Abduction. The first reference I know of for the use of this term and concept in a system of thought is from C.S. Peirce. You may remember Peirce from an earlier post when I discussed his view of pragmatism. His was the view that practicality gave meaning, but not necessarily truth (whereas William James’ form of pragmatism was that practicality could adjudicate truth). The ODP also gives Peirce credit for the use of this term to describe what they call: “the process of using evidence to reach a wider conclusion, as in inference to the best explanation.” Peirce recognized that we could not justify abduction as we can with deduction (i.e. the development of valid forms that, given true premises, true conclusions are necessary), and he even questioned whether abduction could be confirmed through the use of probability. The use of probability will not be discussed as part of this post as it is not critical right now, but as a foreshadow, I do find Bayes formula and Bayesian probability to be a nice match for attempting to confirm and quantify belief in abductive inferences. It is just that most of the time abduction is dealing with so many variables that the use of any quantifiable approach gets intractable, as nicely discussed in Eddy and Clanton’s 1983 paper in NEJM, The Art of Diagnosis: Solving the Clinicopathological Exercise. For students in my courses, you can access this paper through the course wiki, on the LfB page (Links from Blog).
Peirce introduced abduction largely to explain the creative aspect of scientific investigation. He was attempting to answer the question, how do scientists decide which hypothesis to test? His answer was abduction. From making observations we attempt to explain them, from the observations to their causes. They then formulate a hypothesis that could be used to test whether it is in fact true that X causes Y. As an approach this can generally work, but it is not bound to work, in other words, it is not necessary that it works. For example, I observe that my street is wet in the morning. I can do an experiment whereby I take out my hose and spray my street and that causes the street to be wet. But the confirmation of my hypothesis of spraying the street does not rule out the possibility that it had rained for my initial observation of the wet street. Of course, one would easily and rightly argue that the “hose” argument was not ever the best explanation of the wet street. At some point I will have to come up with a better example.
An interesting point is that Peirce was writing pretty close to when Sir Arthur Conan Doyle was writing (circa 1880 - 1910). Doyle (a physician) wrote a series of stories and books about a character he created named Sherlock Holmes. Holmes was inspired by Doyle’s medical school professor, Dr. Joseph Bell, noted for his “power of observation and deduction.” At the time, the terminology abduction, which is what Bell and Holmes was actually doing was not recognized. So, just about every time Holmes points to his “deductions” he is actually pointing to an abduction.
In the initial inference post I introduced the causal component to specifically address abduction. That was not completely necessary, I could have just as easily stuck with antecedents and consequents, and pointed out that abduction is what we do whenever we are committing the deductive fallacy called “affirming the consequent.”
If P then Q Q Therefore P This looks like syllogism, it looks like Modus Ponens to be precise. But, unlike Modus Ponens; this is not valid, it is the fallacy affirming the consequent. (Just a note that I have not written any of the wikipedia posts that I refer to ;) I did however put together a one page summary of the valid deductive logic forms for you if you are interested - see LfB.
Given the use of inference in a knowledge based practice, and given that we accept cause as a critical component to clinical practice, we will stick with the definition for abduction that includes cause:
Abduction: The process of moving from causal relations and some specific effect, to acceptance (or expectation) of a specific cause. (This is what we do when we attempt to determine the cause of a patient’s problem. Since inflammation causes pain; and since the patient has pain (effect), I may infer that they have inflammation (cause). The subtle reversal is full of profound implications though – as we will unpack in the post on abductive inference.
We use abduction in two ways as far as I can tell. As Peirce intended, and as Doyle (Holmes and Bell) intended.
Let’s take the affirming the consequent as our “abductive” form.
If P, then Q (General premise about causal relations) Q (Existence of an effect) Therefore P (Conclusion about a cause)
Abduction ala Peirce (developing clinical research questions) starts with observations of Q and P regularly enough to warrant the testing of the general premise about causal relations - - testing through an inductive inferential process (a research study) whether If P, then Q. For example, a lot of people that get lung disease (effect) also smoke. This raises the question of whether smoking could be a cause. So tests are done to assess the causal association. This is abduction as Peirce developed it.
Abduction ala Doyle (diagnosis or crime solving) starts with the knowledge of the causal relations; it starts with the acceptance of the fact that If P, then Q. When coming across the effect (Q) we then must, based on If P, then Q, consider whether P is the cause or a contributing cause to the existence of P. For example, if myocardial ischemia (P), then there are ST segment changes (Q). We accept this. So when observing ST segment changes we need to consider whether myocardial ischemia is the cause. Even if the general premise is true with certainty (no exceptions, universal), then this abduction does not lead to certain knowledge about the cause as other things can cause ST segment changes (i.e. electrolyte imbalance). This latter part is what we mean when we say we cannot justify abduction. Most of the time our abductions are hampered not only by a lack of justification, but also by the degree of probability of the general premise.
I have now exceeded my daily self imposed word limit. But before I leave. The EBP system of thought certainly addresses diagnostic accuracy. It is where we fret and worry about sensitivities, specificities and likelihood ratios. Numbers generated entirely by empirical observations. I believe we need to broaden the discussion about diagnostic accuracy as part of clinical reasoning by stepping back to what we are doing - abduction - and considering how the numbers of our empirical observation help (or hamper) that process. A complete reliance on the numbers associated with the empirical observations for diagnostic accuracy might be the best evidence of a system of thought so overwhelmed and blinded by its empiricist foundation that it does not realize the limitations imposed by it. The fish that does not recognize it is wet.