Clinical decision support systems to optimize adherence to anticoagulant guidelines in patients with atrial fibrillation: a systematic review and meta-analysis of randomized controlled trials

Background Clinical decision support systems (CDSS) have been utilized as a low-cost intervention to improve healthcare process measures. Thus, we aim to estimate CDSS efficacy to optimize adherence to oral anticoagulant guidelines in eligible patients with atrial fibrillation (AF). Methods A systematic review and meta-analysis of randomized controlled trials (RCTs) retrieved from PubMed, WOS, SCOPUS, EMBASE, and CENTRAL through August 2023. We used RevMan V. 5.4 to pool dichotomous data using risk ratio (RR) with a 95% confidence interval (CI). PROSPERO ID: CRD42023471806. Results We included nine RCTs with a total of 25,573 patients. There was no significant difference, with the use of CDSS compared to routine care, in the number of patients prescribed anticoagulants (RR: 1.06, 95% CI [0.98, 1.14], P = 0.16), the number of patients prescribed antiplatelets (RR: 1.01 with 95% CI [0.97, 1.06], P = 0.59), all-cause mortality (RR: 1.19, 95% CI [0.31, 4.50], P = 0.80), major bleeding (RR: 0.84, 95% CI [0.21, 3.45], P = 0.81), and clinically relevant non-major bleeding (RR: 1.05, 95% CI [0.52, 2.16], P = 0.88). However, CDSS was significantly associated with reduced incidence of myocardial infarction (RR: 0.18, 95% CI [0.06, 0.54], P = 0.002) and cerebral or systemic embolic event (RR: 0.11, 95% CI [0.01, 0.83], P = 0.03). Conclusion We report no significant difference with the use of CDSS compared to routine care in anticoagulant or antiplatelet prescription in eligible patients with AF. CDSS was associated with a reduced incidence of myocardial infarction and cerebral or systemic embolic events. Supplementary Information The online version contains supplementary material available at 10.1186/s12959-024-00614-7.


Introduction
Atrial fibrillation (AF) is the most prevalent arrhythmia worldwide [1][2][3].AF increases the risk for stroke up to fivefold, contributing to up to 25% of all strokes [4,5].Societal guidelines in the U.S recommend using the CHA 2 DS 2 -VASc score to quantify the annual stroke risk and guide oral anticoagulation therapy (OAC) with either direct oral anticoagulants (DOACs) or vitamin K antagonists (VKA).A CHA 2 DS 2 -VASc score of one in men and two in women warrants prescribing OAC to reduce the risk of thromboembolic events.However, the CHA 2 DS 2 -VASc score is not recommended in AF patients with moderate to severe mitral stenosis or mechanical heart valves, where VKA is warranted [1,6].
In a meta-analysis including 28,044 patients, prescribing VKA resulted in a 64% relative risk reduction (RRR) of stroke in patients with AF [7].DOACs, including dabigatran, rivaroxaban, and apixaban, showed at least similar stroke prevention efficacy with a favorable safety profile [8][9][10].Despite the significant RRR of stroke by OAC, there has been underutilization of OAC in AF patients [11][12][13][14][15][16].In an observational study involving 94,474 patients who had experienced an acute ischemic stroke and had a history of AF, it was found that 84% of them had not been prescribed OAC before the occurrence of the stroke [17].
Clinical decision support systems (CDSS) have been increasingly utilized as a low-cost intervention to improve healthcare process measures; however, their impact on improving clinical outcomes remains controversial [18].A randomized clinical trial (RCT) showed that an alert system increased the prescription of deep vein thrombosis (DVT) prophylaxis and reduced thromboembolism rates by 41% among hospitalized patients [19].On the other hand, an alert system did not improve clinical outcomes in hospitalized patients with acute kidney injury [20].
Several RCTs were conducted to study the utility of CDSS and alert systems to improve OAC prescription among AF patients to reduce the risk of stroke and systemic embolism potentially.
We conducted this systematic review and meta-analysis of RCTs to investigate the efficacy of CDSS versus routine care regarding adherence to OAC prescription guidelines and stroke prevention in patients with AF.

Protocol Registration
The study's protocol was registered in PROSPERO with the identification number CRD42023471806, following the Preferred Reporting Items for Systematic Review and Meta-analysis of Interventional Studies (PRISMA) statement [21] and the Cochrane Handbook for Systematic Reviews and Meta-Analysis [22] guidelines.

