Boosted Tree Ensembles for Predicting Postsurgical ICU Mortality Garrick Aden-Buie, Yun Chen, Rashad Kayal, Gina Romero, Hui Yang Dept. of Industrial and Management Sciences Engineering College of Engineering University of South Florida, T ampa, FL INFORMS Annual Meeting 2013, Minneapolis, MN
Motivation MIMIC II Clinical Data Methods Results Outline Motivation MIMIC II Clinical Data Methods Results G. Aden-Buie Boosted Tree Ensembles for Predicting Postsurgical ICU Mortality 1
Motivation MIMIC II Clinical Data Methods Results Outline Motivation MIMIC II Clinical Data Methods Results G. Aden-Buie Boosted Tree Ensembles for Predicting Postsurgical ICU Mortality 1
Motivation MIMIC II Clinical Data Methods Results Trends in Critical Care in US ◮ Critical care beds increased by 6.5% (2000-2005) ◮ Despite 12.2% decrease in hospitals with critical care and 4.2% reduction overall in hospital beds ◮ Constrained ICU capacity ◮ High quality care: safe, effective, equitable patient-centered, timely and efficient (IOM) Halpern, Neil A, and Stephen M Pastores, 2010 “Critical Care Medicine in the United States 2000-2005” G. Aden-Buie Boosted Tree Ensembles for Predicting Postsurgical ICU Mortality 2
Motivation MIMIC II Clinical Data Methods Results Acuity Scores in ICUs ◮ Existing acuity scores ◮ APACHE ◮ SAPS ◮ MPM ◮ SOFA ◮ Aim to compensate for population differences to objectively compare practices across ICUs ◮ Need for patient-specific prognostic models G. Aden-Buie Boosted Tree Ensembles for Predicting Postsurgical ICU Mortality 3
Motivation MIMIC II Clinical Data Methods Results Objective ◮ T o develop a data-driven, patient-specific prognostic model to predict in-hospital death in post-surgical ICU patients. ◮ T o support effective, efficient use of critical care resources G. Aden-Buie Boosted Tree Ensembles for Predicting Postsurgical ICU Mortality 4
Motivation MIMIC II Clinical Data Methods Results Overview ◮ We created and evaluated a gradient boosted trees model using routine patient data recorded during the first 48 hours of an ICU visit. ◮ Uses heterogeneous, routinely-collected data ◮ Requires minimal preprocessing ◮ Effectively addresses sampling and missing information issues ◮ Accurately predicts in-hospital mortality G. Aden-Buie Boosted Tree Ensembles for Predicting Postsurgical ICU Mortality 5
Motivation MIMIC II Clinical Data Methods Results Outline Motivation MIMIC II Clinical Data Methods Results G. Aden-Buie Boosted Tree Ensembles for Predicting Postsurgical ICU Mortality 5
Motivation MIMIC II Clinical Data Methods Results MIMIC II Clinical Data ◮ Physiologic signals and vital signs from patient monitoring and hospital information systems ◮ PhysioNet Computing in Cardiology 2012 Challenge ◮ 12,000 patients divided into 3 sets of 4,000 ◮ Set A: Training ◮ Set B: Validation ◮ Set C: T esting ◮ Inclusion criteria ◮ Age ≥ 16 years ◮ Initial ICU stay ≥ 48hrs http://www.physionet.org/challenge/2012/ G. Aden-Buie Boosted Tree Ensembles for Predicting Postsurgical ICU Mortality 6
