Clinical decision support in cardiovascular medicine
BMJ 2022; 377 doi: https://doi.org/10.1136/bmj-2020-059818 (Published 25 May 2022) Cite this as: BMJ 2022;377:e059818
- Yuan Lu, assistant professor1 2,
- Edward R Melnick, associate professor3 4,
- Harlan M Krumholz, professor1 2 5
- 1Center for Outcomes Research and Evaluation, Yale New Haven Hospital, New Haven, CT, USA
- 2Section of Cardiovascular Medicine, Department of Internal Medicine, Yale School of Medicine, New Haven, CT, USA
- 3Department of Emergency Medicine, Yale School of Medicine, New Haven, CT, USA
- 4Department of Biostatistics (Health Informatics), Yale School of Public Health, New Haven, CT, USA
- 5Department of Health Policy and Management, Yale School of Public Health, New Haven, CT, USA
- Correspondence to: H M Krumholz harlan.krumholz{at}yale.edu
ABSTRACT
Despite considerable progress in tackling cardiovascular disease over the past 50 years, many gaps in the quality of care for cardiovascular disease remain. Multiple missed opportunities have been identified at every step in the prevention and treatment of cardiovascular disease, such as failure to make risk factor modifications, failure to diagnose cardiovascular disease, and failure to use proper evidence based treatments. With the digital transformation of medicine and advances in health information technology, clinical decision support (CDS) tools offer promise to enhance the efficiency and effectiveness of delivery of cardiovascular care. However, to date, the promise of CDS delivering scalable and sustained value for patient care in clinical practice has not been realized. This article reviews the evidence on key emerging questions around the development, implementation, and regulation of CDS with a focus on cardiovascular disease. It first reviews evidence on the effectiveness of CDS on healthcare process and clinical outcomes related to cardiovascular disease and design features associated with CDS effectiveness. It then reviews the barriers encountered during implementation of CDS in cardiovascular care, with a focus on unintended consequences and strategies to promote successful implementation. Finally, it reviews the legal and regulatory environment of CDS with specific examples for cardiovascular disease.
Introduction
Despite tremendous progress in tackling cardiovascular disease over the past 50 years, many gaps in the quality of care for cardiovascular disease remain. Multiple missed opportunities have been identified at every step in the prevention and treatment of cardiovascular disease, such as failure to make risk factor modifications, failure to diagnose cardiovascular disease, and failure to use proper evidence based treatments.1 Furthermore, the value of cardiovascular care in the United States is declining as healthcare costs for cardiovascular disease are continuing to rise and previous decreases in cardiovascular disease mortality at the population level are slowing.2 Consequently, cardiovascular care remains both expensive and far short of an evidence based ideal standard of care.
With the digital transformation of medicine and advances in health information technology, clinical decision support (CDS) tools offer great promise to enhance efficiency and effectiveness of delivery of cardiovascular care. The uptake of electronic health record (EHR) systems in hospitals and office practice settings provides opportunities to collect patient level clinical information efficiently.34 CDS tools integrated with the EHR can analyze patient data and trigger timely, actionable, evidence based recommendations to healthcare teams to support clinical decisions.5
However, to date, the promise of CDS delivering scalable and sustained value for cardiovascular care has not been realized. Previous systematic reviews that are not specific to cardiovascular care have found only a modest effect of CDS, with improvements in process of care of less than 5%.6789101112131415161718 Reviews focused on cardiovascular disease to provide more detailed description of CDS innovations in the field are lacking. Cardiovascular disease is an ideal area to study because it is the leading cause of death and covers prevention, acute care, and chronic care.19 Additionally, CDS interventions that intend to provide valuable information to clinicians often result in other unintended consequences when implemented in a real world clinical environment.20 Few studies have summarized the barriers encountered during implementation of CDS in cardiovascular care and strategies to promote successful implementation. This information is important to guide future implementation of CDS for cardiovascular disease and has not been discussed in detail in previous studies. Also, few studies have summarized the ongoing trials in cardiovascular care and identified novel approaches being tested. Finally, the regulatory environment of CDS is changing. Government agencies such as the US Food and Drug Administration (FDA) have released new policies to provide more streamlined oversight for CDS,21 which has important implications for promoting widespread adoption and implementation of CDS. Understanding emerging questions around the development, implementation, and regulation of CDS is critical to optimize the meaningful use of CDS in cardiovascular care.
Accordingly, the objectives of this review are to provide a contemporary assessment of the effectiveness of CDS on a wide variety of healthcare process and clinical outcomes related to cardiovascular disease, to review the barriers encountered during implementation of CDS in cardiovascular care with a focus on the unintended consequences and strategies to promote successful implementation, to review the ongoing trials in cardiovascular care and identify novel approaches being tested, and to review the current legal and regulatory environment for implementation of CDS with specific examples for cardiovascular disease.
Sources and selection criteria
We identified sources through a search of PubMed for English language articles from its inception to 8 July 2020, using the concepts and search terms shown in table 1. We combined concepts by using the Boolean and Proximity operator “AND” combined search terms within each concept by using “OR”. Our inclusion criteria were any observational studies or randomized controlled trials (RCTs) that implemented CDS in a real clinical setting for use by healthcare providers to aid decision making at the point of care for cardiovascular care. We also searched the reference lists of relevant systematic reviews and meta-analyses and included additional selected observational studies and RCTs from these sources. We excluded studies that described non-electronic CDS, included fewer than 50 participants, described patient facing decision aids that did not involve a provider, were not relevant to cardiovascular conditions, did not quantify the effect of CDS, did not contain primary study data analyses such as study protocols and opinion pieces, or were not published in English.