Data sources & search strategy
PubMed, Web of Science, SCOPUS, EMBASE, and CEN-TRAL were searched by authors (A.M.A. and M.T.A.) through August 2023 without publication date, language, or geographical area restrictions.The search was done using [all field] with a mention of the usage of "alert" and "anticoagulant" in "Atrial Fibrillation" Patients.More details are in (Table S1).

Eligibility criteria
Randomized controlled trials (RCTs) that met all of our PICO inclusion criteria were selected: population (P): AF patients; intervention (I): CDSS, including email alert, notification alert, and electronic alerts; comparison (C): patients treated with usual care or no intervention; outcomes (O): our primary outcome was OAC prescription, while our secondary outcomes were patients prescribed antiplatelets and patients prescribed VKA.Additionally, we assessed hard outcomes, including mortality, major bleeding, clinically relevant non-major bleeding, myocardial infarction, stroke/transient ischemic attack (TIA), and thromboembolic events.Exclusion criteria were as follows: primary studies other than RCTs, duplicate publications, reviews, and conference abstracts.

Study selection
Four reviewers (M.T., A.E., O.A., and M.A.) initially screened the titles and abstracts independently using the Covidence platform.After erasing the duplicates, they independently screened the full texts in accordance with our previous eligibility criteria.

Data extraction
Four reviewers (M.A., M.T., A.E., and O.A.) independently extracted data from the eligible studies.M.T.A. and A.M.A. resolved any conflicts.We used an Excel sheet: summary characteristics (study design, country, number of centers, blinding status, registry number, total participants, intervention details, control, participants were on OAC or not, primary outcome, and follow-up duration), baseline characteristics (number of patients in CDSS and control arms, age, gender (male), CHA 2 DS 2 VASc score, HAS-BLED score, and patients' comorbidities (vascular disease, heart disease, diabetes mellitus, hypertension, stroke/transient ischemic attack (TIA), renal disease, liver disease, and prior bleeding).Additionally, the current study outcomes were the number of patients prescribed anticoagulant (OAC), patients prescribed antiplatelets, patients prescribed vitamin K antagonist (VKA), and proportions of why participants were not on OAC.In addition, hard clinical outcomes such as mortality, major bleeding, clinically relevant nonmajor bleeding, myocardial infarction, stroke/TIA, and thromboembolic events were assessed.

Risk of Bias and Certainty of evidence
Four reviewers (M.A., M.T., A.E., and O.A.) independently used the Cochrane ROB2 tool [23] for quality assessment.The reviewers resolved any conflicts by consensus.We evaluated five domains, assessing the risk of bias due to randomization, deviation from CDSS, missing outcome data, measuring the outcome data, and selecting the reported results.
The Grading of Recommendations Assessment, Development, and Evaluation (GRADE) guidelines [24,25] was used by M.T.A. to evaluate the certainty of evidence for each outcome.

Statistical analysis
RevMan v5.3 was used to run the statistical analysis [26].To pool the results of dichotomous outcomes, we used the risk ratio (RR), while for the continuous outcomes, we used the mean difference (MD), both with a 95% confidence interval (CI).We performed both the Chi-square and I-square tests to evaluate heterogeneity, where the Chi-square test detects the presence of heterogeneity, and the I-square test evaluates its degree.I-square was interpreted In accordance with the Cochrane Handbook (chapter nine) [22].as follows: heterogeneity is not significant for 0-40%, moderate for 30-60%, substantial for 50-90%, and considerable for 75-100%.We considered an alpha level below 0.1 for the Chi-square test to detect significant heterogeneity.A leave-one-out sensitivity analysis was employed to resolve the heterogeneity by excluding each study one time from the pooled analyzed studies.
Rstudio (version 4.2.2) was used to conduct a metaanalysis of prevalence using the random effect model with a 95% confidence interval.The I-square test was used to assess for heterogeneity, with I 2 > 50% considered to be of significant heterogeneity.

Search results and study selection
Our literature search retrieved 3,794 unique records.One thousand-five hundred records were removed as duplicates.After title and abstract screening, 91 studies were eligible for full-text screening.Finally, nine studies were included in this systematic review and meta-analysis.The PRISMA flowchart for study selection is shown in (Fig. 1).We have excluded Guo et al. trial [27] due to differences in the intervention compared to our included RCTs' intervention.Patients could upload reports and pictures of the events, unlike our interventions, which are Electronic Medical Record (EMR) based CDSS.

