This document summarizes a study that used artificial intelligence to develop a dynamic model for predicting thromboelastogram (TEG) outcomes using data from two clinical trials. The model was built using TEG data to estimate patient-specific parameters. A virtual library of over 160,000 simulated TEGs was generated and used to identify patient response trajectories and predict specific TEG outcomes through logistic regression.
This document summarizes a study that used artificial intelligence to develop a dynamic model for predicting thromboelastogram (TEG) outcomes using data from two clinical trials. The model was built using TEG data to estimate patient-specific parameters. A virtual library of over 160,000 simulated TEGs was generated and used to identify patient response trajectories and predict specific TEG outcomes through logistic regression.
This document summarizes a study that used artificial intelligence to develop a dynamic model for predicting thromboelastogram (TEG) outcomes using data from two clinical trials. The model was built using TEG data to estimate patient-specific parameters. A virtual library of over 160,000 simulated TEGs was generated and used to identify patient response trajectories and predict specific TEG outcomes through logistic regression.
AI study - Accelerating availability of clinically-relevant parameter estimates from Thromboelastogram point-
of-care device
Info Domain Key items
SOURCE OF DATA Source of data (e.g., cohort, case-control, RCTs
randomized trial participants, or registry data) Design of the Study of Tranexamic Acid during Air Medical Prehospital Transport (STAAMP) Trial – Level I trauma centres, trauma patients with hypotension and tachycardia, USA. Participant eligibility and recruitment method (e.g., consecutive participants, location, number of Taking the Blood Bank to the Field: The Design and centers, setting, inclusion and exclusion criteria) PARTICIPANTS Rationale of the Prehospital Air Medical Plasma (PAMPer) Trial.
Participant description Trauma patients
Details of treatments received, if relevant Tranexamic acid, prehospital plasma transfusion Study dates Trials used - PAMPer 2016. STAAMP 2014 OUTCOME(S) TO 1. Dynamic Model Construction: BE PREDICTED Dynamic model was built using thromboelastogram (TEG) data.
indicating maximum amplitude (MA) below normal and 466 indicating lysis percent Definition and method for measurement of outcome above normal.
3. Virtual Library Creation:
A virtual library of over 160,000 simulated
RapidTEGs was generated using the dynamic model.
4. Comparison and Trajectory Analysis:
Patient TEG responses were compared
against the virtual library to identify trajectories with the least sum-of-squared error up to specified evaluation times.
Logistic regression was performed using 10
nearest-neighbor trajectories to predict specific TEG outcomes.
Was the same outcome definition (and method for
measurement) used in all patients?
Type of outcome (e.g., single or combined
endpoints) Was the outcome assessed without knowledge of the candidate predictors (i.e., blinded)? Were candidate predictors part of the outcome (e.g., in panel Time or consensus of outcome diagnosis)? occurrence or summary of duration of follow-up Number and type of predictors (e.g., demographics, patient history, physical examination, additional testing, disease Definition characteristics) and method for measurement of CANDIDATE PREDICTORS candidate Timing predictorsmeasurement (e.g., at patient of predictor (OR INDEX TESTS) presentation, Were predictorsat diagnosis, at treatment assessed blinded initiation) for outcome, and for each other (if relevant)? Handling of predictors in the modelling (e.g., continuous, linear, non-linear transformations or categorised) Number of participants and number of SAMPLE SIZE outcomes/events Number of outcomes/events in relation to the number Number of of candidate predictors participants with any(Events missingPer Variable) value MISSING DATA (include predictors Number and outcomes) of participants with missing data for each predictor Handling of missing data (e.g., complete-case analysis, Modellingimputation, or other method (e.g., methods) logistic, survival, neural network, orassumptions Modelling machine learning techniques) satisfied Method for selection of predictors for inclusion in MODEL multivariable modelling (e.g., all candidate DEVELOPMENT predictors, Method forpre-selection based on unadjusted selection of predictors during multivariable modelling (e.g., full model approach, backward Shrinkage or forward selection) of predictor and weights or criteria used regression coefficients (e.g., no shrinkage, uniform shrinkage, penalized Calibrationestimation) (calibration plot, calibration slope, MODEL Hosmer-Lemeshow test) and Discrimination PERFORMANCE (C-statistic, ClassificationD-statistic, measureslog-rank) measures (e.g., sensitivity, with specificity, predictive values, net reclassification improvement) and whether Method useda-priori cut model for testing points performance: were used development dataset only (random split of data, MODEL resampling methods e.g. bootstrap or cross- EVALUATION validation, none) or separate external validation (e.g. In case of poor validation, whether model was adjusted or updated (e.g., intercept recalibrated, predictor effectsmultivariable Final and other adjusted, or new predictors models added) (e.g., basic, extended, simplified) presented, including predictor weights or regression coefficients, intercept, baseline survival, model performance measures RESULTS Any alternative presentation of the final prediction models, e.g., sum score, nomogram, score chart, predictions Comparisonfor specific of the risk subgroups distribution with of predictors (including missing data) for development and validation datasets Interpretation of presented models (confirmatory, INTERPRETATION i.e., model useful for practice versus exploratory, i.e., AND DISCUSSION more research needed) Comparison with other studies, discussion of generalizability, strengths and limitations.