Keywords used in search strategies
For each included study, we extracted the following information into a standardized abstraction spreadsheet: study design, country in which the study was conducted, time period, study setting, sample size, clinical domain of the intervention, intervention and control specifications, study outcomes, unintended consequences reported, user experience reported, barriers to CDS implementation reported, and CDS features associated with effectiveness. Given that studies included in this review compared a wide range of interventions for multiple outcomes, we synthesized the data qualitatively to describe the effects of CDS and identify barriers encountered during implementation. For each subsequent section of the paper, we have focused the discussion on evidence from large RCTs when available and included the results of smaller RCTs and observational studies when of particular interest or when other evidence was unavailable. For the CDS regulation section, we identified and reviewed regulatory documents from government agencies in the US and European Union.
Incidence and prevalence
Cardiovascular disease is a group of heart and blood vessel disorders including coronary artery disease, stroke, heart failure, peripheral vascular disease, and other conditions. Cardiovascular disease is the leading cause of mortality and morbidity globally, accounting for 18.6 million deaths a year, a number that is projected to grow to more than 23.6 million by 2030.19 According to the Global Burden of Disease Study, the incidence of cardiovascular disease worldwide was 55.5 million (684.3 cases per 100 000 people) and an estimated 523 million prevalent cases of cardiovascular disease (6431 cases per 100 000 people) occurred in 2019.1922 From 2010 to 2019, the prevalence of cardiovascular disease increased in almost all non-high income countries; it has also begun to rise in some locations in high income countries where the prevalence of cardiovascular disease had declined for decades. With an aging population and population growth, the number of adults affected by cardiovascular disease worldwide is expected to continue to rise in future years.
Definitions of CDS
A CDS is “any electronic system designed to aid directly in clinical decision making, in which characteristics of individual patients are used to generate patient specific assessments or recommendations that are then presented to clinicians for consideration.”7 CDS tools within EHRs, ranging from simple pop-up alerts to more sophisticated tools incorporating clinical prediction rules, are often used to prompt clinicians to deliver evidence based processes of care, discourage non-indicated care, optimize drug orders, and improve documentation.
Figure 1 illustrates how a clinician would use a specific CDS in a clinical setting. This process is composed of patient data that are exposed to the decision rules, clinical events that trigger the decision rules, alerts that are generated by the decision rules, and actions that are offered as choices to clinicians within the alert context.23
Process of how clinicians would use clinical decision support in a clinical setting. Adapted from McCoy et al23
Overview of studies of CDS interventions in cardiovascular care
We screened 392 unique references from our search and identified 77 studies that met all inclusion criteria (fig 2). Of the 77 studies that evaluated CDS interventions in cardiovascular care, two (3%) were published before 2000, 25 (32%) in 2000-10, and 50 (65%) after 2010 (supplementary table A). Thirty four (44%) studies were conducted in North America, 28 (36%) in Europe, eight (10%) in Asia, six (8%) in Oceania, and one (1%) in South America. Most studies (58; 76%) used a randomized design, with 40 (51%) using a clustered randomized design, allocating intervention status to clinics or provider groups rather than patients. Most studies (59; 75%) took place in outpatient/ambulatory settings, and 18 (23%) studies occurred in emergency department/inpatient settings. The duration of study was ≤12 months for 33 (43%) studies, >12-24 months for 29 (38%) studies, and >24 months for 15 (19%) studies. The clinical areas most commonly targeted were management (30 studies), preventive care (29 studies), diagnosis (10 studies), and screening (eight studies). Whereas most studies assessed CDS in isolation for cardiovascular care, 13 (17%) studies used CDS within a multicomponent approach to implement changes in clinical practice targeted at the patient, provider, organizational, or community levels. These approaches ranged from organizational change such as team based care in which clinicians worked together with pharmacists and nurses to improve healthcare delivery to combining CDS with other implementation strategies such as reminders to patients. Twelve (16%) studies were conducted at a single site, and 65 (84%) studies were conducted at two or more sites. Only one study validated CDS tools in a separate population.24
Flow of studies through review process. CDS=clinical decision support
Effect of CDS on health and health delivery outcomes for cardiovascular disease
As randomized design is considered the most rigorous method for evaluating CDS interventions,2526 we focused on 58 RCTs to evaluate the effect of CDS and barriers to implementation.
Healthcare process outcomes
Overall, 45 of the 58 included RCTs assessed healthcare process outcomes for cardiovascular disease (table 2; supplementary table B). Of these 45 studies, 23 reported a positive effect on improving the healthcare process outcomes of interest, three reported mixed effects, and 19 reported no significant effect. The vast majority (39/45 studies) used usual care or no CDS as the control group without using an active comparator; three studies compared against the same CDS tool with additional features, and three studies compared CDS with other implementation strategies. Among studies that reported a positive effect, common process outcomes measured included CDS recommended preventive service ordered or completed, clinical tests ordered or completed, and treatment prescribed. For example, a multicenter RCT of lipid management that enrolled 105 providers and 64 150 patients from 12 primary care clinics in the US found that CDS with guideline based alerts improved the proportion of participants who were tested for hyperlipidemia by about 5%.27 Another large RCT of 197 Italian general practitioners and 21 230 patients found that the CDS intervention significantly increased the proportion of patients with diabetes treated with antiplatelet drugs or lipid lowering drugs compared with usual care.28 A trial of diagnosis of pulmonary embolism that enrolled 1786 patients from 20 emergency departments in France found that the CDS intervention increased the proportion of patients who received appropriate diagnostic investigations by nearly 20%.29 Among studies that reported no significant effect of the CDS tool being studied on process outcomes, the most common factors identified contributing to the absence of effect were low rate of use of the CDS tool among providers, high baseline rates for the process measure of interest resulting in little room for improvement, improvement of outcomes in the control group because of taking part in the study (that is, Hawthorne effect), and follow-up periods too short to show the effect on outcomes.
