As many if not more women die of heart disease compared to men in the United States. Yet studies consistently show that women with heart disease are more likely to report that doctors do not listen to their concerns [1]. Data also shows that women patients consistently receive less treatments for coronary artery disease as men and are also at higher risk for cardiac emergency hospital visits. This is a sad reality for a disease that kills as many women as men. How did these disparities come to be, in a country that is the leader in cardiac care and research?
The field of cardiology has a track record that is unmatched in medicine. Death rates due to heart disease have fallen dramatically over the last fifty years, declining by a whopping 60% since the peak in the 1960s [2]. The advent of statins and antihypertensives, as well as innovations in invasive procedures and surgical interventions, have saved hundreds of millions of lives all over the world. The field has amassed an impressive history of excellent outcomes that has eluded many other medical specialties. By any measure, cardiology is widely seen as a mature and well-established medical specialty and heralded as a public health success story in medicine. How can it be then that a field with such maturity and success still exhibits such unacceptable disparities when it comes to women?
For one, women still only constitute less than half of all subjects in cardiac clinical trials from 2005 to 2015, with many trials having fewer than 25% women[3]. Much of what we know about heart disease, particularly coronary artery disease (CAD), is a result of clinical trials primarily done on an elderly white male population. For example, let us consider patients with chest pain (angina) symptoms, but who aren't emergency cases because they are not experiencing a heart attack. This chest pain in these patients is called stable angina. Cardiology classifies stable angina into 'typical angina' and 'atypical angina'. Symptoms observed in early cardiac trials, almost exclusively done on men, were the only symptoms considered to be of diagnostic value. For a couple of decades, the field even considered the notion that women enjoy cardio-protective status because of female sex hormones. But more and more women were eventually diagnosed with CAD since the late 1970s. Still, their chest pain symptoms began to be classified as 'atypical angina', and chest pain symptoms primarily associated with men became 'typical angina'. The American College of Cardiology's guidelines on stable angina still considers "chest pain or discomfort or discomfort near the sternum" as the only typical angina symptoms [4]. Symptom descriptions such as "pressure, tightness and burning", not considered to be typical, have no clear guidance for clinical recommendations. Is the association of "typical angina" with men and the confusion around what constitutes "atypical angina" a reason why so many women feel like their symptoms are not taken seriously by doctors? Do men also have atypical symptoms in the same way that women may have typical symptoms? Could it not be that women verbally describe their symptoms differently compared to men and hence are not assigned equal importance compared to men [5]? Has an implicit bias that has worked against women engulfed the field, a possibility that is strengthened by the sobering reality that almost 90% of cardiologists are men [6]?
We conducted an observational trial with a novel design to investigate some of these questions in collaboration with principal investigator colleagues at the Harvard School of Public Health, Brigham & Women's Hospital, and MacMaster University. The study recruited more than 650 eligible patients in a multi-centered trial, both men and women due to undergo a gold-standard coronary angiogram investigation to determine if they had CAD or not. We audio-recorded the natural conversation between the physician and patient and also conducted a separate structured interview with each patient. This structured interview was also audio-recorded. It was vital for us to capture the symptom descriptions of patients in their own words. Too often, patients in clinical trials are given long forms with preconceived checkboxes. This reductionist approach limits our capacity to capture all of their symptoms beyond preconceived symptom categories. The order of conversations was randomized, so each patient randomly saw either the physician or the research assistant conducting the interviews first. Patients then had their coronary angiogram exam. The attending physician then interpreted the results of the exam. We harvested these results and also analyzed the conversation transcripts using machine learning techniques that we developed at MIT along with conventional statistical methods.
The results were striking. Contrary to the conventional dogma, there was no statistical difference between the symptoms reported by men and women. Many men also reported so-called "atypical angina" symptoms. In other words, we showed that the understanding among cardiologists that men tended to have typical angina and women tended to have atypical angina was wrong--they were both just as likely to have either and, even further, the symptoms they experienced were the same, they only described them differently. This work was presented in August 2019 at the European Society of Cardiology, a venue that draws 40,000+ cardiologists from all over the world. There have been other studies following our work that also show that both women and men have the same anginal symptoms. But is the science showing that women and men have the same anginal symptoms enough to change anything for patients on the ground beyond a presentation at a prestigious research venue in cardiology?
