Detecting sepsis in the ICU using machine learning
Kadamangudi Sivakumar, Arjun (2021-06-09)
Detecting sepsis in the ICU using machine learning
(09.06.2021)
Julkaisu on tekijänoikeussäännösten alainen. Teosta voi lukea ja tulostaa henkilökohtaista käyttöä varten. Käyttö kaupallisiin tarkoituksiin on kielletty.
suljettu
Julkaisun pysyvä osoite on:
https://urn.fi/URN:NBN:fi-fe2021070240992
https://urn.fi/URN:NBN:fi-fe2021070240992
Tiivistelmä
Sepsis is a complex syndrome that has been reportedly causing 1 among 5 deaths worldwide. Presently, sepsis can be defined as, “a life-threatening organ dysfunction caused by dysregulated host response to infection”. The definitions of sepsis have been varying over the years and there is no particular gold standard definition as the guidelines are being constantly revised. However, there is a need for appropriate treatment on the diagnosis of sepsis since it can cause organ failure and eventually lead to death.
Culturing the blood samples of the patient to check for the presence of an infection that could lead to sepsis is the most accurate method to diagnose sepsis however, blood culturing takes a minimum of 24 hours to detect the presence of an infection and would take additional days to identify the type of micro-organism for treatment. Hence, few scoring strategies to screen for sepsis such as SIRS and SOFA have been devised. But owing to the complexity of the syndrome, a combination of various data points from the laboratory, medication, vital signs and, patient demographics etc. will be needed to effectively detecting sepsis. Hence, a machine learning model is trained to detect the presence of sepsis in a patient more reliably and in a faster manner than existing conventional rule-based scoring methods.
Based on the availability of critical care data of patients, a machine learning classifier that could combine all the information obtained in a user-specified time window, process the obtained data, and detect the presence of sepsis is being built. The efficiency of the model and reliability in earlier detection is compared with the conventional SOFA scoring method. It could be observed that the gradient boosting model has an AUC score of 0.82 with a 6 hour time window on an unseen patient population, which is higher than the AUC score of 0.74 obtained by SOFA scoring to detect sepsis. By looking for hidden patterns in the data, the classifier would be able to advise on the presence of sepsis that a human eye cannot visualize. This classifier would enable earlier treatment of patients who are found to be septic, thereby reducing the number of deaths caused due to sepsis. This could also directly help in reducing the costs incurred, since non-septic patients if found to be stable could be discharged from the hospital, as ICU costs are the highest hospital costs incurred owing to continuous surveillance of patients.
Culturing the blood samples of the patient to check for the presence of an infection that could lead to sepsis is the most accurate method to diagnose sepsis however, blood culturing takes a minimum of 24 hours to detect the presence of an infection and would take additional days to identify the type of micro-organism for treatment. Hence, few scoring strategies to screen for sepsis such as SIRS and SOFA have been devised. But owing to the complexity of the syndrome, a combination of various data points from the laboratory, medication, vital signs and, patient demographics etc. will be needed to effectively detecting sepsis. Hence, a machine learning model is trained to detect the presence of sepsis in a patient more reliably and in a faster manner than existing conventional rule-based scoring methods.
Based on the availability of critical care data of patients, a machine learning classifier that could combine all the information obtained in a user-specified time window, process the obtained data, and detect the presence of sepsis is being built. The efficiency of the model and reliability in earlier detection is compared with the conventional SOFA scoring method. It could be observed that the gradient boosting model has an AUC score of 0.82 with a 6 hour time window on an unseen patient population, which is higher than the AUC score of 0.74 obtained by SOFA scoring to detect sepsis. By looking for hidden patterns in the data, the classifier would be able to advise on the presence of sepsis that a human eye cannot visualize. This classifier would enable earlier treatment of patients who are found to be septic, thereby reducing the number of deaths caused due to sepsis. This could also directly help in reducing the costs incurred, since non-septic patients if found to be stable could be discharged from the hospital, as ICU costs are the highest hospital costs incurred owing to continuous surveillance of patients.