Clinic-first sepsis recognition in the ICU: a proteomics-guided, parsimonious model with independent validation

Comprehensive Summary

This article aims to define key variables that contribute to potential sepsis infections and integrate this data into a learning model in hopes to alert hospital personnel of these patients. To do this, the authors looked at 55 patients (38 with sepsis and 17 critically ill but non-septic) within 48 hours of critical illness onset and looked at specific plasma proteins through proteomics. The data used includes lab measurements, infection characteristics, and other clinical data and scores. The blood was obtained within 4 hours of enrollment, and the first batch was used to analyze inflammatory markers while the second batch was used to interpret specific proteins. The protein expression was mapped using a heat map and the 12 key septic inflammatory markers can clearly be seen and identified (such as LILRA3, LCN2, and B2M). The final results indicated that sepsis can be predicted using proteins along with other key data such as creatinine. The area under the curve was found to be .73 from the machine learning model, where .5 is random chance, which indicates this learning model was clinically useful in determining patients who are at a high risk for sepsis. There are many markers that have been known to be related to sepsis, however, new inflammatory markers, such as IGFBP6, have yet to be fully explored and their mechanism of action in precipitating sepsis. Overall, using inflammatory markers and other clinical data can yield scores high enough to predict patients who are at a high risk of going into septic shock which can aid in giving physicians and healthcare teams a forewarning about potential life-threatening complications.

Outcomes and Implications

Using the machine learning model and key data points from patients in the ICU, the risk of sepsis can be determined and forewarn the healthcare team to put them on high alert. These novel inflammatory markers used to predict potential sepsis are a new step in the potential long list of warnings and preventative measurements that can come from laboratory measurements, and through the use of these markers, there is the potential to avoid numerous deaths or serious injuries that occur in the hospital. This type of learning model has potential to be incorporated into existing hospital systems, although the use of proteomics may be difficult to make widespread, especially with the data, the proteins, that the model needs to determine which patients are at high risk. However, even if only the high-risk patients are screened and their data used, this can still successfully avoid many cases of sepsis per year in the hospital.