and why protocols are not the answer to good diagnosis and risk management.
Excellent BMJ editorial telling it like it is.
“identifying those febrile young children with the greatest risk for serious infection at the time of clinical presentation is like looking for a needle in a haystack.” Essentially, if you have a child with a fever, there are no great lab tests that will tell you if the child is seriously ill or has a fever from a virus which will sort itself out.
Bayesian reasoning is, I think, the better answer. Some people call it ‘gut feeling‘, but I think this experienced, multi-layered medical response is less voodoo and more a mixture of all kinds of things: knowledge of the family, the previous interactions with the child, how the child looks, responds, what findings there are on examination, progress over time – not just the presence of a fever. Bayesian reasoning is a much more fluid process: you get new information (pink eardrum, two odd-looking spots) and you change the direction of thinking because of it; you do not stick to a linear protocol if you are on the wrong one.
This is important because cheaper medical care tends to be based on less qualified staff following more protocols. If you get onto the wrong protocol, you might not be able to backtrack and get onto the right one. With no lab tests to rely on, identifying a sick child should be done by staff who don’t just use protocols, but deep knowledge.
I’m not involved in clinical decision making, so am reacting as a psychologist interested in reasoning, rather than one at the coal face. In general I share your distrust of protocols. But Bayesian reasoning depends critically on the priors that you operate with. An experienced clinician will have much more accurate priors than a novice. So if you don’t have experienced staff, then protocols are probably the best option.
I agree that it would be bad if protocols were used to justify handing important decision making to inexperienced personnel. But I see them as useful in training, and my order of preference if I was the patient would be :
a) experienced clinician with Bayesian
b) inexperienced clinician with protocol
c) inexperienced clinician with Bayesian
I agree, and have attempted to teach the basics of Bayesian statistics on my Vocational Training Scheme. It is not easy.
absolutely, the problem is that we are building a massive amount of protocols to be run by less well trained practitioners into the NHS. Out of hours services and mental health especially. Instead of experienced doctors being retained at the interface of first contact between NHS and patient, experienced doctors are ‘promoted’ to do more management work, or commissioning. It’s backwards. It’s exactly the same with nursing – the hands on work is devalued, nurses get promoted to do things with less direct patient contact. But it’s exactly here, at that interface, that you need your best people.