Deep learning aid for diagnosing TB in HIV patients
Tuberculosis is the No. 1 cause of preventable death in HIV-positive patients, but is often undiagnosed and untreated. Chest X-ray, which is often used in diagnosis, poses challenges due to atypical radiographic presentation and shortages of radiologists in some regions with high co-infection rates. Researchers at Stanford, Massachusetts General Hospital and the University of Cape Town, South Africa, developed a deep learning algorithm, CheXaid, to diagnose TB using clinical information and chest X-rays from 677 HIV-positive patients with suspected TB and tested its performance as a web-based diagnostic assistant. Use of the algorithm led to a modest but statistically significant improvement in clinician accuracy. However, accuracy was significantly higher when the algorithm was used alone. “These results suggest that deep learning assistance may improve clinician accuracy in TB diagnosis using chest X-rays, which would be valuable in settings with a high burden of HIV/TB co-infection,” the authors wrote. “Moreover, the high accuracy of the stand-alone algorithm suggests a potential value particularly in setting with a scarcity of radiological expertise.” Their work was published online Sept. 9, 2020, in npj: digital medicine.
Self-collected swabs vs. health care worker collected for COVID-19 testing
Researchers at Copenhagen University Hospital reported the results of a study showing that the reliability of self-collected swabs for SARS-CoV-2 testing is comparable to health care worker (HCW)-collected swabs. A total of 109 symptomatic individuals provided mobile phone video-instructed self-collected oropharyngeal and nasal swabs, followed by an HCW-collected oropharyngeal sample. All of the samples were analyzed by the same microbiology laboratory. Nineteen subjects tested positive for SARS-CoV-2. The sensitivity of the self-collected swabs was 84.2%, vs. 89.5% for HCW-collected, with an acceptable agreement of Cohens kappa 0.82, p < 0.001. A questionnaire completed by participants showed that self-reported loss of smell was a strong predictor of a positive test. “The SARS-CoV-2 self-test described here is … a reliable testing method with a low false-negative rate compared to the technique based on HCW-collected swabs and might be used in community testing or settings where HCW time and personal protective equipment need to be economized,” the team concluded. “Future studies should explore the diagnostic accuracy and cost-effectiveness of this method when implemented in a larger and more heterogeneous cohort of patients tested for COVID-19. Their work appeared online Sept. 9, 2020, in Diagnostics.
SCAD vs. plaques in heart attacks
Scientists at the University of Michigan have compared the genomic contributions to atherosclerosis and spontaneous coronary artery dissection (SCAD), and found that genome variants that were associated with a higher risk of SCAD-induced heart attack in their study were associated with a lower risk of atherosclerosis-induced heart attack. SCAD is a form of heart attack that is most common in young women in which the coronary artery wall hemorrhages, most often due to an initial tear that can be due to physical stressors including childbirth and domestic violence. In their work, the Michigan team investigated underlying genomic risk factors and showed links to both fibromuscular dysplasia, which co-occurs with SCAD, and migraine. They also showed that SCAD and atherosclerosis-induced MI had an inverse relationship, that is, genomic loci that were associated with a lower risk of plaque-induced heart attacks in other studies appeared to increase the risk of SCAD. The researchers reported their results in the Sept. 4, 2020, online issue of Nature Communications.
Diagnosing neuroblastoma in children
Neuroblastoma is the most common extracranial tumor in children, accounting for 11% of cancer-related deaths. Detecting and characterizing cell-free DNA (cfDNA) in peripheral blood from neuroblastoma patients could provide a minimally invasive liquid biopsy approach to diagnosis; however, small sample volumes and low cfDNA concentrations make cfDNA analysis challenging. Droplet digital PCR (ddPCR) allows for analysis using low levels of cfDNA. In a study published online Aug. 24, 2020, in The Journal of Molecular Diagnostics, German researchers described two quadruplexed ddPCR assay protocols that can be used in routine clinical settings to assess MYCN and ALK oncogene status using minimal plasma volumes. The quadruplexed ddPCR protocols reliably quantified MYCN and ALK copy numbers in a single reaction, along with two reference genes, NAGK and AFF3, and correctly estimated two mutant allele fraction, ALKF1174L and ALKR1275Q, using cfDNA as input. “We optimized separation of positive and negative droplets to detect two targets in each ddPCR fluorescence channel by adjusting probe and primer concentrations for each target molecule,” the authors wrote. “The quadruplexed assays were validated using a panel of 10 neuroblastoma cell lines and paired blood plasma and primary neuroblastoma samples from nine patients.” The accuracy and sensitivity of the quadruplexed assays corresponded well with those of the respective duplexed assays.