Characteristics of included studies
Nine randomized controlled trials [28][29][30][31][32][33][34][35][36] were included in the meta-analysis with 25,573 AF patients.All the included studies accessed our primary outcome, the number of patients on OAC.The follow-up duration in those studies ranged from three months to 12 months.These studies were conducted in five countries, mainly in the USA (five trials).The summary and baseline characteristics of the included studies are shown in (Tables 1  and 2).More details about the baseline trials' participants' comorbidities and CDSS characteristics are outlined in (Tables S2 and S3).

Risk of Bias and Certainty of evidence
We assessed the quality of included studies according to the Cochrane risk of bias 2 tool, as shown in (Fig. 2).Four included trials had a low risk of randomization process bias (Arts et S4).Certainty of evidence is demonstrated in a GRADE evidence profile (Table 3).

Primary outcome: number of patients on OAC
There was no significant difference in the number of patients prescribed OAC between CDSS compared to routine care (RR: 1.06 with 95% CI [0.98, 1.14], P = 0.16) (Fig. 3-A).The pooled studies were heterogeneous (I 2 = 87%, P < 0.00001).Heterogeneity was not resolved by leave-one-out sensitivity analysis (Table S5).

Discussion
In this systematic review and meta-analysis of nine RCTs involving 25,573 AF patients, we investigated the efficacy of CDSS in oral anticoagulant prescriptions for eligible patients with AF.Key findings include: (1) CDSS was not associated with a significant difference in OAC and antiplatelet prescription rates between CDSS and routine (2) CDSS use was associated with significantly reduced rates of myocardial infarction and cerebral or systemic embolic events (3) There was no significant difference in all-cause mortality, major bleeding, and clinically relevant non-major bleeding between CDSS use and routine care.The Atrial Fibrillation Better Care (ABC) pathway was developed for integrated care for AF patients.It includes a simple approach (avoid stroke, better symptom management, and cardiovascular and comorbidity risk reduction) that guides clinicians through decision-making.In the ABC pathway, prescribing an oral anticoagulant is only one piece of the integrated care approach [37].The ABC pathway has been shown to improve outcomes in patients with AF [38,39].The above approach aligns with AF guidelines, which recommend a patient-centered, holistic approach, necessitating the involvement of multiple stakeholders in AF management decisions.Therefore, CDSS development and application contribute to a more holistic approach to caring for patients with AF, ensuring proper OACs management [40].
Multiple provider-directed interventions have been studied to improve anticoagulation rates among AF patients.For example, email notification to the provider was not associated with increased prescription rates [29].In addition, the Support-AF trial found no benefit to email and inbox notifications [41].Subsequently, electronic health record (EHR)-based CDSS alerts were developed to improve adherence to guidelines and increase anticoagulation rates in eligible AF patients.
Provider-directed EHR CDSS alerts were introduced as a cost-effective intervention to enhance work efficiency and clinical outcomes in inpatient and ambulatory settings.Kawamoto et al. described four essential features of CDSS, including "(a) provide decision support automatically as part of clinician workflow, (b) deliver decision support at the time and location of decision making, (c)  Results demonstrated that process measures were often used as study endpoints rather than patient-related outcomes.128/148 studies assessed healthcare process measures, while only 29/148 assessed clinical outcomes.There was a significant improvement in healthcare process measures, but evidence for clinical outcomes was sparse [18].
We report no significant difference in rates of anticoagulation prescription; however, this finding should be interpreted with caution due to significant heterogeneity    [44].Our results suggested a significant reduction in the incidence of myocardial infarction and cerebral or systemic emboli events.These results align with previous research on CDSS use across different diseases on improving clinical outcomes in other disease states, such as DVT and dyslipidemia [19,43].Additionally, there was no significant increase in bleeding complications.The efficacy and safety outcomes with the use of CDSS were variable.This is likely due to the limited duration of follow-up.The duration of follow-up of 3-12 months in the included studies may be too short to assess the impact on stroke or systemic embolism.Barriers to CDSS tools include alert fatigue, increased number of clicks, time constraints, and clinician burnout [45,46].Arts et al. studied the physicians' perspective of the CDSS; perceived barriers included workflow interruption, increased number of recommendations, and irrelevant recommendations [47].Context-aware CDSS models could help address some of these barriers, possibly by limiting recommendations to a specific encounter [48].
In our study, reasons for not prescribing an OAC included bleeding risk (21%), patient refusal (13%), fall risk (11%), and terminal illness (4%).Incorporation of bleeding and thromboembolism risk scoring tools might be helpful to support clinical decision-making in high bleeding risk patients.Given the high rates of patient refusal, data from the IMPACT-AF trial suggests that patient-directed educational interventions could also lead to a significant increase in anticoagulation rates [49].Patient refusal can be attributed to anticoagulation cost, repeated falls, concerns about bleeding, advanced age, and occupational implications [50].