Summary of effect of clinical decision support (CDS) by outcome
Clinical outcomes
Forty one of the 58 RCTs assessed clinical outcomes of cardiovascular disease. Of these 41 studies, 10 reported a positive effect of CDS on improving clinical outcomes, four reported mixed effects, and 27 reported no significant effect. Common clinical outcomes assessed in these studies included control of cardiovascular risk factors (for example, rates of control of blood pressure and low density lipoprotein (LDL) cholesterol), cardiovascular events, adverse events (for example, bleeding), and hospital admissions and mortality related to cardiovascular disease. Compared with health process outcomes, fewer studies showed a positive effect on clinical outcomes. Among the 10 studies that reported positive effects, seven studies showed a significant reduction in levels of cardiovascular risk factors, including blood pressure and LDL cholesterol, and three studies showed a significant reduction in cardiovascular events, including hospital admissions for cardiovascular disease, major adverse cardiac events, and pulmonary embolism.
Among 58 RCTs of CDS interventions, 23 focused on acute cardiovascular events and 35 focused on chronic conditions. In general, a higher proportion of CDS interventions for chronic conditions reported a positive effect on process outcomes or clinical outcomes compared with CDS interventions for acute cardiovascular disease events. Specifically, 18 (51%) of 35 RCTs that studied chronic cardiovascular conditions reported a positive effect on process outcomes or clinical outcomes, four (11%) reported mixed effect, and 13 (37%) reported no significant effect. Ten (43%) of 23 RCTs that studied acute cardiovascular disease events reported a positive effect on process outcomes or clinical outcomes, two (9%) reported mixed effect, and 11 (48%) reported no significant effect.
User experience and other implementation outcomes
A total of 13 studies assessed providers’ acceptance of and satisfaction with the CDS intervention being studied. Of these, eight reported good user experience, three reported neutral or unsatisfactory user experience, and two mentioned that the user experience results will be published in future publications. The definition of providers’ acceptance was not consistent across studies. For example, in a multicomponent intervention for management of atrial fibrillation in 209 patients in China, acceptability was defined as satisfaction with the CDS tool,30 whereas in another RCT of 39 clinicians and 781 patients in the Netherlands, acceptability was defined as acceptance of the CDS recommendations.31 Only one study assessed clinicians’ knowledge or improved confidence in managing patient care as the outcome of the CDS intervention.30 No study assessed the effect of CDS on clinicians’ workload or efficiency.
Compliance with use of CDS or CDS recommendations was discussed in 24 studies, of which 13 (54%) reported the rate of compliance. Among these 13 studies, the rate of compliance ranged from 6.3% to 90%, with a median of 41%. Studies that reported good user experience were also more likely to report the rate of compliance with CDS. In general, providers were more likely to be compliant with digital dashboards and smart order sets than with best practice alerts. Some studies reported that a large proportion of alerts were ignored by clinicians. For example, in an RCT of stroke prevention that enrolled 39 clinicians and 781 patients from 18 Dutch general practices, clinicians ignored 60% of CDS alerts, for which the most common reasons included lack of time, too many alert notifications, and limitations of the system’s functionality.31 Clinicians acknowledged the potential of CDS for the future of healthcare, but implementing these systems in daily practice for multiple domains remains challenging. Using alert prioritization, user customization, tight EHR integration, and strict selection of alerts might be helpful to reduce alert fatigue and improve the usability of CDS.
Economic outcomes
Six studies assessed economic outcomes and showed that CDS interventions were associated with reduced treatment costs, total costs, or hospital admission expense compared with control groups. Topics covered in these studies included secondary prevention of stroke and vascular disease, prescription of oral anticoagulants, and management of stroke, hypertension, and diabetes. Two studies reported that CDS was cost effective. In an RCT of 1628 patients from 16 primary health center clusters in India, a CDS intervention for managing hypertension was shown to be highly cost effective in resource constrained primary care settings.32 In another RCT of 3391 patients in 55 primary care practices throughout the Netherlands, a multicomponent intervention combining task delegation, CDS, and feedback had a high cost effectiveness ratio for improving cardiovascular risk for patients with type 2 diabetes.33
Health disparity outcomes
Evidence is lacking on the effect of CDS on reducing health disparities in relation to cardiovascular disease and risk factors. Only three of the 58 studies included in this review measured health disparities as an outcome and assessed the effect of CDS on reducing this disparity.343536 In an RCT of 573 patients in three primary care clinics in the Veterans Affairs Medical Center in the US, a nurse administered hypertension management intervention combining CDS and a behavior program significantly reduced racial differences in blood pressure.34 In another large RCT of 38 725 patients in 60 primary healthcare centers in Australia, a CDS tool for cardiovascular disease risk management in primary healthcare reduced the gaps in care among aboriginal populations.35
CDS tools integrated with EHRs hold promise to reduce health disparities by tackling several causes of disparities.37 Firstly, CDS may help to standardize the collection of social determinants of health and inform patient care for the most vulnerable people with higher health risks. Secondly, underserved populations have a greater burden of cardiovascular disease and are more likely to show signs of poor disease management than other populations. CDS tools may help to prioritize the use of actionable patient data for identifying disparities and tailoring improvement efforts for underserved populations. Better clinical care coordination via CDS could also improve clinicians’ adherence to clinical guidelines, detect treatment risks, and consequently facilitate equitable treatment for underserved populations. Thirdly, CDS tools that are culturally tailored may help to enhance patients’ engagement and safety and to reduce adverse outcomes in underserved populations.