The complexity of clinical medicine is truly immense. For one, there's the complexity of biology. Coronary artery disease is, after all, a complex disease with the involvement of several subsystems within the body, spanning the immune, vascular, cardiac, and endocrine systems. Then there is the layer of public health literacy. From the early work of Frank Netter's illustrations graphically depicting heart attacks, most cardiac public health awareness campaigns would consistently describe the victim as an elderly white man. A body of work that emerged later showing that women had atypical symptoms has also entrenched the notion that women experience atypical symptoms. The effect of these public health literacy campaigns on both physicians as well as patients is another complex layer that might have ossified the logic of sex-differences in cardiac symptoms. Many researchers have also built careers examining atypical angina symptoms in women, making sex-differences in anginal symptoms a highly contentious topic with resistance to emerging science. Yet another layer of complexity surrounding guideline-making within the cardiology enterprise is medical practice liability, which many cardiologists will privately admit is a crucial factor for why guidelines are so slow to change. The American College of Cardiology's guidelines on stable angina have not changed since 2012.
While one is painting the complex picture of why the logic of sex-differences in angina symptoms is so deeply ingrained, we must not overlook the role of statistical methods. In most clinical trials, patients receive a preconceived checklist of symptoms to pick. The statistical methods that analyze these symptoms make certain assumptions - that each type of symptom is the same across all patients who report it. But 'radiating backache' in one patient who has a particular set of accompanying symptoms is not equivalent to radiating backache in another patient who has a different set of accompanying symptoms. Furthermore, constructing a checklist of preconceived symptoms is not the same as listening to what the patients have to say about their symptoms in their own words. One could argue that statistical methods may have, over decades, played a substantial role in cementing the lens of typical versus atypical angina.
Given all these systems and factors at play, one finds it hard to implicate any one system or factor as being solely responsible for the status quo. It may be tempting to frame this problem as one of implicit bias hoisted on female patients by male cardiologists. But such a framing, which may be accurate on the surface, is admittedly incomplete and insufficient. Each system in this picture of complexity may not directly be discriminating against women. Still, all the systems interacting with one another have led to disparities in diagnoses and treatment of heart disease in women. Most doctors, both male and female, are trying to do their very best to help their patients. Once one uses a framing of bias, there can be a tacit impugning of deliberate intent, which often results in pushback and a retreat to dogma. But perhaps if we paused and looked at the complexity emerging from the interaction of all the systems at play, we can identify interventions aimed at changing practice on the ground -- to help correctly diagnose and treat more women with heart disease. The scientific investigation is decidedly only one of the many interventions that need to happen to make a practical difference on the ground.
Our world is inherently complex. Whether it is biology, physics, or social science, it is impossible to draw definitive and complete descriptions of even seemingly simple systems. Be it a single-celled organism, a collection of gas molecules, or a small group of people, incontrovertibly causal pictures often elude us. Science frequently attempts to learn the behaviors and properties of systems that do not entirely depend on a full understanding of all of the details. The fields of complexity theory and complex systems science are not new. They attempt to understand the relationships and interactions between various components of systems, with the awareness that systems are adaptive and dynamically changing, whose emergent properties and behaviors are unpredictable, multi-dimensional, and non-linear. The study of complex systems involves an amalgamation of concepts, theories and methods spanning multiple disciplines where emergent behaviors of the system as a whole can often be very different from those of its parts.
One useful rubric to characterize the complex portrait of a system is to think of scale — for instance, the scale of the number of people and cultures and time. For example, a democratic nation with a vast population involving considerable geographical, racial, and ideological diversity may make political decision-making more challenging compared to a democratic country with a small and homogenous society. In medicine, a new cardiac drug that lowers triglycerides and cuts death rates by a third in clinical trials may require the knowledge and understanding of a number of other medical specialties such as the vascular, immune, and endocrine systems for a systemic fuller picture of the drug's mechanistic effects. Even if the drug is safe and efficacious, structural and economic impediments with vast array of medical insurance plans, along with the complex psychology of patient drug-therapy adherence, may make it harder to usher real positive patient outcomes on the ground. These examples illustrate how we might explore the relationship between complexity and time.
It is essential to acknowledge the adaptive and evolving nature of complex systems. A population's ideology can evolve, drug resistance can happen, and a change or tinkering in any component of a complex system doesn't always shift its behavior in intended ways. A disposition that observes, appreciates and attempts to surmise this adaptive nature is crucial to hone and refine. Mathematical models and theories from evolutionary biology to complexity economics offer some languages to describe the adaptive essence of complex systems, but they aren't the only relevant sources. Anthropology, critical theory, and sociology also offer powerful languages and lenses that are helpful in both characterizing the systems and also in designing interventions.
