Building on their prior work aimed at providing early diagnosis to critically ill newborns, researchers at Rady Children’s Institute for Genomic Research in San Diego reported using an artificial intelligence (AI) pipeline to speed up diagnosis of rare genetic diseases, doing so in a record median time of 20.10 hours (Sci Transl Med 2019;11.eaat6177). These efforts in a retrospective analysis of children already diagnosed with genetic diseases matched expert interpretation in 95 children with 97 diseases with 97% sensitivity and 99% precision. Prospectively, the AI pipeline correctly diagnosed 3 of 7 seriously ill infants with 100% sensitivity and precision.
Use of this pipeline, which incorporated a machine learning process and clinical natural language processing (CLNP), could be key in broadly disseminating rapid whole genome sequencing (rWGS) to neonatal intensive care units (NICU), according to the researchers.
Standard rWGS with manual analysis and interpretation of genomic data for diagnosing genetic disorders in newborns typically takes a mean of 16 days, although the researchers in a prior study cut this to 26 hours.
The investigators deployed AI and other new technologies to tighten their rWGS workflow. They used a Nextera DNA Flex Library Prep Kit (Illumina) to manually prepare sequencing libraries directly from blood samples or dried blood spots, which cut out several prep steps. They also performed rWGS with the NovaSeq 6000 sequencer and S1 flow cell (Illumina), which is faster and less labor intensive than their legacy Illumina sequencer. In addition, they used an Illumina hardware and software platform (DRAGEN) “highly optimized for speed, sensitivity, and accuracy” to align and call variants.
To speed up the process of reviewing electronic health records (EHRs) to identify patients’ phenotypes, the research team used CLiX CLNP (Clinithink), which they optimized to extract clinical features from unstructured text in EHRs.
The investigators also wrote scripts to automatically transfer patients’ nucleotide and structural variants from DRAGEN to MOON, an autonomous interpretation software (Diploid) that automates genome interpretation using AI to automatically filter and rank likely pathogenic variants.
Urine Drug Test Results Chart Dramatic Rise in Fentanyl Use
A study of 1 million urine drug test (UDT) results ordered as part of routine care by physicians throughout the U.S. found a 1,850% increase from 2013 to 2018 in the rate of nonprescribed fentanyl positivity in samples that also were cocaine-positive and methamphetamine-negative, and a 798% increase in the rate of nonprescribed fentanyl-positive samples among methamphetamine-positive and cocaine-negative results (JAMA Network Open 2019;2:e192851).
The results underscore that fentanyl, either added surreptitiously to or taken with other drugs, could be a contributing factor in the sharp rise in cocaine- and methamphetamine-related overdose deaths, according to the investigators.
The study involved a random sample of UDT results analyzed for definitive testing by Millennium Health using liquid chromatography-tandem mass spectrometry. Positivity thresholds for fentanyl, norfentanyl, benzoylecgonine, and methamphetamine were ≥2 ng/mL, ≥8 ng/mL, ≥50 ng/mL, and ≥100 ng/mL, respectively.
The median age of the study population was 44, and more than half of patients were women. Test orders came from many different practice settings, with substance abuse disorder treatment centers and pain management practices accounting for 53.5% of specimens. Overall positivity rates for cocaine, methamphetamine, and fentanyl were 4%, 3.1%, and 1.4%, respectively. Positivity rates for nonprescribed fentanyl in cocaine-positive, methamphetamine-negative results rose from 0.9% in 2013 to 17.6% in 2018; and in cocaine-negative, methamphetamine-positive results from 0.9% to 7.9%.
Clinical Assessment, Procalcitonin-guided Antibiotic Therapy Yield Comparable Outcomes
A head-to-head comparison of guideline-based clinical assessment versus procalcitonin (PCT)-guided antibiotic therapy in patients with community acquired pneumonia (CAP) found that the two approaches were about equal in the primary outcome of total antibiotic exposure within 30 days of emergency department (ED) admission (Ann Emerg Med 2019; doi.org/10.1016/j.annemergmed.2019.02.025). “Because [PCT] assessments add time and cost to patient care, these findings support strategies” to help providers better understand and follow clinical guidelines for managing patients with CAP, the authors suggested.
This pragmatic multicenter trial of 370 eligible adult patients in 12 French hospitals randomly assigned 143 patients to clinical assessment and 142 to PCT-guided care. The study enrolled patients given the presumptive diagnosis of CAP while in ED, based on meeting at least two of three respiratory infection criteria. All patients had their PCT levels measured but attending physicians did not see results for patients in the clinical assessment group.
In the runup to the start of the trial, all participating physicians received about 2 hours’ training on the background and use of guideline-directed clinical assessment as well as the PCT algorithm. The latter recommended against or strongly recommended against starting antibiotics when PCT levels were ≤0.25 µg/L or <0.1 µg/L, respectively. Conversely, the PCT algorithm recommended or strongly recommended starting antibiotics if PCT levels were ≥0.25 µg/L or ≥0.5 µg/L, respectively. If antibiotics were not started in the ED after the first PCT result, another PCT measurement took place after 6 to 24 hours, with antibiotic therapy considered using the same cutoffs.
The antibiotic duration was not significantly different between the two groups, with medians of 9 days and 10 days respectively and a P value of 0.21, nor was it significantly different in the modified intention to treat populations that excluded patients who had received antibiotics for other indications or who had Legionnaires’ disease.