Amazon is showing a decided interest in the increasingly hot area of healthcare AI, first by offering a text analytics product and now by partnering to study AI use in a clinical setting.
Late last year, Amazon Web Services announced the availability of Comprehend Medical, which uses natural language processing and machine learning to gather relevant information from unstructured text. AWS says that the software can pull needed information from physician notes, patient health records and clinical trial reports.
Now, Amazon is moving into research on direct clinical uses of AI. It’s given Harvard teaching hospital Beth Israel Deaconess Medical Center (BIDMC) a grant of up to $2 million to look at how machine learning can improve patient care. The research plans, which will extend over multiple years, cover a great deal of ground.
Broadly speaking, BIDMC plans to use AWS machine learning services to look at how such tools can enhance clinical care, streamline operations and eliminate waste. Its first research projects used machine learning to optimize its operating room schedules to improve patient flow in its inpatient setting and improve operational workflow within the ORs. It was just warming up.
More recently, BIDMC has begun a project in which incoming pre-surgical document packages will be scanned as images, processed with TensorFlow on Amazon SageMaker and hosted in BIDMC’s AWS cloud. The machine learning software will then scan EHRs to look for important elements like completed consent forms. If it doesn’t find a consent form, it will insert a signal in the patient record for nurse follow-up.
In addition, BIDMC has kicked off a project which helps improve efficiency in performing surgical procedures. Historically, such procedures have been delayed or rescheduled at times because staff members haven’t been able to locate a completed History & Physical form among records that may be faxed to the hospital. However, BIDMC now uses Comprehend Medical to search through the faxes to identify H&Ps, which saves considerable time.
Meanwhile, BIDMC has a lot more planned for future AI experiments. These include an effort to predict which patients will and which patients will not keep their appointments. Using the Apache MXNet deep learning API and Amazon SageMaker, the medical center will reach out to patients who might miss appointments to help get them to come.
Also, BIDMC has developed a machine learning model built on AWS which detects where operating room schedule changes would improve efficiency, save costs and balance hospital load, as well as predicting the outcome of such changes and how to minimize their impact on patient care.
And I’d be remiss if I didn’t mention yet another healthcare AI project BIDMC has on the drawing board, this one focused on assessing overall risk levels in ICUs and predicting when the hospital can expect an unusually high volume of incoming patients. To do the latter, BIDMC and its academic partners will analyze data such as ED admissions histories, transfers between healthcare institutions, referrals, pre-scheduled surgeries, patient discharges and other variables.
At this point, all I can say is “wow.” Unless its access to the intellectual property generated by these efforts is limited — which I doubt highly — Amazon is sure getting a lot for its money. I can only anticipate that other high-tech firms with no institutional healthcare assets of their own will be doing the same thing soon.