Leveraging electronic health records to streamline the conduct of cardiovascular clinical trials

Abstract

Conventional randomized controlled trials (RCTs) can be expensive, time intensive, and complex to conduct. Trial recruitment, participation, and data collection can burden participants and research personnel. In the past two decades, there have been rapid technological advances and an exponential growth in digitized healthcare data. Embedding RCTs, including cardiovascular outcome trials, into electronic health record systems or registries may streamline screening, consent, randomization, follow-up visits, and outcome adjudication. Moreover, wearable sensors (i.e. health and fitness trackers) provide an opportunity to collect data on cardiovascular health and risk factors in unprecedented detail and scale, while growing internet connectivity supports the collection of patient-reported outcomes. There is a pressing need to develop robust mechanisms that facilitate data capture from diverse databases and guidance to standardize data definitions. Importantly, the data collection infrastructure should be reusable to support multiple cardiovascular RCTs over time. Systems, processes, and policies will need to have sufficient flexibility to allow interoperability between different sources of data acquisition. Clinical research guidelines, ethics oversight, and regulatory requirements also need to evolve. This review highlights recent progress towards the use of routinely generated data to conduct RCTs and discusses potential solutions for ongoing barriers. There is a particular focus on methods to utilize routinely generated data for trials while complying with regional data protection laws. The discussion is supported with examples of cardiovascular outcome trials that have successfully leveraged the electronic health record, web-enabled devices or administrative databases to conduct randomized trials.

Graphical Abstract

A visual representation of how leveraging routinely collected data and technology can streamline all aspects of randomized controlled trials. EHR, electronic health record.

Open in new tabDownload slide

Introduction

While randomized controlled trials (RCTs) have played a central role in advancing cardiovascular medicine, contemporary RCTs have become highly complex and resource-intensive.^1,2 Most cardiovascular RCTs are designed to maximize estimated treatment effect and minimize adverse events (i.e. explanatory design), rather than demonstrate effectiveness in everyday clinical patients and settings (i.e. pragmatic design).³ Explanatory trial designs involve the use of narrow inclusion criteria that select patients most likely to derive benefit from the intervention. These inclusion criteria are often enriched for the primary outcome to ensure an adequate event rate to minimize sample size and shorten the time required to detect a treatment effect.⁴ The need to increase pragmatism in RCTs is being increasingly recognized. The Pragmatic Explanatory Continuum Indicator Summary (PRECIS-2) tool allows investigators to assess the pragmatism of a trial design in each of nine domains: eligibility, recruitment, setting, organization, flexibility in delivery, flexibility in adherence, follow-up, primary outcome, and primary analysis.⁵ The move towards pragmatism in cardiovascular outcome trials (CVOTs) could potentially benefit from leveraging real-world data and technology—which have the potential to reduce workload for investigators, increase the speed of evidence generation, and ameliorate costs, all while yielding more generalizable findings.

The advent and implementation of electronic health records (EHRs) have revolutionized healthcare delivery by allowing better organization and access to patient data, improved adherence to clinical guidelines, and reduced prescription errors, among other advantages.^6–9 The use of EHRs has been adopted in virtually all hospitals in the USA and Europe, as well as many other regions of the world where RCTs are led.¹⁰ Many jurisdictions, in some cases entire countries, have registries that systematically capture consecutive patient data from a range of healthcare settings. Technological devices such as smartwatches, smartphones, and fitness trackers can record heart rhythm, pulse, and other biometric measures, enabling cardiovascular health to be tracked with unprecedented detail and scale.¹¹ Biomedical and healthcare data volumes are increasing faster than most other industries,¹² with data being generated from multiple sources as highlighted in Figure 1.

Integrating RCTs with existing data-generating systems (EHRs, registries, and wearable sensors) can potentially automate and streamline the identification of eligible patients, patient recruitment, informed consent, randomization, follow-up monitoring, outcome ascertainment, and outcome adjudication (Graphical Abstract). This, in turn, can reduce the trial burden on site personnel and participants, and decrease costs.¹³ Access to data routinely collected in clinical care for the purpose of trial recruitment can potentially broaden access to RCT participation. Moreover, the decreased organization and costs associated with using routinely generated data can allow the conduct of trials that compare healthcare policies or treatment strategies (different drug combinations or sequencing), i.e. trials which do not seek a regulatory label and are thus less likely to be industry funded. The conceptual workflow involved in a trial conducted using technology and routinely generated healthcare data is summarized in Figure 2. Despite its potential, the implementation of data systems and technology into the conduct of RCTs has lagged. In this review, we discuss how integration of access to real-world datasets and technology can optimize the conduct of CVOTs, the current barriers to integration, and potential solutions.¹⁴

Patient identification, informed consent, and recruitment

Challenges in conventional cardiovascular trials

Inadequate recruitment is the most important reason for premature termination of CVOTs; up to 80% of conventional RCTs do not reach their planned enrollment targets.¹⁵ To ensure adequate recruitment, enrollment duration and number of participating sites are often increased, each of which increases costs.¹⁶ In most trials, accrual rates are slow at the start, followed by a steep mid-section, and then a plateau which represents trial fatigue.¹⁷ In longer trials, trial fatigue can lead to site discontinuation, patient dropout, and increased cross-over across treatment arms. To deal with these consequences, protocol amendments and changes in the trial endpoint are often required. A study of 836 clinical trials showed that almost half of them (45%) had at least one substantial protocol amendment.¹⁸ The median number of amendments was 2.3 in a Phase 3 clinical trial. Factors associated with amendments included larger sample size, longer patient recruitment durations, and slow accrual rates.¹⁸