Any depiction of a complex system is a reductionist act, a product of continually synthesizing, simplifying, and abstracting our awareness and knowledge. It requires interacting with depth across relevant but siloed academic disciplines and other institutions, combining research knowledge with practical learnings on the ground. This simultaneous and mutually informing approach of 'head in the skies' for generative conceptual insight and 'feet firmly on the ground' for learning practical realities is a rich palette for the continual depiction of a complex system. For example, consider a project that attempts to investigate the use of predictive algorithms in the criminal justice system. Not only is it essential to learn about the various perspectives in fields such as statistics, criminology, and sociology, but it also becomes imperative to learn practical realities on the ground by interacting with law enforcement, the courts, and civil rights organizations. Short of a systemic and historical understanding of the contemporary criminal justice system in the United States and an awareness of the nuances, the resulting insights and interventions can not only be grossly reductionist but also superficially and harmfully techno-solutionist. A willingness to adapt one's knowledge of the complex system with constant learning and the humility that such an understanding may be thoroughly incomplete, become necessary temperamental qualities.
We seek to understand the relationships and interactions between a system’s many components. We do so to intervene with systems to achieve desired outcomes, that is, to shift the systems to some desired state. The process of understanding relationships between different components of a particular system is neither linear nor sequential. A litany of assumptions may undergird an understanding of the behavior of one part of a complex system, and these assumptions may change with further systemic awareness. For example, consider a study that shows that algorithmic predictive risk scores can be used in the criminal system to let more defendants out of jail as they await trial compared to biased human judges [7]. The study concludes that algorithmic predictive risk scores to aid judge decision making are a solution to reduce the scourge of mass pretrial incarceration in the United States. It might be tempting to assume that such a system would work as conceptualized, but a deeper systemic understanding may not be so concurring. Pretrial judges, whose power lies in judicial discretion, may begin releasing defendants but mandate onerous and extreme release conditions such as electronic monitoring and weekly drug tests for even minor charges. The punitive judicial mindset at the heart of mass pretrial detention merely transfers to a new form of harmful pretrial release conditions. Violent crimes committed by defendants released pending trial are rare in the United States. The attempted well-intentioned intervention of introducing algorithmic pretrial risk scores further reinforces the broken frame and failed logic of a justice system built on an irrational fear of mass violence and one that criminalizes symptoms of poverty [8]. This example shows the importance of assumption-checking of intended interventions from a systemic standpoint. Learning the relationships and interactions between the components of a system requires continual questioning of assumptions, probing of candidate interventions, and reflecting on the systemic effects from multiple perspectives. The ambiguity and the discomfort that arises when deliberating on an intended manipulation should be rapturously welcomed, and its impact on the system at large discerned very carefully before elevating it to the status of an intervention.
The relationships between various parts of a system, including the existence of feedback loops, may often require several interventions at the same time. For example, suppose the study showing the promise of pretrial algorithmic risk scores triggers well-intentioned legislative action that seeks to mandate such algorithmic risk assessments at both the state and national levels. Let us also suppose that it triggers newfound academic research within computer science in pursuit of a golden risk score algorithm, as well as increased philanthropic funding for practical operationalization of the idea. A rigorous statistical analysis that investigates and highlights grave concerns with the impact of using algorithmic risk scores in the pretrial arena is unlikely to have any measurable effect on its own. Legislatures can be petitioned, testimonials given, and the voice of civil liberty organizations elevated. Philanthropic institutions can be made aware of the reputational risk of funding an ill-advised technocratic solution that can further entrench the decrepit logic of a failing criminal justice system. In short, the complexity of the system requires a set of carefully crafted complex interventions across its parts to shift the system to the desired state.
Of course, pertinent questions arise. Who decides what interventions to make, and what is the desired state of a system? The desired state for whom and at what time scale? It can feel daunting and often paralyzing to examine these dilemmas, raising complex issues of power, agency, and participation. But as philosophers and non-dual meditative traditions across the world have often taught us over three thousand years, one does not "arrive north" in as much as "walk towards the north," with the possibility that one may often self-deceptively be walking in the wrong direction.