University of Pittsburgh (UPMC) researchers deployed computer algorithms that mined the electronic health records (EHRs) of 63,858 patients to derive four phenotypes of sepsis, marked by demographics, lab values, and outcomes (JAMA 2019;321:2003-17). While acknowledging that more research is needed about these sepsis types, the investigators also suggested that their findings might lead to better treatment for this life-threatening condition.
“Hopefully, by seeing sepsis as several distinct conditions with varying clinical characteristics, we can discover and test therapies precisely tailored to the type of sepsis each patient has,” said first author Christopher Seymour, MD, MSc, an associate professor of medicine at UPMC.
Based on statistical machine learning and simulation tools, the algorithms analyzed 29 clinical variables in patient EHRs to identify the four phenotypes:
- Alpha: The most common type (33%), with the fewest abnormal lab values, least organ dysfunction, and lowest in-hospital death rate (2%);
- Beta: Patients in this type (27%) were typically older and had the most chronic illnesses and kidney dysfunction;
- Gamma: These patients (27%) had elevated measures of inflammation, mostly pulmonary dysfunction, and the second-highest in-hospital death rate (15%);
- Delta: These patients (13%) typically were the sickest, often with liver dysfunction and shock. 85% were admitted to intensive care, and 32% died in hospital.
The authors developed and validated the algorithm and their findings in three patient groups and assessed reproducibility, correlation with biological parameters, and clinical outcomes in one patient group and in other recently completed international clinical trials involving sepsis care. The patient groups included 20,000 UPMC patients recognized to have sepsis within 6 hours of hospital arrival from 2010-2012, 43,000 UPMC sepsis patients from 2013-2014, and 583 patients at 28 U.S. hospitals who developed sepsis due to pneumonia.
The researchers used simulation models to apply the four sepsis types to three randomized clinical trials—ACCESS, PROWESS, and ProCESS—and found that doing so would have changed the overall picture of the trials. For example, early goal-directed therapy—a hallmark of sepsis care—benefited Alpha type patients but worsened outcomes for Delta type patients.
Direct Platelet Autoantibody Testing Bests Indirect in Ruling-in Immune Thrombocytopenia
Asystematic review and meta-analysis of platelet autoantibody testing in immune thrombocytopenia (ITP) concluded that this testing has high sensitivity but low specificity in assessing patients for ITP but that using an optical density (OD) >3 standard deviations (SD) above normal improved sensitivity without compromising specificity (J Thromb Haemost 2019;17:787-94). Based on the latter finding, the authors suggest that OD >3 SD above normal should be established as a threshold to improve standardization of these assays across laboratories.
While guidelines state that autoantibody testing is not useful for diagnosing ITP, the authors also suggest, based on their findings, that it helps rule-in ITP.
In a literature review the investigators included 18 studies involving 1,170 ITP patients that met the criteria of having at least 20 ITP patients, using direct testing to measure autoantibodies against glycoprotein (GP) IIbIIa or GP IbIX of the IgG isotype bound to the platelet surface or indirect testing detecting GP-specific platelet autoantibodies in plasma or serum, as long as the latter also reported direct assay results.
The authors’ pooled estimates for sensitivity and specificity of direct testing and indirect testing were 53% and 93%, and 18% and 96%, respectively. In studies that used a cutoff of OD >3 SD as the threshold for a positive test, the pooled sensitivity and specificity for direct and indirect testing were 58% and 94%, and 21% and 96%, respectively.
In the subset of six studies that used both direct and indirect testing methods, the authors found that direct assays are more sensitive than indirect tests, consistent with a 2012 report that recommended direct assays over indirect ones on account of their improved sensitivity.
Trial Design Described for Assessing Analytical, Clinical Performance of High-sensitivity Cardiac Troponin Assays
Researchers have described a trial design for assessing the analytical and clinical performance of high-sensitivity cardiac troponin (cTn) I assays in the U.S., including the Siemens Healthineers’ Atellica TnIH, ADVIA Centaur TNIH, Dimension EXL 200 THNIH, and Dimension Vista 1500 TNIH systems (Contemp Clin Trials Commun 2019;14:100337). They did so based on Food and Drug Administration requirements for patient enrollment and in accordance with at least five Clinical and Laboratory Standards Institute guidance documents on different aspects of assessing a test’s performance.
The investigators determined the assays’ 99th percentile upper reference limits in a healthy population by recruiting patients at least 22 years old from 12 sites across the U.S., excluding those with various lifestyle or health risks.
They assessed the assays’ clinical performance in emergency departments (EDs) by enrolling subjects at least 22 years old at 29 sites who presented as patients to EDs with symptoms of possible acute myocardial infarction (AMI) and who were enrolled within 1.5 hours of their first clinical blood draw. Study adjudicators classified subjects for AMI status based on the Third Universal Definition of Myocardial Infarction.
The researchers considered a battery of analytical performance factors for each assay, such as detection capability; limit of blank, detection, and quantitation; analytical measurement range; and high dose hook effect.
In assessing the assays’ clinical performance, the researchers tested the hypotheses that each assay had a sensitivity for AMI ≥90% or <90%. They also calculated four measures of diagnostic performance, including sensitivity, specificity, positive predictive value, and negative predictive value.