A common reason for recruitment difficulties in conventional trials is the use of narrow eligibility criteria. While trials often select patients at increased risk for the outcomes of interest, they also commonly exclude the very highest risk patients, such as older adults and those with extensive and advanced comorbidities, due to the competing risks of death and elevated safety/tolerability concerns. For example, more than 25% of heart failure (HF) trials have excluded patients by an arbitrary upper age limit, and over 80% exclude patients with one or more serious or advanced comorbidity.¹⁹ Given that elderly patients with multiple comorbidities make up a substantial proportion of patients with cardiovascular disease, this compromises the generalizability of the results to the broader population of afflicted patients.

Other common reasons for poor recruitment include limited availability of routinely collected data on potentially eligible trial participants to inform study design and recruitment strategies, and a lack of clarity in understanding the trial objectives and inclusion criteria among site investigators.²⁰ Moreover, current recruitment practices often result in eligible patients being missed, due to study coordinators not being notified that there is a potential patient eligible for the study.

Role of routinely generated data and technology

The integration of routinely collected data into standard RCT operations provides a potential solution to streamline the process of pre-screening and obtaining informed consent, while facilitating enrollment of a more diverse and representative patient population. The EHRs or registries can be prospectively queried to identify sites with high volumes of trial-eligible patients.^20,21 In addition to identifying the volume of trial-eligible patients, EHRs and administrative databases can be queried to estimate event rates for outcomes of interest. For example, Rapsomaniki et al.²² developed a model using an EHR cohort of >100 000 patients with stable coronary artery disease which could reliably predict all-cause mortality (c-index: 0.81) and the composite of non-fatal myocardial infarction or coronary death (c-index: 0.78). This has an important implication for clinical trials—the estimated event rates from EHR can assist with the calculation of power.

During the recruitment phase, systems can be put into place to automatically pre-screen patients for eligibility for trials. This process is relatively straightforward if patient characteristics of interest are available in a structured and standardized form within the EHR. In the absence of structured data, clinical narratives (i.e. free text) in the EHR can be analysed using natural language processing to identify potentially eligible patients. Natural language processing is a subfield of artificial intelligence that enables the computer to analyse free text using a collection of processing algorithms based on statistics, syntactic, and/or semantic rules. It has well-established utility for pre-screening patients and is increasingly being used to pre-screen patients for cardiovascular disease trials.²³

Alerts within EHRs can be used to systematically inform providers of patients potentially eligible for a trial during routine clinical encounters. At the same time, the provider can be provided with a short description of the trial. In 2011, a pilot RCT was conducted to compare weight-based insulin dosage with a sliding scale insulin dosing regimen in hospitalized patients for the control of hyperglycaemia.²⁴ For this RCT, the Veterans Affair Healthcare System was optimized to automatically identify eligible patients (those for whom insulin was being ordered) and notify treating providers of potential recruitment for the trial. If a potentially eligible patient was identified by the system, the treating provider would be presented with a short summary of the trial and then given the option to enrol the patient. With agreement of the treating provider and consent of the patient, the research team would randomize patients. Strategies such as these can considerably reduce the need for additional research personnel and trial-specific training for site investigators to encourage recruitment. They can also ensure that potentially eligible patients are not missed—thereby improving the inclusiveness of trials and speed of enrollment.

Sometimes, clinical trials seek to recruit participants with a specific severity of symptoms. For example, almost all CVOTs use the New York Heart Association (NYHA) classification as part of the screening criteria. Unfortunately, the NYHA class is often not recorded in charts and even if it were, the known interphysician variability in its assessment can minimize its validity.^25–28 The growing availability to the internet through smart devices can enable the collection of validated patient-reported health status, which has greater reproducibility, sensitivity, and prognostic value than clinician-assigned assessments. This can help identify sufficiently symptomatic patients to meet the goals of the RCT without the requirements and costs of a trial-specific face-to-face visit. Additionally, if self-reported health status suggests that the patient requires clinical care, the smartphone application can notify both the patient and the treating physician regarding this.^29,30

Once eligible patients are identified from EHRs or registries, consent may be obtained remotely. Patients can be informed about risks, benefits, and financial disclosures via videos that use patient-centred language and minimal medico-legal jargon. This was exemplified in the Aspirin Dosing: A Patient-Centric Trial Assessing Benefits and Long-term Effectiveness (ADAPTABLE) trial, which used a US Food and Drug Administration (FDA)-compliant electronic informed consent method.^31,32 Potentially eligible patients (identified through EHRs) were provided with a link to the trial web portal, along with a personalized code to access it. Within the portal, participants were presented with an informational video about the trial and a consent form. The consent form was adapted locally at each site and approved by the institutional review board of the site. An on-site, in-person consent process was available for those who were uncomfortable with technology or had no internet access. The FDA requires verification of the identity of the person signing the consent form, but does not mandate a particular method for verification. Possible verification methods include the use of information from official identification such as birth certificate or driver’s license, or the use of security questions.³³