The US federal health information technology strategic plan 2015-20 calls for research evidence on how health information technology can reduce disparities in the quality, accessibility, and safety of healthcare and long term support services.38 The National Institutes of Health also released funding opportunities for leveraging health information technology to tackle minority health and health disparities.39 Given the national attention on health equity and promise of CDS for reducing disparities in underserved populations, studies are urgently needed to examine the effect of adopting CDS on health disparities in quality of care, cost, and health outcomes.
Features associated with CDS effectiveness
Four of the 58 included trials directly compared a given CDS intervention with the same system with additional features (for example, providing additional information on patients’ symptoms), but they did not show a significant beneficial effect of the additional features.31404142 Although evidence is lacking from the studies included in this review, several previous reviews have identified design features that are closely related to the success of CDS. A large systematic review analyzing 70 RCTs of CDS interventions found that four features—including automatic provision of CDS as part of workflow, providing CDS at the time and location of decision making, providing recommendations in addition to assessments, and using computer based CDS—were independent predictors for improving clinical practice.7 Other large systematic reviews echoed these findings and identified additional features associated with positive effects of CDS. Specifically, CDS interventions were found to be more likely to succeed when they automatically prompted users,43 minimized the need for manual input of patient data, increased the specificity and sensitivity levels of CDS advice,44 provided advice for patients in addition to practitioners, and were evaluated by their developers.45 A recent systematic review of 108 RCTs showed that CDS requiring acknowledgment and documentation of reasons for not following the recommendations was associated with about 5% larger effects than CDS without this feature.6 The ability to execute the desired action through the CDS and considering alert fatigue in designing or delivering CDS were also associated with larger effects, but the incremental changes for these features were relatively small (fig 3). Free standing, non-EHR integrated CDS could complement EHR integrated CDS when the integration is difficult and data inputs are not readily available from the EHR. For example, simple tools for risk stratification such as the atherosclerotic cardiovascular risk score calculator are commonly provided as free standing tools that require manual input of parameters. However, given the development of health information technology, increasing effort has been made to integrate these tools into the EHR to provide actionable insights to providers.46
Clinical decision support (CDS) features associated with larger effects on outcomes
Barriers to successful dissemination and implementation of CDS
Despite the great potential for CDS to improve care and patient outcomes in cardiovascular care, our review of the literature and previous systematic reviews consistently found low use of CDS by providers across studies. The slow uptake of CDS in clinical practice highlights the need to understand barriers to successful dissemination and implementation of CDS in real world settings.
Of the 58 included RCTs, 31 reported barriers encountered during implementation of CDS interventions as assessed by post-intervention surveys of or interviews with providers. The most common barriers were time and resource constraints, lack of compatibility with workflow, alert fatigue, technical problems with the CDS or EHR system, discordance between local guidelines and CDS recommendations, lack of trust in CDS recommendations, and complexity of real world clinical management of cardiovascular disease (table 3). Notably, most of these barriers pointed to an underappreciation of the complex sociotechnical environment of the real world clinical settings in which the CDS was implemented.54 Other types of barriers mentioned in the studies were providers’ fear of doing harm, lack of awareness and training of providers, lack of financial incentive, and lack of buy-in by providers. As a result of these barriers, many CDS interventions failed to deliver the “five rights” of CDS (the right information to the right people in the right intervention formats through the right channels at the right points in the workflow),55 thereby limiting their effect on improving patient care.
Common barriers encountered during clinical decision support (CDS) implementation and strategies to tackle barriers
Time and resource constraints
Twelve studies reported time and resource constraints and lack of compatibility with workflow as barriers to implementation of CDS. In an RCT of hypertensive disorders in pregnancy that enrolled 2286 patients in 16 Dutch hospitals, clinicians reported that the use of CDS was time consuming and not well integrated into daily practice routines.47 Ultimately, the CDS was used in only a quarter of eligible patients. Likewise, in an RCT of asthma and angina management that included 8365 patients from 60 general practices in north east England, use of CDS was low owing to the challenge of integrating the CDS tool into clinical encounters in which busy practitioners manage patients with multiple, complex conditions.56 In another large RCT of an intervention combining a point-of-care device for testing lipids and glycated hemoglobin with a web based CDS tool for assessment of cardiovascular disease risk, 13 638 patients from 20 general practices in New Zealand were enrolled. Nurses reported feeling so time pressured with their workload that having to wait for the test results and the sequential analysis of cardiovascular risk by the machine was not good use of their time.57
Alert fatigue
Seven studies reported alert fatigue (clinicians’ tendency to ignore repeated alerts) as a major negative unintended consequence when implementing CDS in real world settings. This low signal-to-noise ratio for alerts was likely due to violations of one or more of the five rights of CDS mentioned above, particularly not providing the right information at the right time in the workflow. For example, in a large RCT of drug management in cardiovascular patients at high risk that enrolled 197 general practitioners and 21 230 patients in Italy, the main reason reported for discontinuing the use of CDS was a large number of alerts.28 In the CDS-AF study, 14 134 patients from 43 primary care clinics in Sweden were enrolled to study the effect of CDS on improving adherence to guidelines for anticoagulant therapy in patients with atrial fibrillation. Primary care physicians ignored the recommendation or made a decision that the patient would not benefit from therapy given the false alerts for patients without appropriate treatment and the additional workload imposed by the CDS.58 In another RCT of intraoperative hypotension that included 3156 patients in the Hillcrest Hospital in Cleveland, USA, anesthesiologists reported that they ignored the CDS because the alert did not provide actionable recommendations (for example, anesthesiologists were already aware of hypotension and doing their best to treat it) or the responses were ineffectual even if the alerts provided additional information (for example, anesthesiologists were distracted by the tool). Although the studies included in this review did not report the number of alerts providers received, a study published in JAMA Internal Medicine in 2016 found that primary care providers receive an average of 77 notifications every day.59 Although not all of those notifications are generated by the CDS systems—many came from laboratories, pharmacies, or other physicians—this large number of notifications suggests that adding CDS alerts to an inbox already overflowing with input can do more harm than good. Besides alert fatigue, another unintended consequence reported in the studies was distraction from current workflow. Inappropriate alerts added to cognitive load can lead to reduced efficiency and increased number of medical errors.6061
Difficulty following recommendations
Ten studies reported discordance between local guidelines and CDS recommendations, lack of trust in CDS recommendations, and complexity of real world clinical management of cardiovascular disease as barriers to CDS implementation. A study of 1493 patients from 15 primary care practices in the UK assessed the effect of a CDS tool for atrial fibrillation in primary care. Post-visit survey results found that more than half of the providers discussed anticoagulation treatment with their patients, yet only 6% of them actually made a change in therapy at that visit given the nuance of real world clinical situations.49 The most frequent documented explanations were patients’ preferences, anticoagulation therapy managed by another specialist, and concerns about increased risk of falls in older patients. In the IMPART trial of venous thromboprophylaxis in acutely ill medical patients, which enrolled 1085 patients from 10 hospitals in Switzerland, CDS was used by only 30% of physicians; even when the CDS was used, only 75% of physicians followed the recommendation made by the tool.48 The major reasons reported were providers’ distrust of CDS recommendations and discordance between local policies of thromboprophylaxis and the CDS recommendations.
