Synthetic computational demonstration
Research simulator
This interactive page demonstrates the sequence labs → score → sigmoid probability → threshold band. It deliberately does not impersonate the trained XGBoost model.
Not a patient risk calculator. Coefficients on this page are synthetic visualization parameters. Do not enter real patient data for clinical decision-making.
Demonstration output
Synthetic score
—
Synthetic sigmoid output
—
Demonstration band
—
\[p_{demo}=\frac{1}{1+e^{-S_{demo}}}\]
These outputs are intentionally labelled synthetic. The actual XGBoost model consists of hundreds of nonlinear tree rules.
Why the simulator is not built from gain weights
Feature gain indicates importance, not direction or a fixed coefficient. For example, the learned gain ranking currently places calcium first, but the model may use calcium at several different thresholds and in interactions with other features.
\[\mathrm{Gain}(x_j)\neq\beta_j\quad\text{and}\quad S(X)\neq\sum_j \mathrm{Gain}(x_j)x_j\]
The actual feature-gain table is available on the computational model page.
Synthetic equation used only for UI demonstration
\[S_{demo}=-2.0+0.03(WBC-10)+0.003(CRP-50)+0.002(\alpha HBDH-160)+0.08(Glu-7)-0.9(Ca-2.2)-0.025(ALB-38)+0.06(PT-13)+0.02(APTT-33)\]
Scientific boundary.
This equation is not fitted to the cohort and is not a surrogate clinical score. A research-grade browser predictor would require the frozen tree ensemble, preprocessing, missing-data handling and probability calibration to be versioned and validated together.