From an investigator perspective, remote e-consent for research allows contact with a much larger number of patients compared with the traditional practice of approaching patients encountered in routine clinical practice. From the patient perspective, bringing the trial to the patient, rather than the patient to the trial, is more convenient and can facilitate engagement with relatives to help make an informed decision on whether to participate. The potential downside of video-based consent and recruitment is that it may be challenging for certain patients. For example, non-native English speakers may not be fully ‘informed’ prior to making a decision, unless translation is provided. Similarly, elderly patients who are unfamiliar with new technology may find this process difficult.

Data collection

Challenges in conventional cardiovascular trials

Conventional RCTs involve extensive data collection, with multiple protocolized trial-specific visits, diagnostic tests, and surveys routinely required to gather data. More than a fifth of these procedures, although complementary to the trial integrity, are not directly related to the core objectives of the RCT.^34,35 For industry-sponsored RCTs being conducted for product registration, partial or complete source data verification is conducted—a practice wherein the collected data are checked by study monitors to ensure that they are complete and consistent with study participant source records. Source data verification is expensive and can consume a large proportion of the trial budget.^34,35 Moreover, source data verification is responsible for a major chunk of the overall carbon footprint of clinical trials.^34,36 However, it remains unclear if source data verification yields any benefit.^34,35 The FDA does not require extensive source data verification and recommends limiting verification to essential data points or when anomalies are suspected.³⁷ Despite its unproven utility, it is commonly used due to a conservative interpretation of regulations and the notion that it preserves integrity of trial results.³⁴

Role of routinely generated data and technology

The use of routinely collected clinical data can reduce the burden of trial-specific data collection. Patients’ demographics, clinical course, diagnostic test results, treatment regimens, and clinical events of interest are typically updated in the EHRs or registries with every in-person or telehealth visit with the provider. In many countries, data from hospital records are systematically coded and transferred to federal administrative datasets. With the appropriate technological infrastructure in place, this data collection may be standardized and captured for use in clinical trials. For investigators and stakeholders, automated data standardization and capture from EHR, registries, and digital devices can reduce the need for personnel and sites to support data collection and reduce the risk of transcribing errors. Estimates show that automated data capture to populate RCT case report forms can result in a >5% reduction in the cost of cardiovascular clinical trials.^34,35 Automated data collection may also allow much longer and detailed follow-up than conventional trials, increasing sensitivity to capture important rare or delayed adverse effects. Importantly, to make routinely generated data optimally ‘research-accessible’, technological infrastructures, processes, and codification of terms need to be broadly generalizable and ideally harmonized, such that any cardiovascular RCTs could capitalize on such access.

Data derived from wearable devices, smartphones, and other technologies in everyday use can be used to remotely track specific parameters of cardiovascular health of enrolled participants. Large-scale e-registries that gather cardiovascular data already exist. A pertinent example is the MyHeart Counts study, which obtains data on physical activity, fitness, and sleep via a smartphone application.³⁸ Almost 50 000 participants across the USA voluntarily consented and were enrolled into this study within 8 months since the launch of the application. Similarly, the Health eHeart study—which enrolled over 200 000 participants—also collects cardiovascular health and fitness data remotely through a smartphone application (https://www.pcori.org/research-results/2015/health-eheart-alliance). In addition, a subset of participants is given gadgets, such as wearable sensors and Bluetooth-linked blood pressure measuring devices, to gather more in-depth data. The Apple Heart Study is an example of a technology-based, pragmatic study design where a smartwatch was used to detect irregular rhythm in >25 000 trial participants, and linkage to administrative data is being used to assess the clinical outcome benefits of being exposed to the device.³⁹ The notifications for irregular pulse are transmitted through an application to study investigators following which participants were sent an electrocardiogram patch to characterize the irregular rhythm, determine if atrial fibrillation was present, and among those with atrial fibrillation, classify it as paroxysmal or continuous. The Apple Heart Study, MyHeart Counts, and Health eHeart highlight the potential of technology to enhance the process of obtaining consent, collecting data, and recruiting large sample populations. Moreover, data gathered from commonly used devices and wearables in the community may provide detailed insights that may not be gleaned from clinical evaluation. This was demonstrated by the results from the Nitrate’s Effect on Activity Tolerance in HF With Preserved Ejection Fraction (NEAT-HFpEF) trial,⁴⁰ which studied the effect of isosorbide dinitrate on activity levels in patients with HF and preserved ejection fraction. Two concurrent outcomes were used to evaluate physical activity—the conventional 6-min walk distance and daily physical activity level measured by wearable activity monitors with high-sensitivity accelerometers. Isosorbide dinitrate (all doses vs. placebo) was found to have no effect on physical activity when measured using the 6-min walk distance, but daily activity was significantly reduced by isosorbide dinitrate when measured using the wearable sensors. This discrepancy highlights that conclusions from RCTs are often based on limited data that are gathered during office visits and in restricted/selected patient cohorts, and the use of technology may increase sensitivity by obtaining more complete and continuous information, without placing additional burden on the participants. Observational studies conducted using wearable technologies would exclude participants who cannot afford or do not have access to smartphones and wearable devices—which may create further inequities in healthcare research. However, this is less of a concern in RCTs, where the required devices are often provided to enrolled participants.

