Integrating Electronic Health Records into the Study of Heart Failure: Promises and Pitfalls

This article refers to ‘Prognostic burden of heart failure recorded in primary care, acute hospital admissions, or both: a population-based linked electronic health record cohort study in 2.1 million people’ by S. Koudstaal et al., published in this issue on pages 1119–1127. Heart failure (HF) represents a global epidemic: close to a million new cases are identified annually and over 8 million people are expected to be living with the chronic disease by 2030.1 In the USA alone, over 1 million patients are hospitalized with a primary discharge diagnosis of HF and roughly 30% of these patients die within 1 year.2 This alarming burden of disease, coupled with high rates of hospital and post-discharge care utilization, has resulted in a substantial economic footprint. The total cost of HF-related care in the USA is expected to rise to $70 billion by 2030.1 Unfortunately, despite these adverse trajectories,3 research investment and scientific advancements for patients with HF remain sluggish. For instance, only two novel drugs (combination sacubitril–valsartan and ivabradine) have been approved by the US Food and Drug Administration (FDA) Center for Drug Evaluation and Research as new molecular entities over the last 5 years. The high cost of drug development and the inability to efficiently track patients living with HF have posed major challenges to therapeutic development and effective clinical care in this space. The advent and increased utilization of integrated electronic health record (EHR) systems may improve the capture of real-time data in HF patients. We discuss the promises and pitfalls of EHR-linked data in identifying gaps in care at a population level, improving individual patient care, and optimizing the clinical trial enterprise. Contemporary registries and trials offer snapshots of patient profiles in specific care settings (for instance, exclusively inpatient or outpatient),4 and thus may not be representative of the overall HF population. In this issue of the journal, Koudstaal et al.5 present a population-based, EHR-linked study from the CALIBER (Clinical Research using Linked Bespoke Studies and Electronic Health Records) programme evaluating the distribution of HF patients across care settings. This platform bridges data streams from four different national health data sources informing primary care, acute care, and mortality patterns. Of the approximately 2.1 million inhabitants of England included in this analysis, ∼4% were determined to have incident HF during the period from 1997 to 2010.5 A quarter of these identified patients carried an HF diagnosis in the primary care setting alone and were never hospitalized, a third were hospitalized for HF without concurrent established primary care, and a quarter were seen in both primary care and hospitalized settings (the remaining patients were identified by death certificate only). Unfortunately, these HF patients, including those who are never hospitalized, consistently experience a malign 5-year prognosis.5 The database is robust and the study benefits from a very large sample, the inclusion of linked inpatient and outpatient data, and excellent follow-up (at a median of 1.7 years). The study unveils important gaps in HF care and identifies potential areas for quality improvement, including better transitions in care,6, 7 greater attention to seemingly stable outpatients with HF,8 and closer follow-up and monitoring of patients recently hospitalized for HF.9 On an individual patient level, newer EHR algorithms are incorporating alerts to treating clinicians with regard to guideline-directed drug or device eligibility. Similarly, EHR triggers may identify patients who are prescribed contraindicated or potentially harmful therapies, such as non-steroidal anti-inflammatory drugs. Appropriate use criteria could be incorporated into EHR systems to provide clinicians with real-time feedback regarding ordering practices of specific diagnostic interventions, including right heart catheterizations. Prompts regarding the availability of patient care documents and shared decision-making tools may facilitate real-time clinical interactions. Links to pharmacy records may identify patients at risk for poor adherence based on low refill rates. Ultimately, as with antithrombotic therapies, specific risk scores can be calculated using available clinical and laboratory parameters for estimated risks and benefits of various HF therapies. Contemporary clinical trials in HF are costly to operate and are becoming increasing complex and lengthy. These factors may limit trial size and ability to expeditiously complete protocols. Strategies that leverage EHR-linked data may streamline the clinical trial process and thus allow for more seamless patient identification, event ascertainment, and participant follow-up. Recent large, ongoing trials are directly applying EHR-based solutions to address these barriers to efficient clinical trial conduct.10 Registry-based randomized controlled trials, including the recently completed iFR-SWEDEHEART (Instantaneous Wave-free Ratio vs. Fractional Flow Reserve in Patients with Stable Angina Pectoris or Acute Coronary Syndrome; ClinicalTrials.gov