Using a Clinical Decision Support System to Improve Anticoagulation in Patients with Nonvalve Atrial Fibrillation in China's Primary Care Settings: A Feasibility Study

Background To primarily investigate the effect of using a clinical decision support system (CDSS) in community health centers in Shanghai, China, on the proportion of patients prescribed guideline-directed antithrombotic therapy. This study also gauged the general practitioner (GP)'s acceptance of the CDSS who worked in the atrial fibrillation (AF) special consulting room of the CDSS group. Methods This was a prospective cohort study that included a semistructured interview and a feasibility study for a cluster-randomized controlled trial. Eligible patients who sought medical care in the AF special consulting rooms in two community health centers in Shanghai, China, between April 1, 2020, and October 1, 2020, were enrolled, and their medical records from the enrollment date, up to October 1, 2021, were extracted. Based on whether the GPs in the AF special consulting rooms of the two sites used the CDSS or not, we classified the two sites as a software group and a control group. The CDSS could automatically assess the risks of stroke and bleeding and provide suggestions on treatment, follow-up, adjustment of anticoagulants or dosage, and other items. The primary outcome was the proportion of patients prescribed guideline-directed antithrombotic therapy. We also conducted a semistructured interview with the GP in the AF special consulting rooms of the software group regarding the acceptance of the CDSS and suggestions on the optimization of the CDSS and the study protocol of the cluster-randomized controlled trial in the future. Results Eighty-four patients completed the follow-up. The mean age of these subjects was 75.71 years, the median time of clinical visits was six times per person, and the follow-up duration was 15 months. The basic demographics were similar between the two groups, except for age (t = 2.109, p = 0.038) and the HAS-BLED score (χ2 = 4.363, p = 0.037). The primary outcome in the software group was 8.071 times higher than that in the control group (adjusted odds ratio (OR) = 8.071, 95% confidence interval (2.570–25.344), p < 0.001). The frequency of consultation between groups was not significantly different (p = 0.981). It seemed that the incidence of adverse clinical events in the software group was lower than that in the control group. The main reason for dropouts in both groups was “following up in other hospitals.” The GP in the AF special consulting rooms of the software group accepted the CDSS well. Conclusions The findings indicated that it was feasible to further promote the CDSS in the study among community health centers in China. The use of the CDSS might improve the proportion of patients prescribed guideline-directed antithrombotic therapy. The GP in the AF special consulting room of the software group showed a positive attitude toward the CDSS.


