Since the long-held potential for technology to assist in clinical decision-making has not come to fruition, researchers argue that the next generation of computerized support software requires more robust data to effectively identify disease patterns within specific patient populations.
Physicians are trained to rely on probabilities to diagnose patients, but humans are generally poor performers when it comes to probabilistic reasoning, according to a viewpoint published in the Journal of Medical Internet Research by two informatics researchers, including one with the IBM TJ Watson Research Center. So far, technology has failed to fill in those knowledge gaps, in part because clinical-decision support systems are still ineffective at consistently recognizing patterns.
Artificial intelligence—including IBM’s Watson—has been heralded as a potentially monumental technology to assist physicians with clinical decision-making. Cognitive computing has already been used in some hospitals to combat physician burnout, improve oncology care, diagnose skin cancer and predict admission and readmissions.
But two major reasons that artificial decision support software is still ineffective is because the available data lacks specificity and it fails to align with clinician workflow. Instead of relying on a “bag of findings” approach that lacks standardization, the authors argue that the next generation of decision-support tools need to capture a “richer clinical picture” by incorporating physician-generated crowdsourced data along with demographic parameters like age, sex, race and location that can pinpoint higher probabilities of a certain disease.
“The collective experience of physicians worldwide can be stored in a structured knowledge base made available to support pattern recognition-based diagnosis,” the authors wrote. “Similarity analysis could support this process by computing the degree of match between a patient’s pattern and patterns of other patients in the knowledge base who already have a diagnosis.”
But the authors also acknowledged that this approach “requires major healthcare stakeholders to make substantial, prolonged and coordinated efforts.” Although crowdsourcing data quality, a lack of real-time data sharing capabilities and medical complexities serve as significant limitations to this approach, successful implementation could lead to fewer diagnostic errors, better medical education, enhanced disease surveillance and better care for those in resource-limited areas of the world.