Customize your JAMA Network experience by selecting one or more topics from the list below.
Chronic or acute hypervolemia affects more than 6 million US patients, but there’s no practical way to precisely measure this harmful buildup of fluid to guide treatment. For patients undergoing hemodialysis, physicians typically rely on a physical examination and changes in body weight to monitor fluid status.
A technique called magnetic resonance imaging (MRI) relaxometry, while reliable, is expensive and unsuitable for routine use. Now, scientists have developed a portable magnetic resonance (MR) sensor that can be placed directly against the body, where it takes the same relaxometry measurements as a traditional MRI. It works faster and more cheaply than MRI machines by estimating extracellular fluid buildup from a single pixel instead of a whole image.
“Similar sensing methods have been used by the oil industry and by airport security [but] no one had yet applied this approach to the human body,” said researcher Lina A. Colucci, PhD, of the Massachusetts Institute of Technology in Cambridge.
The 11-lb sensor costs around $1000. It detected fluid changes after dialysis in the lower legs of 5 patients with end-stage renal disease in a recent proof-of-concept study reported in Science Translational Medicine. The sensor performed comparably with bioimpedance measurements but, unlike the latter, factors such as sweat or electrode placement don’t affect its readings. The MRI outperformed both bioimpedance and the MR sensor.
Next, Colucci’s team wants to conduct a larger trial using a more sensitive version of the sensor that should detect early fluid overload with a single measurement, as was demonstrated with MRI. The work could open the door for many more portable, point-of-care MR diagnostics, she said.
Other work suggests that it also soon may be possible to continuously assess the risk of kidney injury in hospitalized patients, reducing the need for dialysis.
A new deep learning model predicted 55.8% of inpatient episodes of acute kidney injury up to 48 hours before they could be diagnosed clinically. The tool, recently described in Nature, also predicted 84.3% and 90.2% of kidney injuries that led to dialysis within 30 and 90 days, respectively. The model had a ratio of 2 false predictions for every true-positive prediction, but most of the false-positives were in patients with existing chronic kidney disease.
Researchers at the University College London, working with experts at the US Department of Veterans Affairs (VA), trained and tested the model using separate sets of electronic health records from more than 700 000 adult VA inpatients. The approach more accurately predicted acute kidney injury in men than in women, who comprised just more than 6% of the patients in the overall data set. Future training and evaluation are needed to address this limitation.
In an accompanying viewpoint, Eric Topol, MD, of the Scripps Research Translational Institute in La Jolla, California, said the model “stands out by providing a prediction that might enable effective clinical intervention.” A fifth of post-admission acute kidney injury cases are avoidable, according to one estimate.
Although the researchers looked at acute kidney injury in this study, they suggested that deep learning approaches could be used to predict the risk of future patient deterioration across medical conditions.
Abbasi J. Advances in Fluid Assessment and Kidney Injury Prediction. JAMA. 2019;322(10):918. doi:10.1001/jama.2019.13930
Create a personal account or sign in to: