Researchers at the University of California, San Francisco, have developed a strategy for prefetching relevant prior images with almost perfect accuracy, a step they believe can improve their interpretations and may reduce their legal liability for missed diagnoses.
The most common radiology malpractice issue involves failure to read a relevant prior image, which usually occurs because those images are not available, said Katherine P. Andriole, Ph.D., lead researcher. In a filmless environment, prefetching relevant priors from the archive to be read alongside the new images has been tried as a solution, but this has proved extremely difficult to implement. It has never been as successful as an experienced fileroom clerk.
Other studies have considered the problem, the researchers said: Just pulling the prior two studies of any type yielded an 83% likelihood of getting the relevant prior. Pulling the prior four studies of any type produced one relevant prior 91% of the time. Retrieving all studies for a given patient from the past month produced a relevant prior 80% of the time.
The innovation in the latest study was the formation of metagroups such as gastrointestinal, chest, and abdomen that could represent in a small number of categories the hundreds of examinations performed by a typical radiology department. Tables incorporating the metagroups are included in the algorithm to help establish the relevance.
An example: A patient is given a CT chest exam for possible esophageal cancer. The prefetch mechanism looks for other CT chest exams, chest x-rays, and esophagrams.
In the study, the sensitivity of the prefetch algorithm was found to be 98.3% and the specificity 100%. Time required to obtain the prefetched images averaged 9.5 minutes. The maximum number of relevant priors was three, with the most relevant chosen first.
The system is not trouble-free, Andriole said. If the exam information is not specific — "CT angio," for example — and does not specify a body area, the system will prefetch CT angiograms and may ignore relevant priors.
Overall, however, the system appears to pull the most relevant priors and reduce the number of missed relevant priors along with the time spent chasing them, she said. It has also reduced network traffic by decreasing the number of priors selected and transmitted to the viewing stations.
