How to understand medical decisions

Reasoning by probable elimination - a transparent rationale

Doctors often assemble combinations of features that predict a diagnosis in a ‘logical’ way.  For example, a patient with sudden onset of severe central chest pain might have a coronary thrombosis, an unstable coronary circulation, angina, indigestion from inflammation of the oesophagus or a pulmonary embolus (a blood clot blocking vessles to the right side of the heart).  If the ECG showed features that occurred often patients with a heart attack but rarely in those with an unstable coronary circulation, angina, indigestion or a pulmonary embolus, the probability of all the latter become low and the probability of a heart attack becomes high.  This is also the same reasoning process used to show that the probability of the different causes of non-replication were low, so that the probability of replication was high.


The important feature of this reasoning process is that the list of possibilities generated by the initial evidence e.g. chest pain should be as complete as possible.  The other feature is that the finding used to differentiate between the possibilities must occur commonly in at least one diagnostic possibility and uncommonly (ideally never by definition) in at least one of the other diagnostic possibilities.  Under these circumstances, the combination of chest pain and ECG findings occur commonly in those with a heart attack and uncommonly in all the other possibilities.


The reason that the combination of chest pain and the ECG change might occur commonly in patients with a heart attack is as follows:  If chest pain occurs in a high proportion (e.g. 80%) of patients with a heart attack and the ECG findings also occur in a high proportion (e.g. 90%) of patients with a heart attack then both findings must occur in at least (90 + 80 - 100) = 70% of those with a heart attack.  


The reason that the combination of chest pain and the ECG change occurs uncommonly in patients with a indigestion and other possibilities is as follows: If the ECG change only occurs in 1% of those with indigestion then the combination of chest pain and the ECG change must occur in 1% or fewer patients with indigestion.


The minimum proportion of patients with a coronary artery thrombosis in those with the combination of chest pain and the ECG change can be calculated using Bayes theorem.  The details of this and ways of calculating probability estimates are shown in Chapter 13 of the 3rd edition of the Oxford Handbook of Clinical Diagnosis.  


Evidence based diagnosis and decisions

The purpose of medical knowledge is to select appropriate treatments.  This is done by applying the diagnostic process.  A diagnostic term can thus be regarded as the title of an explanation as to why a particular treatment should help a patient with a particular combination of symptoms, examination findings and test results.  However, in order that diagnoses can be agreed upon and shared, it is important that there are rules regarding when a diagnosis can be used and cannot be used.  These are the ‘sufficient’ criteria (that specify when a diagnosis can be used) and the ‘necessary’ criteria, which when absent, indicate that the diagnosis cannot be present and is 'excluded' or 'ruled out'.


The diagnosis of diabetic ‘microalbuminuria’ is can be confirmed if and only if there are at least 2 albumin excretion rates (AER) of greater than 20mcg/min out of three tests and if some necessary criteria for other diagnoses are absent (e.g. absence of white cells in the urine that would happen in an urinary infection).  If the first two tests give AER results of 25 and 30mcg/min then these would help provide ‘sufficient criteria’ to confirm the diagnosis.  However, if the average of three AER results are less than 40mcg/min, less than 1% go on to develop diabetic nephropathy, so there imay be no need to treat these patients even though they do have the diagnostic criteria for ‘microalbuminuria’.    


Necessary criteria are of important practical help in day to day medicine.  If a necessary criterion is absent, e.g. that an AER is less than 20mcg/min on two occasions, then the diagnosis (e.g. of micro-albuminuria) can be excluded.  Sufficient criteria often used in research, but they are not always used consciously in practical day to day medicine.  Even if a sufficient criterion is present, a treatment may not be used, for example, because the diagnosis may be too mild to warrant treatment. However a suffcient criterion does make the doctor consider giving the treatment.


In the absence of a 'necessary' or a 'sufficient' criterion so that a diagnosis cannot be confirmed or excluded, even if a diagnosis is probable and may pose a great danger to the patient, then its treatment may be given ‘just in case’.  In order to make a decision to treat, the doctor has to estimate the probability of a satisfactory outcome if a treatment is given and also to estimate the probability of a satisfactory outcome if the treatment is not given (or that something else is given).  This is discussed in detail elsewhere (see Decision options, outcomes and evaluation).



© Huw Llewelyn 2016