Re-examining physician-scientist training through the prism of the discovery-invention cycle

The linear assumptions of “bench-to-bedside” research

The discovery-invention cycle can be considered both a theory of scientific progress as well as a theory of innovation. In applying this framework to medicine, we take the latter perspective and consider that at its core, innovation consists of novel amalgams of existing concepts, skills, methodologies, and resources, which in turn give rise to fundamentally new ones (Baregheh et al., 2009; Fagerberg et al., 2005). Theoretically, the greater the number, diversity, and depth of the clinical and scientific training of individuals, the greater the possibilities for novel combinations of concepts, skills, methodologies, and resources. We see the original vision of the MD/PhD as being in this spirit—to increase innovative outcomes at the interface between basic science and clinical research.

Physician-scientists are meant to play the unique role of bridge builders in the vast landscape of the basic and clinical sciences (American Academy of Arts and Sciences, 2013; Zemlo et al., 2000). What is the nature of these bridges and how should we train future leaders to help construct these key edifices? The notions of “bench to bedside” and “bedside to bench” hide the complex realities which underlie both of these research and development pathways. A ‘pure’ example of formulating testable hypotheses from clinical observations is John Snow’s discovery of the causes of cholera outbreaks in 19^th century London, even before the advent of the germ theory (Johnson, 2006). Likewise, it is expected that physician-scientists enable the subsequent steps of translational research, which will give rise to novel therapies from bench research.

Unfortunately, both sides of this process invariably involve complex detours that hamper the attempts of the well-trained physician-scientist. One example of the tortuous nature of both the “bench to bedside” and “bedside to bench” trajectories is manifest in psychiatric genetics. Schizophrenia and mood disorders are characteristically polygenic, thus preventing the formulation of biological pathogenesis hypotheses from classical Mendelian genetic approaches (International Schizophrenia Consortium et al., 2009). It was the advent of powerful chemical DNA synthesis tools, DNA hybridization on chips, DNA sequencing technologies, robotics, novel statistical tools, and the computer power to analyze and store genomic and clinical data from thousands of patients and controls that has opened the door to formulate biological hypotheses of the underlying disease mechanisms (Geschwind & Flint, 2015; Sullivan et al., 2012). While it is not our aim to undertake an analysis as thorough as Narayanamurti and Odumosu, there is no shortage of clinical breakthroughs whose trajectory was as meandering and boundary crossing as the advances in solid state physics and NMR summarized above. Consider the example of statins, drugs that are widely used to treat hypercholesterolemia. Their discovery and clinical applicability needed the convergence of knowledge about natural product chemistry, cholesterol metabolism, and the genetics of hypercholesterolemia disorders, to mention just a few (Stossel, 2008). In both of these cases, schizophrenia and statins, fundamental science, technological / product development, and managerial / organizational insights have been key enablers of research progress.

On the one hand, many will point out that these observations are hardly surprising. To some degree, the diversity of efforts required to actualize a given medical therapy—basic or applied, academic or industrial—is a well understood phenomenon. On the other hand, if so, is this reality reflected in how we train our physician-scientists? And what of the expected career trajectories of dual-degree graduates? Does the ideal physician-scientist career path optimally take advantage of the unique nature of their training? What policy changes can be implemented to fully realize the intellectual potential of our medical establishment and our arsenal of dual-degree graduates?

The need for qualitative analysis

As we discussed in the introduction, there is a sizeable body of metascientific research analyzing the impact of MD/PhD training. Most recently, in the United States, the Association of American Medical Colleges conducted a thorough and thoughtful investigation into nearly 50 years’ worth of career outcomes of dual-degree graduates (Akabas et al., 2018). To assess the institution-wide impact, they examine a wide variety of measures including employment at an academic center, presence in major grant databases, and percentage of protected research time, among others. The AAMC study is an impressive first-pass analysis characterizing the landscape of dual-degree career outcomes. Given the complex, multi-faceted nature of physician-scientist training and practice, their report paves the way for subsequent investigations that can build on this foundation.

One of the conclusions of the AAMC study is that many questions remain in understanding the nature of the research landscape of physician-scientists. As an example, although the authors went into the study expecting a bimodal distribution of research time (i.e. with most being primarily clinicians or pure researchers), they discovered that there is in fact a continuum of research effort. Beyond a high-level description of their research activities, a major open question remains as to how researchers are taking advantage of their training apart from maintaining parallel job descriptions in varying proportions.