Technical problems with CDS or EHR system
Five studies reported technical problems with the CDS or EHR system as a barrier to CDS use. In a large RCT of a multifaceted intervention to prevent venous thromboembolism that enrolled 15 351 patients from 27 hospitals in France, the authors were able to implement plug-in reminders in only two centers. Two centers had no EHR system, four had an EHR system incompatible with the plug-in, and five had hospital policies that did not allow implementation of the plug-in.50 In another study of cardiovascular risk reduction that included 7914 patients from 12 primary care clinics in Minnesota, USA, implementation of CDS was delayed because the CDS containing the URL to display the CDS tool took longer to develop than anticipated.62 Notably, despite the investment in EHR and health information technology, adoption and maintenance of EHR systems involve high operational costs.63 Many small sized and rural hospitals may not be able to afford to invest in EHR systems or health information technology infrastructure, leading to inequalities in accessing and using CDS services. This highlights the digital divide as a determinant of widening inequities in cardiovascular care and outcomes. A critical need exists to bolster the funding of initiatives and passage of legislation to advance the accessibility and utility of digital health technologies, particularly for underserved populations.
Strategies to promote successful implementation
Despite multiple national efforts to advance adoption and implementation of CDS,64656667 35 of the 58 RCTs included in this review did not explicitly mention adoption of any specific implementation strategies. Among the 23 studies that explicitly mentioned specific implementation strategies, the most common strategies were integrating CDS into clinical workflow and the EHR, audit and feedback, using more precise triggers to reduce alert fatigue, and conducting ongoing provider training (table 3). Other strategies included obtaining support from local leaders, conducting local consensus meetings, providing financial incentives, engaging front line providers in the design of CDS, and culturally adapting the CDS intervention to the study population. Many of these strategies pointed to more deliberate processes for stakeholder engagement and buy-in and continuous usability testing, which are endorsed by multiple professional organizations and federal agencies.64656667
Notably, successful studies that reported positive effectiveness of CDS and good user experience have applied specific strategies to optimize dissemination and implementation of CDS. In an RCT of 7914 patients from 12 primary care clinics in Minnesota, USA, strategies used to achieve high rates of CDS use included training of primary care physicians, engaging primary care physicians and nursing leaders for workflow integration, providing monthly feedback of CDS use rates to intervention clinic managers and primary care physicians, and providing compensation to clinics that sustained CDS use rates of more than 75% of targeted patients.62 The trial showed significantly positive change in absolute 10 year cardiovascular risk in clinics using CDS compared with clinics in the usual care arm and reported high provider satisfaction. In another trial of hypertension management in adolescents that enrolled 31 579 patients from 20 primary care clinics within a large midwestern US health system, the CDS tool was designed to be integrated with clinical workflow.52 Provider alerts were delivered only when at least two blood pressure measurements were recorded at a visit and the average was at or above the 95th centile for age and sex based on the clinical guideline for hypertension management in children and adolescents.68 Audit and feedback to encourage nurses to repeat blood pressure measurements and providers to open the CDS along with in-person training sessions were conducted at intervention sites. As a result, most providers using the CDS reported that it was useful for pragmatic decision making. In the TORPEDO trial of 38 725 patients in 60 primary healthcare centers in Australia, strategies used to facilitate implementation included workflow integration, alignment with usual decision making processes in the patient consultation, provision of treatment recommendations rather than just assessments, and repeated audit and feedback with explicit recommendations.35
Current legal and regulatory environment for CDS
To advance toward widespread adoption and implementation of CDS, recognizing and managing external factors—such as the policy, legal, and governance factors that affect the process of developing, disseminating, and implementing CDS interventions—are important. An appropriate regulatory framework for CDS should seek to achieve an optimal balance between promoting technology innovation and protecting patients. To achieve this goal, the EU and the US use different approaches.69 In this section, we outline the legal and regulatory landscape for CDS in the US and EU and describe developments that are taking place.
US approach
The current legal and regulatory landscape for CDS in the US is largely affected by the FDA’s medical device regulations. According to the FDA’s 2019 draft guidance on CDS,70 if a healthcare professional can independently review the basis for a recommendation, a CDS is not considered a medical device and hence not subject to oversight by the FDA. Specifically, the FDA used criteria from the 21st Century Cures Act and defined a non-device CDS as software that meets all four of the following criteria: not intended to acquire, process, or analyze a medical image or a signal from an in vitro diagnostic device or a signal acquisition system; intended for the purpose of displaying, analyzing, or printing medical information about a patient; intended for the purpose of supporting or providing recommendations to a healthcare professional about prevention, diagnosis, or treatment of a disease or condition; intended for enabling a healthcare professional to independently review the software recommendations to make a clinical diagnosis or treatment decision for patients.71 A device CDS, also known as software as medical device (SaMD),72 is a CDS that does not meet one or more of the four criteria above. The FDA applies a risk based approach to provide oversight on SaMD as described below.
