From 3M Health Information Systems
AI Talk: Kidney injury, empathetic tech and product recalls
This week’s AI Talk…
Predicting Kidney Injury
The Wall Street Journal had an article this week on Google’s Deepmind division getting closer to a solution that predicts various kidney conditions. Closer is the operative word, as the current prediction rate for kidney injury is just a little over 50 percent! For really severe conditions the prediction goes to 90 percent—but in these cases the physician is most likely already aware of the risk involved. Obviously, the solutions need to be fine-tuned and rigorously tested before they can be deployed in real hospital settings. Current detection of kidney injury happens after the fact using lab tests that measure creatinine levels—and tools such as these can be quite useful. One more deep learning solution that shows promise but needs to be evaluated further!
Empathetic care from tech
This article in FierceHealthcare pointed out that only 17 percent of psychiatrists believe one can receive empathetic care from tech. Not really a ringing endorsement of technology. The results shared in this article are based on surveying 800 psychiatrists in 22 countries. For the most part, they seem fairly skeptical of tech’s ability to help mental health patients. More than the survey and its results, a pointer in the article to the data collected by National Institute of Mental Health shocked me. According to the data collected by this institution, 20 percent of adults (47 million) and 50 percent of adolescents are afflicted with some mental health condition! The statistics are indeed enough to boggle the mind. Despite what the survey seems to suggest, there is significant evidence of technology helping certain categories of patients afflicted with various mental health conditions. And despite the sentiment of these psychiatrists, the venture capital community has voted with their pocketbooks in support of tech initiatives. According to an article in the same magazine, mental health and wellness companies raised $321 million in the second quarter of 2019 alone!
I saw a reference in AMIA’s daily download (deluge) of current events that describes an interesting technique to identify product recalls. Boston University researchers linked Amazon reviews to the FDA recall database of food products and figured out the correlation between negative reviews in Amazon and product recalls! They used a machine learning approach dubbed BERT, which stands for Bidirectional Encoder Representation Transformation. BERT was published by Google last year and has transformed the machine learning landscape. It’s a way to learn from a large corpus of data in an unsupervised fashion, and then what is learned can be transferred to specific problems like classifying which food should be recalled given the product reviews. What I find interesting is not their primary result—the correct identification of about 75 percent of the products that were recalled by FDA from reviews—but their follow-on claim. The same technique can be used to identify foods that should be considered for recall and have not yet come under FDA scrutiny.
I am always looking for feedback and if you would like me to cover a story, please let me know. “See something, say something”! Leave me a comment below or ask a question on my blogger profile page.
V. “Juggy” Jagannathan, PhD, is an AI Evangelist with four decades of experience in AI and Computer Science research.