Clinicians are faced with a dilemma – the need to make decisions based on a universal set of evidence and experience that usually does not explicitly include that individual. My understanding of the clinician’s dilemma germinated while working toward my professional Master’s in Physical Therapy and became clear during graduate course work in epidemiology. I didn’t have a chance to write about it and propose some vague abstract solutions until 2005,[i] and didn’t propose tangible solutions until 2014 which are embedded into a curriculum I developed for a new Doctor of Physical Therapy (DPT) program at Plymouth State University (2015-2017) and were then published in 2018.[ii] And to be clear, no one has solved this dilemma. At best we have some inkling of the types of reasoning that make it less poignant, or at least enable a clinician to have a rationale for decisions. There’s a gulf between a clinical researcher saying “Your practice is not evidence based”, and the clinician saying in response “Your research isn’t practice based”. A hardline stance of evidence-based practice not including mechanistic causal reasoning and only including (or giving strong priority to) randomized controlled trials that focus on outcomes as the primary determinant of the best treatment is not practice based.[iii]
The CauseHealth[iv] project considers the problem in their book Rethinking Causality, Complexity and Evidence for the Unique Patient[v]. Clinicians must be able to inquire and reason about unique situations and consider what, whether, how, when, and to what extent clinical practice guidelines and evidence from systematic reviews and randomized controlled trials apply in that particular situation. Clinicians consider mechanisms as causes underlying particular situations even when they are part of a unique arrangement of a complex system and when the observation of prior situations exist, or the ability to repeat the situation is limited.
In physical therapy this requires a depth and breadth of physiological understanding that starts with the core concepts and proceeds to integrated, complex, mechanistic causal relationships. Physical therapists “diagnose and treat individuals of all ages, from newborns to the very oldest, who have medical problems or other health-related conditions that limit their abilities to move and perform functional activities in their daily lives” (APTA).[vi] A core component to the knowledge used in practice for this profession is physiological knowledge. Beyond understanding pathophysiology, physical therapists must be able to reason through the consequences of various situations – when is physiology as expected vs. not as expected and how does a set of expectations (or non expectations) influence our understanding of the current particular situation. In other words, clinicians are reasoning through causal models, either implicitly or explicitly. And often, much of what happens at the causal model level of knowledge for practice includes physiology.
My writing and teaching promote the use of causal models as representations of knowledge for clinical reasoning, and the use of graphical causal models for the clear articulation and sharing of such knowledge. This approach is helpful for the consideration of how universal concepts learned through an empirical process and thought to be true for a population can be applied in a particular situation. When teaching DPT students how to use physiology in clinical reasoning we approach causal models of physiological mechanisms as an interpretive lens for the clinician’s dilemma.
Clinical research utilizes statistical inference to estimate, from a sample, what a population characteristic or cause-effect relationship may be. The cause-effect relationship may be intervention-outcome, or it may be exposure-disease. The patient in front of a clinician was usually not in the original sample. The question then becomes, is this patient part of the population that this study (or these studies) represent via statistical inference? And is this patient part of that population in a manner that is important given the physiological mechanisms involved in the cause-effect relationship? This is a physiological question. We immediately can consider whether the inclusion and exclusion criteria of the research includes or excludes this particular patient. Those are obvious reasons to question whether the patient you are working with is part of the same population to which these studies inferred. We naturally look at age, sex, comorbidities and severity of the situation. All of these considerations imply variation in the underlying physiological state of the particular patient from the inferred population. But even if the particular patient is similar to the inferred population on all of these considerations, underlying physiological assumptions based on the mechanisms remain and should be considered.
For example, research demonstrates that electrical stimulation of the major skeletal muscles involved in walking is causally associated with positive outcomes in people with chronic heart failure such as maximum oxygen consumption (VO2max), six minute walk distance (6MWD) and even, to a lesser extent, health related quality of life (HRQOL).[vii] Figure 1 depicts the simple graphical causal models that the clinical research (randomized controlled trials) has investigated as part of an evidence-based practice empirical approach to understanding the relationship between interventions and outcomes (made with DAGitty).[viii] Even when assuming a particular patient would be included (based on meeting inclusion criteria and not meeting exclusion criteria for these studies), there are several very poignant physiological mechanisms when considering the use of electrical stimulation in practice that impact the probability of the intended outcome.
No one assumes that electrical simulation directly improves health related quality of life, or six minute walk distance, or even oxygen consumption. There are physiological and even psycho-physiological, behavioral and cultural mechanisms involved in the connection between electrical stimulation and these three outcomes, and these three outcomes are very likely connected to one another. Figure 2 is one possible graphical causal model that fills in some of the possible mechanisms.2
The clinician is working with many competing hypotheses, and is “faced with all sources of variation at the same time and must deal constantly with the full burden of the complex system.”1 Let’s take a closer look at the many causal assumptions of the model in Figure 2. As a graphical causal model, the first thing to realize is that all edges in the graph with an arrow encode the knowledge that one variable causes the other (and the lack of an arrow implies no causal connection). This does not have to be a definite causal connection; in fact, most of them are probabilistic and can be stated as conditional probabilities. For example, this graph encodes the knowledge that ES acts as a cause on muscle function. Which can be stated as a conditional probability: the probability of improved muscle function given ES is greater than the probability of improved muscle function given no ES. The model in Figure 2 also includes additional interventions since ES would rarely be considered the only intervention available. In fact, the patient in the condition where ES is the only intervention available probably would not be in the inferred population (for example, there are no studies on the use of electrical stimulation with patients with chronic heart failure that were not ambulatory or were unable to do other forms of exercise). This model includes aerobic training (AT), resistance training (RT), ES, adaptive equipment (AE), inspiratory muscle training (IMT), all as possible interventions for improving 6MWD, VO2max, and HRQOL in people with HF.
