SUMMARY

• The predictive value of sepsis clinical scoring systems is controversial but needs further improvement.

• Machine learning methods have recently emerged in the use of predicting disease identification, diagnosis and treatment.

• There is a current lack of comprehensive understanding of machine learning models of predictive variables.

BACKGROUND

• Various clinical scoring systems for sepsis exist in order to help clinicians assess for sepsis and predict poor outcomes, specifically mortality.

• Unfortunately, the predictive value of most of these tools remain unsatisfactory.

• Machine learning has good predictive performance and has been shown useful in various predictions of disease prevention, diagnosis, treatment and prognosis for readmission and death.

• Recently 2 major publications have comprehensively analyzed existing data of machine learning systems in determining the diagnostic accuracy and mortality risk prediction in sepsis.

RESULTS

Early Sepsis Recognition – L.M. Fleuren et al Intensive Care Med (2020) 46:383–400 https://doi.org/10.1007/s00134-019-05872-y

• 24 publications with 130 machine learning models covering ICU, ED and hospital wards.

• Retrospective studies with substantial heterogeneity.

• AUROC was used as the main performance metric.

• Number of features used within the models ranged from 2 to 49.

• Individual machine learning models have a high predictability of sepsis onset over traditional scoring tools.

• Well controlled studies are needed to validate this initial information.

• Risk of death was 24-34% lower with continuous infusion vs intermittent.

Machine Learning for Prediction of Sepsis Related Mortality -Y Zhang et al BMC Medical Informatics and Decision Making. 2023;23:283. Doi.org/10.1186/s12911-023-02383-1.

• 50 studies with 104 machine learning models, 125 modeling variables.

• 1.9 million patients; 270,361 (14.02%) mortality rate, with 247,519 in hospital deaths.

• 38 additional studies on the accuracy of SOFA and qSOFA in predicting short term mortality risk in sepsis patients.

• Machine learning models demonstrated favorable accuracy in predicting sepsis mortality with in the hospital and up to 1 month post hospitalization.

CONCLUSIONS

• An improved sepsis assessment tool would enable clinicians to promptly diagnose and formulate treatment decisions.

• Machine learning models appear to possess improved useability in terms of sepsis identification as well as mortality prediction.

• Selection of appropriate modeling variables is a key factor in improving their predictive accuracy.

To receive articles like these in your Inbox, you can subscribe to Sepsis Program Optimization Insights.

Erkan Hassan is the Co-Founder & Chief Clinical Officer of Sepsis Program Optimization where he designs & oversees the implementation of solutions to optimize sepsis programs.

To discuss your organization’s Barriers of Effective Sepsis Care, contact Erkan by phone (844) 4SEPSIS (844-473-7747), email (erkan@spo.icu), or video chat.