Seven-steroid TabPFN V2 model

Online calculator based on preoperative peripheral plasma steroids for predicting PASO-defined complete success after adrenalectomy

Outcome: PASO-defined complete clinical and biochemical success

Input variables

All seven values are required. Check the units: values entered in the wrong unit are the most common error.

Connecting to the model…

The individual prediction report appears here.

Enter the seven steroid concentrations and select Generate report.

Model performance

Training
out-of-fold		ValidationProspective

Study cohort

PatientsComplete success

Observed complete success by outcome stratum

TrainingValidationProspective

Percentage with complete success (n/N, Clopper–Pearson exact 95% CI).

How the report is computed

Model
Frozen TabPFN V2 classifier (tabpfn 9.1.0, four ensemble members) using the natural logarithm of the seven steroids, with the 196 training patients as context. The locked threshold 0.424 was derived from training out-of-fold predictions (Youden index).
Outcome stratum
Low (<0.38), intermediate (0.38–0.65) and high (≥0.65) predicted probability, with the observed rates of complete success and Clopper–Pearson exact 95% CIs.
Contributing steroids
Exact Kernel SHAP over all 128 coalitions of the seven steroids, with the same 100 training patients as background as in the study. Contributions add up to the predicted probability.
Prediction stability
The model was fitted on 200 bootstrap samples of the training patients in the verified study environment, using the same weights and settings (the first 200 of the 2,000 samples used in the study). These fitted bootstrap models are frozen and reused to predict each new patient. The 95% uncertainty interval is the 2.5th–97.5th percentile of these predictions. The classification instability index (CII) is the proportion of bootstrap models that classify the patient differently from the main frozen model; a CII above 0.2 is reported as unstable.
Applicability
Each steroid is compared with the training 1st–99th percentiles, and the profile with the training patients by Mahalanobis and nearest-neighbour distances on log-standardized values (above the training leave-one-out 95th percentile: borderline; above the 99th: outside).
Your data
Entered values are used only to compute the report. They are held in server memory, never written to disk or logs, and discarded within two hours.