MAP–lactate risk surface
Explore the risk surface
Enter values or load a synthetic example, then select Calculate risk.
Estimate mortality risk over the next 24 hours using 10 clinical variables at 6, 24, 48, or 72 hours after ICU admission.
Enter values or load a synthetic example, then select Calculate risk.
MIMIC: patient-level five-fold cross-validation, with 10 clinical variables selected by RFE within each training fold. eICU: external evaluation of the frozen model refitted on all MIMIC data.
Metrics use a reporting threshold of 0.5. Ideal calibration: intercept 0, slope 1. Predictions have not undergone additional recalibration.
Same patients, outcomes, and folds. This descriptive comparison does not establish noninferiority or clinical equivalence.
Developed for the study cohorts of adults with sepsis in the ICU. Availability of retrospectively coded comorbidities at prediction time requires verification. No additional probability recalibration or prospective clinical validation has been performed.
Inputs are sent to the server for calculation and are not saved by the application. The hosting platform may retain connection and operational logs. Do not enter identifying patient information.
Ten clinical variables plus MOP are encoded as 18 columns. Selected variables may differ across folds; the reported cross-validation fixes the variable count and is not the separate nested variable-count selection procedure. Limitations of standard SMOTE interpolation of categorical variables and MOP remain. Full-cohort reference-stability and downstream analyses have not been completed for this reduced model.
MAP, mean arterial pressure; ICU, intensive care unit; SOFA, Sequential Organ Failure Assessment; RFE, recursive feature elimination; SMOTE, synthetic minority oversampling technique.