Background
Atrial fbrillation (AF) is one of the most common cardiac arrhythmia diseases [1]. Tere were more than 7.90 million patients with AF in China over 45 years of age, with a prevalence of 1.8%. Nearly 3% of people over 75 years of age had AF in China in 2020 [2,3]. Due to the rapidly aging population, it was estimated that the prevalence of AF would increase at least 2.5 times in the next 50 years [4]. Tus, the increasingly high prevalence of AF brought a heavier disease burden on families and society. A systematic review published in 2011 indicated that the medical expense caused by AF management was 10,100-14,200 dollars per person in the United States and 450-3,000 euros per person in Europe [5]. Tromboembolic events, especially ischemic stroke, are the main issues of great expenses [1]. Te risk of ischemic stroke in patients with AF was four to fve times higher than that in patients without [3], with 24.2% of patients with nonvalve AF in China experiencing strokes [6].
As China progressed in primary care and implemented a two-way referral system, most patients with AF would receive antithrombotic medical services in community health centers in the future. Te movement also corresponded with the view of "integrated care and stratifed therapy" highlighted by some scholars, which meant that patients could get access to comprehensive management in primary care, such as risk assessment, AF treatment, international normalized ratio (INR) monitoring, health education, and treatment of comorbidity diseases, with superior hospitals responsible for the treatment of complications, emergencies, and operations [7]. However, at present, China's primary care settings are not equipped with sufcient capabilities and enthusiasm to undertake AF management [8,9], due to heavy workloads already assigned to general practitioners (GPs) and the lack of knowledge of AF and anticoagulation treatment. Tus, many patients sought healthcare in superior hospitals or remained in unmanaged states [8]. For instance, the proportion of appropriate anticoagulation treatment in tertiary hospitals (the highest level of hospitals in mainland China) was 9.6%-68.4% but was only 4.0%-28.2% in nontertiary hospitals [6], and in the community health centers in Shanghai, the proportion dropped to 4.24%-12.64% [10][11][12]. To move AF management into primary care, the frst step should be strengthening the GPs' AF management abilities. Te development of clinical decision support systems (CDSSs) may help overcome such problems.
CDSSs, to some extent, are typical examples of artifcial intelligence applications in healthcare settings, in which they are supported by computer algorithms. Tey can help primary care providers quickly make individualized and correct clinical decisions by combining patient data with guidelines in short consultations [13]. CDSSs signifcantly improved the GPs' work efciency and quality, as well as reduced their workloads. As early as 2002 [14], scholars proposed the idea of applying CDSSs to AF management. AF antithrombotic management-associated CDSSs were then applied in some developed countries [15] and appraised by most GPs [16][17][18]. Due to the short follow-up duration, inadequate sample size, and imperfect design of CDSSs, many studies, however, reported that CDSSs might improve the appropriate prescription of antithrombotic agents and lower the incidence of adverse events but had no efects on reducing the incidence of thromboembolism [19][20][21][22][23][24]. Diferent from developed countries, the proportion of appropriate antithrombotic treatment in patients with AF is lower in China [25], and patients can seek medical care in any hospital they prefer. To our knowledge, no similar studies have been published in China's primary care settings until now.
We evaluated an AF antithrombotic managementassociated CDSS, specifcally designed for GPs, and wanted to carry out a cluster-randomized controlled trial to confrm its utility. Tis study was a prospective cohort study with a semistructured interview and a feasibility study for the cluster-randomized controlled trial. We aimed to primarily investigate the efect of the CDSS in community health centers in Shanghai, China, on the proportion of patients prescribed guideline-directed antithrombotic therapy. We also sought to evaluate the GP's acceptance of the CDSS who worked in the atrial fbrillation (AF) special consulting room of the CDSS group.

Study Design.
Tis study was a prospective cohort study with a semistructured interview. We designed and reported the study according to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement [26]. Te Ethics Committee of Zhongshan Hospital in Shanghai, China, approved the study (approval number. B2022-009R).

Settings and Population.
We selected the Fenglin Community Health Center as the software group, as it had eligible servers to install the CDSS and was willing to participate in the study. Considering the population of residents and the GP in AF special consulting rooms, we selected the Beicai Community Health Center, which is also in Shanghai, China, as the control group. Te GPs in the two sites included were an associate chief physician and an experienced attending physician, respectively. Patients who sought medical care in the AF special consulting rooms of the two community health centers between April 1, 2020, and October 1, 2020, were potential participants. Patients who were older than 18 years of age, diagnosed with any type of AF or atrial futter (ICD-10 codes are shown in Supplementary Material) during a consultation or within their medical history, and who were willing to participate in the study were included in this study. Individuals with valvular AF or prosthesis valves and those who were pregnant or breastfeeding were excluded.

