Why clinicians are natural bayesians

Editor—According to Gill et al, clinicians are natural bayesians.¹ Their reasoning about patients is intuitive, probabilistic, and reiterative. Such subjective, context dependent reasoning is integral to clinical judgment and useful when diagnosing rare diseases.

In the same issue, the fictional Dr House is described.² He, too, must be bayesian. He solves rare cases intuitively through flashes of grim insight, this occurring against a backdrop of reiteration: obscure diagnoses are hurled back and forth. He is also likened to fictional detectives, which is interesting because the detective model of clinical judgment has already been described.³ At its core is the collection of evidence about the patient's condition by the clinician, who interprets and reinterprets this in a context dependent way, such evidence being given due weight and accepted or rejected according to the coherence it brings to the diagnostic picture.

It therefore seems that the detective and bayesian models of clinicians are closely similar, if not the same. Both consider information presented by the patient. The detective model expresses the probability of this being relevant in terms of non-numerical “weight,” whereas the bayesian model expresses this numerically.

Also, both models focus on the individual patient. The detective model is compared to the scientific model, which provides a well validated evidence-base for practice generally but does not allow for the unique circumstances of the individual patient. The bayesian model is compared to the limitations of decision making based on frequency statistics and associated algorithms.

The scientific approach, with its reductive abstractions and generalisations based on frequency statistics, causes the clinician to lose sight of the individual patient who is pushed to the margins of attention. Hurrah for bayesianism—its insistence on attention to the patient as context returns the patient to the centre of the clinician's considerations.