The characterization of the intervention (exposure, cause) in this model is discrete (yes/no), but it does not need to be discrete; it can be continuous and can include any of the considered important parameterizations of ES. Also, the effect muscle function is discrete but can be continuous and include any of the important parameterizations of muscle function. In other words, the causal model can encode as much of the ontological information about reality (its variables) as the user would have it encode. As attributed to George Box, “All models are wrong, some models are useful.”
The mathematical and logical implications of the causal model go on to include the multivariable considerations such as the chain rule of conditional probabilities (VO2max), identification of confounders (balance as a confounder), blocking variables (anaerobic threshold), and adjustment sets (6MWD).
My point here is to answer the question—“Isn’t this the same as concept map?” No. Causal models depicted as graphs are based on graph theory and adhere to a set of logical and mathematical rules that allow logical and mathematical implications to be proposed and tested. But they do share concepts. We could say that all causal models are concept maps, but not all concept maps are causal models; therefore, they are not equivalent since equivalence implies bidirectional implication.
Concept maps of physiological mechanisms are great teaching and learning tools. The next step, to use physiology as an interpretive lens for the clinician’s dilemma, is to consider encoding them as graphical causal models. In fact, this is the logical step from the core physiology concept of causation.
Another consideration for the clinician is that no single study has confirmed these causal connections all at the same time. But, a corpus of studies has tested these causal associations. The model in Figure 2 represents knowledge for practice; practicing based on this model is an example of using physiology to help in reasoning through whether to use an intervention with a particular patient. For example, if a particular patient has a problem with balance unrelated to muscle function then the probability of ES improving their 6MWD and even HRQOL is likely lower than in a particular patient without such a problem. And if a particular patient’s problem is mostly from the direct reduction in cardiac output associated with chronic heart failure, then a change in muscle function from ES may have less of an impact than in a particular patient with stronger contribution of muscle function in their reduction in oxygen consumption. And if the particular patient has low inspiratory muscle strength (IS), then IMT may be the best approach to start with – despite the fact that there are no clinical trials that investigate the intricacies of when to use ES vs. IMT. Thus, causal models of physiological mechanisms are an interpretive lens for applying clinical research in clinical practice. And this involves reasoning through causal models of complex physiological mechanisms.
The question is not whether this is already being done in practice (because it is, though usually implicitly not explicitly). The question is how are we teaching future clinicians and students? Is there a way to teach it that expedites the transition from classroom reasoning to clinic reasoning? Effective teaching often includes pulling back the curtain and explicitly revealing that which has been implicitly occurring. When a student asks me how their Clinical Instructor was able to come to some particular conclusion, the answer is usually that they were implicitly reasoning through some assumed causal model. Causal models can explicitly bridge the gap between learning physiology from a standard medical physiology textbook, doing a case study in a clinical course, and seeing a patient in a clinic.
The next step in my journey of using causal models for clinical pedagogy is the relationship between narratives, stories and causal models. If causal models are a more complex depiction of the reality underlying evidence-based causal claims; then narratives and stories are a more complex depiction of the reality underlying causal models. If you’re interested in discussing this further, please let me know.
Thank you to all of my colleagues (which includes all of the DPT students) at Plymouth State University for trusting this vision enough to take a chance on a new DPT program; and thank you to my closest dialogue partners in this and my upcoming work in the causal structure of narratives, Drs. Kelly Legacy and Stephanie Sprout (Clinical Assistant Professors of Physical Therapy); and Dr. Elliott Gruner (Professor of English/Director of Composition).
[i] Collins SM. Complex System Approaches: Could They Enhance the Relevance of Clinical Research? Physical Therapy. 2005;85(12):1393-1394. doi:10.1093/ptj/85.12.1393
[ii] Collins SM. Synthesis: Causal Models, Causal Knowledge. Cardiopulmonary Physical Therapy Journal. 2018;29(3):134-143.
[iii] Howick JH. The Philosophy of Evidence-Based Medicine. John Wiley & Sons; 2011.
[iv] CauseHealth Blog https://causehealthblog.org/ (Accessed 10/15/2021)
[v] Anjum RL, Copeland S, Rocca E. Rethinking Causality, Complexity and Evidence for the Unique Patient: A CauseHealth Resource for Healthcare Professionals and the Clinical Encounter. Springer Nature; 2020.
[vi] American Physical Therapy Association https://www.apta.org/your-career/careers-in-physical-therapy/becoming-a-pt (Accessed 10/15/2021)
[vii] Shoemaker MJ, Dias KJ, Lefebvre KM, Heick JD, Collins SM. Physical Therapist Clinical Practice Guideline for the Management of Individuals With Heart Failure. Physical Therapy. 2020;100(1):14-43. doi:10.1093/ptj/pzz127
[viii] Textor J, van der Zander B, Gilthorpe MS, Liśkiewicz M, Ellison GT. Robust causal inference using directed acyclic graphs: The R package “dagitty.” International Journal of Epidemiology. 2016;45(6):1887-1894. doi:10.1093/ije/dyw341
Figure 1: Simplified Causal Associations Tested in Clinical Trials (Abbreviations: ES, electrical stimulation; 6MWD, 6-minute walk distance; VO2_max, maximum oxygen consumption; HRQOL, health-related quality of life)
Figure 2: Complex Causal Associations Necessary for Clinical Practice (Abbreviations: AT, aerobic training; (a-v)O2, arteriovenous oxygen difference; IMT, inspiratory muscle training; IMS, inspiratory muscle strength; PaO2, partial pressure of oxygen in the arterial blood; RT, resistance training; Ve, minute ventilation; VQ matching, ventilation perfusion matching )