Synthetic data only. Educational use only. UNMIRI is a developer preview, not a medical device, and is not for clinical decision-making.
Tumor-agnostic TMB-high + MSI-H
The starting input
A Tempus xT sample. Mismatch-repair-deficient colorectal adenocarcinoma, but the same biology drives a tumor-agnostic recommendation across many cancers. Sample at data/sample_reports/tempus/tempus_xt_sample.pdf.
What the parser extracts
- Variants: MLH1 splice-region, MSH2 frameshift (germline-confirmed)
- Biomarkers: TMB 21 mut/Mb (high), MSI-H, MMR-deficient
- FDA CDx flag: pembrolizumab tumor-agnostic indication
Tempus formats biomarkers as a top-of-report banner. The parser anchors on that and emits structured biomarker entries with page and char-offset provenance.
What the evidence join adds
- openFDA: pembrolizumab tumor-agnostic indication for MSI-H or dMMR solid tumors after progression on standard therapy
- CIViC + ClinVar: MLH1 / MSH2 pathogenic Lynch syndrome alleles
- ClinicalTrials.gov: 100+ MSI-H trials, including basket designs
- AMP/ASCO/CAP tier: I-A
What the CDS surface produces
The tumor-agnostic predicate is a deterministic OR over the biomarker set. Engine 2 surfaces the pembrolizumab option, a Lynch-syndrome counseling pointer (because both variants are in MMR genes and germline-confirmed), and the matching trial list. Sticky educational-use banner. No LLM in the rendering path.
Why this case matters for buyers
TMB and MSI are biomarkers, not variants. Vendors disagree about how to compute and report them. Different cutoffs, different denominators, different tables. UNMIRI doesn't recompute them. It normalizes the vendor's reported value with provenance and marks the vendor's methodology block. The CDS layer reads the normalized biomarker, not a vendor string.