Echo Dot (3rd Gen) - Smart speaker with Alexa - Charcoal

Use your voice to play a song, artist, or genre through Amazon Music, Apple Music, Spotify, Pandora, and others. With compatible Echo devices in different rooms, you can fill your whole home with music.

Buy Now

Wireless Rechargeable Battery Powered WiFi Camera.

Wireless Rechargeable Battery Powered WiFi Camera is home security camera system lets you listen in and talk back through the built in speaker and microphone that work directly through your iPhone or Android Mic.

Buy Now

Machine Learning Takes On Sepsis


Machine Learning Takes On Sepsis 1
Startup Rubicon Health is using AI and machine learning to improve detection of sepsis, a life-threatening infection that kills one in three stricken adults.

By Deborah Borfitz, Senior Science Writer

An early-stage startup is endeavoring to help clinical teams deliver high-value critical care with precision, and its first target is sepsis—a life-threatening complication of infection that kills one in three stricken adults and one in 10 kids. Its chief tactics are to “organize behavior at the bedside” and make machine learning a member of the team, according to James Courtney Fackler, founder of Rubicon Health as well as director of pediatric critical care medicine and associate professor of anesthesiology and critical medicine at John Hopkins Medicine.

Fackler led a session at the 2019 Next Generation Dx Summit on integrating artificial intelligence into clinical care for infectious diseases. He works full time in a pediatric intensive care unit (ICU) but is also an avid reader who believes in human intuition, he says, referencing the books Sources of Power (Gary Klein) and The Book of Why (Judea Pearl).

As a physician, it’s difficult to be precise when faced with 350 clinical data elements on a child coming out of surgery, Fackler says. The human brain can generally handle five to seven pieces of information at a time and, if it’s well-trained, perhaps 12 to 15.

In a 40-bed pediatric ICU, such as the one at Johns Hopkins, a physician might be looking at 14,000 data streams and “some of it is analog and some of it is wrong,” says Fackler. How well a diagnostic device improves patient care involves all 150 people on the team.

Machine learning could be playing a pivotal role to help solve some of the more perplexing problems, including earlier diagnosis of sepsis, he adds. “Mortality increases 8% per hour if antibiotics are delivered late.” Better patient outcomes also require care coordination.

“Goal-directed therapy is bad,” Fackler says. “It does not work; the goals are wrong.” The central problems are that sepsis is a complex condition and needed treatment is delivered late.

Rubicon Health is currently working on a machine learning algorithm for predicting sepsis, which has an 8% to 10% prevalence rate in the pediatric ICU, says Fackler. The goal is to do better than current methodologies that typically have a positive predictive value of around 19%—up to one in three cases “at best.” Asking additional questions not in the electronic health record (EHR) will save a life one in five times, he adds.

The idea behind the algorithm is to help the computer-clinician dyad know when a child is sick and chart out what the next 10-12 steps should be, including administration of an antibiotic within one hour, says Fackler. Rubicon is using a version of collaboration hub Slack to create a place for care teams to hold virtual huddles.

The MITRE Corporation was sent a dataset and, based on prescribed criteria, came up with a positive predictive value for sepsis at different times slices in the 24-hour period preceding its onset, says Fackler. “We’re still only in the 20% range. That’s not good enough, but it’s a decent place to start.”

In an article published earlier this year in Scientific Reports, Fackler and his colleagues discussed how they used publicly available retrospective data to make an earlier prediction of impending septic shock by applying three different machine learning techniques to the EHR data of 15,930 adult patients, producing a median early warning time of seven hours.

The paper introduces the notion of patient-specific positive predictive value. The detection method estimates the confidence with which a positive prediction is made, providing more reliable and actionable information to clinicians than would an alert alone, Fackler says.

Of the eight potential paths a clinician might take—based on machine-generated true and false positives and true and false negatives together with the doctor’s clinical judgement to treat or not to treat in each of those cases—Rubicon hopes to push systems toward the two where diseased people get treated. “The false negatives are what keep me awake at night,” says Fackler.

“Early diagnosis, if it leads to early therapy, is the only thing that will bend the cost curve,” Fackler says, noting he is making a case for team intelligence. “As a clinician you need to know when to turn the machine off and do things by hand.”

Read More


Please enter your comment!
Please enter your name here