FDA’s risk based approach for SaMD oversight
In 2017 the FDA adopted the International Medical Device Regulators Forum framework in its risk based approach to SaMD regulation.73 This framework characterizes risk of SaMD on the basis of two main factors: the significance of information provided by the SaMD to the healthcare decision and the state of the healthcare situation or condition. As shown in table 4, the categorization of risk occurs across a continuum, ranging from the lowest risk software functions of informing clinical management for non-serious conditions (that is, “inform × non-serious”) to the highest risk software functions of diagnosing or treating critical conditions (that is, “diagnose/treat × critical”).
International Medical Device Regulators Forum framework for software as medical device (SaMD) risk categorization
According to the risk of SaMD, the FDA classifies devices into three distinct classes (class I low risk devices, class II moderate risk devices, and class III high risk devices) and regulates them accordingly (table 5): the higher the risk, the stricter the control. Most of the class I devices and all class II devices are regulated by the FDA through the pre-market notification (PMN) pathway (commonly known as 510(k) application). This is an expedited pathway that allows devices to obtain marketing authorization if the sponsor can show substantial equivalence to an existing legally marketed device. Class III devices are devices involving the greatest risk and, therefore, are regulated by the FDA through the more stringent pre-market approval (PMA) pathway. This pathway requires extensive scientific evidence including technical, non-clinical laboratory, and clinical investigations to demonstrate a device’s safety and effectiveness before FDA approval. The FDA also has another pathway, de novo pre-market review, for new low to moderate risk devices without an equivalent predicate device. The de novo pathway provides an opportunity to reclassify a novel class III device to a class I or class II device, which is then subject to less stringent regulation. Table 5 shows specific examples of CDS tools in cardiovascular care for each device class.
Summary of regulatory policy for clinical decision support (CDS) for which intended user is healthcare professional
CDS tools that are designed, developed, and validated using the health system’s unique knowledge of protocols, performance characteristics, and means of analysis to provide clinical information to clinicians in the health system are regulated by the FDA through a different pathway than the traditional PMN/PMA/de novo pathways for other SaMD. Assuming that the in-house developed CDS tools are not distributed commercially, they are considered as practice of medicine similar to the laboratory developed testing services, and hence are subject to the Clinical Laboratory Improvement Amendments, overseen by the Centers for Medicare and Medicaid Services.74 Examples of CDS tools recently approved by the FDA for cardiovascular disease care are shown in supplementary table C.
FDA’s new software precertification program
Concern exists that the traditional device approval/clearance pathway discussed above is not suited to the faster, more iterative design and development cycle of devices, which takes weeks to months rather than the months to years long cycle of more traditional medical products. The FDA released the Digital Health Innovation Action Plan in July 2017,75 announcing that it was reimagining the approach to digital health medical devices. In 2018 the FDA started the Software Precertification Pilot Program,76 which is considered a voluntary alternative to the traditional PMN/PMA pathway to provide more streamlined oversight of digital devices and accelerate their time to market. This program aims to shift the focus from solely pre-market evaluation to evaluating both the product and the company by continuously monitoring the real world performance of its products once they have reached the market. Although the program is still in the development phase, the current version has four interdependent components77: excellence appraisal and precertification, review pathway determination, streamlined pre-market review process, and real world performance (table 6).
Four interdependent components of US Food and Drug Administration (FDA)’s Software Precertification Program
To respond to the covid-19 pandemic and to provide more efficient oversight in the midst of the pandemic, the FDA has also suspended requirements for certain “lower risk device” software. For the subset of “higher risk” digital health devices, the FDA is using the emergency use authorization process to help to expand access to medical products for use during the pandemic.
EU approach
The EU is reforming the legal framework for medical devices with several new legislative reforms (General Data Protection Regulation (GDPR), Cybersecurity Directive, and Medical Devices Regulation). Although reform is a gradual process, the GDPR and the Cybersecurity Directive enacted in May 2018 have already begun to have an impact.
The definition of a medical device in the EU includes any kind of software intended by the manufacturer to be used for human beings for the purpose, among others, of diagnosis, prevention, monitoring, treatment, or alleviation of disease.78 This definition has been endorsed by the MEDDEV guidelines drafted by the European Commission to guide stakeholders in complying with legislation related to medical devices.79
The regulatory framework for CDS tools in the EU is largely affected by three directives on medical devices created in the 1990s.788081 These directives require manufacturers to comply with several essential requirements depending on the risk classification of the device and to ensure that the produced devices are fit for their intended purpose. Whether the essential requirements have been met can be assessed either by the manufacturer or by an independent accredited certification organization appointed by the competent authorities of EU member states.
Recognizing that the existing directives do not fit with new, evolving technologies, the EU issued a new Medical Devices Regulation in May 2017.82 This new regulation, officially applied from May 2020, extends the scope to include a wider range of products, extends the liability in relation to defective products, strengthens requirements for clinical data and traceability of the devices, provides more rigorous monitoring of independent certification organizations, and improves transparency through making information relating to medical devices available to the public. Different from directives that require national legislation to implement their purposes, the new regulation is applied directly in EU member states without the need for national legislation to implement.
