 Bases de données complexes et nouveaux outils prédictifs: - MIMIC-II - Super ICU Learner Algorithm (SICULA) Project PIRRACCHIO R, Petersen M, Carone.

Slides:



Advertisements
Similar presentations
Grant review at NIH for statistical methodology Jeremy M G Taylor Michelle Dunn Marie Davidian.
Advertisements

Surgery volume and operative mortality: A re-examination using fixed-effects regression Amresh Hanchate, PhD Section of General Internal Medicine Boston.
Donald T. Simeon Caribbean Health Research Council
Milestones from the Past / A Spotlight on the Future Quality Improvement Operations Management Research Randall Wetzel, MD, MBA - Chief Executive Officer,
Principal Component Analysis Based on L1-Norm Maximization Nojun Kwak IEEE Transactions on Pattern Analysis and Machine Intelligence, 2008.
Gall C, Katch A, Rice T, Jeffries HE, Kukuyeva I, and Wetzel RC
Improving the quality of medical and surgical care NCEPOD Dr Marisa Mason.
Simulation of “forwards-backwards” multiple imputation technique in a longitudinal, clinical dataset Catherine Welch 1, Irene Petersen 1, James Carpenter.
Collected from work by The Joint Commission and Abt Associates for CMS/ONC.
MOLEDINA-1 CSE 5810 CSE5810: Intro to Biomedical Informatics The Role of AI in Clinical Decision Support Saahil Moledina University of Connecticut
Informative Censoring Addressing Bias in Effect Estimates Due to Study Drop-out Mark van der Laan and Maya Petersen Division of Biostatistics, University.
Glycemic Control in Acutely Ill Patients Martin J. Abrahamson, MD FACP Associate Professor of Medicine, Harvard Medical School Senior Vice President for.
Model Assessment, Selection and Averaging
Prediction Models in Medicine Clinical Decision Support The Road Ahead Chapter 10.
From last time….. Basic Biostats Topics Summary Statistics –mean, median, mode –standard deviation, standard error Confidence Intervals Hypothesis Tests.
Propensity Score Matching A Primer in R 1 David Zepeda Assistant Professor Supply Chain & Information Management Center for Health Policy.
A Cost-Effective Agent for Clinical Trial Assignment Princeton K. Kokku Lawrence O. Hall Dmitry B. Goldgof Eugene Fink Jeffrey P. Krischer.
Prediction Methods Mark J. van der Laan Division of Biostatistics U.C. Berkeley
Variable Selection for Optimal Decision Making Lacey Gunter University of Michigan Statistics Department Michigan Student Symposium for Interdisciplinary.
Using Regression Models to Analyze Randomized Trials: Asymptotically Valid Tests Despite Incorrect Regression Models Michael Rosenblum, UCSF TAPS Fellow.
Validation of predictive regression models Ewout W. Steyerberg, PhD Clinical epidemiologist Frank E. Harrell, PhD Biostatistician.
TRAUMA SYSTEM Mazen S. Zenati, M.D, MPH, Ph.D. University of Pittsburgh Department of Surgery and Epidemiology.
Diagnostic Indicators of Anxiety and Depression in Older Dizzy Patients in Primary Care J Geriatr Psychiatry Neurol 2011;24(2) Maarsingh OR, 1 Dros.
) Benchmarking Critical Care Outcomes: Using data to drive effectiveness and efficiency Thomas L. Higgins MD MBA Vice Chair for Clinical.
Deltex Medical Group plc AGM 28 April 2011 Ewan Phillips.
Cumulative Sum (CUSUM) charts for medical student peripheral venous cannulation; development of a difficulty-adjusted CUSUM Dr Harry Murgatroyd SpR Anaesthesia.
Development and evaluation of software to support prescribing and drug supply management in the treatment of MDR-TB in Peru. Fraser H, Choi S, Jazayeri.
EUropean Best Information through Regional Outcomes in Diabetes Risk Adjusted Diabetes Indicators Fabrizio Carinci Technical Coordinator The BIRO Academy.