Exposure.
Te exposure in this study was a CDSS for nonvalve AF anticoagulation management, which was developed by our team and Ping An Healthcare and Technology Co., Ltd., Beijing, China. Te decision-making processes of the CDSS were developed based on the latest AF guidelines in China at that time [27]. Shenzhen Wang'an Computer Safety Checking & Measuring Technology Co., Ltd., Shenzhen, Guangdong Province, China, examined the system's security, and developers tested the stability and algorithm verifcation. Te rationality of decisions was assessed by cardiological experts. Before use, we embedded the CDSS into the hospital information system (HIS) of the community health center in the software group. When one patient was diagnosed with AF during consultation for the frst time, the GP opened the CDSS. After entering the patient's related information, such as personal details, medical or drug history, and laboratory or image fndings (Figure 1), the CDSS evaluated the risk of strokes using the CHADS 2 score and CHA 2 DS 2 -VASc score and the risk of bleeding using the HAS-BLED score, ORBIT score, and ATRIA score and provided treatment and follow-up suggestions according to the risks of the patient (Figure 2). Although the likelihood of program errors was low, we also added a red exclamation mark behind the scores of each risk evaluation ( Figure 2). When the GP clicked the mouse on the red exclamation mark, they could see the explicit causes of these scores. During follow-up visits, the GP opened the follow-up visit interface of the CDSS and registered information, such as recent stroke and recent bleeding ( Figure 3). Te CDSS provided suggestions on whether to refer the patient to superior hospitals and how to adjust the dosage or frequency of anticoagulants ( Figure 4). Te GP then rendered fnal decisions according to those suggestions.

Outcomes.
Te primary outcome was the proportion of patients prescribed guideline-directed antithrombotic therapy. It was defned as the agreement percentage between the GPs' anticoagulation prescriptions for patients and the guideline's recommendations, including prescribing anticoagulation drugs to correct patients or recommending a referral to superior hospitals for patients with anticoagulation indications and refraining from anticoagulation for patients without such indications or with contraindications.
Te secondary outcome was the frequency of consultation. It was defned as the average number of patients in each group, seeking medical care in community health centers due to AF. We also reported a composite of bleeding, stroke, death for AF or AF complications, and hospitalization for AF or AF complications. Strokes included ischemic stroke and transient ischemic attack (TIA). Referring to the Guidelines of Stroke Prevention in Chinese Patients with Atrial Fibrillation (2017), we classifed bleeding events as minor bleeding and major bleeding [28]. When such adverse events occurred, we recorded the time of onset and ceased the follow-up.

Data Collection.
We collected participants' medical records, from the enrollment date to October 1, 2021. We frst extracted the demographic information of each subject from their health records, including their gender, age (y), systolic and diastolic pressure (mmHg), medical history, such as hypertension, diabetes, and cancer, current drugs, whether the patient was taking antiplatelet or nonsteroidal anti-infammatory agents (NSAIDs), CHA 2 DS 2 -VASc score, HAS-BLED score, and anticoagulation contraindications if any.
How and when patients would revisit their GPs were determined by the subjects themselves after considering suggestions from their GPs. We extracted data on AF treatment from the subjects' medical records after the follow-up. Te data collected included anticoagulants they    International Journal of Clinical Practice 3 used, INR values of patients with warfarin, and adverse clinical events during the follow-up. Te occurrence of these events was identifed according to their medical records, diagnostic records, and any related available auxiliary examinations. For some information that was not accessible via the Internet or due to loss to follow-up, data collection was performed by contacting the patients via telephone. At the end of the study, we interviewed the GP in the AF special consulting room of the software group to inquire about the acceptance of the CDSS and some rational improving ideas about the CDSS and the next cluster-randomized controlled trial. Te interview lasted approximately 20 minutes. Te interview was recorded (by audio and verbally) after obtaining informed consent from the GP. Two researchers summarized the conversations and independently extracted the key information related to the study. Disagreements were resolved through discussion or by a third researcher. For subjects with new oral anticoagulants (NOACs), loss to follow-up was defned as the last follow-up visit date which was at least 12 months before the ending date (October 1, 2021). For patients receiving warfarin, that was defned as the last follow-up visit date was at least 1 month before the ending date [29].

Sample
Size. According to the study by Whitehead et al. [30], the recommended sample size of a feasibility study was ten in each group, if the estimated efect in the formal research of a randomized controlled trial was ≥0.7 and the power (1-β) was 0.80.