Our primary takeaway from applying the lessons of the discovery-invention cycle to medical research is that there is a significant need for in-depth qualitative analysis to better characterize whether dual-degree training is truly being utilized. Even if we can establish beyond the shadow of a doubt that MD/PhDs are deeply involved in both clinical and research activities, there is still the lingering question of whether or not they are leveraging their unique and extensive training. If we were to eliminate the dual-degree trajectory entirely, would the corresponding research output vanish, or would it be accomplished in different ways by pure MD and PhD researchers? Or is it that the true contribution of MD/PhDs is to facilitate dialogue between researchers and clinicians through higher-level administrative and organizational activities? The AAMC study is an impressive cross-sectional examination of the entire cohort of MD/PhD trainees giving policymakers high-level structural insights and subsequent analyses can be more targeted and examine smaller groups of individuals in significantly greater depth. Examining the roles, activities, and intellectual philosophy of physician-scientists even from a single academic organization (school, department, or division) would give significant insight into the impact of the training.

Open data, machine learning, and data science

The application of data science to public corpora of scholarly publishing is allowing researchers to investigate novel hypotheses about the scientific process itself. These new research trends are being embraced in the life sciences (see, for example, the Meta-Research collection in the journal eLife). We can now analyze complex datasets to study the evolution of scientific language since the 19th century or the frequency of medical reversals in the literature on clinical trials, and use the results to inform policy decisions (Herrera-Perez et al., 2019; Plavén-Sigray et al., 2017).

Likewise, we argue that modern computational techniques may have a unique role in augmenting the qualitative, anthropological analyses we have argued for in this piece. Indeed, quantitative techniques have a long history in the study of innovation (Hall & Jaffe, 2012; National Academies of Sciences, Engineering, and Medicine, Division of Behavioral and Social Sciences and Education, and Committee on National Statistics, 2017). Imagine that we have a fully-digitized corpus of biomedical literature that can be used by data scientists. Such a resource would consist of aggregated articles from biomedical journals along with the tools for conducting large scale text searches, visualization, and the training of machine learning models. Is it possible that lying hidden in such datasets are signatures of research uniquely enabled by MD/PhD training?

One way to approach this question is from a purely exploratory standpoint. What are the patterns that emerge when looking at the research output of MD/PhD investigators? Are there differences in the distributions of journals that they publish in relative to pure MDs or PhDs? What about the distribution of topics? What about correlating trends in academic publishing from MD/PhDs with other sources, such as biomedical news or patent databases?

Another way to approach this question is the paradigm of classification (Bishop, 2006). Suppose for every research publication in our dataset, we extract the educational qualifications of the authors (since many medical journals require this, we might even restrict to those papers that list the qualifications of authors). We can then use supervised learning to create models that predict whether a dual-degree graduate co-authored the research study. To simplify the analysis, we could focus on the first author, senior author, or perhaps train a model to predict how many MD/PhD degree holders were involved with the study. How accurate might such a model be? If it turns out that no such model can be trained to perform better than chance, does that imply that the research output of MD/PhDs could just as well have come from pure MDs or PhDs? On the other hand, if it turns out that a model can be trained to achieve high classification accuracy, does it imply the converse? Or might there be some latent structural explanation subtly biasing the model? These are all tantalizing questions that would require deeper investigation and which would add significant depth and new dimensions to the types of questions identified above.

Even more powerful would be to combine corpora of biomedical literature with anonymized medical records and even patent databases. For example, are there differences in documentation patterns between pure MDs versus MD/PhDs? What about in the standard of care? One would hope that there are no differences in routine cases, but what about in difficult situations where there is no standard of care? Are MD/PhDs able to leverage their knowledge and experience with the scientific literature to impact patient care? Analogous questions might be asked using combinations of the biomedical literature with patent databases to understand the role of MD/PhDs in the process of translating discovery to invention.

We make this proposal with some amount of trepidation and emphasize that the types of analysis proposed here need to be anchored to solid, anthropological investigation. Without a deep understanding of the context and culture of MD/PhD training, purely data-driven investigations might easily be misinterpreted and give credence to erroneous policy decisions.

For traditional MD and MD/PhD pipelines

For existing integrated MD/PhD programs, shifting the start of the PhD until after a full year of clinical rotations. For the traditional MD to residency track, expanding the pool of residency/fellowship-based PhD programs. In both cases, encouraging the PhD student to have both a true research mentor and a true physician-scientist mentor.

One of the most significant drawbacks of the linear model of research and development is a failure to adequately identify bottlenecks in the convoluted path from discovery to invention. Unfortunately, the current model of integrated MD/PhD training solidifies this model by sequentially training students in the pre-clinical sciences, followed by basic research, followed by clinical rotations. In the vast majority of cases in the US, students begin their PhD after having completed the pre-clinical years and taken Step 1 of the United States Medical Licensing Examination (USMLE), but before their required third-year clerkships.