Effect of Hypertension and Dyslipidemia on glycemic control among Type 2 Diabetes patients in Thailand Dr. Mya Thandar Dr.PH. Batch 5 1.
ProteinShop: A Tool for Protein Structure Prediction and Modeling Silvia Crivelli Computational Research Division Lawrence Berkeley National Laboratory.
Time – Immortal Bias in the analysis of “Influenza and COPD Mortality Protection as Pleiotropic, Dose-dependent effects of statins” by Floyd J, Frost et.
April 4 Logistic Regression –Lee Chapter 9 –Cody and Smith 9:F.
Effect of Hypertension and Dyslipidemia on glycemic control among Type 2 Diabetes patients in Thailand Dr. Mya Thandar DrPH Batch 5 1.
RBC transfusions in critically ill patients TMR Journal Club March 1, 2007 Maggie Constantine.
Empirical Efficiency Maximization: Locally Efficient Covariate Adjustment in Randomized Experiments Daniel B. Rubin Joint work with Mark J. van der Laan.
Chapter 11 Statistical Techniques. Data Warehouse and Data Mining Chapter 11 2 Chapter Objectives  Understand when linear regression is an appropriate.
Ilona Verburg Nicolette de Keizer Niels Peek
Wins/Losses and Errors/Ties: Quality of Care for Acute Myocardial Infarction in the VA Health Care System Laura A. Petersen, M.D., M.P.H. 1 Sharon-Lise.
Raghavan Murugan, MD, MS, FRCP Associate Professor of Critical Care Medicine, and Clinical & Translational Science Core Faculty, Center for Critical Care.
Making Every Contact Count Sarah McCormack 20 th October, 2015.
Selecting Employees DeNotra Geddis April 11, 2005.
Super Learning in Prediction HIV Example Mark van der Laan Division of Biostatistics, University of California, Berkeley.
IMPORTANCE OF STATISTICS MR.CHITHRAVEL.V ASST.PROFESSOR ACN.
Rehospitalization Analytics: Modeling and Reducing the Risks of Rehospitalization Chandan K. Reddy Department of Computer Science, Wayne State University.
Achieving Glycemic Control in the Hospital Setting (Part 2 of 4)
Resource Allocation in Hospital Networks Based on Green Cognitive Radios 王冉茵
NTU & MSRA Ming-Feng Tsai
1 Module 1 Introduction: The Role of Gender in Monitoring and Evaluation.
< 회기-강동 합동 컨퍼런스> Systemic Inflammatory Response Syndrome criteria in Defining Severe sepsis Kirsi-Maija Kaukonen, M.D., Ph.D., Michael Bailey, Ph.D.,
Physiological Data Analysis of Neuro-Critical Patients Using Markov Models By Shashwat Bhoop sb3758.
EECS6898 Final Project Mortality Predictions in ICU Yijing Feng yf2375.
Evaluating health outcomes: the experience of a national evaluation programme Luigi Pinnarelli Rome, 15-16/10/2012.
Jason P. Lott, Theodore J. Iwashyna, Jason D. Christie, David A. Asch, Andrew A. Kramer, and Jeremy M. Kahn Am J Respir Crit Care Med Vol 179. pp 676–683,
Kelci J. Miclaus, PhD Advanced Analytics R&D Manager JMP Life Sciences
Bootstrap and Model Validation
Incorporating co-morbidity: understanding inequality to improve the value of targeted public health strategies Authors: David Jeffries & Warren Stevens.
Logistic Regression APKC – STATS AFAC (2016).
Statistical Approaches to Support Device Innovation- FDA View
Adoption of Health Information Exchanges and Physicians’ Referral Patterns: Are they Mutually Reinforcing? SAEEDE EFTEKHARI*, School of Management, State.
AACN Certification A Mark of Excellence.
S1316 analysis details Garnet Anderson Katie Arnold
Public Health Physician, Lecturer Critical Care Medicine,
What is Patient Blood Management?
Postoperative neonatal mortality prediction using superlearning
Model generalization Brief summary of methods
Clinical prediction models
Analysis on Accelerated Learning Cohorts
Sabrina M. Figueiredo1,3, Alicia Rozensveig3, José A. Morais2, Nancy E
Chaoran Hu1,4, Xiao Tan2,4, Qing Pan3, Yong Ma4, Jaejoon Song4
Public Health Implications
Presentation transcript:

