“This Week,” an editorial feature in Nature[1], The editor asks what the BBC, IPCC and the CIA have in common. As an example The CIA in Mar 1951, warned of a “serious possibility” that Russia would invade Yugoslavia. Kent, the intelligence analyst was dismayed that nobody seemed to agree on just what that meant. Correcting the problem, the CIA dropped the term possible, replacing it with almost certain and almost certainly not. The article goes on to examine the problems that the Intergovernmental Panel on Climate Change (IPCC) has with its estimates of global warming.
The same problem exists with medical diagnosis. Serious retrospective studies indicate an alarming error rate in initial diagnosis. There are a number of causes: Reimbursement for lab studies requires a relevant diagnosis. Reimbursement for the visit requires a definitive diagnosis, a so-called 5-digit code as opposed to a broader category of 3 or 4 digits. -- It’s an ugly system. Discounted reimbursement, often more than 50% encourages snap diagnosis. Additionally, many encounters hide conditions that are not on the clinician’s list of commonly encountered diseases, problems that require research and head scratching with studies that may or may not be revealing. After all, what good is a test if you already know the answer? Moreover, what good is a diagnosis if you don’t consider the possibilities? Differential diagnosis has become a lost art. Some form of quantitative measurement needs to accompany a medical diagnosis along with the other possibilities. Intriguingly, statistics may be a better answer
When a doctor attends a clinical pathological conference (CPC), he or she hears from a number of persons associated with the case, the clinician, the radiologist, the pathologist, any number of consultants and often a visiting expert. They seldom agree and this takes place after the autopsy.
How then should we state a diagnosis? Fuzzy at best, should we say, “almost certain” or “almost certainly not?” Statistics serves a better quantitative role. If we compare a thousand instances of a diagnosis with the final-outcome, we can equate the degree of fuzziness in quantitative statistical terms.
Furthermore, in real-time, we can offer these statistics --- taken in relation to the signs, symptoms, physical findings and tests --- to the clinician as a differential diagnosis with numeric probabilities. The doctor, nonetheless, must choose based on the individual. In doing so, the diagnosis will have some validity.
Statistics can provide enough documentation to overcome the denials from a third party provider, allowing for reasonable studies and testing before jumping to conclusions.
No comments:
Post a Comment