Arts & Architecture

AITION: Interactive Medical Data Analytics (The JIA Case)

Description
AITION: Interactive Medical Data Analytics (The JIA Case) Yannis Ioannidis The UoA Team (Harry Dimitropoulos, Nicholas Economides, Herald Klapi, Eva Sitaridi, Omiros Metaxas, Manolis Tsangaris) 1 Health-e-Child
Published
of 27
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Related Documents
Share
Transcript
AITION: Interactive Medical Data Analytics (The JIA Case) Yannis Ioannidis The UoA Team (Harry Dimitropoulos, Nicholas Economides, Herald Klapi, Eva Sitaridi, Omiros Metaxas, Manolis Tsangaris) 1 Health-e-Child IGG-UoA-DISI, Jan 15th, 2009 AITION in JIA - Outline JIA Problems JIA Data Problem #1: Understanding Data Dependencies Problem #2: Outcome Predictors and Treatment Response Future Steps 2 Health-e-Child IGG-UoA-DISI, Jan 15th, 2009 JIA Problems Improve JIA classification (clustering) with the identification of more homogeneous groups according to their clinical and biologic features taking into consideration disease duration, short term / long term outcome, response to treatment, etc. general Find early predictors of poor outcome allowing for the early identification of patients who are most likely to develop joint damage and thus need aggressive treatment any Identify sensitive markers of JIA activity and damage progression to assess the disease-modifying effect of drugs in clinical trials. regular clinical procedures 3 Health-e-Child IGG-UoA-DISI, Jan 15th, 2009 Analyze Dependencies JIA Problems Analyze snapsot data to discover dependencies between different variables Find Outcome Predictors Analyze prognostic factors and treatment response 4 Health-e-Child IGG-UoA-DISI, Jan 15th, 2009 76 patients (9 with 2 Visits) 37 Variables related to JIA Data Demographic / Clinical / Laboratory / Medication Can be categorized according to Use: Predictor / Outcome Vertical Integration: Demographics, Clinical, Lab Want more: Genetics (IL2RA, MEFV), Imaging (Poznanski Score): Medical Record: Subjective, Objective, Assessment, Planning Preprocessing and Cleaning Restoring missing values Checking for inconsistencies & outliers Discreteizing 5 Health-e-Child IGG-UoA-DISI, Jan 15th, 2009 Problem #1: Analyze Data Dependencies 17 related variables capturing different medical areas Demographics: Sex, AgeOnset Subjective & Objective Clinical Data: Disease Duration, GlEvalDiseaseActivity, JointsWithLOM, CHAQ, JointsWithActiveDisease, JADI-E, GeneralEvaluation Assessment: ILARClass Laboratory: ESR, CRP, ANA Treatment Plan (Medication): JointInjection, OnGoingDMARDs, OnGoingSTEROID, OnGoingBIOLOGICS 6 Health-e-Child IGG-UoA-DISI, Jan 15th, 2009 Result: A graph model of variables and probabilities 7 Health-e-Child IGG-UoA-DISI, Jan 15th, 2009 Zoom In: Weighted model Pairwise dependency analysis Sex AgeOnset DisDur ILARClass GlEvalDisAct JntsLOM JntsActDis JADIE GenEval CHAQ ESR CRP ANA JntInjection OnDMARDs OnSTEROID OnBIOLOGICS edge marginals based on weighed model averaging Health-e-Child IGG-UoA-DISI, Jan 15th, 2009 Qualitative dependency analysis & Model Building Finding most important dependencies and.... independencies: e.g. Sex, ANA are almost uncorrelated and excluded from the final model important dependencies ANA dependencies Sex AgeOnset DisDur ILARClass GlEvalDisAct JntsLOM JntsActDis JADIE GenEval CHAQ ESR CRP ANA JntInjection OnDMARDs OnSTEROID OnBIOLOGICS edge marginals based using weighed model averaging Health-e-Child IGG-UoA-DISI, Jan 15th, 2009 Vertical Integration imaging demographics Poznanski s Score lab genes MEFV IL2RA clinical 10 Health-e-Child IGG-UoA-DISI, Jan 15th, 2009 A priori knowledge encapsulation Manual changes to the model