6. Diagnosis as Strategic Reasoning
The framework we want to outline in this section does not aim to rival the probabilistic approach in formal rigor. Rather, we want to outline a more flexible, general framework for thinking about diagnostic reasoning, drawing on Hintikka’s (1998) suggestion that abduction can be understood as in terms of strategic reasoning . Hintikka’s proposal is based on an analogy between game-theoretic reasoning and scientific inquiry. Knowing how to play a game such as chess involves at least two kinds of knowledge. The definitory rules tells us what kinds of moves are allowed and what the consequences are of those moves. The strategic principles , by contrast, tell us what would be a wise or an unwise move in a given situation, i.e. whether the move is likely to help us achieve the goals of the game.
In medical diagnosis, the definitory rules involve knowledge of how of a given diagnostic hypothesis should be evaluated in light of current evidence, what the available tests are, how the hypothesis should be re-evaluated in light of different possible test results and what the potential consequences are for the health of the patient. Based on this, the clinician has to adopt a diagnostic strategy . By this we mean an overall strategy for how to generate hypotheses, select a differential diagnosis and prioritize hypotheses for testing. As Hintikka emphasizes (1998, 513), one usually has to evaluate entire strategies, rather than individual steps. This is because it is often only possible to evaluate the importance of potential consequences—e.g. providing clues for hypothesis generation—within the context of a broader strategy.
We propose to see diagnostic reasoning as two-tiered. Individual moves (ordering a given test, choosing to stop generating new hypothesis, etc.) are justified in terms of whether they contribute to an overall strategy. The crucial choice then concerns which strategy to pursue. Diagnostic strategies can be thought of as analogous to the strategies a seasoned chess player might employ.13 Choosing a chess strategy depends, in part, on what kind of opponent you think you are up against together with knowledge of the definitory rules. Similarly, the choice of a diagnostic strategy will be informed by what kinds of diseases the physician thinks are most likely. In some very simply cases it may be possible to represent this in an explicit decision-theoretic model. In this sense the threshold approach is not incompatible with the broader framework proposed here. However, for the reasons given above, in many cases a physician’s reasoning about what the best strategy is in the given clinical context will not be adequately captured by any generally applicable formal model. This does not mean that we cannot say something about what kinds of considerations are involved in this type of reasoning.
In our case study, we identify three crucial choices of strategy. The first is the initial choice to pursue the ACS hypothesis before systematically generating new hypotheses. The emergency room clinician is trying to achieve a quick resolution, to spare the patient from potential harm of leaving the condition untreated and from unnecessary testing. This choice is in part justified by what diagnosis she thinks is most likely. In the chess analogy, a player may push for a quick checkmate because she thinks the opponent is likely not to recognize a certain trap. Similarly, the clinician recalls the most common cause of the presenting symptoms (ACS) and knows of tests which, if positive, would quickly and conclusively verify the hypothesis. This strategy in turn justifies not systematically generating further hypotheses.Given the strategy of achieving a quick resolution, it was reasonable to stop generating new hypotheses once she had identified what was—given the evidence—the most likely cause and realized that this could be quickly and reliably tested. Spending more time generating hypotheses in this context would have been unnecessary.
Unfortunately, many opponents—whether chess players or diseases—will not be defeated by such a direct strategy. When the initial tests fail to confirm the ACS diagnosis, putting the clinician in a more uncertain situation, she adopts a new strategy. Since she no longer has a clear view of what the likeliest diagnosis is, her priority shifts to ruling out the most serious threats, hoping in the process to discover—or better: create—a “strategic opening”, that is, a clue which could lead to the correct diagnosis. Adopting this strategy, she attempts to systematically recall the most dangerous alternative causes of chest pain. She chooses a test which is both highly reliable for ruling out such alternative (pulmonary embolism) and might enable future moves, by producing clues for further hypothesis generation. Her decision to request a cardiologist consultation at this stage also makes sense in light of this strategy, since his expertise would (i) complement her ability to think of relevant hypotheses and (ii) enable him recognize the relevance of any emerging clues.
Finally, at the concluding stage of the case, the cardiologist adopts a strategy for generating hypotheses that is focused on the salient clue—a dilated aortic root—brought out by the CT-scan and the echocardiogram, as well as the puzzling diastolic murmur. Due to the worsening state of the patient there is no time for further testing of the hypotheses thus generated. Whether to retain the ACS diagnosis or to adopt a newly generated hypothesis has to depend on his clinical judgement of which hypothesis best ‘fits’ the clinical picture. He therefore chooses a strategy of thinking quickly through a range of hypotheses, counting on his experience to allow him to recognize the correct hypothesis when he ‘sees’ it. Given his specialization as a cardiologist, and the severe constraints of the situation, considering possible aortic syndromes with a focus on explaining the dilated root and the murmur was a reasonable—and as it turned out successful—way of implementing this strategy.