An estimated 64 million people suffer from heart failure (HF) globally. Americans have a 20% lifetime risk of developing HF. Although recent progress in the development of new treatments has led to a modest decrease in morbidity and mortality, mortality for heart failure patients remains unacceptably high at 50% within the first five years of diagnosis.1 Different clinical presentations early in disease onset can lead to delays in diagnosis and treatment and, as a result, worse outcomes.
New research strategies are needed to identify patients earlier and to understand disease progression in different clinical phenotypes. This would enable more effective treatment pathways and more personalized approaches for addressing unmet needs.
How Real-World Data can help
Real-world data (RWD), the healthcare data routinely collected in practice, is changing how research is conducted. Examples of RWD include electronic health records, insurance claims, laboratory results, and patient-reported outcomes (PROs). When linked together in a standardized and cloud-based architecture, RWD can be accessed by researchers to provide insights into disease progression, treatment safety and effectiveness. For example, the OM1 Heart Failure Premium Dataset links RWD on more than 180,000 heart failure patients and includes information on medications, labs, left ventricular ejection fraction (LVEF), New York Heart Association (NYHA) class, social determinants of health (SDoH) and more. These data can be used for conducting studies and performing analyses. This comprehensive dataset also provides a base for applying artificial intelligence and other advanced analytics methods and tools for predicting outcomes and patient identification.
A changing treatment landscape
In the last decade, heart failure treatments took a leap forward with new drug approvals that have different mechanisms of action, such as sodium-glucose cotransporter-2 (SGLT2) inhibitors, which had originally been designed to treat type 2 diabetes, and soluble guanylate cyclase (sGC) stimulators. Early this year, another new drug application was filed for a cardiac-specific myosin activator, which aims to help the heart pump blood more efficiently.2
Medical devices and digital health technologies have also advanced in recent years and are increasingly being used in conjunction with pharmaceutical treatments. Left ventricular assist devices (LVAD) help the heart pump blood and newer implantable pressure-sensing devices are advancing how symptoms are evaluated and managed by clinicians. For example, a doctor can better gauge a patient’s pulmonary artery pressure and adjust medications as needed, hopefully preventing worsening heart failure.3
The American College of Cardiology (ACC), American Heart Association (AHA), and the Heart Failure Society of America (HFSA) released new HF treatment guidelines in 2022 to incorporate some of these newer treatment options based on disease progression and comorbidities.4
Most treatments are aimed at targeting reduced ejection fraction (HFrEF) patients. That leaves potentially 50% of patients with preserved (HFpEF) or midrange (HFmEF) ejection fraction with fewer treatment options. Also, HFmrEF patients typically have a clinical profile and prognosis closer (in general) to patients with HFpEF than those of HFrEF. More attention is being focused on the HFpEF patients, and a recent randomized trial reported a reduction in hospitalization in HFpEF patients treated with an SGLT2 inhibitor (even those patients who did not have diabetes). Understanding the trajectory and prognosis of specific subsets of HF patients will clearly help guide more effective management.
RWD can help stratify patients by LVEF, identify new phenotypes, and assess similarities and differences in characteristics and outcomes among HFpEF, HFmrEF, and HFrEF patients.The data can be used to explore patient journeys, changes in characteristics, and gaps in care and to identify patient subtypes who are most likely to respond to certain treatments. The insights garnered can then be used for evaluating protocol feasibility, clinical trial recruitment, and as supporting evidence for regulatory submissions or label expansion into additional HF populations.
Another consideration for new treatments and interventions is that payers may hesitate to provide coverage and, in most cases, do seek more data around value and comparative effectiveness in areas such as clinical benefit, quality of life (QoL), and healthcare utilization – especially when comparing new higher cost treatments versus lower cost (often generic) well-accepted treatments that have been around for years. Being able to demonstrate that a new treatment reduces hospitalizations or shortens length of hospital stays in the real-world setting (and not just in a randomized controlled trial) can be powerful evidence for securing coverage. These last two outcomes are of particular interest to payers as well as to patients and providers because HF is a leading cause of hospitalizations and the mortality following hospitalization for patients with HF is over 20% at 1-year.
RWD can help with evaluating treatment effectiveness broadly in the heart failure patient population and within different clinical phenotypes. AI models can help with predicting risk, such as which patients are most likely to be hospitalized or re-admitted. For example, the OM1 Medical Burden Index (OMBI) score, which is a standardized measure of the combined effect of current and prior conditions and treatments on current health status, has been applied to the OM1 HF dataset to successfully predict unplanned hospitalization for 8 out of 10 HF patients.5 Similar models have also been developed to predict readmission and 12-month survival. Risk stratification models can also be used at the point of care to proactively intervene and prevent important medical events.
