Combining Biomarkers with EMR Data to Identify Patients in Different Phases of Sepsis

Abstract

Sepsis is a leading cause of death and is the most expensive condition to treat in U.S. hospitals. Despite targeted efforts to automate earlier detection of sepsis, current techniques rely exclusively on using either standard clinical data or novel biomarker measurements. In this study, we apply machine learning techniques to assess the predictive power of combining multiple biomarker measurements from a single blood sample with electronic medical record data (EMR) for the identification of patients in the early to peak phase of sepsis in a large community hospital setting. Combining biomarkers and EMR data achieved an area under the receiver operating characteristic (ROC) curve (AUC) of 0.81, while EMR data alone achieved an AUC of 0.75. Furthermore, a single measurement of six biomarkers (IL-6, nCD64, IL-1ra, PCT, MCP1, and G-CSF) yielded the same predictive power as collecting an additional 16 hours of EMR data(AUC of 0.80), suggesting that the biomarkers may be useful for identifying these patients earlier. Ultimately, supervised learning using a subset of biomarker and EMR data as features may be capable of identifying patients in the early to peak phase of sepsis in a diverse population and may provide a tool for more timely identification and intervention.

Conflict of interest statement

Ishan Taneja, Bobby Reddy Jr. and Samuel Waschpress work for Prenosis Inc.. Jackson Winter worked at Prenosis at the time of submission but has since left and is now a graduate student at the University of Illinois at Urbana-Champaign. All other authors report no competing interests.

Figures

Figure 1

Early to Peak Phase Sepsis Illustration. The red curve represents a hypothetical patient’s severity of sepsis as a function of time, where the apex of the curve corresponds to their worst case state. A patient is considered to be in the early to peak phase of sepsis if they exist on a point on the curve that is not shaded. We marked the boundaries of peak phase sepsis to illustrate the fact that a patient is in the peak phase of sepsis if they are within some tolerance of their worst case state.

Figure 2

Normalized feature coefficients outputted by SVM for clinical adjudication label set. The absolute value of each feature coefficient in SVM corresponds to its relative importance.

Figure 3

SVM w/feature selection performance as a function of time for clinical adjudication label set. (A) ROC curves are displayed for various feature sets. The EMR data that is used is constrained to up to 48 hours before the biomarker measurement and 1 hour after. (B) A plot of the AUC as a function of the number of hours of EMR data used post biomarker measurement.

Figure 4

Normalized feature coefficients outputted by Random Forest for SOFA score label set.

Figure 5

Heatmap for clinical adjudication label set. The x-axis corresponds to which category the patient was adjudicated to be in (see Materials and Methods) and the y-axis corresponds to the rank of the patient according to the probability outputted by SVM. For each patient, a line is plotted. The x coordinate of the line corresponds to which category the patient is labelled to be in and whose y coordinate corresponds to the rank of the patient is plotted. The color of this line is based on the probability that the patient is in early or peak phase according to SVM. The mapping from probability to color is displayed at the right of the figure. The vertical dotted white line separates the septic (categories 2–5) from the non-septic patients (categories 1, 6–11). The horizontal dotted line represents the patient whose probability of having sepsis was 0.50 according to SVM. The upper left quadrant represents the false positives, the upper right quadrant represents the true positives, the lower left quadrant represents the true negatives, and the lower right quadrant represents the false negatives. The black background corresponds to empty entries in the heatmap.

Figure 6

Feature coefficient outputted by SVM as a function of category.