Limitations
Our review has the following limitations.Firstly, variations in baseline characteristics were noted among different study populations.Secondly, there was notable heterogeneity among studies in the effect size of various outcomes, including the number of patients on anticoagulants, all-cause mortality, and major bleeding.Thirdly, there was notable heterogeneity in CDSS interventions among different studies, which presents a valid concern when interpreting pooled meta-analysis results.The variation in CDSS interventions could explain some conflicting study results well.Fourthly, there was a considerable difference in study weights, which may significantly influence the contribution of certain studies to the pooled results.Given the aforementioned limitations, our study provides a systematic review to accurately interpret the  results of individual studies.Moreover, true heterogeneity is expected in prevalence estimates due to differences in the time and place where the included studies were conducted.I 2 statistics may not be discriminative and should be interpreted with caution in this case.In case of substantial heterogeneity, planned sensitivity analysis can help elucidate the factors associated with the variability among estimates [51].Additionally, hard clinical outcomes were exclusively assessed by the same research group, Piazza et al., in AF-ALERT and AF-ALERT2, with the analysis involving a smaller patient cohort (n = 643).Moreover, challenges in CDSS implementation include a lack of medical informatics expertise in certain centers.

Conclusion
Our meta-analysis underscores CDSS's potential to reduce the incidence of myocardial infarction and cerebral or systemic embolic events in patients with AF.However, we report no significant difference in the rate of prescribing OAC and antiplatelets, all-cause mortality, major bleeding, or clinically relevant non-major bleeding.These insights can guide clinicians in optimizing CDSS use in AF management.

Funding
We received no funding for this study.
Open access funding provided by The Science, Technology & Innovation Funding Authority (STDF) in cooperation with The Egyptian Knowledge Bank (EKB).

Fig. 1
Fig. 1 PRISMA flow chart of the screening process provide actionable recommendations, and (d) use a computer to generate the decision support."[42].CDSS were studied in different clinical conditions with variable efficacy in improving clinical outcomes.Kucker et al. demonstrated increased use of DVT prophylaxis and reduced DVT and pulmonary embolism incidence with CDSS alerts (HR = 0.59, P = 0.001)[19].Van  Wyk et al. showed improved dyslipidemia screening and treatment with CDSS alerts [43].On the other hand, Wilson et al. found no improvement in hospitalized patients with acute kidney injury [20].Bright et al. conducted a large systematic review, including 148 trials assessing the efficacy of CDSS.

Fig. 2
Fig. 2 Quality assessment of risk of bias in the included trials.The upper panel presents a schematic representation of risks (low = green, unclear = yellow, and high = red) for specific types of biases of each study in the review.The lower panel presents risks (low = green, unclear = yellow, and high = red) for the subtypes of biases of the combination of studies included in this review

Fig. 3 Fig. 4
Fig. 3 Forest plot of the primary outcome (prescription of OAC) with the secondary outcome (prescription of antiplatelet and VKA), RR: risk ratio, CI: confidence interval

R
.G., and M.T.A. wrote the final manuscript.B.A. supervised the project.All authors have read and agreed to the final version of the manuscript.
al. 2017, Ashbumer et al. 2018, Bajorek et al. 2016, and Chaturvedi et al. 2018), three had some concerns (Karlsson et al. 2018, Piazza et al. 2019 and Piazza et al. 2023), and two had a high risk (Kapoor et al. 2020 and Silbemagel et al. 2016).All the included studies had a low risk of deviations from intended intervention bias, missing outcome data bias, measurement of the outcome bias, and selection of the reported result bias.Author judgments are further clarified in (Table

Table 1
Summary characteristics of the included RCTs.

Table 2
Baseline characteristics of the participants N., number; SD, standard deviation; N/A: not available

Table 3
GRADE evidence profileamong the included studies.An observational study by Osterland et al. reported no significant change in the trend of anticoagulant use before and after implementation of best practice advisory in eligible ambulatory AF patients Piazza et al. 2019 and Piazza et al. 2023 had some concerns of selection biaas, while Kapoor et al. 2020 and Silbemagel et al. 2016 had a high risk of selection bias b.I-square test CI: confidence interval; RR: risk ratio Explanations a. Karlsson et al. 2018,