ID NCT02166736) trial,11 enroll patients within large, well-phenotyped, often EHR-linked registries to expedite participant identification. The Patient-Centered Outcomes Research Institute's National Patient-Centered Clinical Research Network launched its first large pragmatic trial, ADAPTABLE (Aspirin Dosing: A Patient-Centric Trial Assessing the Benefits and Long-Term Effectiveness; ClinicalTrials.gov ID NCT02697916),12 which aims to randomize 20 000 high-risk patients to low- vs. high-dose aspirin. The informed consent, randomization process, and initial data collection are all facilitated by the use of EHRs and web-based platforms. The National Heart, Lung, and Blood Institute (NHLBI)-sponsored pragmatic INVESTED (Influenza Vaccine to Effectively Stop Cardiothoracic Events and Decompensated Heart Failure; ClinicalTrials.gov ID NCT02787044) trial13 is testing the effects of two formulations of the influenza vaccine on cardiopulmonary endpoints. Certain trial networks, including the Veterans Affairs Consortium, are leveraging coordinated EHR systems to follow patients and expedite the event adjudication process. These innovative emerging approaches may optimize the assimilation of EHR-linked data into various aspects of clinical trial conduct and execution. Although these EHR-linked data offer great promise and important insights into overall care patterns, several limitations regarding the validity and integrity of these data should be recognized. Heart failure is an inherently challenging entity to study using ‘big data’. It often presents with dyspnoea and is at times difficult to distinguish from other disease processes in older adults with multiple co-morbidities. There is no clear laboratory or echocardiographic standard for its diagnosis, and medications used for HF are used in other cardiovascular and non-cardiovascular conditions. Even clinically based assessments remain imperfect and lack complete inter-provider concordance,14 especially for HF with preserved ejection fraction (HFpEF). Big data science often relies heavily on the administrative coding of disease processes and problem lists, which are frequently incomplete and imprecise.14 Indeed, patients included in CALIBER were significantly older (median age: ∼80 years) and had higher rates of significant co-morbidities than patients enrolled in contemporary clinical trials (Table 1).15, 16 In CALIBER, regardless of care setting, the reported rates of beta-blocker use (∼25–35%) and mineralocorticoid receptor antagonist use (<25%) are substantially lower than rates observed in contemporary practice. Taken together, misclassification in CALIBER and other large EHR-linked databases may bias the characteristics and prognoses of identified HF cases to those of a general cohort of elderly patients. Several emerging analytics and approaches are being applied to improve EHR-based identification and phenotyping of key disease cohorts.17 In a recent retrospective cohort study, a machine-learning algorithm using free text was superior to standard problem list-based case identification of HF and showed higher comparative sensitivity and predictive accuracy.14 Endpoint ascertainment is also fraught with challenges using EHR systems. As in the CALIBER experience, HF is often listed as a cause of death on death certificates, even in patients who lack a previous history of HF. Unfortunately, examination of death certificates and medical records alone, without corroboration with next of kin and autopsy reports, may be subject to significant error.18 Recent pragmatic trials, such as INVESTED, are thus employing less specific endpoints (all-cause mortality and cardiopulmonary hospitalization) that are less prone to misclassification and may be more easily identified using EHR systems. Finally, at least in the USA, the lack of interoperability of EHR systems adds another barrier to broad-scale nationwide efforts towards coordinated data acquisition. Future epidemiological studies and clinical trials will continue to utilize EHR-linked data sources to define HF and potentially target interventions moving forward. The potential for integrated EHR-linked data to complement clinical trials has captured the interest of multiple scientific parties, and has even been the subject of a recent US FDA guidance.19 There is hope that EHR-linked data may identify key gaps in HF care at the population level, improve patient-level care, and promote efficiency and cost savings in clinical trials. The ultimate goal is to tease apart the signal from the noise and to truly hone in on HF patients in an era of big data. Given the rising global burden of HF, the complexity of the syndrome, and the lack of objective metrics to guide its definition, a more nuanced approach to the application of EHR-linked data is required, especially in HFpEF. Conflict of interest: J.B. has received research support from the National Institutes of Health, the European Union, the Health Resource Services Administration and the Food and Drug Administration, and has been a consultant for Amgen, AstraZeneca, Bayer, Boehringer Ingelheim, Bristol-Myers Squibb, Cardiocell, Janssen, Novartis, Relypsa, ZS Pharma, Medtronic, Merck, CVRx and Gilead. All other authors report no conflicts of interest.