Once data have been captured from EHRs or registries, algorithms may be applied to identify suspicious data patterns in clinical trials. The algorithms can potentially identify anomalous values at least at two levels: at the participant level and site level. Potentially erroneous data at the participant level can then be verified by manually cross-checking the case report forms with the source data. If one or more sites have data that deviate considerably from the overall pattern, the sites can be assessed for systematic errors in data collection or entry, which can then be rapidly addressed. This risk-based monitoring approach to clinical trial surveillance can allow the early detection of site failure, fraud, and data inconsistencies or incompleteness—while simultaneously reducing the costs associated with source data verification.⁴¹

Recently, it has been shown that linking genetic biorepositories with EHR can facilitate the conduct of genome-wide association studies (the identification of various genetic variants associated with a phenotype) and phenome-wide association studies (the identification of various phenotypes associated with a genetic variant).^42,43 Moreover, clinical outcomes and treatment allocation may be improved if the identified gene-phenotype associations are returned to the EHR and used to support clinical decisions (e.g. by incorporating genetic information into prognostic risk scores).^44,45 This has important implications for the conduct of embedded clinical trials. If EHRs linked to biorepositories are used to conduct RCTs, the data can be used to identify patients with genotypes/phenotypes, which are most likely to respond positively to the intervention. This can potentially allow precise treatment allocation, which maximizes efficacy and safety, and minimizes futile use of the intervention.⁴⁶

Barriers to the implementation of routinely generated data, possible solutions, and current progress

Despite the immense potential of using existing information systems for clinical trials, several barriers need to be overcome before this process can become a standard (Figure 3).

Inaccurate and incomplete codification of routinely generated data

Incomplete codification of data in EHRs and registries represents one of the biggest barriers for EHR-embedded research. Provider notes within the EHR are extensively and almost exclusively in the form of free text.⁴⁷ Formal diagnosis lists, and problem lists that map onto objective clinical codes, are often insufficiently curated in the EHR, greatly compromising their utility for screening and follow-up of patients for research. In addition to being incomplete, the codified data are often inaccurate. The International Classification of Disease (ICD) diagnosis and procedure codes, a comprehensive system established by the World Health Organization to harmonize and codify clinical and procedural annotations, has been adopted by healthcare systems across the globe. However, there is no common imperative to document these codes (at least in the USA), and the codes have questionable sensitivity and specificity, as highlighted by Fawcett et al.⁴⁸ in their study of diagnosis codes for infective endocarditis. Only 44% of total cases assigned with endocarditis codes actually had endocarditis, while 24% of patients with possible or definite endocarditis were not assigned their respective ICD codes.⁴⁸

Potential solutions and current progress

To allow appropriate use of structured/coded EHR data for research, completeness and accuracy of structured data need to improve. To overcome incomplete availability of structured data, EHRs should be designed to curate data that are more complete for use in research. Optimization of these systems for research would require guidelines as well as incentives for manufacturers. The inaccuracy of codified data can be addressed by continued evolution in training and practice for clinicians to accurately document diagnoses in problem lists and history domains. In the absence of accurately codified EHR data, evaluation of clinician notes (free text) may be a more reliable method to determine diagnoses. However, in RCTs, clinician notes are traditionally reviewed by personnel—which adds to the workload and costs. Artificial intelligence tools, such as natural language processing, have the potential to parse unstructured text (including but not limited to clinician notes) in the EHR and convert it to structured data. Clinicians may use different terms to refer to the same thing (such as ‘hypertension’ and ‘elevated blood pressure’)—natural language processing can accumulate similar terms under the same code. The structured data may then be used for research purposes.⁴⁹ Although the accuracy of natural language processing–derived data has increased greatly over the past decade, there remains room for improvement, and models should be internally validated before use for research. Research studies utilizing structured EHR data should follow best practices, such as those laid out in the recently published CODE-EHR framework.⁵⁰

Patient privacy and legal concerns

Conducting embedded RCTs requires sharing of patient data across healthcare systems and with investigators, inevitably placing patient privacy at risk. This raises the critical question about how patient data can be shared for the successful conduct of RCTs, while maintaining patient confidentiality and data security. The use of interoperable EHRs for clinical care is already underway, especially in Europe. For example, in the UK, almost 75% of health systems have some means in place to electronically access patient data from other healthcare systems. The eHealth Digital Service Infrastructure (eHDSI) is an infrastructure that further allows exchange of data across countries in the European Union (EU). By 2025, patient summaries and prescription information should be exchangeable across 25 countries in the EU. Both the General Data Protection Regulation (GDPR) in the EU and the Health Insurance Portability and Accountability Act of 1996 (HIPAA) in the USA require measures to ensure security of health information. These measures include limiting access to authorized individuals, encryption of stored data, and an ‘audit trail’ feature (which records who accessed a patient’s information, what changes were made, and when).^51,52 While these measures are sufficient for clinical care, policies on data sharing for research are less simple.

