TacroDose AI: Simulated Clinical Scenarios for Tacrolimus Dosing

300 structured simulated clinical scenarios were used to evaluate a protocol-constrained large language model for tacrolimus dosing recommendations. The dataset, authored by Benjamin Bizer and uploaded in 2026, underpins a study showing that prompt refinement improved model protocol adherence from 72.9% to 91.7% and reduced sentinel errors from 13.9% to 1.0%.

Use Cases

• Benchmarking clinical LLM adherence to institutional guidelines based on simulated scenarios.
• Analyzing error patterns in AI-generated medication dosing recommendations.
• Training or fine-tuning models for immunosuppressive drug management in transplant care.
• Studying the impact of prompt engineering on model reliability and reproducibility in a medical context.

Strengths

• Dataset size is explicitly defined as 300 structured simulated clinical scenarios.
• Performance metrics for the evaluated model are provided with specific percentages (e.g., 91.7% adherence).
• The data is associated with a peer-reviewed research methodology described in the source PDF.
• License is clearly defined as CC-BY-4.0, permitting reuse with attribution.

Limitations

• Row count and column-level documentation for the simulated scenarios are unknown.
• Data is simulated and not derived from real-world patient records, which may limit generalizability.
• The 3.7 MB file size suggests the dataset is relatively small in scale.

Provenance

Source

Benjamin Bizer via figshare.

Collection Method

Structured simulated clinical scenarios generated for model evaluation.

Freshness

Last updated 2026-05-21 05:42:26.