Data privacy considerations for CDS using patient data
Developers and implementers of CDS tools using patient data should be aware of the various laws that protect health information privacy.83 In the US, the Health Insurance Portability and Accountability Act of 1996 (HIPAA) established national standards for the protection of patients’ health information.84 Development of CDS using data from patients who have consented to their use is permitted, and the HIPAA requires that patients be given the right to direct any covered entity to transmit a copy of their medical records to a designated entity of the individual’s choice.85 However, several exceptions to the requirement of patient consent under HIPAA exist. For example, protected health information may be used by CDS without patients’ consent to support selected quality improvement activities, clinical guideline development, and population based activities relating to reducing costs.86 Besides federal laws, several states in the US have their own data privacy protection laws and regulations that must be followed. Thus, companies that plan to operate in multiple states need to ensure their compliance with each state’s laws.
In the EU, the GDPR enables people to control their own health data. In particular, the GDPR has regulations that outline that people have the “right to an explanation” when it comes to machine learning algorithms.8788 In practice, providers need to inform patients that a CDS or machine learning algorithm was used in clinical decision making. The “right to an explanation” means that people have the right to obtain information about the logic involved in the automated decision making system, its significance, and any resulting consequences. They also have the right to object to any decision made about them if that decision is based purely on automated processing.89 Diagnostic tools previously not considered medical devices may be considered medical devices under the new regulations if they have a purpose of “prediction and prognosis.”90
Guidelines
High quality CDS studies are needed to learn the best ways to apply CDS systems to achieve important improvements in healthcare delivery and outcomes; however, limited guidance is available for designing and reporting CDS studies. A recent article summarized the key problems encountered in previous CDS studies and proposed 13 recommendations for research and reporting of CDS studies (fig 4).91 If adopted, these recommendations could help to improve the quality of CDS studies and ultimately fulfill the CDS promise of more efficient and effective care. Eventually, CDS interventions, especially in high stakes settings, should be evaluated with the same rigor as other therapeutics.
Recommendations for research and reporting of clinical decision support (CDS) studies. EHR=electronic health records. Adapted from Kawamoto K, McDonald CJ. Ann Intern Med 2020;172:S101-991
In addition to reporting guidelines, guidance on best practices in dissemination and implementation of CDS based on real world experience is also lacking. The 2017 National Academy of Medicine on Optimizing Strategies for Clinical Decision Support highlighted the importance of disseminating best practices for CDS.67 The report proposed convening expert groups to cultivate, plan, and direct the publication of actionable implementation guides that draw on existing efforts to delineate best practices in implementation and platform integration approaches for different delivery systems; CDS management approaches for organizing multi-stakeholder CDS implementation and governance committees and for clinicians and health systems of various sizes/resources; and usability recommendations for usable, practical, workflow supportive CDS for various situations.
Emerging advances
To describe advances in CDS, we searched ClinicalTrials.gov by using “clinical decision support” and “clinical decision” as keywords and restricting disease conditions to cardiovascular diseases and trial status to “recruiting” or “active, not recruiting.” We identified 29 ongoing CDS trials with a focus on cardiovascular conditions, such as heart failure, stroke, myocardial infarction, atrial fibrillation, diabetes, hypercholesterolemia, hypertension, and cardiac arrhythmia (supplementary table D). These ongoing trials include several advances in CDS.
Developing and testing multilevel interventions that incorporate CDS to improve cardiovascular care—The RICH LIFE Project is examining the effectiveness of a multilevel intervention in improving blood pressure control and reducing its disparities among 1890 adult patients with uncontrolled hypertension and cardiovascular risk factors at 30 primary care practices in Maryland and Pennsylvania.92 The multilevel intervention includes standardized blood pressure measurement training, audit and feedback of blood pressure control rates at the practice provider level, community health worker referrals, subspecialist curbside consults, and an ongoing virtual workshop for organizational leaders in quality improvement and reduction of disparities.
Leveraging CDS to reduce disparities in quality of care and outcomes of cardiovascular disease—The CV Wizard study is testing whether an innovative, point-of-care, web based CDS system will reduce disparities in control of cardiovascular risk factors and in rates of heart attacks and strokes among the low income, racially/ethnically diverse Americans who receive primary care at safety net community health centers.93
Testing implementation models of delivering CDS in diverse settings—The ADHINCRA study is testing the feasibility of a multilevel, nurse led, mobile health enhanced intervention in patients with uncontrolled hypertension in Ghana.94
Integrating multiple data sources (for example, data from wearables, sensors, genetic testing, pharmacy claim) with EHR data to augment the ability of CDS to support clinical decision making for cardiovascular disease—In the PAVS study, patients are given a Fitbit pedometer and their Fitbit account can sync to their MyChart account. After that sync, patients’ step counts are available for their cardiologists to review as needed.95 Patients are assessed by the physical activity vital sign during check-in for their appointment.
Applying best practices in CDS design, including human centered design to customize CDS and to improve usability—In a study led by the University of Colorado Health, researchers are using user centered design to customize a commercially available CDS tool in the EHR to improve prescribing of β blockers for patients with heart failure.96
Using advanced analytics and automated algorithms to inform disease risk assessment, diagnoses, and treatment decision making for cardiovascular disease—In the ABC-AF Study, researchers are evaluating whether personalized treatment by decision support, based on the biomarker based risk prediction (ABC scores) guided strategy, reduces the occurrence of the composite outcome of stroke or death in patients with atrial fibrillation.97
Using innovative digital healthcare technologies (for example, cloud based platform) to facilitate information sharing and shared decision making between patients and providers—In the AIM-POWER study, researchers are assessing the effectiveness of the cloud based CDS platform to manage initiation and titration of guideline directed medical therapy outside of normal clinical encounters.98 Cloud computing is the use of computing resources that are delivered as a service over a network.99 As one of the emerging strategic information technologies, cloud computing is promising for its cost efficiency and its potential to provide quality information services in the healthcare industry.100 Centralized, cloud based solutions could provide standardized and centralized knowledge artifacts to standardize care recommendations across diverse practice settings at low cost.