The GDPR in the EU and the Common Rule (a policy by the Department of Health and Human Services) in the USA regulate the use of personal patient data for research. Both allow the use of personal patient data for research if the data are appropriately de-identified or if the patients have provided informed consent for research. However, the definitions of ‘appropriate de-identification’ and ‘consent’ should be noted. The GDPR is more stringent than the Common Rule. For example, in order to waive the requirement for consent, GDPR requires data to be anonymized—the highest form of de-identification. Anonymization involves the removal of all personal data, including direct identifiers (e.g. name, telephone number, and social security number) as well as indirect identifiers (e.g. date of birth, gender, ethnicity, and location). Moreover, GDPR does not consider data as anonymized if personal data are coded and can be reversed to uncover the identity of patients. Additionally, GDPR requires that consent is obtained from patients for the specific research project for which their personal data will be used, rather than ‘broad consent’. The GDPR does recognize that it may not always be possible to fully identify the purpose for which the data will be utilized, and in these cases allows the purpose to be described at a more ‘general level’ while obtaining consent—however, it does not specify the level of generality that is acceptable.⁵³

In the USA, HIPAA outlines a set of 18 specific identifiers that must be removed or coded in order for data to be considered de-identified.⁵⁴ Moreover, the Common Rule and HIPAA do not consider coded information as ‘identifiable private information’ as long as steps are taken to prevent the investigator from obtaining means to link the code to the patient’s identity. Furthermore, the Common Rule allows for ‘broad consent’ from individuals for the storage, maintenance, and secondary research use of their identifiable private information.⁵⁴ Both the GDPR and the Common Rule do not consider silence, pre-ticked boxes, and inactivity as consent. For example, the consent statement should read ‘please tick here if you consent to sharing your personal data for this study’, rather than ‘please tick here if you refuse to share your personal data for this study’.⁵³

These differences have important implications for the conduct of embedded trials. For example, in the USA, investigators may be able to use coded patient data to pre-screen patients and invite them to participate in a trial while maintaining their anonymity. Once a patient consents to participation in a trial and processing of their personal data, the data can be uncoded to identify the participants. Pre-screening would be more difficult in the EU, where reversibly coded patient data are protected by the GDPR.

Potential solutions and progress

Sharing and utilizing patient data for research while complying with legal requirements are complex, but possible. In the EU, the Electronic Health Records for Clinical Research (EHR4CR) project exemplifies this. The EHR4CR is a sustainable research platform that links the EHR systems of 34 academic and pharmaceutical partners in Europe. The software ‘InSite’ was developed to allow researchers to query the EHRs of all hospitals in the EHR4CR system in order to estimate the number of eligible patients for a clinical trial at each site, assess the feasibility of the trial protocol, and fine-tune the protocol if necessary. The researchers query completely anonymized data, making this solution compliant with GDPR. The EHR to electronic data capture (EHR2EDC) builds on the success of EHR4CR. Once a patient consents to participation in a trial, the EHR2EDC platform exports relevant data from the patient’s EHR to the trial investigator’s system. This can minimize or eliminate the need for manual data collection—reducing costs, preventing errors, and saving time.

While the EHR4CR is a useful tool for protocol development, the UK has also established ‘Secure Data Environments’ that can allow the conduct of EHR-enabled RCTs while remaining GDPR compliant. The use of such a system was successfully employed in the High-Sensitivity Troponin in the Evaluation of Patients with Suspected Acute Coronary Syndrome (High-STEACS) trial.⁵⁵ High-STEACS was a stepped-wedge cluster-randomized trial in which 10 hospitals in Scotland were randomly assigned either early or late implementation of a high-sensitivity cardiac troponin I (hs-cTnI) assay to diagnose myocardial infarction. The Community Health Index is a population healthcare register that includes all individuals living in Scotland. For this trial, each patient’s Community Health Index number was used to link multiple data sources—including EHRs, databases containing investigation results and treatments, and national registries containing hospitalization and mortality records. The linked data were held securely in the ‘NHS safe haven’ of each health system and used for screening, enrollment, and outcome adjudication of each patient. These data were then anonymized (thus making them shareable under GDPR regulations) and transferred to a national analytical platform for analysis. The High-STEACS trial showed that implementation of hs-cTnI assays did not reduce the risk of re-infarction or death at 1 year. This trial was unique in that patient written informed consent for research was not required throughout, as determined by Scotland, a Research Ethics Committee, the Public Benefit and Privacy Panel for Health and Social Care, and by each NHS Health Board.⁵⁵ No informed consent for research was required for enrollment as the intervention was implemented at the hospital level, and no informed consent for research was required for data sharing as data were anonymized before it left the health system. The High-Sensitivity Cardiac Troponin on Presentation to Rule Out Myocardial Infarction (HiSTORIC) trial, which evaluated the use of hs-cTnI to rule out myocardial infarction early, was designed in a similar manner to avoid the need for patient informed consent for research.⁵⁶

