Evaluating the Accuracy of Multiple Blood Tests to Diagnose Sepsis in Adult Burn Patients

Background and Rationale Sepsis is a leading cause of mortality in patients with severe burns, yet its early diagnosis is challenging. The systemic inflammatory response triggered by the burn injury itself often mimics the clinical signs of sepsis, confounding diagnosis and leading to delayed or unnecessary antibiotic use. Conventional biomarkers like C-reactive protein (CRP) and procalcitonin (PCT) lack specificity in this population, as they are elevated by the sterile inflammation from the burn itself. This study aims to overcome these limitations by evaluating the diagnostic utility of a multi-biomarker panel, analyzing their dynamic changes over time, and leveraging machine learning to develop a more accurate and interpretable diagnostic model for sepsis in burn patients.

Study Design and Objectives This is a single-center, prospective observational cohort study conducted at the Burn Intensive Care Unit (BICU) of Hallym University Hangang Sacred Heart Hospital. The study will run from September 2025 to December 2027.

Primary Objective: To evaluate the diagnostic accuracy (sensitivity, specificity, AUC) of a multi-biomarker panel-comprising Serum Amyloid A (SAA), Procalcitonin (PCT), Presepsin (PSP), and CRP-for diagnosing sepsis in adult burn patients, using the Sepsis-3 criteria as the reference standard.

Secondary Objectives: To compare the diagnostic performance of the multi-biomarker panel against individual biomarkers and standard clinical severity scores (e.g., SOFA score).

To assess the panel's ability to predict disease severity and mortality (28-day and 90-day).

To develop a predictive model for sepsis using machine learning and interpret the model's decisions using SHAP (SHapley Additive exPlanations) analysis.

Methods and Analysis Data and Sample Collection: For enrolled patients, comprehensive clinical data-including demographics, burn characteristics (%TBSA, inhalation injury), severity scores (SOFA, APACHE II), and microbiological results-will be collected. Blood samples for biomarker analysis will be collected upon ICU admission (baseline), at the time of clinical suspicion of sepsis, and serially thereafter (daily if possible) for a minimum of seven days to capture the temporal dynamics of the biomarkers.

Biomarker Panel: The core panel includes SAA, PCT, PSP, and CRP, supplemented by routine hematological markers like WBC, platelet count, NLR, and PLR.

Statistical and Advanced Analytical Plan:

Diagnostic Accuracy: Traditional statistical methods, including ROC curve analysis and DeLong's test, will be used to compare the diagnostic performance of individual and combined biomarkers.

Dynamic Analysis: To analyze the longitudinal biomarker data, generalized estimating equations (GEE) or linear mixed models (LMM) will be employed. This will allow for the quantitative assessment of biomarker trends (e.g., rate of change) and their association with sepsis diagnosis and prognosis.

Machine Learning and Interpretation: A key component of this study is the development of advanced predictive models using machine learning algorithms (e.g., Random Forest, XGBoost). To address the "black box" nature of these models, we will utilize SHAP analysis to provide transparent interpretations. This will quantify the precise contribution of each biomarker to individual patient predictions, offering deeper clinical insights into the pathophysiological state.

Sample Size: A total of 165 patients will be recruited. This sample size was calculated using G*Power software to achieve 80% power at a significance level of 0.05, assuming an expected sensitivity of 90% and specificity of 80% for the multi-biomarker panel in ROC analysis.