Statistical Analysis.
Categorical data were reported as frequencies and percentages (n, %), and chi-square tests were used to detect signifcant diferences. Continuous data were reported as the mean ± standard deviation (SD) and compared using the t-test if they adhered to the Gaussian distribution. Otherwise, the data were reported as the median (interquartile range (IQR)) and compared using the Mann-Whitney U test. A logistic regression model was used to detect the efect of the exposure on the primary outcome. Considering the small sample size, we only calculated the number and incidence of composite adverse clinical events in each group and in total. To simplify the analysis, we recorded the follow-up time in the unit of month. Regardless of the start and end dates, we recorded the onset time as this month if events occurred in the former month or as the next month if they occurred in the latter. We presented the reasons for loss to follow-up in a pie diagram. All statistical analyses were performed using IBM SPSS Statistics, version 26.0 (SPSS Inc.). Two-tailedp < 0.05 was considered statistically signifcant.

Basic Subject Demographics.
From April 1, 2020 to October 1, 2020, a total of 115 patients with AF or atrial futter sought medical care in the AF special consulting rooms of the two community health centers. Eight patients met at least one of the exclusion criteria. Sixty-six subjects were in the software group, and 53 had complete records.
Forty subjects were in the control group, and 31 had complete records. Te fowchart of the study is shown in Figure 5.
Among the 84 subjects who completed the study, 46 (54.8%) were male. Te average age was 75.71 years. Te CHA 2 DS 2 -VASc scores of more than 95% of subjects were ≥2. Te basic demographics were similar between the groups, except for age (79.96 ± 7.025 years in the software group and 73.58 ± 7.205 in the control group, t � 2.109, p � 0.038) and the HAS-BLED score (the number of subjects with the HAS-BLED score ≥ 3 was 33 in the software group and 12 in the control group, respectively; χ 2 � 4.363, p � 0.037). Te median time of clinical visits and the median follow-up duration were six times per person and 15 months, respectively, with no statistical diferences between the two groups. Te basic demographics of the subjects are shown in Table 1.

Te Proportion of Patients Prescribed Guideline-Directed
Antithrombotic Terapy. A total of 52 subjects in both groups received anticoagulants at the end of the study, with 40 in the software group and 12 in the control group. After removing subjects with anticoagulation contraindications, the proportion of patients prescribed guideline-directed anticoagulants was 75.5% in the software group and 38.7% in the control group. Te only subject whose CHA 2 DS 2 -VASc score was lower than 2 but was prescribed with anticoagulants was in the software group, and his CHA 2 DS 2 -VASc score was one. In the software group, 19 individuals received NOACs and 21 received warfarin, while in the control group, all 12 individuals received NOACs ( Figure 6). Table 2 shows the factors associated with antithrombotic treatment. After adjusting for gender, age, hypertension, heart failure, diabetes, stroke/TIA/thromboembolism history, CHA 2 DS 2 -VASc score, and HAS-BLED score, the primary outcome in the software group was 8.071 times higher than that in the control group (adjusted OR � 8.071, 95% CI (2.570-25.344), p < 0.001), and the anticoagulation rate of patients receiving antiplatelets or NSAIDs was 0.165 times of those not receiving them (adjusted OR � 0.165, 95% CI (0.051-0.529), p � 0.002).

Adverse Clinical Events.
Adverse events occurred in seven subjects during the study, and the incidence of all subjects was 8.3%. Among them, three were in the software group and four were in the control group. Te incidence rates were 5.7% and 12.9% in the software group and the control group, respectively.

Causes of Participants Lost to Follow-Up.
Twenty-two subjects were lost to follow-up, with 13 in the software group and nine in the control group. Te dropout rate was observed with no statistical diference between the two groups. Signifcant diferences were observed in the variables of stroke/TIA/thromboembolism history (χ 2 � 4.935, p � 0.026), CHA 2 DS 2 -VASc score (χ 2 � 6.034, p � 0.012), and HAS-BLED score (χ 2 � 11.221, p � 0.001) ( Table 3). Five subjects in each group reported that they followed up in other hospitals regularly, and two in the control group believed that they did not need to revisit for follow-up, as they experienced no uncomfortable symptoms (Figure 7).

