Electronic Alert System for Improving Stroke Prevention Among Hospitalized Oral‐Anticoagulation‐Naïve Patients With Atrial Fibrillation: A Randomized Trial

We developed a computer‐based electronic alert system for identifying consecutive hospitalized OAC‐naïve patients with AF and tested the hypothesis that such an alert system would improve OAC prescription.

The alert system automatically identified hospitalized patients with AF without an active OAC prescription in the electronic order entry system. The alert system was incorporated into the electronic medical chart and order entry system of the University Hospital in Bern, Switzerland. It recognized AF by permanently searching diagnosis lists and physician notes of the entire electronic patient chart database for free text entries of AF or its various abbreviations. Alerts were issued 24 hours after the onset of hospital stay if the following 4 criteria for an individual patient were present: (1) AF detected by search criteria; (2) no active prescription order for anticoagulants, including unfractionated or low‐molecular‐weight heparin, fondaparinux, direct oral anticoagulants, or vitamin K antagonists; (3) at least 1 electronic drug prescription order other than an anticoagulant had to be in place in the order entry system; and (4) the patient was randomized to the alert group. Once the criteria were fulfilled, the alert was issued in the electronic patient chart. The alert was visible to physicians and nurses, but only physicians were enabled to respond to the alert. In the first alert screen (Figure 1), the physician was notified that this patient had suspected AF without an active OAC prescription. In addition, the physician was asked to confirm the presence of AF. The physician in charge had the option to complete the CHA₂DS₂‐VASc score electronically or to reject the alert if there was no AF. If AF was present but the physician was unable to complete the CHA₂DS₂‐VASc score, he or she was able to postpone the action 3 times. During this time, the alert remained active. After 3 times of rejecting the alert, the physician in charge was informed that the alert would permanently disappear from the electronic patient chart. If the physician in charge agreed to calculate the CHA₂DS₂‐VASc score, a new screen with the CHA₂DS₂‐VASc score items opened (Figure S1). The system automatically entered the information for the score items sex and age. The system also calculated the score once the remaining items of the score were entered. In men with a calculated score <1 and women with a score <2, no further information was provided to the physician and the alert disappeared. For all other patients with increased CHA₂DS₂‐VASc score, an additional screen opened, containing the current recommendations from the European Society of Cardiology for stroke prevention in patients with AF with the following text.¹¹

The alert system was tested and adjusted in a passive run‐in phase in collaboration with the IT Department of the University Hospital (Inselspital) Bern. Instructions for use of the alert system were provided in the electronic medical chart and order entry system. In addition, the heads of the medical departments were asked to inform their medical staff about the study.

From September 2014 until October 2015 at the University Hospital Bern, we randomly assigned 1707 patients in a 1:1 fashion to the alert group (n=877) and to the control group (n=830) where no alert was issued. Randomization was performed electronically by automatically generating a number between 1 and 65 535 for each eligible patient. Patients with odd numbers were randomized to the alert group, whereas patients with even numbers were randomized to the control group.

All hospitalized patients aged ≥18 years with AF but without an active OAC prescription in the order entry system were included. There were no exclusion criteria.

The study (ClinicalTrials.gov. Identifier: NCT02455102) was approved by the institutional review board at the University of Bern. Informed consent by the patients was waived for the following reasons: (1) The study did not involve an intervention to patients but to physicians. The intervention served to remind the responsible physician to assess the stroke risk in patients with AF and to consider the prescription of preventive measures if an increased risk of stroke was present. Therefore, the alert was regarded as a clinical decision support system to help the physician to comply with current international consensus guidelines. However, the responsible physician solely carried the responsibility for ordering or not ordering measures to prevent stroke in patients with AF; (2) there was no direct or indirect patient contact during the study (chart review from the hospital stay only; and (3) a consent procedure in control group patients without alert was regarded as unethical and would have confounded the outcome.

As this was a hospital‐wide quality improvement initiative, which involved all departments except pediatrics, approval was also obtained from the hospital management.