In the USA, the Health Information Technology for Economic and Clinical Health (HITECH) Act was passed in 2009, which led to the widespread adoption of EHRs across US healthcare systems through financial incentives and penalties for non-compliance.⁵⁷ The HITECH Act also pushes for health information exchange between healthcare systems. The National Institutes of Health (NIH) Health Care Systems Research Collaboratory was founded in 2012, with the aim to provide a platform for designing and implementing EHR-embedded RCTs. In 2014, the Patient-Centered Clinical Research Network (PCORnet) was launched.⁵⁸ This is a program that links the EHR systems of nine large clinical research networks. The ADAPTABLE was the first trial to employ the PCORnet tool. The trial applied computational phenotype modelling, an efficient method of converting study inclusion criteria into a query, to EHRs across multiple participating sites to identify potentially eligible participants, provide trial invitation from a central centre, negotiate informed consent virtually, and collect all trial data from EHR documentation.³¹

Thus, EHR4CR and PCORnet demonstrate the feasibility of developing reusable digital platforms that can optimize the conduct of trials while complying with regional laws. The ‘TriNetX’ platform applies these principles on an international level. TriNetX is a solution that allows researchers to query the EHRs of several healthcare organizations across several countries, including the USA, Japan, Singapore, Malaysia, Israel, and India. This is made possible due to its federated nature—i.e. data are retained within institutional boundaries, and researchers can access only anonymized patient data. In 2019, TriNetX acquired the InSite platform, thus expanding its network to include European healthcare organizations that are part of the EHR4CR system.

Gatekeepers and costs

Even within the same country, health systems may have different EHRs, most often with limited capacity for integrated data handling and analyses.⁵⁹ In a survey in the USA, one-quarter of health information exchange leaders believed that health systems force hospitals and individual providers to adopt particular EHR systems instead of creating an interoperable interface between existing EHRs,⁶⁰ motivated by financial interests and market dominance. Half of the respondents also reported EHR vendors intentionally making the interface not amenable to information exchange and charging high fees for data sharing. Thus, these EHR vendors act as ‘gatekeepers’ to data sharing.⁶¹ Moreover—although EHR-embedded trials have been touted as economically advantageous—the initial investment required for setting up appropriate EHR infrastructure with seamless interoperability between different healthcare systems to facilitate CVOTs conduct is big. Additional costs may be incurred due to the requirement of personnel with deep systems knowledge and big data analytic skills—which is crucial for the operation and analysis of EHR-embedded trials.

Potential solutions and progress

To allow the development of interoperable EHR systems, gatekeepers must be brought to the table for discussions and swayed towards the idea of integration. National level policies, along with incentives for gatekeepers, may assist in overcoming this barrier.⁶¹ In addition to the EHR4CR and PCORnet systems discussed earlier, another example of efficient data sharing is the SWEDEHEART registry, which records all patients admitted to a coronary care unit with a suspected or definite acute coronary syndrome in Sweden and Iceland.^62,63 This registry was leveraged to conduct the Thrombus Aspiration in ST-Elevation Myocardial Infarction in Scandinavia (TASTE) trial, as well as the Bivalirudin versus Heparin in ST-Segment and Non–ST-Segment Elevation Myocardial Infarction in Patients on Modern Antiplatelet Therapy (VALIDATE) trial. These trials used the registry for all aspects of the trial, including screening, enrollment, capture of baseline characteristics, follow-up data, and outcome acquisition. EuroHeart, an initiative of the European Society of Cardiology (ESC), is a continuous registry being developed across ∼20 countries in Europe.⁶⁴ National registry programmes will be set up in each country using a common data set of standardized variables, and the local infrastructure will belong to the respective participating country. The aim of the project is to provide infrastructure to support observational and randomized research, as well as safety surveillance of new cardiovascular drugs and devices. This project is a step up from ESC EuroObservational Research Programme (EORP), which aimed to promote and monitor the implementation of new treatments and adherence to ESC guidelines—but was criticized for its substandard data quality and coverage of common cardiovascular diseases.⁶⁴

Moreover, fees to utilize services of these interoperable systems can be steep, with exact costs depending on the services rendered. For example, EHR4CR services can range from identification of potentially eligible patients to complete infrastructure support of a trial, with corresponding estimated costs ranging from 50 000 to 500 000 Euros (as of 2012).⁶⁵ However, ultimately, the efficiency gains achieved by clinical trial sponsors (i.e. the person-time saved) can substantially reduce the cost of conducting clinical trials. As per their website, TriNetX (which now includes the EHR4CR network) has allowed researchers to test over 39 000 trial protocols and present over 10 000 trial opportunities to healthcare organizations.⁶⁶

Varying data formats across electronic health record systems

The EHRs are built using a wide variety of data models, with the coding format varying from one EHR system to another, i.e. different EHR systems often use different codes for the same data point. Thus, most analyses focus on data from a single database, using a single analysis method customized to the underlying data model and its terminologies. Thus, even after data sharing between EHRs, the heterogeneity in coding across systems poses a major challenge for the use of the data for research purposes. However, this challenge can be overcome, as discussed below.