Te Qualitative Interview.
Two topics were discussed during the qualitative interview, including the acceptance of the CDSS and suggestions to improve the CDSS and the protocol for the next cluster-randomized controlled study. Te GP in the AF special consulting room of the software group accepted the interview.

Acceptance of the CDSS.
Te GP felt that the CDSS was helpful in her daily work and strengthened her confdence to prescribe necessary anticoagulants to patients. It was also useful when she taught patients about the precautions of anticoagulation. She also indicated that the CDSS might assist GPs with a poor command of AF management.
"Connected with the hospital information system (HIS), the CDSS was convenient to use at any time. It helped me ranging from making a treatment decision for an initial

Discussion
Tis study was a feasibility study of a cluster-randomized controlled trial which we are carrying out. Te fnding indicated that using the CDSS in China's primary care settings might improve the proportion of patients prescribed guideline-directed antithrombotic therapy. Te GP in the AF special consulting room of the software group showed a positive attitude toward the CDSS. Tromboembolic prophylaxis, through standardized anticoagulation, can signifcantly reduce the risk of strokes in patients with AF [31], as well as the elevated risk of disability and mortality caused by AF. Standardized anticoagulation plays an essential role in AF management. However, in China, the proportion of appropriate antithrombotic treatments administered to patients with AF was lower than that in most other countries [25]. Tis problem was even severe in primary care settings. One reason was that GPs in China lacked the knowledge to prescribe  antithrombotic treatments and were worried about bleeding in patients with AF [9,28]. Te CDSS used in this study functioned to assess the risks of stroke and bleeding and provided suggestions on AF management, like treatment decisions, follow-up time, adjustment of anticoagulants, or dosage administered. Te CDSS might help GPs manage patients with AF and enhance the GPs' capabilities and confdence in the feld. Tus, it was well accepted by the GP in the AF special consulting room of the software group. Te fnding also indicated that the CDSS might improve the proportion of patients prescribed guideline-directed antithrombotic therapy and helped GPs provide scientifc and normative health education to patients with AF. Te proportion of appropriate anticoagulation treatments was 75.5% in the software group and 38.7% in the control group. Tere were 12 patients in the software group  International Journal of Clinical Practice and 17 patients in the control group with high-stroke risks but were not prescribed with anticoagulants. Te common reasons for no anticoagulation treatment in both groups were mainly old age, complex comorbidities, and strong rejection from patients themselves. Besides these, the GP in the control group also said that he was not sure if it was necessary to prescribe anticoagulants for those whose CHA2DS2-VASc scores were two and those who were taking antiplatelets or NSAIDs. Due to the few adverse clinical events during the study, we used a composite outcome indicator. It seemed that the incidence of adverse clinical events in the software group was lower than that in the control group, but this potential still needed to be further validated with a larger sample size in the clusterrandomized trial.
Considering the results in this study, we estimated that it should be feasible to promote the CDSS in more community health centers. In the subsequent randomized controlled study, we adjusted the protocol in the four aspects below. First, the dropout rate was high in the current study (21.5%), and one reason was that patients followed up in other hospitals. To reduce dropout rates, we will add inclusion and exclusion criteria to avoid the loss to followup. For example, we require that subjects be registered with a GP for at least 1 year and exclude patients whose expected survival time would be less than 1 year or those who are unable to cooperate with the study due to any reason. Second, considering that the GP in the AF special consulting room of the software group viewed the CDSS well, and patients seeking healthcare in AF special consulting rooms were few, we plan to carry out the formal trial in all general practice clinics. We want to recruit a sufcient number of participants and investigate whether using the CDSS remained efective or not in more community health centers. Terefore, in the next trial, we will set the inclusion and exclusion criteria for GPs and train them in AF management and the use of the CDSS. After training, GPs will be able to recruit patients with AF in the study. Tird, based on the interview with the GP in the AF special consulting room of the software group, we will update the CDSS. Apart from its existing functions, the latest version also provides referral suggestions and tips on the interactions between anticoagulants and other drugs or food. Te program also supports the assessments of the risk of falling and balancing (Tinetti Scale and Self-Rated Fall Risk Questionnaire), cognition (Mini-Mental State Examination), and emotional and psychological disorders (Geriatric Depression Scale-15 and the Confusion Assessment Method). In addition, we modifed the record methods for the warfarin dosage. Fourth, GPs will be invited to complete a questionnaire on antithrombotic treatments three times before and after the training, as well as at the end of the study. Within the frst week and the fourth, eighth, and twelfth months after enrollment, patients will be required to fnish the satisfaction questionnaires. Trough questionnaires, we hope to reduce dropouts to some extent and evaluate the efect of the CDSS in more directions. Te revision for the protocol and the CDSS was approved by the Ethics Committee of Zhongshan Hospital and registered in the clinical trial registry (approval number. B2021-579(2) and registration number. ChiCTR2100052307) [32].
Te study had some strengths. First, as far as we knew, the study was the frst one in China to identify the efect of an AF antithrombotic treatment-related CDSS on patients with AF in primary care. Second, the data in this study were extracted from the medical records of patients or via telephone contacts. Te data were reliable and close to the real environment of China's primary care. Tird, the CDSS that we used was validated by algorithm engineers and cardiological experts and was highly praised by GPs in the AF special consulting room of the software group.
Tere also existed some shortcomings in this cohort study. First, the age and HAS-BLED score of the subjects at the baseline were not the same between the two groups. Second, the preferences for treatment choices of the subjects and GPs might infuence the fnal treatment decisions, especially in regard to anticoagulant selection and for patients whose risk of stroke was one. However, we could not estimate the magnitude of such efects. Tird, the data in this study were extracted from the medical records of each patient, and we did not have access to their socioeconomic information, such as their educational levels, incomes, and family support. However, in a previous study by Lunde et al. [33], the association between socioeconomic factors and AF decreased with increasing age. Fourth, the high dropout rate in this study might infuence the results. However, in the two factors that were associated with the primary outcome in the study, we did not detect any diference between subjects who lost to follow-up and subjects who completed the follow-up. Fifth, the subjects were those who sought medical care in AF special consulting rooms of the two community health centers. Te GPs in AF special consulting rooms should be more interested and experienced in AF anticoagulation treatment than most GPs. Terefore, the results in this study might not be able to be generalized to all GPs or all patients with AF in China's primary care.