The primary end point of the study was the rate of adequate OAC prescription at hospital discharge, defined as prescription of any of the recommended drug regimens in OAC‐naïve men with CHA₂DS₂‐VASc score ≥1 and in OAC‐naïve women with CHA₂DS₂‐VASc score ≥2. Patients were considered OAC‐naïve if they were not receiving OAC within 30 days prior to randomization. The secondary end point was the use of the CHA₂DS₂‐VASc score calculation tool by the physician in charge. We also collected data to calculate the HASBLED score. This score indicates bleeding risk and includes the risk factors hypertension, abnormal renal or liver function, stroke, bleeding history or predisposition, labile INR, elderly, and drugs or alcohol abuse.¹⁷

For sample size calculation, we assumed a 30% rate of adequate OAC prescription in the alert group and a 20% rate in the control group. Using a power of 90% and a 2‐sided alpha of 5%, at least 412 OAC‐naïve AF patients per group were required to reject the null hypothesis. During the recruitment phase, we continuously monitored all randomized patients whether they were OAC naïve and had confirmed AF by medical record review. We planned to terminate the patient recruitment phase once at least 412 OAC‐naïve AF patients per group were eligible for analysis.

Data for baseline characteristics and the primary and secondary end points are presented as absolute numbers and percentages or as means and standard deviations for categorical or continuous variables, respectively. P‐values for differences between the groups with regard to end points are calculated with χ² tests. All P‐values are 2‐sided and P‐values <0.05 were considered significant. For the primary end point, we additionally calculated the relative risk for ordering adequate OAC prescription comparing the alert and control groups. All analyses were performed with IBM SPSS Statistics for Windows, Version 21.0 (IBM Corp, Armonk, NY).

A total of 889 OAC‐naïve AF patients were eligible for analysis (Figure 2). Mean (SD) age was 73.9 (11.3) years. Overall, 359 (40.3%), 48 (5.4%), and 31 (3.5%) patients had paroxysmal, permanent, and persistent AF, respectively, whereas in 451 (50.7%) patients, the type of AF was unknown. Both groups were balanced with respect to baseline characteristics with the exception of a higher rate of systemic hypertension, a trend toward a higher rate of transient ischemic attack, and a trend toward a lower rate of renal dysfunction in the alert group (Table 1). The most frequent reasons for hospital admission were acute coronary syndrome or other cardiovascular disease (30.0%), cancer (10.6%), nonpulmonary infection (7.9%), stroke (6.9%), and heart failure (6.4%) (Table 2). The mean duration of the hospital stay was 9.4 (10.8) days.

Very few patients had a CHA₂DS₂‐VASC score of 0, and there were only 10 women with a CHA₂DS₂‐VASC score of 1 (Table S1). Overall, 856 (96.3%) of the patients were OAC candidates based on the CHA₂DS₂‐VASC score. There was no difference in the proportion of OAC candidates based on the CHA₂DS₂‐VASC score between the alert (443; 97.4%) and the control (413; 95.2%) groups (P=0.083). On the other hand, 391 (44.0%) patients had a HASBLED score of ≥3. There was no difference in the proportion of patients with HASBLED score of ≥3 between the alert (208; 45.7%) and the control (183; 42.2%) groups (P=0.287). There were only 32 (3.6%) patients with a HASBLED score ≥5 (Table S2). There was no difference in the proportion of patients with HASBLED score of ≥5 between the alert (16; 3.5%) and the control (16; 3.7%) groups (P=0.892).

Overall, 100 (22.0%) patients from the alert group versus 69 (15.9%) from the control group received an adequate OAC prescription (Table 3) (relative risk 1.38; P=0.021). In patients who did not receive an adequate OAC prescription at discharge, aspirin 100 mg per day was ordered in 199 (56.1%) patients from the alert group and 183 (50.3%) patients from the control group (P=0.12); an adenosine diphosphate‐receptor antagonist was ordered in 81 (22.8%) versus 81 (22.3%), respectively (P=0.86); dual antiplatelet therapy was ordered in 55 (15.5%) and 63 (17.3%), respectively (P=0.51). OAC or antiplatelet therapy was prescribed in 325 (71.4%) patients from the alert group versus 271 (62.4%) from the control group (P=0.004).

In 48 (10.5%) patients from the alert group, physicians used the electronic entry system to calculate the CHA₂DS₂‐VASc score. Among these, 19 (39.6%) calculations were identical as compared with data obtained from discharge letters and 16 (37.5%) calculations differed by 1 point. Only 1 patient judged to have an increased score according to information from the discharge letter was classified as a low‐risk patient by the physician in charge. In patients from the alert group, an OAC prescription was present in 11 (22.9%) patients whose physicians used the CHA₂DS₂‐VASc score calculation tool and 89 (21.9%) patients whose physicians did not use it (P=0.87).