Potential solutions and current progress

An analysis across multiple disparate databases must either tailor the analysis to accommodate each of the underlying data models and terminologies or code the data into a common format. For example, in the USA, the PCORnet community developed the Common Data Model, which standardizes codes for data points across multiple EHR systems into a common format. This allows researchers to use a single query to obtain a particular data point from the EHRs of millions of patients. International consortia such as the Observational Medical Outcomes Partnership (OMOP) and Clinical Data Interchange Standards Consortium (CTAC) have helped make progress towards standardization of the format of routinely generated data for research and regulatory submissions.^67,68 The Academic Research Consortia, which are collaborations between clinical experts, various academic research organizations, and the US FDA, aim to develop consensus definitions and standard nomenclature for use in cardiovascular clinical trials.⁶⁹ Expansion of such efforts to create standardized data definitions for EHRs can play an important role in the advancement of embedded trials.

Exposure fatigue due to electronic health record–based alerts

The EHR-based alerts regarding patients potentially eligible for an RCT—although an attractive option to increase and simplify recruitment—can lead to exposure fatigue and lack of response from providers. Results from a cross-sectional study evaluating the utility of EHR alerts for potential trial eligibility demonstrated that providers became less likely to respond to alerts after several weeks of exposure to EHR-based alerts. Although the decline was statistically significant, it was slow, and response rates remained reasonably high (30%–40%) even after 36 weeks of exposure.⁷⁰

Potential solutions and current progress

Exposure fatigue to EHR alerts can be reduced by increasing the specificity of alerts, wherein physicians are alerted when the likelihood of the patient’s eligibility for a trial is relatively high. A systematic approach to assessment and reduction of alert burden has been proposed—and has been successfully carried out by large healthcare systems such as Geisinger Health and Penn Medicine.⁷¹ Moreover, human factor principles can be applied when designing alerts, with care given to the format, content, legibility, and the colour of alerts. Additionally, alerts should be designed to be customizable by the physician, giving them the flexibility of turning off alerts for some or all clinical trials or reducing the intensity of alerts. Indeed, it has been shown that user-configurable alerts have greater acceptance among users.⁷²

Absence of health-related quality of life data in routinely collected data

Assessment of health-related quality of life (HRQoL) data, a patient-relevant outcome, is a crucial component of pragmatism in RCTs. Over the past three decades, the science of validly collecting patients’ experiences of their disease has evolved, and the FDA has even qualified some questionnaires as clinical outcomes assessments, meaning that drugs and devices can be approved if significant improvements in patients’ health status can be demonstrated, even if there is no difference in clinical events. However, HRQoL is typically not assessed in a clinical encounter and therefore is not accurately represented in EHRs, registries, and administrative datasets.

Potential solutions and current progress

Growing internet connectivity, combined with widespread smartphone use, can allow collection of patient-reported outcomes through ‘apps’—a rich, important, and low-cost option for outcomes collection. The recently completed Canagliflozin: Impact on Health Status, Quality of Life and Functional Status in HF (CHIEF-HF) trial used the Kansas City Cardiomyopathy Questionnaire as the primary outcome, and the entire trial was conducted completely during the Covid-19 pandemic without a single face-to-face visit.⁷³

Challenges with outcome adjudication

Traditionally, central committees have been established to adjudicate outcome events using clinical annotations and trial-specific definitions pre-specified by charter for each trial. While event adjudication is widely considered to be the best practice, there is conflicting evidence on whether adjudication reduces misclassification of events.^74,75 The use of routinely collected data to ascertain outcomes has the advantage of being automated, independent, and highly cost effective. However, the accuracy of automated outcome adjudication using the non-standardized free text in EHR and claims data has been called into question.

Potential solutions and current progress

The most appropriate method to overcome the challenges with automated outcome adjudication would be to make EHR data available in a standardized format. Countries in the EU have started to unify EHRs and employ standardized diagnostic coding, and accuracy of standardized coding is improving as these data are increasingly used for healthcare quality improvement and reimbursement. Furthermore, guidelines have helped to standardize definitions and the use of diagnostic coding in the delivery of RCTs is gaining popularity. Results from systematic reviews suggest that effect size estimates are unaltered when analysing site investigator–reported events or routinely collected data.⁷⁴ Studies comparing results ascertained using events identified by claims data vs. those confirmed by trial event adjudication committees show that both methods yield similar findings of all-cause death. However, findings from claims data often differ from adjudicated data for non-fatal and non-procedural cardiovascular outcomes—likely due to the absence of these data in structured format or due to non-standardized definitions for these outcomes.^76,77 In the absence of structured data, nuanced trial outcomes may be determined using machine learning algorithms which analyse free text in EHRs. Results from few recent studies have shown that natural language processing is able to ascertain clinical outcomes with ∼90% accuracy.^78–80 Once further refined, tested, and adopted, techniques such as this can potentially eliminate the need for adjudication of outcomes by central clinical trial committees.

