Diagnostic Accuracy of Cerebrospinal Fluid Presepsin vs. Procalcitonin in Post-Neurosurgical Bacterial Ventriculitis/Meningitis: A Machine Learning Analytical Approach.

This study aimed to evaluate the diagnostic accuracy of cerebrospinal fluid presepsin and procalcitonin in patients who had undergone neurosurgery between November 2023 and December 2024 enrolled on the basis of specific guidelines. Cerebrospinal fluid presepsin and procalcitonin levels were evaluated via ELISA. Machine learning models were implemented to assess the diagnostic accuracy. A total of 120 patients were included in the study and categorized into three different groups. Machine learning model: random forest model was implemented for ROC curve analysis and the model had an accuracy of 94.5%. The optimal presepsin cut-off value for discriminating between the infectious and non-infectious group was 1729 pg/ml. The specificity and sensitivity for presepsin was 0.875 and 0.632, respectively, and the AUCs for all groups were greater than 0.800 in the random forest model. The specificity and sensitivity for PCT were 0.458 and 0.789, respectively, and the AUCs for the confirmed and probable groups were 0.810 and 0.800 respectively. The variable importance plot revealed presepsin to be the second most useful parameter in model prediction. The random forest model has good performance in predicting infections among neurosurgical patients. CSF presepsin clearly distinguished the three groups, and the median PCT levels were similar across the three groups. The optimal cut-off for PCT is not suggestive compared with presepsin. CSF presepsin is a better biomarker than CSF PCT in diagnosing post-neurosurgery patients and can be implemented in routine diagnostic procedures.