(Apologies if my views on logical inference are overly simplistic -- I'm a radiologist by profession, very far from a philosophy major. My goal is to understand what medical decisions are and where they are coming from -- not only to understand my own decisions, but to be able to explain it to my resident physicians and students.)
I'm working on a structured framework of medical practice. There are two facets of medical practice where we use logical inference: figuring out a diagnosis (inferring pathological processes from observed features) and choosing the most appropriate intervention for the patient.
Figuring out a diagnosis (in my mind) is simple enough.
- Let's say we observe that the patient has chest pain.
- We know that chest pain occurs in pneumonia, myocardial infarction (heart attack), and a rib fracture. So we use abduction to infer that all three conditions are possible causes of the chest pain in our patient.
- Now we use deduction (modus ponens) to infer what additional features we need to differentiate between these three conditions. For example, if our patient had pneumonia, we know that she would also have cough and fever. If, on the other hand, she had a rib fracture, we know that she would have had a recent fall and bruising.
- After deducing these additional features, we go back and observe some more to find out if she does or does not actually have them.
- And now we can use deduction (modus tollens) to exclude hypotheses that did not gain supporting features, and induction to infer that the hypothesis that received the most support, is in fact the cause of the patient's chest pain. If, for instance, the patient hasn't had trauma and has no bruising, then we exclude rib fracture, and if she indeed has cough and fever, then pneumonia is the best explanation to the observed features.
This makes sense to me, and I can also explain it to my students when I teach. I can use the familiar (if overly simplistic) triad of rule, cause and effect -- a known condition (in the abstract) is the rule, a presumed condition (in the patient) is the cause, and observed features are the effects:
- Given an observed feature and known condition, we abduce a presumed condition.
- Given a known condition and a presumed condition, we deduce (modus ponens) additional supporting features.
- Given a presumed condition and absent supporting features, we deduce (modus tollens) that what the patient has does not reflect our known condition. Or given a presumed condition and observed supporting features, we induce that this presumed condition reflects our known condition.
Cool. But can I also apply the same ideas to choosing the best intervention?
In medical practice, we can direct our interventions (drugs, procedures, or even active monitoring) to observed features (for instance, give paracetamol for fever, or pain medication for pain) or presumed conditions (for instance, give antibiotics for pneumonia). In my mind, the observed feature or presumed condition would be the cause, and the intervention (in the patient) would be the effect. And the rule would be our knowledge about the intervention (in the abstract).
The slight glitch here for me is the reversal of cause and effect. When we diagnose, we see the effect, we know a number of rules, and we want to find a cause. When we intervene, we have a cause, we know a number of rules, and we want to find the most appropriate effect.
Intuitively, the steps are very similar.
- Let's say we observe that the patient has chest pain.
- We know about interventions that have chest pain as an indication. For instance, paracetamol and ibuprofen are both options. Is this still abduction, when I infer possible interventions here?
- We know that paracetamol and ibuprofen both have contraindications and we need additional information about the patient to decide which would be most appropriate in the given situation. For instance, we would prefer paracetamol, if we would observe a stomach ulcer. Or, we would prefer ibuprofen, if we would observe liver failure. Are we still using deduction here?
- Now we go back to the patient and make sweet observations.
- And based on our observations we decide on the most appropriate intervention. Is this still induction?
I'm eagerly waiting for your insight and will happily clarify if some ideas here were a bit vague.
