As far as I can tell, IBM's Watson has not lived up to the expectations of many healthcare providers which was probably predictable, given the exuberance of its marketing campaign. Too many promises. However, a recent article about its use in creating a Patent Synopsis caught my eye. This would appear to be a potentially useful tool given the complexity of EHRs and the inability of users to often extract usable information from it (see: IBM Watson is taking on EHR qualms with a new AI tool). Below is an excerpt from this article:
AI-enabled programs that help parse through data have the potential to minimize physicians' headaches with EHRs and ultimately slash costs. Patient Synopsis simplifies data extraction, which can help minimize medical errors resulting from physician burnout. Patient Synopsis pulls out clinically relevant information from a patient's EHR in seconds — including current illnesses, medications, and medical history — and organizes it into 12 categories....By eliminating the need for doctors to dig through mounds of patient data, hospitals could avoid these errors and save a huge amount of money....The tool could also help doctors expedite treatment plans and clamp down on unnecessary testing. Patient Synopsis offers comprehensive, easy-to-access patient summaries, which could grant physicians a clearer view of the best plan of action for treatments and more quickly send patients onto the road of recovery. This could also help prevent doctors from ordering tests that might not be needed to guide effective treatment....
Watson...struggled to retain major hospital clients in 2018, citing softening demand....But its AI-enabled solution for EHRs could help the tech titan lay claim to more healthcare partners, considering 92% of regular hospitals in the US have implemented EHR systems and likely all share in the struggle of managing the burdensome content they contain. Landing Patient Synopsis in Hardin Memorial Health puts IBM in competition with other tech giants offering solutions to combat EHR-related troubles. Leading cloud vendors...have all jumped in to offer solutions to EHR woes: For example, Microsoft's Healthcare Next program seeks to mitigate the burden of inputting EHRs for physicians, and Google's Cloud for Healthcare platform is designed to simplify the process of pulling insights from patient data stored in the cloud to improve clinical decision-making....[IBM] joined forces with other tech leaders last August to tackle interoperability issues plaguing the healthcare system, which often stem from difficulties in sharing EHR data among providers. Now that IBM's doubling down on curbing EHR troubles, it will likely eye more hospital partnerships to emerge as a front-runner in the crowded market.
Although it may not be obvious to all observers, the goal of IBM's Patient Synopsis project, extracting clinically relevant information from the EHR "soup", is a daunting task. Clinical documentation with EHRs is commonly entered using drop-down menus with a controlled vocabulary, making computerized synopsis relatively easy. A similar story pertains to medications and numerical data from the labs. However, surgical pathology and radiology reports, although labeled with diagnostic codes, may harbor subtleties in text that can be important. Such textual reports require computerized natural language processing (NLP) to extract meaning. I have posted a number of previous notes about this topic (see, for example, Assessing Drugs Using "Real World Evidence" in Addition to Clinical Trials; AI Allows Computers to "Read" EHR Records and Make Predictions; Amazon Launches New Medical Record Language Processing Service). Because the EHR record for an individual patient can be complex, the risk of the creation and availability of a patient synopsis generated by Watson is that clinicians may come to depend on it and miss some of the information contained in the larger EHR record. Hence and at least initially, the Patient Synopses will probably be published with a warning label.
Comments