Abstract
Sepsis must be diagnosed quickly to avoid morbidity and mortality. However, the clinical manifestations of sepsis are highly variable and emergency department (ED) clinicians often must make rapid, impactful decisions before laboratory results are known. We previously developed a technique that allows the measurement of the biophysical properties of white blood cells as they are stretched through a microfluidic channel. In this study we describe and validate the resultant output as a model and score-the IntelliSep Index (ISI)-that aids in the diagnosis of sepsis in patients with suspected or confirmed infection from a single blood draw performed at the time of ED presentation. By applying this technique to a high acuity cohort with a 23.5% sepsis incidence (n = 307), we defined specific metrics-the aspect ratio and visco-elastic inertial response-that are more sensitive than cell size or cell count in predicting disease severity. The final model was trained and cross-validated on the high acuity cohort, and the performance and generalizability of the model was evaluated on a separate low acuity cohort with a 6.4% sepsis incidence (n = 94) and healthy donors (n = 72). For easier clinical interpretation, the ISI is divided into three interpretation bands of Green, Yellow, and Red that correspond to increasing disease severity. The ISI agreed with the diagnosis established by retrospective physician adjudication, and accurately identified subjects with severe illness as measured by SOFA, APACHE-II, hospital-free days, and intensive care unit admission. Measured using routinely collected blood samples, with a short run-time and no requirement for patient or laboratory information, the ISI is well suited to aid ED clinicians in rapidly diagnosing sepsis.