An explainable machine learning algorithm for risk factor analysis of in-hospital mortality in sepsis survivors with ICU readmission

BACKGROUND AND OBJECTIVE Patients who survive sepsis in the intensive care unit (ICU) (sepsis survivors) have an increased risk of long-term mortality and ICU readmission. We aim to identify the risk factors for in-hospital mortality in sepsis survivors with later ICU readmission and visualize the quantitative relationship between the individual risk factors and mortality by applying machine learning (ML) algorithm. METHODS Data were obtained from the Medical Information Mart for Intensive Care III (MIMIC-III) database for sepsis and non-sepsis ICU survivors who were later readmitted to the ICU. The data on the first day of ICU readmission and the in-hospital mortality was combined for the ML algorithm modeling and the SHapley Additive exPlanations (SHAP) value of the correlation between the risk factors and the outcome. RESULTS Among the 2970 enrolled patients, in-hospital mortality during ICU readmission was significantly higher in sepsis survivors (n = 2228) than nonsepsis survivors (n = 742) (50.4% versus 30.7%, P<0.001). The ML algorithm identified 18 features that were associated with a risk of mortality in these groups; among these, BUN, age, weight, and minimum heart rate were shared by both groups, and the remaining mean systolic pressure, urine output, albumin, platelets, lactate, activated partial thromboplastin time (APTT), potassium, pCO2, pO2, respiration rate, Glasgow Coma Scale (GCS) score for eye-opening, anion gap, sex and temperature were specific to previous sepsis survivors. The ML algorithm also calculated the quantitative contribution and noteworthy threshold of each factor to the risk of mortality in sepsis survivors. CONCLUSION 14 specific parameters with corresponding thresholds were found to be associated with the in-hospital mortality of sepsis survivors during the ICU readmission. The construction of advanced ML techniques could support the analysis and development of predictive models that can be used to support the decisions and treatment strategies made in a clinical setting in critical care patients.

[1]  Jürgen Bajorath,et al.  Interpretation of Compound Activity Predictions from Complex Machine Learning Models Using Local Approximations and Shapley Values. , 2020, Journal of medicinal chemistry.

[2]  Li-feng Li,et al.  Early goal-directed and lactate-guided therapy in adult patients with severe sepsis and septic shock: a meta-analysis of randomized controlled trials , 2018, Journal of Translational Medicine.

[3]  Peter Szolovits,et al.  MIMIC-III, a freely accessible critical care database , 2016, Scientific Data.

[4]  K. Ho,et al.  Machine learning for the prediction of volume responsiveness in patients with oliguric acute kidney injury in critical care , 2019, Critical Care.

[5]  Mical Paul,et al.  Early diagnosis of bloodstream infections in the intensive care unit using machine-learning algorithms , 2020, Intensive Care Medicine.

[6]  Scott M. Lundberg,et al.  Explainable machine-learning predictions for the prevention of hypoxaemia during surgery , 2018, Nature Biomedical Engineering.

[7]  K. Langa,et al.  Long-term cognitive impairment and functional disability among survivors of severe sepsis. , 2010, JAMA.

[8]  T. van der Poll,et al.  Incidence, Risk Factors, and Attributable Mortality of Secondary Infections in the Intensive Care Unit After Admission for Sepsis. , 2016, JAMA.

[9]  D. Harrison,et al.  Relationship Between Peak Lactate and Patient Outcome Following High-Risk Gastrointestinal Surgery: Influence of the Nature of Their Surgery Elective Versus Emergency , 2016, Critical care medicine.

[10]  Shamim Nemati,et al.  An Interpretable Machine Learning Model for Accurate Prediction of Sepsis in the ICU , 2017, Critical care medicine.

[11]  C. Horkan,et al.  Functional Status in ICU Survivors and Out of Hospital Outcomes: A Cohort Study , 2016, Critical care medicine.

[12]  Joel Dunning,et al.  Statistical primer: sample size and power calculations—why, when and how? , 2018, European journal of cardio-thoracic surgery : official journal of the European Association for Cardio-thoracic Surgery.

[13]  Jie Xu,et al.  The practical implementation of artificial intelligence technologies in medicine , 2019, Nature Medicine.

[14]  Bin Huang,et al.  Opening the black box of neural networks: methods for interpreting neural network models in clinical applications. , 2018, Annals of translational medicine.

[15]  M. Shankar-Hari,et al.  Risk Factors at Index Hospitalization Associated With Longer-term Mortality in Adult Sepsis Survivors , 2019, JAMA network open.

[16]  Rafael Viégas Cremonese,et al.  Neutrophil-lymphocyte ratio in the early diagnosis of sepsis in an intensive care unit: a case-control study , 2019, Revista Brasileira de terapia intensiva.

[17]  R. Bellomo,et al.  Mortality related to severe sepsis and septic shock among critically ill patients in Australia and New Zealand, 2000-2012. , 2014, JAMA.

[18]  P. Pronovost,et al.  Long-term mortality and quality of life in sepsis: A systematic review* , 2010, Critical care medicine.

[19]  A. Mebazaa,et al.  Protracted immune disorders at one year after ICU discharge in patients with septic shock , 2018, Critical Care.

[20]  M. Shankar-Hari,et al.  Epidemiology of sepsis and septic shock in critical care units: comparison between sepsis-2 and sepsis-3 populations using a national critical care database , 2017, British journal of anaesthesia.

[21]  Aldo A. Faisal,et al.  The Artificial Intelligence Clinician learns optimal treatment strategies for sepsis in intensive care , 2018, Nature Medicine.

[22]  Systolic blood pressure variability in patients with early severe sepsis or septic shock: a prospective cohort study , 2017, BMC Anesthesiology.