Contemporary guidelines and regulations

In recent years, both international and national guidelines and regulatory bodies have become more amenable to the idea of RCTs layered onto routine clinical practice. In March 2015, international reporting guidelines for studies conducted using routinely collected health data were published that encourage use of the REporting of studies Conducted using Observational Routinely-collected health Data (RECORD) checklist.⁸¹ In April 2021, the International Council of Harmonization released an early draft of the updated version of the 1996 Good Clinical Practice guidelines, which advocates for use of electronic consent and calls for incorporation of technology in existing healthcare infrastructure to enable use of a greater variety of data sources.^82,83 In May 2021, an extension of the Consolidated Standards of Reporting Trials (CONSORT) guidelines was published in order to standardize the elements that should be recorded and reported in RCTs conducted using routinely collected data.⁸⁴

In Europe, in June 2021, the European Medicines Agency (EMA) released a draft version of guidelines for the use of computerized systems and electronic data in clinical trials.⁸⁵ This draft is currently undergoing revisions based on public consultation, and an implementation date will soon be announced. In December 2021, the Medicines & Healthcare products Regulatory Agency (MHRA) published a guidance on the use of real-world data in clinical studies to support regulatory decisions.⁸⁶ Moreover, in August 2022, the ESC and the BigData@Heart consortium together published the CODE-EHR best practice framework for the use of structured EHRs in clinical research.⁵⁰

In the USA, in September 2021, the FDA issued requisites that trialists must adhere to when utilizing EHR and claims data for generation of evidence for new medical therapies, in order to support regulatory decision-making.⁸⁷ Moreover, in December 2021, the FDA issued a draft guidance for the use of digital technologies (such as wearable sensors and smartphones) for remote data acquisition in clinical investigations.⁸⁸

Examples of ongoing and completed trials that leverage routinely generated data and technology

Examples of ongoing and completed trials that leverage routinely generated data and technology are discussed in Supplementary material online and summarized in Tables 1 and 2.

A call to action

The Clinical Trials Transformation Initiative (CTTI) has laid out five goals for the conduct of clinical trials, to be met by 2030. Figure 4 summarizes how leveraging routinely generated data can help cardiovascular trials achieve these goals. Similar goals and solutions have been proposed by expert panels in the UK and USA, which represents convergent thinking among academics across continents and across disciplines.^101,102

Conclusion

Conducting cardiovascular RCTs using routinely generated healthcare data from EHRs, registries, and administrative databases has the potential to streamline all aspects of trial conduct, including the identification of eligible patients, recruitment, consent, randomization, follow-up, and outcome ascertainment. The EHR-embedded trials can reduce costs, thus allowing the enrollment of larger and more heterogeneous participant populations and longer follow-ups. Using routinely collected data to enhance enrollment of patients with broader eligibility criteria can also allow a shift towards pragmatism by yielding findings that are more generalizable to routine clinical practice. However, to successfully utilize routinely generated data to streamline cardiovascular RCTs, there is a need for technological innovations which allow capture of accurate data from diverse databases and standardize these data such that they can be used for analysis. Importantly, technological infrastructures which make routinely generated data ‘research-worthy’ need to be transferable such that multiple cardiovascular RCTs can be embedded into healthcare system through them over time. Moreover, these systems should align with patient privacy and data protection standards. Despite these barriers, certain cardiovascular trials have successfully utilized EHR systems to conduct large-scale trials. Given the high burden of cardiovascular disease and the fact that cardiovascular medicine has one of the highest outputs of RCTs annually, it is in a unique position to lead the adoption of EHR-embedded trials.

Supplementary data

Supplementary data is available at European Heart Journal online.

Data availability

No new data were generated or analysed in support of this review article.

Funding

All authors declare no funding for this contribution.

Author contributions

Muhammad Shahzeb Khan (Conceptualization [lead], Project administration [lead], Supervision [equal]), Stefan K. James (Methodology [equal], Supervision [equal], Writing—original draft [supporting]), Stefan D. Anker (Investigation [equal], Writing—original draft [equal], Writing—review & editing [equal]), Matthew T. Roe (Investigation [equal], Supervision [equal], Writing—review & editing [equal]), John A. Spertus (Investigation [equal], Supervision [equal]), Sunil Rao (Investigation [equal], Supervision [equal], Writing—review & editing [equal]), Gregg C. Fonarow (Supervision [equal], Writing—original draft [equal], Writing—review & editing [equal]), Robert J. Mentz (Investigation [equal], Supervision [equal], Writing—original draft [equal], Writing—review & editing [equal]), Ziad A. Ali (Supervision [equal], Writing—review & editing [equal]), Nicholas L. Mills (Supervision [equal], Writing—original draft [supporting], Writing—review & editing [supporting]), Sadiya S. Khan (Conceptualization [supporting], Supervision [equal], Writing—original draft [Supporting]), Muthiah Vaduganathan (Investigation [supporting], Supervision [equal], Writing—original draft [supporting]), Stephen J. Greene (Supervision [equal], Writing—original draft [supporting], Writing—review & editing [equal]), Harriette G.C. Van Spall (Supervision: Supporting; Writing—original draft: Supporting; Writing—review & editing: Equal), Talha Khuwaja (Investigation [equal], Writing—original draft [equal], Writing—review & editing [equal]), Muhammad Shariq Usman (Conceptualization [equal], Investigation [equal], Methodology [supporting], Writing—original draft [lead]), Javed Butler (Investigation [equal], Supervision [lead], Writing—review & editing [lead]), and Darren K. McGuire (Investigation [equal], Supervision [lead], Writing—review & editing [lead]).

References

Author notes