Next-gen technology raises hopes for true clinical decision support
One of the many long-held promises of health IT has been based on the contention that computers would one day be able to partner with clinicians in the diagnosis of disease. While, with AI developments such as IBM’s Watson, significant strides have certainly been made, many stakeholders lament what they consider the slow pace of progress.
Writing recently at the Journal of Medical Internet Research, two informatics researchers, Amos Cahan, MD, and James J Cimino, MD, of the IBM TJ Watson Research Center and the Informatics Institute at the University of Alabama at Birmingham, respectively, offer an analysis of the current status of computer-aided diagnosis support systems (DSSs), as well as their assessment that the technology may finally exist to begin seriously realizing the long-held dreams of clinicians.
In their view, there are two core reasons why DSSs are still ineffective. First, they note, despite the significant increase in digitally available data, that data still lacks specificity. “Current DSSs cannot efficiently match patients and diseases on patterns,” they argue, “since they rely on a unidimensional projection of clinical information; typically, the system uses a vector of “findings” (symptoms, signs and laboratory results) provided by the user to generate a differential diagnosis. Some systems differentiate between acute and more prolonged processes, but none are able to cluster findings based on their course in time. Using this ‘bag of findings’ approach makes DSSs agnostic to key clinical clues. For instance, chest pain and dyspnea appearing during physical exercise strongly suggest angina, whereas shortness of breath with subsequent chest pain may suggest pneumothorax or a pulmonary infarction. From a DSS point of view, these conditions are indistinguishable, as both have the same findings: ‘chest pain’ and ‘dyspnea’.”
The other problem, they note, is that DSSs do not align well with clinicians’ work flow. At best, “a few DSSs now offer variable degrees of direct connectivity to the electronic health record (EHR), (while) some can extract data from the EHR using natural language processing tools, although this may adversely affect performance.”
Despite the ongoing challenges, the writers contend that the technology may finally be available to assist clinicians more effectively than ever before. “Industries have figured out...that big data becomes transformative when disparate data sets can be linked at the individual person level,” they note, and technology is now ripe to enable the development of next-generation DSSs (NGDSSs).
The rest of their piece is dedicated to mapping out how NGDSSs should be able to help clinicians develop what they call a “structured presentation pattern” of patient diseases.
“A structured pattern can be thought of as a model, which can represent physician knowledge and reasoning in a machine-interpretable format,” they explain. “A structured pattern should ideally represent key symptoms and signs associated with a particular patient’s presentation and their temporal and semantic interrelations. This allows for translation of a list of findings (symptoms and signs) into multiple distinct structured patterns according to the temporal course of the disease and other relations between findings. Through this approach, a differential diagnosis constructed by NGDSSs is likely to be more specific than one based on a list of findings.”
Are we there yet? Not really, as the authors acknowledge that their vision ‘requires major healthcare stakeholders to make substantial, prolonged and coordinated efforts.” At the same time, they conclude, “EHR systems have become ubiquitous; powerful computers enable sophisticated analytics; the Internet can connect physicians from around the globe in real time; and human-computer interaction technologies have ripened. Taken together, there is both a real need for NGDSSs and the technology to meet it.”