A Deep Learning-Based Sepsis Estimation Scheme

摘要

The objective of this research is to design and implement a machine learning (ML) based technique that can predict cases of septic shock and extreme sepsis and assess its effects on medical practice and the patients. The study is a retrospective cohort type, which is used to algorithmic deduction and validation, along with pre- and post-impact assessment. For non-ICU cases, the algorithm was deduced and validated for specific periods. The classifiers used for the study have been deduced and validated by employing electronic health records (EHR), which were silent initially but alerted the clinical personnel concerning the sepsis prediction. For training the classification system, the chosen patients should have had ICD and the latest codes concerning extreme sepsis or septic shock. Moreover, the patients should have had positive blood culture during their interaction with the hospital, where there were indications of either systolic blood pressure (SBP) or lactate levels. The classification algorithms demonstrated a 93.84%, 93.22%, 95.25% accuracy, sensitivity and specificity respectively. The pattern used for clinical detection, in the context of the alerting system, led to a small but statistically significant increase in IV usage and lab tests. The values used for the alerting system were found to have no statistically significant difference in the context of different ICU wards since data from the laboratory tests serve as the primary early indicator of septic shock by confirming the presence of toxins.