This setup results in students choosing their PhD topic having had a limited exposure to the actual practice of clinical medicine. Consequently, students are not adequately prepared to identify a research path that is informed by experience with medicine itself. Shifting the start of PhD training to after the completion of third-year clerkships would give students substantially better preparation to choose a PhD area that would best serve their vision of being a physician-scientist in the context of an area of clinical scholarship that they wish to embark upon.

At some schools that have moved to an accelerated format for the pre-clinical years, in which Step 1 of the USMLE is taken after only 1–1.5 years of coursework, MD/PhD students are beginning their PhD after all or some of the clinical clerkships. See, for example, the MSTP program at Duke University, which has a one-year pre-clinical curriculum, or Baylor University, which has a 1.5-year pre-clinical curriculum. We endorse this path and our recommendation stands for schools that have maintained the traditional two-year format for pre-clinical training. For schools that have adopted a 1.5-year pre-clinical curriculum, but where MD/PhD students begin their graduate studies immediately after Step 1 of the USMLE, we recommend completing six months of clinical clerkships along with their traditional MD classmates prior to beginning their PhD, and perhaps also including 1–2 months of elective rotations in the areas relevant to the research they expect to pursue. An additional benefit of this arrangement is that students would be able to maintain their clinical skills during their PhDs. For example, an afternoon in an outpatient clinic once or twice a month would allow them to remain in contact with the clinical world while in the depths of their PhD research. It would also provide a valuable point of structure and regularity in the otherwise open-ended (and often soul searching!) years of graduate training.

The motivation behind expanding the pool of residency/fellowship-based PhD programs is similar in spirit. By residency/fellowship-based PhD programs, we are conceiving of a curriculum in which the majority of a student’s research is done during dedicated years separate from clinical training. We are skeptical that quality research can be conducted entirely in parallel with the grueling demands of residency and fellowship. See the Stanford ARTS Program or the UCLA STAR Program for outstanding models of research curricula integrated with post-graduate medical training. In both cases, the motivation is to ensure that students have greater exposure to clinical medicine before beginning their graduate careers. Moreover, residency/fellowship-based PhD programs open the door to many more students whose interest in a research career took a longer time to develop than those who decided early in their college careers apply to an integrated MD/PhD program.

Residency/fellowship-based PhD programs also carry the benefit that students are more mature than had they began their PhDs soon after completing their undergraduate degree. Such students are likely to bring a more informed perspective to their graduate studies. Compared to typical first-year graduate students, they will already have substantial exposure to the scientific and clinical literature and will have been surrounded by research for many years, even if they have not yet been immersed in it. Moreover, because they are close to being fully-trained physicians, they can continue their clinical responsibilities during their PhD years in outpatient clinics or while moonlighting.

For residency/fellowship-based PhD programs, many will correctly point out that the elephant in the room is length of time it takes to train and the corresponding financial implications. Most residents interested in pursuing a PhD did not have the benefit of a subsidized MD/PhD program, and thus are saddled with significant loans from medical school. They are also at a time in life when they may have a new marriage, are planning a family, etc., with competing demands on time and other resources. It becomes very difficult for most individuals in this situation to turn down a starting job with a well-remunerated salary, and taking several more years to pursue the PhD. There will likely need to be innovative, but perhaps costly solutions to deal with these realities. However, we remain optimistic that such solutions are possible. Individuals frequently make career decisions that are far from optimal financially, but which give them a deep satisfaction in their life choices. Even a career in academic medicine, particularly in the setting of a dual-income household, is more than livable, and we suspect that a small, but sizeable student body would be enthusiastic to participate in such a set of programs.

In both integrated MD/PhD and residency/fellowship-based PhD settings, appropriate forms of mentorship would be necessary to have maximal impact. Specifically, it should be encouraged for students to have both a research mentor and a physician-scientist mentor.

For non-traditional MD candidates

Creating a pipeline in which PhDs in other subjects or professionals of any kind with advanced training are recruited to the medical pipeline. For non-traditional MD candidates, ensuring a culture of mentorship and freedom that will allow them to develop their vision for integrating their prior experiences into a medical career.

One of the aims of viewing physician-scientist training from the lens of the discovery-invention cycle is to enable medical culture to be flexible in the face of a rapidly evolving landscape and emerging interfaces with other subjects. We believe that a sensible approach to creating such innovation and agility that would require minimal administrative overhead is to recruit PhDs as well as professionals from other disciplines to the standard medical pipeline. There is a long history of individuals who have made the decision to transition from a PhD to an MD. However, what we are advocating is a more ambitious and targeted recruiting culture within the medical establishment itself. The significant challenge, therefore, is how to discern those who are committed to a career as a physician-scientist, which is the minority, from those who have found they no longer see themselves pursuing a career in research and want a new direction in life.

