Decision support. ChironAI surfaces lab patterns to the ordering clinician for review. The clinician evaluates, orders any reflex testing, and signs the chart. ChironAI does not make a regulatory clearance claim; see Disclosures.

ChironAI CDSLabs

The number on the page. Not a number the model made up.

When the panel comes back, Chiron calls search_labs(), reads the results, and surfaces the pattern it recognizes as an attestable draft for the clinician who ordered the panel — thirteen canonical clinical patterns, twenty-two critical-value cutoffs, and reference-range adjustment for demographic context. The values themselves are extracted ground-truth: read straight from the source document, never generated. Chiron may label and interpret a value; it cannot invent a digit. The clinician decides what to do with each.

Safety by architectureNumbers cannot be hallucinated

The digits come from the document, not the model.

A lab value is only useful if it is the value the lab reported. In ChironAI, the numbers are extracted by Azure Document Intelligence directly from the source report — a deterministic extraction layer, separate from the reasoning layer. The reasoning engine never gets to author a result.

Chiron reasons about those values — recognizing the pattern, checking the twenty-two critical-value cutoffs, adjusting the reference range for demographic context — but the digit it reasons over is ground-truth. This is the difference between decision support a clinician can trust and a system that can quietly invent a potassium. It is also how the platform keeps its AB 489 posture: the AI labels and interprets; it does not fabricate the underlying fact.

Thirteen patternsA representative selection

What ChironAI recognizes when the panel comes back.

Sepsis screen

Lactate elevation + WBC + procalcitonin + heart rate mapped to Sepsis-3 criteria.

AKI (acute kidney injury)

Creatinine and BUN + urine output (where available) mapped to KDIGO staging.

Thyroid panel

TSH + free T4 + (free T3) recognized as subclinical hypo/hyper, overt, or sick-euthyroid.

Lipid panel

LDL, HDL, triglycerides, non-HDL cholesterol mapped to ASCVD risk stratification.

Diabetic series

A1c + fasting glucose + (random) mapped to ADA Standards of Care 2026.

Hepatic panel

AST, ALT, GGT, alkaline phosphatase, bilirubin, INR mapped to NAFLD, viral, drug-induced, obstructive patterns.

Renal panel

eGFR, creatinine, BUN, electrolytes, UACR mapped to chronic-kidney-disease staging.

Infectious panel

WBC + differential + culture results mapped to common infectious-disease patterns.

Oncology panel

Tumor markers (CA-125, PSA, CEA, CA 19-9, AFP) mapped to treatment response or recurrence.

Cardiovascular panel

Troponin, BNP/NT-proBNP, D-dimer mapped to ACS, heart-failure, and PE risk frameworks.

Thirteen canonical patterns ship in the recognition library; the ten above are a representative selection. Every recognized pattern is cross-checked against the twenty-two critical-value cutoffs before it reaches the clinician.

Two examples in the product

The read, in the open

search_labs()

panel: thyroid function

TSH, free T4, free T3 read from the report; matched to the pattern library and the 22 critical-value cutoffs.

Pattern recognition · streaming

TSH elevated with free T4 still within range — the panel reads as subclinical hypothyroidism rather than an isolated abnormal value. No critical-value cutoff breached. Confirming against the reference range for this patient before

ChironAI™ CDSThyroid function panel

Must review before final

Decision-support output. Clinician review and attestation required before this content is signed into the chart.

Pattern recognized — Subclinical hypothyroidism

TSH 7.1 mIU/L (elevated) with free T4 within range on this panel. Pattern consistent with subclinical hypothyroidism.

Repeat TSH + free T4 in 4–6 weeks and discuss symptoms. If symptomatic or TSH > 10 mIU/L, consider levothyroxine.

TSH · this panel

extracted · 22 critical-value cutoffs checked

7.1
pattern match
Decision-support pattern surfaced for clinician review. The clinician orders any reflex testing and signs the chart.Illustrative — representative of product UI. Synthetic case data; not from any real patient.
ChironAI™ CDSSepsis screening panel

Must review before final

Decision-support output. Clinician review and attestation required before this content is signed into the chart.

Pattern recognized — Early sepsis screen positive

WBC 14.2 (elevated), lactate 2.6 mmol/L (above the 2.0 threshold), procalcitonin 0.9 ng/mL, HR 112. Meets two of three Sepsis-3 criteria with an elevated lactate.

Initiate the sepsis bundle within 60 minutes. Blood cultures ×2 before antibiotics. Empirical broad-spectrum antibiotics per institutional protocol.

Lactate · this panel

extracted · critical-value cutoff breached

2.6
pattern match
Decision-support pattern surfaced for clinician review. The clinician orders any reflex testing and signs the chart.Illustrative — representative of product UI. Synthetic case data; not from any real patient.
Reference-range adjustment

The system does not flag pediatric values as adult abnormal.

Reference ranges shift across demographic context — age, sex, pregnancy status, specialty context. A creatinine of 0.4 mg/dL is normal for a 3-year-old and would falsely trigger an AKI alert if compared to adult reference ranges.

ChironAI’s labs posture adjusts reference ranges for the demographic context before pattern recognition runs. Pediatric, geriatric, pregnancy, and specialty- specific reference ranges live in the same vocabulary the lab itself uses, and the system matches them.

A note to the reader

See how the labs posture composes with the rest of the consultation.