Using smartphones to detect AF

More than 6 million Americans have atrial fibrillation (AF), yet the condition remains underdiagnosed. Given that smartphones are ubiquitous, it has been proposed that smartphone applications might detect AF, but their overall accuracy is unclear. In a systematic review and meta-analysis, researchers in the U.S., U.K. and Australia assessed the accuracy of smartphone apps that use the device’s camera to measure the amplitude and frequency of the user’s fingertip pulse to detect AF. The team included 10 primary diagnostic accuracy studies, with 3,852 subjects and four smartphone apps. The overall meta-analyzed sensitivity and specificity was 94.2% and 95.8%, respectively. The positive predictive value (PPV) for camera apps detecting AF in asymptomatic subjects aged 65 and older was between 19.3% and 37.5%, and the negative predictive value (NPV) was between 99.8% and 99.9%. Both PPV and NPV increased for individuals 65 and older with hypertension. “The modeled NVP was high for all analyses, but the PPV was modest, suggesting that using these applications in an asymptomatic population may generate a higher number of false-positive than true-positive results,” the team wrote. “Future research should address the accuracy of these applications when screening other high-risk population groups, their ability to help monitor chronic AF, and, ultimately, their associations with patient-important outcomes.” Their work appeared online April 3, 2020, in JAMA Network Open.

A study of AI algorithms in mammography assessment

Researchers conducted an international crowdsourced challenge to foster the development of artificial intelligence (AI) algorithms focused on screening mammography interpretation, with the aim of determining how well algorithms performed compared with radiologists. Algorithm accuracy for breast cancer detection was assessed using under the curve and algorithm specificity vs. radiologist specificity with radiologists’ sensitivity set at 85.9% (U.S.) and 83.9% (Sweden). Two cohorts of 144,231 screening mammograms from 85,580 U.S. women and 166,5778 exams from 68,008 Swedish women were used to train and validate the algorithms. The top-performing algorithm achieved under the curve of 0.0858 and 0.903 for U.S. and Sweden, respectively, and 66.2% and 81.2%, respectively, at the radiologists’ sensitivity – below the 90.5% (U.S) and 98.5% (Sweden) specificity observed in community practice radiologists. Combining top-performing algorithms with U.S. radiologist interpretations, however, yielded a higher under the curve of 0.942, with specificity of 92.0% and no change in sensitivity. “While no single AI algorithm outperformed radiologists, an ensemble of AI algorithms combined with radiologist assessment in a single-reader screening environment improved overall accuracy,” the team concluded. “This study underscores the potential of using machine learning methods for enhancing mammography screening interpretation.” The researchers believe their study is the first in AI and mammography benchmarking requiring teams to submit algorithms to challenge organizers, allowing for unbiased and reproducible evaluation. As a stipulation of the DM DREAM challenge, the fully documented algorithms are freely available for use in future studies of automated and semiautomated mammography assessments. The study was published online March 2, 2020, in JAMA Network Open.

Stopping tau in its tracks

Investigators at the University of California at Santa Barbara have identified the low-density lipoprotein receptor-related protein 1 (LRP1) as a key to controlling the internalization and spread of tau protein in Alzheimer’s disease (AD). Tau aggregates are a critical feature of advanced AD, and once tau aggregation has begun, the protein can travel from neuron to neuron in a prion-like fashion. The authors used CRISPR to knock down various members of the low-density lipoprotein receptor (LDLR) family, as previous work had shown that proteins that interact with LDLRs were important for the spread of tau. They showed that LRP1 took up both tau oligomers and fibrils via endocytosis. “Our results identify LRP2 as a key regulator of tau spread in the brain, and therefore a potential target for the treatment of diseases that involve tau spread and aggregation,” the authors wrote. They published their study in the April 2, 2020, issue of Nature.

No Comments