What are some concrete ways in which present day tech can improve the productivity of the healthcare sector?
In a nutshell, we are focusing on ways of relieving frontline workers of the burdens of documentation and allowing them to concentrate more on patients' needs. In 2018 we embarked on a major project with Nuance to develop the Dragon Ambient eXperience (DAX) - a solution that automatically documents doctor-patient conversations and builds a clinical encounter note that is later integrated within all the major electronic health record systems. It enhances the quality of care and the patient experience, increases provider efficiency and satisfaction, and improves financial outcomes with clinical documentation that writes itself. In addition, our solutions have the ability to find possible errors in prescriptions and identify the probability for prior authorization in Medicare.
We also developed PubMedBERT, which is the primary tool for reading medical research texts. Given that about 4,000 new medical research papers are released each week in the field of oncology alone, it is impossible for anybody to keep up with so much reading. PubMedBERT has been used by 700 cancer specialists around the world to surface in an automatized way all the studies that would be the most relevant to their case.
In which ways can the cloud technologies help change the course of healthcare as we know it?
The biggest advantage of the cloud is that data is under the control of the organizations that produce it, while also being fit to be aggregated and shared. For example, radiological images - which are normally put in separate servers - can be shared over the cloud and linked to nationwide electronic health records. In that way, radiologists can treat data together, checking on other data to increase the value of the types of data they have.
For the past six years we have been working with Adaptive Biotechnologies to develop a machine learning technology that can separate all the T-cells from a blood sample, analyze the DNA and produce a printout with information about every disease that your body is coping with. This piece of tech can be used to identify the immune signatures of many different diseases to improve diagnosis and monitoring. At the start of the pandemic, a company gathered data from Vo' village in Italy based on blood samples taken from two thirds of the population. Because the data was shared in the cloud, they were able to take all those samples, aggregate them with data from other sample sets, and produce a new diagnostic test (T-Detect) for Covid-19. Presently, the same infrastructure is being used to identify and validate disease signatures across several therapeutic areas, including autoimmune disorders, cancer, and infectious diseases.
How does technology translate into an actual protocol to develop precision medicine?
As it turns out, the code employed in AI learning is very similar to the code used for translating languages. Similarly, the topic identification that we can find in search engines has the same basic principles as precision medicine. These ways of analyzing DNA and the genetics of a disease are shockingly similar to the kinds of problems we are solving with machine learning for understanding and translating languages or identifying topics in documents.
Are there areas in which human intuition may be more effective than algorithms and analyzing data?
When we build a car, we do it in a modular way so that changing one part does not affect the others. In human biology, this is not possible because everything is interlinked in unexpected ways that we do not fully understand yet. It is highly important to understand that these are mere tools and that doctors' experience, and instincts cannot be replaced that easily. However, it still is incredible to have an AI system that can spark new ideas when medical questions arise.
All in all, AI will not make treatment decisions by itself, but it will help doctors consider more options, and give them a chance to review information and to make a better call, especially for difficult medical cases.
All these technologies seem like the solution to everything, but when will we see them being applied in a widespread context?
Although there is no fixed timetable for when these technologies will become widespread, we like to think of them as "inevitable futures," so we are doing everything in our power to start that research today. With both human benefits and other incentives motivating future development, the trick lays in finding the best path from here to there, as it can have many twists and turns. More to the point, if we can reduce by just one or two years the time it takes to go from molecule candidate to a new approved drug, we could save billions of dollars in economic value. There is huge incentive to make this technology move faster and become more practical.
In general, the public in the U.S. and Europe have expressed complaints about how far behind healthcare is from where they imagine it could be in terms of digital technologies, but the reality states otherwise. Just 15 years ago, only 15% of health records were digital, nowadays we reached an incredible 98%. I think precision medicine will follow in the same path and in ten years from now we will wake up to machine learning and the cloud creating personalized treatments.