are allowed 11 Health-e-Child IGG-UoA-DISI, Jan 15th, 2009 A priori knowledge encapsulation Before model building we can define layers for automatic node ordering and causal discovery Layer 1 Layer 2 Layer 3 Layer 4 Layer 5 12 Health-e-Child IGG-UoA-DISI, Jan 15th, 2009 Feature Selection using dependency analysis Find the set of nodes shielding a given node from the influence of the other nodes 13 Health-e-Child IGG-UoA-DISI, Jan 15th, 2009 Feature Selection using dependency analysis..my parents and the parents of my children knowing these nodes I can infer anything regarding me and my children 14 Health-e-Child IGG-UoA-DISI, Jan 15th, 2009 AITION Reasoning using Inference in Graphs Evidence: observation of specific state Task: compute the posterior probabilities for query node(s) given evidence. x1 x1 x1 x8 x1 x5 Y x2 x5 x5 x8 x8 x8 Diagnostic inference Predictive inference Intercausal inference Mixed inference 15 Health-e-Child IGG-UoA-DISI, Jan 15th, 2009 DEMO 16 Health-e-Child IGG-UoA-DISI, Jan 15th, 2009 Problem #2: Outcome Prediction Find predictors of poor outcome allowing for the early identification of patients who are most likely to develop joint damage and thus need aggressive treatment, and evaluate treatment response 17 Health-e-Child IGG-UoA-DISI, Jan 15th, 2009 Predictors Outcome Prediction & Vertical Integration demographics imaging genetics SEX AgeOnSet clinical DisDur JntLOM JntActDis GenEval Poznanski IL2RA MEFN CHAQ ANA CRP ESR ILAR GlbActDis JADI lab Medication NSAID STEROID DMARD BIOLOGIC GlbActDisOut JntLOMOut CHAQOut JntActDisOut ESROut CRPOut GenEvalOut PoznanskiOut JADIOut Outcome BOXValidatedOut FINALOut 18 Health-e-Child IGG-UoA-DISI, Jan 15th, 2009 a simple example Based on HeC JIA visit2 data (9 patients) Predictors selected from visit1: AgeOnset,ILARClass,GlEvalDisAct,GenEval Medication Selected from visit1: OnDMARDs,OnSTEROID Outcome computed using visit2: GlEvalDisActOut, GenEvalOut, BOXValidatedOutcome (based on JointsActDisease, JointsWithLOM, CHAQ, GenEval,ESR, CRP GlobalEvalDiseaseActivity) 19 Health-e-Child IGG-UoA-DISI, Jan 15th, 2009 Temporal modeling for outcome prediction Init with visit 1 model 20 Health-e-Child IGG-UoA-DISI, Jan 15th, 2009 Temporal modeling for outcome prediction Compute outcome using visit2 data 21 Health-e-Child IGG-UoA-DISI, Jan 15th, 2009 Temporal modeling for outcome prediction Full temporal model building 22 Health-e-Child IGG-UoA-DISI, Jan 15th, 2009 Temporal reasoning: Inferring outcome 23 Health-e-Child IGG-UoA-DISI, Jan 15th, 2009 Temporal reasoning: Inferring outcome 24 Health-e-Child IGG-UoA-DISI, Jan 15th, 2009 Temporal reasoning: treatment response Even though GlbEvalDisAct is improving BOXValidated outcome is getting worse 25 Health-e-Child IGG-UoA-DISI, Jan 15th, 2009 A different way of modeling temporal data. Each visit s internal dependencies are represented as edges between node of the same layer (intra slice topology) Temporal dependencies between visits are represented as edges between node belonging to different layers (inter slice topology) Intra slice topology Inter slice topology OnSet Visit Visit 1 year later Visit 3 years later 26 Health-e-Child IGG-UoA-DISI, Jan 15th, 2009 Semantics Incorporation: Future Steps Use ontologies to impose constraints in node ordering / layering describing extracted knowledge with its associated uncertainty in a principled, structured, sharable, and machine-understandable way Clinical Evaluation: Extend experiments and clinical evaluate the results HeC Integration: Integration with other HeC subsystems 27 Health-e-Child IGG-UoA-DISI, Jan 15th, 2009
Search
Related Search
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks