Table 1.
Baseline characteristics of subjects in each of the three cohorts.
Fig 1.
Representative scatter plots of cell size versus aspect ratio and visco-elastic inertial focusing (VEIR) from a healthy donor, a subject adjudicated as non-septic SIRS 2+ from the low acuity cohort, and a subject adjudicated as septic from the high acuity cohort (S2 Dataset).
As a visual aid, cells with high AR and high VEIR are hued in red. For the purposes of this plot only, the cutoff for “high AR” was arbitrarily chosen as all the points defined by AR ≥ 0.3 + 2.5 x (Cell Size), and the cutoff for “high VEIR” was arbitrarily chosen as 9 μm.
Fig 2.
Box plots displaying cell count, cell size, AR, and VEIR from healthy donors (n = 72), subjects with non-septic SIRS 2+ (n = 323), and subjects with sepsis (n = 78), with features stratified by cell type (lymphocytes, neutrophils, and monocytes).
For these experiments, data from subjects presenting to the ED with 2 or more criteria for SIRS who were later adjudicated as non-septic or as septic were drawn from both the high acuity and low acuity cohorts. Mean values were computed for each cell type within a single test run. A p-value < 0.001 was observed using one-way analysis of variance for each feature for each cell type across the three cohorts, except for mean cell size for lymphocytes, where p-value was 0.002. The p-value from a post-hoc unpaired Student’s two-sided t-test comparing two columns at a time are indicated as follows: * p < 0.05, ** p < 0.01, *** p < 0.001, **** p < 10−10. Box-plot elements: center line, median; box limits, upper and lower quartiles; whiskers, 1.5x interquartile range; points, outliers.
Fig 3.
(A, top panel) Histogram of WBC counts in healthy donors (n = 72), subjects with SIRS 2+ (n = 323), and subjects with sepsis (n = 78). (A, bottom panel) Scatter plots of VEIR and Aspect Ratio in neutrophils and monocytes from the subset of healthy donors (n = 62), subjects with SIRS 2+ (n = 134) and subjects with sepsis (n = 13) that have normal WBC counts between 4 x 106 and 12 x 106 cells/mL. The lines on each graph indicate the decision boundary and the margins for a linear Support Vector Machine classification algorithm that is able to separate subjects between septic and non-septic using mean VEIR and mean AR features. (B) Scatter plots of AR and VEIR in neutrophils and monocytes from subjects with normal WBC counts. The lines indicate the linear regression between Aspect Ratio or VEIR and cell count.
Fig 4.
(A) Receiver Operating Characteristic (ROC) curves showing the performance of the ISI as a diagnostic binary classifier to distinguish septic vs. non-septic subjects for the initial training set (n = 238, adjudicated sepsis incidence of 25.2%), the sequestered testing set (n = 69, adjudicated sepsis incidence of 17.4%), and the full training set (n = 307, sepsis incidence of 23.5%). Also shown are logistic distributions of ISI values for (B) the high acuity cohort (n = 307, adjudicated sepsis incidence of 23.5%), (C) the low acuity cohort (n = 94, sepsis incidence of 6.4%), and (D) the cohort of healthy donors (n = 72).
Table 2.
Diagnosis of subjects in the high acuity cohort separated into those with high ISI (Red Band) and those with low ISI (Green Band).
Fig 5.
Trends in severity of illness across ISI interpretation bands for subjects in the high and low acuity cohorts, as measured by (A) SOFA, maximum over three days following ED presentation, (B) APACHE-II, (C) hospital free days for the subpopulation of subjects admitted to the hospital, (D) Admission to the ICU.
Hospital free days were defined as 28 days less the in-hospital length of stay, where in-hospital mortality leads to a value of zero. Box plots: lines in the boxes, medians; the box ends, interquartile ranges (IQR); whiskers, 1.5x IQR; diamonds, outliers. Bar graphs: Bars, percentages; error pars, 95% confidence intervals. p-values were obtained from an unpaired two-sample Welch’s t-test (except for hospital free days, where the Mann–Whitney U was due to the non-normal distribution), with the null hypothesis that the mean of the two samples are equal. p-values reported as * p < 0.05, ** p < 0.01, and *** p < 0.001.
Fig 6.
Kaplan-Meier survival stratified by ISI interpretation band for (A) all evaluable subjects (n = 307) and (B) a subpopulation of subjects adjudicated as infected (n = 102) in the high acuity cohort.
Plots were created with right-censoring, with the assumption that subjects discharged from the ED or hospital survived ≥ 30 days in the absence of evidence to the contrary (e.g., return to the ED, discharge to hospice, or other indication in the electronic health record, which was reviewed after 30 days, that the subject died). Shading depicts 95% confidence intervals.