Thursday, March 26, 2015

Thinking about Thinking Fast and Slow

Today I led a teaching session for the 3rd and 4th year clerks within the Internal Medicine Department. My teaching topic was "How do we think: Approach to Differential Diagnosis". My objective in picking this topic was, by going through cases, to demonstrate that there are many different ways that clinicians develop differential diagnoses depending on the familiarity with the case (or learner expertise), the urgency of the situation, and individual preference. Each has their pros and cons, and can be equally as useful, depending on the context. I had hoped that by going through this exercise, the students would become more aware of their individual thought processes (thinking about how they think) and to expand their toolbox so that, in future, they might have an increased variety of approaches to use.

I walked through cases from the simple (45yo man with fatigue and microcytic anemia) to the more complex (5yo with failure to thrive, oral ulcers, and joint pain), picking and choosing an approach for each. I started with the most basic 'Deadly and Common' approach - most useful for those with experience in a familiar and urgent situation. We then spent a significant portion of time walking through mnemonics. We discussed how they can be straightforward and automatic (such as TAILS for microcytic anemia), or, a tool for developing a more broad differential when we are less familiar or our initial attempt has failed (i.e. the systems/etiology oriented VITAMIN CDE). We also discussed approaches such as physiologic which can become more schematized as experience grows (such as the approach to an AKI) or anatomic (approach to abdominal pain). The latter was used more as a demonstration of the subconscious process of how we would come up with the questions to conduct a focused history. Hybrid matrix approaches such as published in Sacher & Detsky's paper were also discussed.

I enjoyed the exercise of talking about how we think with my colleagues, but equally as enjoyable were their questions afterwards. They echoed so closely my own questions as to whether one approach was better than the other, whether there was danger in automatic pattern-recognition or heuristic thinking (like using TAILS), and whether I had discovered in my research evidence of how learners go from more System 2 type thinking to System 1 thinking with greater expertise. I was particularly gratified by one individual's comment that he enjoyed the exercise of becoming more aware of what he was doing and reflecting on it, rather than 'just doing it'.

These comments prompted me to read Diagnostic Reasoning: Where We've Been, Where We're Going by S.M. Monteiro and G. Norman (aside: it seems like I'm reading ALL of Dr Norman's papers; he's very prolific). They contrast in their article the perspective made popular by Kahneman's best-selling novel, Thinking Fast and Slow from which the System 1 and System 2 thinking come, and a more psychological-derived theory that medical diagnosis is more of an exercise in categorization and memory retrieval.

In Thinking Fast and Slow, Kahneman describes a 'Default-Interventionalist' model which talks about System 1 (Fast) thinking being the default mode and which relies on quick, heuristic thinking, and System 2 (Slow) thinking, which is a more logical, deductive process requiring an increased cognitive load (see previous post). System 2 is, according to his book, preferential as it in theory is less prone to cognitive biases that heuristic thinking has traditionally been associated with, but is less of a natural process to us; we are encouraged in his novel to develop strategies to 'slow down' our thinking, which in theory would result in more accurate reasoning. As Monteiro and Norman point out however, there is a flaw to this assumption which is that our brain is even capable of 'choosing' to slow down our thinking (most of us do it without thinking, that IS the point after all) and that we can only do one at a time. Several studies have also shown that the assumption that slower thinking will result in more accurate solutions is false.

Further, the argument that System 1 thinking is inherently susceptible to cognitive bias may not actually be a bad thing. As Monteiro and Norman discuss, directed history taking to confirm a working diagnosis may look like 'confirmation bias' when the initial hypothesis was correct, but is actually probably the most efficient way to come to a diagnosis; certainly if we received an answer to the question that did not match our working diagnosis, I think most physicians would take a step back to re-evaluate their thinking. I think of this 'backward reasoning' approach as one of the most rapid examples of the Scientific Method I've ever seen. They additionally point out that System 2 thinking may be just as susceptible to these 'cognitive biases' as System 1 thinking; in this manner they cast doubt on the idea of System 2 thinking being superior to System 1.

Their contrasting perspective discusses a Parallel-Competitive model which describes dual processing that operates in a simultaneous manner in the context of memory models of categorization and recognition. Categorization relies on remembering prior knowledge of either learned cases or lived experiences to which we are constantly comparing the current clinical presentation to find 'which fits best'. This operates in a simultaneous process as Recognition itself which allows us to see the patterns and thereby compare them to our encoded memories. This theory is a perfect example of why Mixed Learning (as discussed in my previous post) can be so effective; by encoding a variety of presentations simultaneously we learn better how to differentiate between them rather than memorizing each independently.

Relating this all back to my presentation today, it's interesting to see that I was encouraging learners to engage in more System 2 type thinking as an exercise to become aware of what they were doing in an automatic System 1 thinking process, and, as a way of approaching a problem which is more unfamiliar, or when their previous attempt had failed. I honestly think that both an heuristic pattern-recognition approach to diagnosis and a more thorough broad-based approach are both useful, and that we should actively engage in using our brains in both ways so as to be able to use the tools as appropriate. I don't think that we would be very efficient as doctors if we simply relied on System 2 thinking exclusively, especially in emergent situations. However, I also find significant value in thinking about a more parallel-processing model. I personally have always thought of my memory as a library: there are many bookcases, and on each is a set of shelves, on each shelf is a set of books, and in each book there is information. Recognition serves me well as a way of knowing that I have the book in my library, but categorization allows me to find it when I need it. By continuing to see patients, I strengthen my book-retrieval system and become more adept at not just diagnosis, but remembering the proper management of a condition.

Certainly, going forward, it's been very valuable to think about how learners at different stages of training have different experiences of how they think about something. I'm more mindful now of when different approaches can be used and for whom they are most useful. I've always found it easiest to teach when I'm aware of what I'm doing subconsciously, rather than just doing it; as we're all going to be teachers someday, it behooves us to think about how we think more often.

~LG

No comments:

Post a Comment