Motivation MIMIC II Clinical Data Methods Results MIMIC II Clinical Data ◮ Physiologic signals and vital signs from patient monitoring and hospital information systems ◮ PhysioNet Computing in Cardiology 2012 Challenge ◮ 12,000 patients divided into 3 sets of 4,000 ◮ Set A: Training ◮ Set B: T esting ◮ Inclusion criteria ◮ Age ≥ 16 years ◮ Initial ICU stay ≥ 48hrs http://www.physionet.org/challenge/2012/ G. Aden-Buie Boosted Tree Ensembles for Predicting Postsurgical ICU Mortality 6
Motivation MIMIC II Clinical Data Methods Results Input Variables ◮ Up to 41 variables recorded per patient ◮ 5 general descriptors ◮ 36 time series variables G. Aden-Buie Boosted Tree Ensembles for Predicting Postsurgical ICU Mortality 7
Motivation MIMIC II Clinical Data Methods Results General Descriptors Variable Mean S.D. Age 64.5 yrs 17.1 Height 169.5 cm 17.1 Weight 81.2 kg 23.8 Gender Male : 56.1% Female : 43.8% ICU T ype Medical : 35.8% Surgical : 28.4% Cardiac surgery : 21.1% Coronary : 21.1% In-Hospital Death 13.85% G. Aden-Buie Boosted Tree Ensembles for Predicting Postsurgical ICU Mortality 8
Motivation MIMIC II Clinical Data Methods Results Time Series Variables 36 variables describing ◮ Arterial Blood Gasses ◮ Overall Condition ◮ Cardiac Biomarkers ◮ Renal Function ◮ Blood Count ◮ Serum Electrolytes ◮ Consciousness ◮ Ventilation Support ◮ Hepatic Function ◮ Vital Signs G. Aden-Buie Boosted Tree Ensembles for Predicting Postsurgical ICU Mortality 9
Example Patient: Survived Patient 133659 −− Outcome: 0 Female Age: 46 Weight: 220lbs Height: 5' 10" BMI: 31.63 kg/m2 ICUType: 1:Coronary Care Albumin ALP ALT AST Bilirubin Cholesterol 4.50 58.50 59.50 120.50 1.00 −0.50 4.25 58.25 59.25 120.25 0.75 −0.75 4.00 58.00 59.00 120.00 0.50 −1.00 ● ● ● ● ● 3.75 57.75 58.75 119.75 0.25 −1.25 3.50 57.50 58.50 119.50 0.00 −1.50 SaO2 TroponinI TroponinT MechVent BUN Creatinine −0.50 −0.50 100 ● ● 14 ● 0.800 ● 3.2 −0.75 −0.75 0.775 12 ● 99 3.0 −1.00 −1.00 0.750 2.8 10 98 0.725 −1.25 −1.25 2.6 ● 8 97 0.700 ● ● ● ● ● −1.50 −1.50 Glucose HCO3 HCT K Lactate Mg −0.50 26 39 4.1 2.1 ● ● ● ● ● ● ● 99 25 4.0 ● 2.0 ● −0.75 38 24 3.9 1.9 ● 96 −1.00 ● ● 23 3.8 1.8 37 ● 93 −1.25 22 ● 3.7 ● 1.7 36 90 ● 21 ● ● 3.6 ● 1.6 ● −1.50 Na PaCO2 PaO2 pH Platelets WBC −0.50 −0.50 −0.50 220 140 ● ● ● 11 139 −0.75 −0.75 −0.75 200 ● 138 10 ● −1.00 −1.00 −1.00 180 137 ● 9 −1.25 −1.25 −1.25 136 8 160 ● 135 ● ● ● −1.50 −1.50 −1.50 DiasABP NIDiasABP MAP NIMAP SysABP NISysABP 90 ● 80 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● 80 ● ● ● ● ● ● ● ● ● 100 ● ● ● ●●● ● ● ●●● 75 ● ● ● ● ● ● ● ● 85 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 100 ● ● 60 ● ● ● 75 ● ● ● 95 ● ● ● ● ● ●● ● ● ● ● ● 80 ● ● ● ● ● ● ● ● ● ● 50 ● ● 70 ● 90 ● ● ● 40 ● ● ● ● ● ● ● ● ● ● ● ● ● ● 75 ● ● ● ● ● 50 ● ● 85 ● 65 ● ● ● ● ● ● ● ● ● ● ● 25 ● ● ● 20 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 70 ● ● ● 80 60 ● ● ● ● ● ● 0 ● ● 0 ● 75 ● 0 ● 55 65 GCS HR Temp Urine.Sum FiO2 Weight 15.50 −0.50 100.50 ● ● ● ● ●●● 80 110 36.75 ● 15.25 ● ● ● ● −0.75 100.25 ● 100 60 ●● ●● ●● 36.50 15.00 ● −1.00 100.00 ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● 90 ● ● ● ● ● 40 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 36.25 14.75 ● ● ● ● ● ● −1.25 99.75 80 ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ● 20 ● ● ● ● ● ● ● ● ●● ● 70 ● 36.00 ● ● 14.50 −1.50 99.50 0 1000 2000 0 1000 2000 0 1000 2000 0 1000 2000 0 1000 2000 0 1000 2000 Time
Recommend
More recommend