1. Introduction
How should we conceive of and evaluate the process of medical diagnosis? Saying that it is a matter of inferring the correct disease from the available signs and symptoms is too sparse. Clinical diagnosis is anexperimental science based on observation, hypotheses, and testing. It is a dynamic process that involves observation, diagnostic conjectures and testing, possibly leading to new or revised conjectures. For the clinician, it should always be a reflexive process subject to revision.
Consider for example the following scenario: A 54-year-old man with no previous history of chronic disease suffers sudden substernal chest pain and is rushed to an emergency room. His presenting symptoms also include tachycardia (abnormally rapid heart rate), shortness of breath and sweating. The challenge a clinician faces in cases like this is not just to evaluate the likelihood of different possible causes of these symptoms; she also has to select which hypotheses to actively consider in the first place, which to prioritize for further testing, which can be put aside for the time being and when to initiate treatment on the basis of a given hypothesis. Additionally, all of these decisions presuppose that the relevant hypotheses have been generated and introduced into the diagnostic inquiry. The clinician does not start out considering every possible cause of chest pain known to medicine; rather, she needs to decide when and how to generate new diagnostic hypotheses, as well as when to stop.
In this paper, we present a framework for understanding the different kinds of reasoning underlying medical diagnosis as it occurs in clinical practice. Our starting point is the observation that, in addition to evaluating the likelihood of candidate diagnostic hypotheses in light of the evidence, the process of medical diagnosis involves two distinct types of reasoning, namely: (i) reasoning concerned with generating new candidate hypotheses and (ii) reasoning about which hypotheses should be pursued, i.e. prioritized for testing and further consideration. That these forms of reasoning are crucial to understanding inquiry was argued by C. S. Peirce in his writings on the form of reasoning he calledabduction . Following recent commentators (Upshur 1997; Stanley and Campos 2013, 2015; Chiffi and Zanotti 2015), we believe that Peirce’s mature account of abduction provides important lessons for understanding diagnostic reasoning. Specifically, we argue that recent Peirce scholarship, which construes abduction in terms ofstrategic reasoning , provides a promising framework for analyzing diagnostic reasoning.
Our aims in presenting this framework are primarily normative: we want to explicate the reasons which underlie diagnostic reasoning in realistic clinical situations, rather than necessarily describing the psychological processes clinicians go through in diagnosis. The best psychological description may often be that the clinician makes a quick, intuitive judgment, perhaps based on some unconscious heuristic. By contrast, our framework aims to explicate the factors which make such judgments reasonable in a concrete, clinical situation. Despite this normative scope, our aims in this paper are not prescriptive in the sense of recommending whether existing practices can or should be improved. Rather, our main aim is to be able to explicate diagnostic reasoning as it occurs in current practice.
A unified, normative framework for understanding clinical reasoning is currently lacking from the methodological literature. On the one hand, when hypothesis generation is addressed (e.g. Kassirer, Wong and Kopelman 2010, Ch. 13) it is mainly discussed from the perspective of cognitive psychology without an underlying normative framework. On the other hand, the probabilistic approach to clinical-decision-making currently popular in the medical literature—the so-calledthreshold approach —while normative, does not address the question of hypothesis generation. As we shall argue, because of the way hypothesis generation and reasoning about pursuit are intertwined, this neglect means that threshold models, in their current form, fail to capture all relevant reasons for pursuing a hypothesis.
Our discussion proceeds as follows. We start, in Section 2, by outlining our understanding of Peircean abduction and, Section 3, explaining how these ideas apply to medical diagnosis. In Section 4, we then use this framework to analyze a clinical case study. In Section 5 we return to our criticism of threshold models. Finally, in Section 6, we defend the strategic reasoning interpretation of abduction as a framework for analyzing diagnostic reasoning.