtained from the Human Research Ethics Committees at Beijing Anzhen Hospital. Ethic review boards in each participating hospital agreed their participation. The data collected will be held centrally in a secure database, only de-identified information can be used for analysis. Informed consent from individual patients was sought for their agreement to participate in long-term follow-up and usage of their blood sample for further analysis. The study was performed according to the ethical principles for medical research involving human subjects specified in the Declaration of Helsinki.

The CHADS₂ schema and CHA₂DS₂-VASc schema will be compared and validated in these Chinese AF patients, after excluding patients receiving anticoagulation therapy or ablation therapy during follow up. Logistic regression will be used to evaluate the predictive value of the various risk schemata in discriminating between patients who develop the outcome of interest and those who do not. The c-statistic will be estimated to determine the predictive ability for stroke of each schema. The 95 % confidence interval will be estimated for the c-statistic using a non-parametric bias-corrected bootstrapping method. To assess the effect of individual risk factors on the occurrence of events, multivariable logistic regression will be used with the baseline covariates as independent variables. According to the result of the multivariable logistic regression analysis, the score for each risk factor will be adjusted for the best fitted classification schema. If necessary, new risk factors may also be added to generate a new stroke risk classification schema for Chinese AF population. The new schema will be validated on the validation population and the C-statistics will be calculated using the new risk schema. In addition, the net reclassification index will also be used to evaluate predictive value of the prognostic score.

To compare the ablation and non-ablation therapies, propensity score matching for receiving catheter ablation will be calculated for each patient based on a multivariable logistic regression model. Baseline characteristics variables which are assumed to be associated with the probability of having catheter ablation will be included in the model as independent variables. Based on the propensity scores, ablation and non-ablation patients will be matched on a 1:1 basis with a combination of the nearest neighbor algorithm and caliper algorithm [12]. For each ablation patient, a non-ablation patient with the smallest propensity score difference, and smaller than 0.02, will be matched without replacement. Patients not matched will be excluded from the analysis.

Unadjusted event rates for the composite primary outcome of time to first all-cause mortality, stroke and peripheral embolism for the propensity-matched ablation and non-ablation groups will be estimated using the Kaplan-Meier method and compared between groups with the log-rank test. As sensitivity analyses, the inverse probability of treatment weighting using propensity score will also be used to compare the survival distributions between ablation and non-ablation group.

For the objective of developing the optimal risk prediction model for Chinese AF population, a minimum of 200 events (i.e. 10 times of the variable number) are required to accommodate 20 variables in the prediction model. The analysis is planned to be done in 2015 to accumulate enough events for analysis. For that purpose, about 5000 patients without anticoagulation therapy are required for an evenly splitted training cohort and validation cohort. For the objective of evaluating the effectiveness of ablation therapy, the following assumption is made: the accumulate ischemic stroke and intracranial hemorrhage is 10 % in optimally anticoagulated patients and 7 % in ablation patients in five years, 1800 matched patients in each arm are required to provide 90 % power to detect the difference (α = 0.05). As those two groups of patients are quite different, we amplified the sample size by 3 times to guarantee enough matched patients. In total, 16000 patients are required for the main analyses and considering a possible 20 % withdraw and loss of follow up, we up round the sample size to 20000.

17467 patients had been enrolled as at June 30, 2015. The majority (84.0 %) came from tertiary hospitals where most AF patients are managed. The missing rates for the key variables are listed in Table 3. There are no missing data among the mandatory variables which is controlled by the EDC system. Out of range variables were very rare (<0.1 %) for those with range check. Patient follow up rates at 6, 12, 18 and 24 months are 93, 91, 87 and 85 % respectively.

Participating hospitals of this study are all hospitals in urban and semi-urban area of Beijing, the capital of China. However, about half of the registied patients come from all around the country, which increases the representativeness of the study population. In addition, the majority of patients are registred from tertiary hospitals, but this reflects current practice of Chinese medical care as the mechanisms of referral are not fully established and patients usually go direct to terriary hospitals without referral from primary health care providers. Despite these limitations, we feel that we are unlikely to overtly bias conclusions for the prespecified main analysis of this study.