The challenge of co-morbidities
One challenge in understanding heart failure progression and treatment pathways is the plethora of co-morbidities that often accompany it, such as obesity, chronic kidney disease, diabetes, chronic obstructive pulmonary disease (COPD), sleep disorders, anemia, cognitive dysfunction and depression. Some of these comorbidities are caused by shared risk factors and others may be caused by heart failure itself or vice versa. In addition, different comorbidities seem to align with different clinical phenotypes. For example, COPD, diabetes, anemia, and obesity tend to be more prevalent in HFpEF patients. The interplay between different conditions is complicated and plays into the difficulties with diagnosing and managing the disease. These comorbidities also decrease QoL and increase mortality risk.6
Large real-world datasets can help quantify the heterogeneity and numerous comorbidities of different HF populations. Researchers can also use large datasets to help with patient identification of unrecognized common or rare conditions, such as transthyretin amyloidosis and obstructive hypertrophic cardiomyopathy.
Patient populations and equity in research
Heart failure affects some patient populations more than others. The elderly (over 65 years old), men, and black patients tend to be at higher risk of developing heart failure and experiencing complications. In the United States, mortality from HF patients over 65 years old is more than two-fold higher in black compared with white patients. Additionally, black men have a 33% greater risk and black women have a 50% greater risk of hospitalization for heart failure.7
Unlike clinical trials, real-world data networks can include patients regardless of geography and care setting, which expands who can be included in studies. These automated large datasets are more inclusive, reduce selection bias, and allow for analyses into how different patient populations respond to treatments, and assess the potential contribution of gaps in care that may exist for those specific populations.
Socioeconomic (SES) factors, such as income level, educational attainment, and employment status, have been shown to have a measurable and significant effect on cardiovascular health including heart failure.SES factors can play into lifestyle choices that increase risk and also determine the patients who have access to or are likely to seek treatments, all of which can drastically impact outcomes.8
For the reasons listed above, SDoH data are growing in importance for our understanding of racial and other disparities in health utilization and outcomes for a variety of diseases. For heart failure, SDoH data can be linked to real-world data networks, like the OM1 HF Premium Dataset, and enable deeper explorations into how socioeconomic and other factors are related to disease progression and clinical outcomes. These insights can then help determine potential appropriate interventions and treatment pathways for subgroups of patients based on factors not typically studied during clinical development.
Considerations with data
While RWD has many benefits when conducting research and analyses, one of the key challenges is missing data, values that are not available and that would be helpful for analysis. Missing data is inevitable in clinical research, can be a source of bias, and can lead to loss of statistical power and precision. No universal method for handling missing data exists due to the wide variation across studies.
A common pitfall in handling missing data in real-world studies is simply dropping variables with a large proportion of missing data. An example is that labs like creatinine, sodium, and potassium may be missing for many patients in an analysis of predicting hospital admission or mortality. Simply dropping these variables may lower the predictive power of the model, or even result in a misleading model.
Good practices in handling missing data include differentiating between missing data and data not intended to be collected. It is possible that some clinically important data are collected for a subset of patients only. With missing lab results, a missing indicator variable may be included in the predictive models. In general, data imputation is discouraged. Likelihood analytic approaches are important alternatives but do not work for all regression models. If data imputation is needed, multiple imputation is preferred. However, the assumptions associated with the multiple imputation should always be checked.
Machine learning is a much more powerful method for handling missing data. While clinical data might not be in the structured data fields, it could be available in clinical notes, also known as unstructured data. Studies show that approximately 80% of healthcare data is unstructured. For example, LVEF and NYHA data that are not in structured fields may be extracted from clinical notes by machine learning and abstraction. Being able to include extracted data from notes or additional data sources can greatly improve the completeness and quality of the data if done appropriately. Transparency and traceability of the data and reporting on methods used are also important considerations, especially if the intent is to use RWD for regulatory purposes.
Finally, RWD can continue to be linked to additional existing data as needed and can also be set up to prospectively collect additional data variables, such as PROs to add depth and completeness depending on the study goals.
The future of HF research
Although treatment pipelines are strong and are expected to continue delivering improvements over the next few years, unmet need, especially in certain patient populations, still remains high. Linked real-world data that leverages advanced analytics offers a new way to view and understand heart failure, how it progresses, how it affects different patients, and how subgroups of patients respond to treatment.With these new insights will come opportunities to optimize treatment pathways and clinical outcomes in ways not previously possible.