Healthcare in the United States is complex and expensive. There is a need for more evidence to inform decisions that lead to better, more efficient, more affordable care (Alper and Grossmann 2015). Patients, care providers, researchers, administrators, payers, regulators, and the public agree that the delivery of healthcare should be informed by high-quality scientific evidence regarding the risks and benefits of treatments. Yet, high-quality evidence—generated by randomized controlled trials and disseminated through clinical practice guidelines—is lacking for many therapeutic areas (Tricoci et al 2009; Roos et al 2011; Wright et al 2011; Koh et al 2013; Feuerstein et al 2014; Neuman et al 2014; Fanaroff et al 2019).

When they lack high-quality evidence, clinicians must determine treatment options by making educated guesses based on personal judgment and knowledge of the patient rather than expert consensus (Tricoci et al 2009). Clinicians and patients often simply do not have enough evidence to effectively inform clinical decisions. For example, in cardiology, which arguably has one of the most robust evidence bases among medical specialties, most treatment recommendations are founded upon lower-quality clinical trials, observational studies, or expert opinion (Tricoci et al 2009).

When we survey the US clinical trials enterprise from a broad perspective, we see that the kinds of clinical trials needed to provide high-quality evidence to support treatment decisions are, for the most part, not being conducted. Most clinical trials are too small to provide sufficient statistical power to definitively answer clinical questions, fail to address critical treatment priorities, or have shortcomings in design and execution that limit their usefulness (Califf et al 2012; Pasquali et al 2012; Alexander et al 2013; Goswami et al 2013; Hirsch et al 2013; Lakey et al 2013; Todd et al 2013; Witsell et al 2013; Inrig et al 2014; Subherwal et al 2014). Moreover, the data from many clinical trials are not reported in timely and transparent ways (Anderson et al 2015; Ramachandran et al 2021; DeVito and Goldacre 2021). Adding to these complications, there has been a steady drumbeat of revelations indicating that many findings published in the peer-reviewed literature are unreliable (Ioannidis 2005; Ioannidis 2016; Open Science Collaboration 2015; Le Noury et al 2015).

In modern medicine, clinical research has generally been kept separate from the delivery of routine patient care. As a result, research data are collected using standalone systems. These systems are designed to ensure that the information gathered during research activities is valid and complete. However, maintaining separate systems for research and clinical care comes at a significant cost. In a landmark essay, Ioannidis (2005) sounded an alarm about growing concerns that results obtained from traditional approaches to clinical research may not apply to “real-world” situations. Clinical research is often conducted under artificial conditions with volunteers who may not reflect the complexity of the populations of patients who live with the given disease or condition. Growing awareness of the limitations of traditional approaches to clinical research spurred interest in creating “learning health systems” (Platt et al 2024).

Learning Health Systems and Embedded Clinical Research

In recent years, clinicians, researchers, and healthcare system leaders have advocated for the development of learning health systems, in which tools such as computing power, connectivity, team-based care, and systems engineering techniques produce a culture of continuous learning at lower cost (Institute of Medicine 2013). Approaches to fostering learning health systems continue to evolve, with scholars bringing attention to the importance of health disparities (Parsons et al 2021), trust and shared decision-making (Kelley et al 2015), and real-world data (Butler-Henderson et al 2025), among other issues.

Ideally, clinical trials would be embedded within a system of healthcare delivery where evidence is rapidly and continually fed back into clinical care, and clinical care itself would inform the further development of medical evidence (Embi 2019; Simon 2020). At the same time, the widespread use of electronic health records (EHRs) and advances in information technology and informatics are creating opportunities to combine large, complex datasets (“big data”) in ways that until now were almost unimaginable. As systems for managing data continue to improve in US healthcare systems, the availability of electronic data and advances in artificial intelligence are likewise increasing rapidly.

There is a need for “a different context to clinical research that could speed the discovery and implementation of evidence-based advancements to healthcare delivery. Pragmatic clinical trials (PCTs) are a promising type of trial conducted within real-world health care delivery systems” (Tuzzio and Larson 2019).

Pragmatic clinical trials embedded in healthcare systems represent one approach to building a learning health system to inform real-world practice with digital health data collected at the point of care. Embedded pragmatic trials have the potential to inform policy and practice with high-quality evidence at reduced cost and increased efficiency compared with traditional clinical trials (Platt et al 2024).

The next sections introduce characteristics of embedded pragmatic clinical trials—with examples drawn from the NIH Collaboratory Trials—and point readers to resources available in this Living Textbook that describe best practices for the design, conduct, and dissemination of embedded pragmatic trials.