Actively identifying and recruiting highly trained professionals (PhDs or otherwise) has the advantage that individuals bring a mature worldview to medical school and set of novel skills that can inform their clinical training and research focus. A small number of schools have created accelerated MD programs for biological science PhDs. See for example, the 3-Year PhD-to-MD-Program at Columbia University. Moreover, in situations where there is an immediate need for insights from a field outside of medicine, trained professionals can begin to make small contributions from the very beginning and are no more than four years away from becoming resident physicians. On the other hand, recruits to the standard MD/PhD pipeline are many years away from being able to contribute their knowledge base to medicine, even if their PhD topic has been strategically chosen to address a current need. Certainly, some creativity will be necessary to execute such a recruitment process, given the personality, skill set, and passion required for success in clinical medicine. However, considering that the demands and rewards of medicine are well understood by the educated public, and given the long history of individuals in medicine who have gone through significant career changes, we are confident that such a targeted recruitment process is achievable by sufficiently motivated medical school admissions committees.

As one contemporary example, we mention the intensely controversial discussions surrounding electronic health records (EHRs). Many believe that EHRs have negatively impacted the job satisfaction of healthcare workers as well as patient satisfaction. Indeed, research articles have specifically examined the role that EHRs play in increasing burnout among physicians (Arndt et al., 2017; DiAngi et al., 2016; Shanafelt et al., 2016). However, among the many voices in this discussion, we observe that one voice is notably absent: that of professional software engineers and user interface designers who could lend insights into why the situation in medicine has had such a disappointing outcome when we are surrounded by high-quality software in other arenas. Recruiting professional software engineers to the medical pipeline would allow these highly skilled individuals to contribute their insights to medical culture and establish the foundations for future advances at the intersection of software, medicine, and artificial intelligence.

The importance of mentorship in this process cannot be overstated. Unlike mathematics or theoretical physics, where one can be a prodigious scientist at a young age, the human and institutional complexity of medicine means that years of experience are typically required before an individual can speak with authority and generate mature insights. For those who are entering medical training after having established themselves in another field, being mentored by someone experienced in medicine can play a significant role in finding avenues to allow their existing knowledge base to become intertwined with medicine.

There are several exciting developments in science publishing that can be powerful enablers of this process. The use of “pre-prints” as part of the life cycle of idea generation, a practice that has been commonplace in some subjects for nearly 30 years, and which is now becoming more common practice in biomedicine, opens the door for experienced medical students, such as those with PhDs in other areas, to write scholarly articles on topics of medical interest but from a novel and integrated perspective that become part of the citable scientific literature. Likewise, the emergence of “open science,” and large-scale, distributed scientific collaborations may be particularly valuable ways for trained professionals to identify topics and projects of relevance to medicine that may not yet exist at their home institution (Nielsen, 2011; Sarma & Faundez, 2017).

For both traditional and non-traditional candidates

Encouraging a culture that values the MD as a terminal degree for individuals interested in non-clinical career paths such as medical data science, product development, consulting, science journalism, program management, venture capital or other positions that would advance the discovery-invention cycle across the life sciences.

An implicit assumption in analyses of the integrated MD/PhD program is that ideal outcomes involve individuals who maintain dual clinical and research responsibilities. However, this assumption rests on a linear view of bench-to-bedside research in which the role of the physician-scientist is to make testable hypotheses at the bench based on observations made at the bedside. If the reality of this pathway is far more complex, then perhaps it is not simply the specifics of training that should be reconsidered, but also the ideal career trajectory that individuals aspire to. Indeed, is it possible that success stories of dual-degree training may involve careers with neither research nor clinical responsibilities?

The dramatic growth of knowledge across the entire spectrum of the life sciences and the critical role played by novel technologies suggests that there are many important roles outside of academia that could be played by traditional dual-degree graduates as well as the non-traditional variants we have suggested above. Taking advantage of their strong skills as scientific generalists, they would play critical roles in overseeing product development at startups, managing portfolios of life science startups at venture capital firms, or leading data science efforts at health systems. The role of the physician-scientist training in these contexts is not necessarily to directly enable linking fundamental research in biology to patient care. Rather, it is to help manage the burgeoning institutional complexity of the life sciences by bringing significant intellectual depth that encompasses diverse facets of a biomedical education.

There are many individuals who have created non-traditional career paths building on MD or MD/PhD training without pursuing additional post-graduate experiences (Moawad, 2013). However, there may be significant value in a more systematic approach to training and encouraging such individuals. One possibility might consist of a 2–3 year program that encompasses a medical internship along with rotations with companies spanning different verticals in the life sciences. Such a post-graduate education would allow individuals to continue to strengthen their core clinical competencies while also giving them a survey of the many dimensions required to realize the bench-to-bedside vision.