CDS that targets non-physician providers to improve care for cardiovascular disease—Researchers in the Veterans Administration are testing the effect of a multifaceted intervention with pharmacist and interactive voice response to improve adherence to clopidogrel after percutaneous coronary intervention.101
Conclusion
Despite a rapid increase in publications of CDS studies in cardiovascular care over the past two decades, this review shows marked heterogeneity in the design, implementation, and evaluation of CDS interventions. Evidence shows the positive effectiveness of some CDS interventions in improving cardiovascular disease care process outcomes, such as screening and preventive care services, ordering recommended clinical tests, and prescribing recommended treatments to mitigate the risk of cardiovascular disease. The context in which the CDS is applied may strongly influence the results, highlighting the concomitant importance of implementation science. The results on improvement in cardiovascular disease related clinical outcomes are more mixed, and evidence is lacking on the effectiveness of CDS for improving other implementation outcomes, economic outcomes, and health disparities as they relate to cardiovascular disease. The uptake of CDS in cardiovascular disease care remains slow, and barriers to implementation exist at multiple levels. Many of these barriers are due to a lack of adequate understanding of the end users’ needs, a lack of EHR integration, and problems with usability. Although the search strategy in this review may have limited some of the articles included, our results are consistent with the broader CDS literature.
To advance the development, evaluation, and implementation of CDS in cardiovascular care, we propose several recommendations as follows. Firstly, in the field of cardiology, crucial data elements (for example, imaging data) are often not available as structured data in electronic health records. Therefore, future CDS interventions should have the ability to integrate both structured and unstructured data elements by using more advanced techniques such as natural language processing. Integrating multiple data sources (for example, data from wearables, sensors, genetic testing, pharmacy claim) with EHR data is also important to augment support for clinical decision making in complex cardiovascular disease. Moreover, best practice in CDS design, including the combination of human factors engineering and iterative user centered design, is essential to improve usability and minimize user burden in clinical practice. Such an approach will provide a better understanding of what the right information is and when and how it should be delivered to the right person. It will also allow CDS developers to incorporate patients’ perspectives into the recommendations so that CDS systems are configured in such a way that patients have confidence in their healthcare professionals.102 Furthermore, more research is needed to study the unintended negative consequences of CDS use on clinical workflow, patient-clinician and clinician-clinician communication, and health outcomes. Pilot testing should be conducted before full implementation to avoid unintended consequences. Finally, broadening adoption of CDS relies on collaboration among multiple sectors and stakeholders. Industry, including EHR and CDS vendors, could have a leadership role in helping to identify and test standards; knowledge from frontline providers is crucial to understanding priorities, usage, workflow, and design; and professional societies are positioned to disseminate best practices and guidelines. Convening an authority for standard setting might offer a vehicle for customers’ voices to motivate changes with vendors, as well as to facilitate collaboration between societies, multiple vendors, and different specialties to develop building blocks for improvement.
Taken together, the available evidence shows that CDS has yet to realize its potential in cardiovascular care. Learning from more than two decades of evidence and establishing guidelines and best practices in designing, implementing, evaluating, and reporting CDS interventions are crucial to fulfill the promise of CDS to enhance care delivery, accelerate system-wide continuous learning, and improve healthcare outcomes for cardiovascular disease.
Glossary of abbreviations
CDS—clinical decision support
EHR—electronic health record
FDA—Food and Drug Administration
GDPR—General Data Protection Regulation
HIPAA—Health Insurance Portability and Accountability Act
LDL—low density lipoprotein
PMA—pre-market approval
PMN—pre-market notification
RCT—randomized controlled trial
SaMD—software as medical device
Research questions
What is the effectiveness of clinical decision support (CDS) on patient centered outcomes, economic outcomes, and health disparity outcomes related to cardiovascular disease?
Can cardiovascular CDS achieve long term sustainability?
Can CDS be effective for other members of the healthcare team (such as nurses and pharmacists)?
How can CDS for cardiovascular diseases be expanded to accommodate multiple comorbid conditions simultaneously, and how can new CDS products be integrated with existing electronic health record systems?
Considering implementation of CDS for cardiovascular diseases in real world settings ranging from small physician practices to large health systems and across a variety of workflows, which implementation models are more efficient than others and in what settings?
Footnotes
Series explanation: State of the Art Reviews are commissioned on the basis of their relevance to academics and specialists in the US and internationally. For this reason they are written predominantly by US authors
Contributors: YL, ERM, and HMK conceived and designed this review. YL did the search and selected the studies for inclusion. YL drafted the manuscript, and ERM and HMK edited and approved the final version.
Competing interests: We have read and understood BMJ policy on declaration of interests and declare the following interests: HMK has received personal fees from UnitedHealth, Element Science, Aetna, Reality Labs, F-Prime, Siegfried & Jensen Law Firm, Martin/Baughman Law Firm, and Arnold and Porter Law Firm and grants from Johnson & Johnson; he is a co-founder of HugoHealth, a personal health information platform, and co-founder of Refactor Health, an enterprise healthcare artificial intelligence-augmented data management company, and he has contracts from Centers for Medicare & Medicaid Services Contracts, through Yale New Haven Hospital, to develop and maintain performance measures that are publicly reported outside the submitted work; YL is supported in part by the National Heart, Lung, and Blood Institute (K12HL138037) and the Yale Center for Implementation Science; ERM is supported in part by the National Institute On Drug Abuse of the National Institutes of Health under Award Number UH3DA047003. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.
Patient involvement: No patients were asked for input in the creation of this article.
Provenance and peer review: Commissioned; externally peer reviewed.