 Bases de données complexes et nouveaux outils prédictifs: - MIMIC-II - Super ICU Learner Algorithm (SICULA) Project PIRRACCHIO R, Petersen M, Carone M, Resche Rigon M, Chevret S and van der Laan M Division of Biostatistics, UC Berkeley, USA Département de Biostatistiques et informatique Médicale, UMR-717, Paris, France Service d’Anesthésie-Réanimation, HEGP, Paris

 The Data

Upcoming Medical Data  « Big data »  p >>> n  Génomic, radiomic, …  I2B2 data centers:  Informatics for Integrating Biology & Bedside  Boston: MIT – Harvard

MIMIC-II  Publically available dataset including all patients admitted to an ICU at the Beth Israel Deaconess Medical Center (BIDMC) in Boston, MA :  medical (MICU), trauma-surgical (TSICU), coronary (CCU), cardiac surgery recovery (CSRU) and medico-surgical (MSICU) critical care units.  Data collection started in 2001  Patient recruitment is still ongoing.  Patients charts, beat-by-beat waveform signal, biology, notes …. Lee, Conf Proc IEEE Eng Med Biol Soc 2011 Saeed, Crit Care Med 2011

MIMIC-II  Access to the Clinical Database:  On-line course on protecting human research participants (minimum 3 hours)  For all participants  Basic Access Web interface :  Requires knowledge of SQL  User friendly for databases specialists  Limited size of the data export  Root data export (.txt) (20Go)

 Adapted Prediction Algorithms We need new models for ICU mortality prediction !

Motivations for Mortality Prediction  Improved mortality prediction for ICU patients in remains an important challenge:  Clinical research: stratification/adjustment on patients’ severity  ICU care: adaptation of the level of care/monitoring; choice of the appropriate structure  Health policies: performance indicators

Currently used Scores  SAPS, APACHE, MPM, LODS, SOFA,…  And several updates for each of them  The most widely in practice are:  The SAPS II score in Europe Le Gall, JAMA 1993  The APACHE II score in the US Knauss, Crit Care Med 1985

Currently used Scores  SAPS, APACHE, MPM, LODS, SOFA,…  And several updates for each of them  The most widely in practice are:  The SAPS II score in Europe Le Gall, JAMA 1993  The APACHE II score in the US Knauss, Crit Care Med 1985 PROBLEM: fair discrimination but poor calibration

Why are the current scores performing that bad ?  4 potential reasons for that:  Global decrease of ICU mortality  Covariate selection  Geographical disparities  Parametric Logistic regression => Which means we acknowledge assuming a linear relationship between the outcome and the covariates

Why are the current scores performing that bad ? WHY would we accept that ???  We have alternatives !  Data-adaptive machine techniques  Non-parametric modelling algorithms

Super Learner  Method to choose the optimal regression algorithm among a set of (user-supplied) candidates, both parametric regression models and data- adaptive algorithms (SL Library)  Selection strategy relies on estimating a risk associated with each candidate algorithm based on:  loss-function (=risk associated with each prediction method)  V-fold cross-validation  Discrete Super Learner : select the best candidate algorithm defined as the one associated with the smallest cross-validated risk and reruns on full data for the final prediction model  Super Learner convex combination : weighted linear combination of the candidate learners where the weights are proportional to the risks. van der Laan, Stat Appl Genet Mol Biol 2007

van der Laan, Targeted Learning, Springer 2011 Discrete Super Learner (or Cross-validated Selector)

Discrete Super Learner  The discrete SL can only do as well as the best algorithm included in the library  Not bad, but….  We can do better than that !

Super Learner  Method to choose the optimal regression algorithm among a set of (user-supplied) candidates, both parametric regression models and data- adaptive algorithms (SL Library)  Selection strategy relies on estimating a risk associated with each candidate algorithm based on:  loss-function  V-fold cross-validation  Discrete Super Learner : select the best candidate algorithm defined as the one associated with the smallest cross-validated risk and reruns on full data for the final prediction model  Super Learner convex combination : weighted linear combination of the candidate learners where the weights weights themselves are fitted data- adapvely using Cross-validation to give the best overall fit van der Laan, Stat Appl Genet Mol Biol 2007

van der Laan, Targeted Learning, Springer 2011 Discrete Super Learner (or Cross-validated Selector)

Asymptotical Properties The combination has Oracle properties: Performs asymptotically at least as well as the best choice among the library of candidate algorithms if the library does not contain a correctly specified parametric model Achieves the same rate of convergence as the correctly specified parametric model otherwise van der Laan, Stat Appl Genet Mol Biol 2007

Results

SAPS II

Super Learner 1

Super Learner 2

Conclusion  I2B2: new exciting perspective for clinical research  Need to get rid of “old good” regression methods !  As compared to conventional severity scores, our Super Learner - based proposal offers improved performance for predicting hospital mortality in ICU patients.  The score will evoluate together with  New observations  New explanatory variables  SICULA : Just play with it !!