In a recent study published in Scientific Reports, researchers evaluated the additional predictive value of molecular immune markers with the Predict Sepsis screening tools developed previously by the same research team.
Sepsis, a life-threatening organ dysfunction caused by a dysregulated host response to infection, is a vital condition that contributes significantly to global mortality. Early treatment is crucial for reducing mortality and improving outcomes. However, current screening tools for sepsis identification in the prehospital setting are inadequate.
The authors of a study developed three ‘Predict Sepsis’ screening techniques based on symptoms and vital signs, but these tools are not suitable for use in ambulances. Additionally, they do not include parameters reflecting the underlying pathophysiology, which is essential for early identification and treatment in the prehospital setting.
About the study
In the present prospective cohort study, researchers evaluated the additional estimative value of molecular immune markers compared to the priorly developed pre-hospital screening methods.
The study included 551 adult non-trauma ambulance patients suspected of infections judged by ambulance personnel based on clinical experience and presenting symptoms. The study was conducted from 2017 to 2018, as a subset of the Predict Sepsis study, in Stockholm.
Initially, 15 genes [caspase 1 (CASP1), PYD and CARD domain-containing (PYCARD), interleukin-1B (IL1B), NLR family pyrin domain containing 3 (NLRP3), IL18, IL6, tumor necrosis factor (TNF), IL1RN, IL10, hypoxia-inducible factor 1-alpha (HIF1A), major histocompatibility complex, class II, DR alpha (HLA-DRA), endothelial PAS domain protein 1 (EPAS1), spi-1 proto-oncogene (SPI1), NF-kappa-B inhibitor alpha (NFKBIA), and sirtuin 1 (SIRT1)] and 74 inflammatory proteins were analyzed among 96 screening cohort individuals, among whom 46 had sepsis.
The screening cohort underwent univariate analysis, multivariate analysis, machine learning, and a literature review.
Subsequently, 12 molecules, including nine mediators [C-C motif chemokine ligand 11 (CCL-11), CCL-19, CCL-24, CCL-27, IL-17A, IL-17AF, TNF, interleukin-1 receptor antagonist protein (IL-1Ra), and C-X3-C motif ligand 1 (CX3CL1)] and three genes [NLRP3, HIF1A, and EPAS1], as probably synergistic estimators, were analyzed combined with priorly devised screening techniques using clinical variables in the prediction cohort, comprising 455 individuals, among whom 271 were septic and 184 were non-septic.
For machine learning-based model prediction, seven alternative methods with layered cross-validation were applied. Posterior distributions for the mean area under the receiver operating characteristic curve (AUC) values and discrepancies in AUCs were analyzed to compare model performances.
Variable value permutations, scoring losses of categorization as a measure, and weights specific to the models were used to assess model variable relevance.
Blood samples were obtained in the ambulance to quantify the levels of circulating immunological mediators by electroluminescence and extract ribonucleic acid (RNA) and complementary deoxyribonucleic acid (cDNA) for determining gene expression by real-time polymerase chain reaction (PCR). Sample purity and concentration were assessed by spectrophotometry.
Seventy-five percent of the data obtained was used for model training, and the remaining 25% was used for testing. The PubMed database was used for evaluating the literature on the molecular variables with the highest permutation importance. Only original research conducted on human participants and published in English within the previous 10 years was included.
Results and discussion
Comparing the predictive value of screening methods with as well as without new molecular indicators and their associations showed that the molecular marker inclusion had no effect. This might be attributed to the substantial multicollinearity between the immune mediators and clinical measures, most notably for temperature, which seemed to capture the majority of the informative variance of numerous molecular markers, including IL-1Ra.
Prediction models using only the molecular-type variables showed AUC values ranging from 0.65 to 0.70. Among the immunological molecular markers, IL-1Ra and TNF were the most critical predictors. In addition, serum levels of CCL19, IL-17A, TNF, and CX3CL1 were statistically significantly higher among septic individuals compared to non-sepsis individuals. According to the authors, neither CCL19 nor CX3CL1 have been evaluated previously as sepsis predictors.
IL-17AF expression was low in septic as well as non-septic individuals, whereas IL-17AF expression was high among those with sepsis included for screening in comparison to those without sepsis, while it was found to decrease among septic individuals of the prediction group. The disparity in serological IL-17AF levels among individuals included for model screening and predictions may have resulted from heterogeneity in the groups.
In total, 26 individuals were misclassified by the models, indicating that septic individuals must be stratified based on their clinical phenotypes or characteristics to facilitate early diagnosis. Individuals misclassified as septic demonstrated higher Glasgow Coma Scale (GCS), higher systolic blood pressure values, milder fever, and lower IL-17A and IL-1Ra levels.
Overall, the study findings showed that combining molecular biomarkers with clinical variables did not improve predictive values, likely because of the high multicollinearity observed between the immunological mediators and body temperature.
The findings indicated a need for classifying septic individuals with prior knowledge of molecular and clinical parameters to diagnose sepsis early and therefore facilitate prompt treatment.