As sophisticated as computers are now, isn’t about time we began using them more to help with the diagnosis of disease? Physicians tend to be diagnosticians, and primary care physicians need to have a massive breadth of knowledge these days in order to correctly diagnose the multitude of disorders in patients that may walk through their doors. The same goes for ER physicians. Currently, new doctors are relying more and more on information at their fingertips rather than information remembered. Perhaps relying even more on computers than we already do makes sense. Currently, we routinely use simple computer algorithms in clinical laboratory testing. Things like test results above an AMR causing the computer to direct the instrument to dilute and repeat the assay on that sample before reporting a result. Or diagnostically, a negative monospot test for Epstein Barr Virus (EBV) on a child under 4 years of age can be programmed to automatically order an EBV IgM and IgG, since the utility of the monospot test is unclear in that age group. This sort of “reflex” testing is already in use, and requires no operator intervention.
Here’s an example of a diagnostic approach that could be used: A sick infant comes in to the ED and has blood work run immediately. The initial results show a low pH, low bicarbonate and high pCO2. When a software program sees that combination of results, it could reflexively order more tests based on the differential diagnoses associated with a high anion gap metabolic acidosis (for example, ordering a blood glucose to detect diabetic ketoacidosis). If that is ruled out, the software then looks at the next most common cause of metabolic acidosis, and so on. The computer would not be diagnosing the child; the software would simply be ordering the appropriate next step tests to allow a rapid diagnosis, and probably doing it faster than the average multi-tasking ED doctor.
Software-based diagnostic systems exist and are on the market. So why are we so slow to adopt these systems into everyday use? We should let technology help us as much as we can. Software-based diagnostic systems have not been shown to be better than humans for diagnosing (http://www.nejm.org/doi/full/10.1056/NEJM199406233302506), and may never be. However, I would opine that they are faster than humans at deciding what tests to order based on lab results, or on a combination of lab results and clinical signs and symptoms. Using them this way would then leave the human to human interactions and the final diagnosis to the doctor and his patient when he has all the necessary test results at hand.
-Patti Jones PhD, DABCC, FACB, is the Clinical Director of the Chemistry and Metabolic Disease Laboratories at Children’s Medical Center in Dallas, TX and a Professor of Pathology at University of Texas Southwestern Medical Center in Dallas.
One thought on “Computers and Diagnostic Algorithms”
Well, that’s right.
How can we see the fact there so little studies published on this subject. Some recently are from Indian scientists, or other are date from 10 or 20 years ago?
It seems a little not realistic that scientists can give picture, and computer can now recognize any animals, or objects (with new NN algorythms and also with the help of huge image databases as ImageNet). Even if those data-scientists don’t know really why it works, it works.
The thing is also how can we get access to millions of diagnosis done by physicians. If any input, give me some hints.
Last thing is we talk about Software-based diagnostic systems which help or support physicians. But in other areas of expertise, face recognition, game playing, drive a car or even artistic creation, it might happen in several years that expert will only be able to help/confrim what computers will done or evaluate.