Synthetic-data demo · PDPA consent & IRB approval are required before any real-data use.details
More soon

Predict · β-Thalassemia

v1.0.0-rc3 · model-lockedResearch preview
A
β-Thalassemia · patient prediction

Patient SYN-025 · HSCT-MUD

Predicted on 2026-07-25 11:22:25 · Model v1.0.0-rc3
Model card
Readiness
Redesign protocol
Rationale: alloantibodies indicates current arm not viable as-is.
MDT action: Redesign protocol and stabilize before proceeding
Stabilization phase first; re-stage in 3–6 months
Served by the in-app heuristic (v1.0.0-rc3) — the fitted-model service was not used. Point estimates come from a hand-built sigmoid; ranges are illustrative, not data-derived confidence intervals.
Probabilistic outputs
Locked endpoints
Research preview
P(transfusion independence · 24 mo)
0.46
Illustrative range [0.33, 0.57]illustrative
  • HGB-207 trial-level (lovo-cel) · 0.89@ 24 mo
    first curative attempt (autologous gene therapy) · transfusion-dependent non-β0/β0 TDT (Locatelli et al., NEJM 2022)
  • CLIMB-131 long-term (exa-cel) · 0.98@ ≥24 mo sustained TI (mean follow-up 40.5 mo)
    first curative attempt (CRISPR autologous) · transfusion-dependent β-thalassemia (Frangoul et al., EHA 2025 update; range 13.6–70.8 mo)
Trial anchors are context for orientation, not the model's prediction for this patient.
criteria · CLIMB THAL-111 (Locatelli et al., NEJM)
P(VOD by d100)
0.07
Illustrative range [0.00, 0.16]illustrative
  • EBMT registry (busulfan-conditioned) · 0.14@ day +100
    allogeneic HSCT, myeloablative busulfan · paediatric haemoglobinopathy EBMT cohort (Corbacioglu et al., BMT 2018)
  • Sykora et al., BMT 2023 — treosulfan arm · 0.040@ day +100
    allogeneic HSCT, treosulfan-based conditioning · paediatric non-malignant indications (PMC10781637)
    reduced SOS/VOD vs busulfan
  • Lüftinger et al., Ann Hematol 2022 — OS anchor · 0.95@ 24 mo (OS, not VOD)
    allogeneic HSCT, treosulfan vs busulfan · thalassemia, n=772 (EBMT)
    OS context only — busulfan arm 92.7%
Trial anchors are context for orientation, not the model's prediction for this patient.
criteria · EBMT 2023 refined (Kanate et al., BMT 2024)
EBMT 2023 refined
probable0.04
clinical0.02
proven0.01
SOFA grade · 0
P(graft failure by d365)
0.03
Illustrative range [0.00, 0.10]illustrative
  • CIBMTR — haemoglobinopathy allo-HSCT · 0.080@ day +365
    allogeneic HSCT (matched related/unrelated) · paediatric β-thalassemia (CIBMTR public summary)
Trial anchors are context for orientation, not the model's prediction for this patient.
criteria · CIBMTR operational definition
P(chronic GvHD · d+180)
chronic GvHD — NIH consensus grade ≥ moderate
0.10
Illustrative range [0.04, 0.17]illustrative
  • CIBMTR matched-related (paediatric haemoglobinopathy) · 0.20@ day +180
    allogeneic HSCT, MSD/MUD · paediatric β-thalassemia / SCD
Trial anchors are context for orientation, not the model's prediction for this patient.
criteria · NIH consensus grade ≥ moderate
P(chronic GvHD · d+365)
chronic GvHD — NIH consensus grade ≥ moderate
0.14
Illustrative range [0.08, 0.22]illustrative
  • CIBMTR matched-related (paediatric haemoglobinopathy) · 0.20@ day +365
    allogeneic HSCT, MSD/MUD · paediatric β-thalassemia / SCD
Trial anchors are context for orientation, not the model's prediction for this patient.
criteria · NIH consensus grade ≥ moderate
Continuous predictions
Time-to-event & magnitude
Research preview
Engraftment day
13.8 d
Illustrative range [11.8, 16.8]illustrative
Expected HbF · 12 mo
28.4 %
Illustrative range [21.4, 35.4]illustrative
Predicted transfusion burden · 24 mo
2.66 units
Illustrative range [1.36, 4.86]illustrative
Top-K risk dimensions
Ranked feature attributions
Research preview
  • MODERATE
    BCL11A HbF QTL dose · 1 / 2
    Higher dose of HbF-raising allele favours transfusion independence.
  • LOW
    Cardiac T2* · 28.8 ms
    Low cardiac T2* indicates iron loading; affects conditioning safety.
  • LOW
    Liver iron concentration · 6.2 mg/g
    Elevated LIC raises VOD risk and slows engraftment under busulfan conditioning.
Illustrative composition weights
illustrative
not a real ensemble · single calibrated XGBoost per endpoint
c10.50
c20.10
c30.40
Decision capture · synthetic-data demo
local-only

Record (a) the recommendation shown, (b) the MDT decision, and (c) a reason for any override. Persisted to this browser only; export as JSON from the Decisions page.

shown · redesign-protocol
Redesign protocol and stabilize before proceeding
All entries are clearly marked synthetic. No PHI is collected.
Traceability
Every prediction on this page is regenerable from a single command line.
model v1.0.0-rc3Research preview
tcp predict --predictor bthal --model models/v1.0.0-rc3.tar.gz --input patients/SYN-025.json --out out/SYN-025.pred.json
This interface presents outputs of a research-preview prediction model. It is not a regulated decision-support tool. Predictions come from one calibrated XGBoost classifier per endpoint, fitted on a synthetic-labeled cohort (no PHI; labels anchored to published base rates) and evaluated on a held-out synthetic test split. Not pre-registered; not externally validated.
Research preview · not for clinical use · synthetic data · model v1.0.0-rc3
criteria versions