Conclusions
From this study, we drew the primary conclusion that the use of the CDSS might improve the proportion of patients prescribed guideline-directed antithrombotic therapy. GPs in AF special consulting rooms of the software group had a positive attitude toward the CDSS. Te fnding indicated that it was feasible to further promote the CDSS in the study among community health centers in China, and a scientifc, randomized, controlled trial was needed to verify the effectiveness of the CDSS in China's primary care.

AF:
Atrial fbrillation CDSS: Clinical decision support system CI: Confdence interval GP: General practitioner HIS: Hospital information system HR: Hazard ratio ICD-10: International Classifcation of Diseases

Data Availability
Te datasets generated and/or analyzed in the current study are available from the corresponding author on reasonable request.

Ethical Approval
All methods were performed in accordance with the relevant guidelines and regulations (Declaration of Helsinki). Te confdentiality of all subjects was guaranteed, and the study protocol was approved by the Ethics Committee of Zhongshan Hospital of Fudan University (approval number. B2022-009R). As a retrospective study, the data did not concern personal privacy or had any commercial interests. Te ethics committee approved the application for waiving the signed consent form.

Consent
Subjects lost to follow-up and the GP in the software group who was invited to the face-to-face interview were required to sign the informed consent form at the beginning of the survey. If they refused in-person consent, we proposed a telephone interview and obtained verbal informed consent. Individuals rejecting both interview formats were labelled as "rejected."

Disclosure
Te funding bodies played no role in the design, collection, analysis, or interpretation of the data or editing of this manuscript.

Conflicts of Interest
Te authors declare that they have no conficts of interest.