A prototype artificial intelligence (AI) tool showed that just three inputs had the most predictive power for discerning which patients with COVID-19, the illness caused by the novel coronavirus SARS-CoV-2, would likely develop acute respiratory distress syndrome (ARDS) (Comput Mater Con 2020;63:537-51). Based on data from 53 patients treated at two hospitals in China, this predictive analytics system found that elevations in alanine aminotransferase (ALT) and hemoglobin, along with patient-reported myalgia predicted risk of ARDS with up to 80% accuracy.
Surprisingly, many of the clinical features associated with COVID-19 such as ground glass opacities on chest computed tomography, fever, cough, lymphopenia, and dyspnea did not distinguish risk of disease progression and were not highly predictive. Patients’ viral load (cycle threshold) also did not prove to be predictive.
Moreover, the patients’ ALT and hemoglobin values were only modestly elevated. The median ALT value at the time of presentation at hospital was 24 U/L (range, 15-40.5 U/L; reference range, 9-50 U/L). The median hemoglobin level was 13.7 g/dL (range, 12.9-14.4 g/dL, reference range, 12.8-16.5 g/dL). Other features, including sex, temperature, age, and levels of sodium, potassium, and creatinine, and lymphocyte and white blood cell counts, added modestly to prediction.
“The model highlights that some pieces of clinical data may be underappreciated by clinicians,” wrote the investigators in Wenzhou, China, and in New York City. They added that features don’t have to be causal to be predictive.
In their feature engineering and statistical analysis, the researchers employed entropy, which measures how much information a feature encapsulates; information gain—the amount of information acquired after knowing the value of the feature; Gini index, a measure of the impurity of a dataset; and Chi-Squared statistics, indicating how dependent two variables are.
The authors speculated that myalgia “could represent generalized inflammatory and cytokine response not captured well by other indicators.” The slightly elevated hemoglobin levels could be linked to smoking, which has been associated with increased hemoglobin values, or to male sex.
Major Genetic Testing-guided Trial Falls Just Short of 1-Year Event End point
The much-anticipated TAILOR-PCI trial assessing genetic testing to guide antiplatelet therapy after percutaneous cardiovascular intervention failed to meet its primary end point of a 50% reduction in adverse cardiovascular events at 1 year. However, the largest trial to explore the clinical utility of detecting CYP2C19 *2/*3 loss of function allele carriers prior to starting antiplatelet therapy showed a 34% reduction in a composite of major cardiovascular events at year 1. TAILOR-PCI also found a statistically significant 40% drop in the total number of events per patient who received genetically guided treatment compared with those who received standard therapy. These outcomes were presented at the virtual American College of Cardiology/World Congress of Cardiology meeting (20-LB-20309-ACC).
“Although these results fell short of the effect size that we predicted, they nevertheless provide a signal that offers support for the benefit of genetically guided therapy,” said co-principal investigator Naveen Pereira, MD, professor of medicine at the Mayo Clinic in Rochester, Minnesota.
In a post hoc analysis, the researchers found a nearly 80% reduction in the rate of adverse events in the first 3 months of treatment in participants who received genetically guided care versus those who received standard care.
Subjects were randomized to receive either standard care—75 mg daily of clopidogrel—or genetic testing-guided care. Those who were determined through genetic testing to be CYP2C19 *2/*3 carriers (35%) received 90 mg of ticagrelor twice daily; otherwise, participants in the genetic testing arm of the trial received clopidogrel. There were 1.6% major or minor bleeding events at the end of 1 year in participants in the standard care arm and 1.9% in carriers in the guided-treatment group.
Little Concordance AMONG Noninvasive Methods for Identifying NASH
Three noninvasive methods for identifying patients with nonalcoholic steatohepatitis (NASH) agree in only 18% of cases, under-scoring the need for better noninvasive means of recognizing this condition, according to an abstract accepted for the Endocrine Society’s annual meeting (SUN-606).
The investigators used data from the National Health and Nutrition Examination Survey III (NHANES III) to compare three noninvasive methods of identifying NASH: the NASH liver fat score, the HAIR score, and the Gholam score.
The HAIR score incorporates the presence of hypertension, alanine transaminase (ALT) levels, and insulin resistance. The NASH liver fat score is based on the presence of metabolic syndrome, type 2 diabetes, and levels of serum insulin, ALT, and aspartate aminotransferase (AST), while the Gholam score uses AST and a diagnosis of type 2 diabetes.
The investigators identified NHANES III participants who had moderate to severe hepatic steatosis, as determined by ultrasound. In all 1,236 subjects were determined to have NASH by at least one noninvasive method, but the three methods all identified NASH in just 18% of cases. Two methods agreed in 20% of cases.
The three methods all identified significant risk factors for NASH as being overweight or obese, having elevated AST or ALT levels, and having raised C-peptide, serum glucose, or serum triglycer-ide levels. However, the methods disagreed on the significance of other risk factors.