Bias and equity auditor

Compares error rates across patient groups and languages.

In audits Layer 2 · Measure
Takes inOutputs with consented demographic or proxy data
The agentEquity auditor
Hands backDisparity report with CIs

Checked against clinicians

Clinician review of flagged groups; small-subgroup CI discipline.

Where a person decides

Findings reviewed by governance committee.

Works on

Chat, scribes, voice, payer, predictive.

Delivered inside a Quality Audit today. Reviewed .

The problem it solves

An average score can hide a group that gets worse care: a language, an accent, an age band, a dialect.

How it works

01

Split results by group

The agent breaks errors down by language, accent, age, sex and other agreed groups.

02

Test the gaps

It checks whether differences are larger than chance, using intervals.

03

Ask whether it's clinical

Gaps go to clinicians to decide if there is a clinical reason.

An example, illustrative

Before

Overall error rate: 8%.

With the agent

Hindi-English calls: 15% (interval 11 to 20%). English: 6%.

What it needs from you

Case data with group labels you're allowed to use.

Its limits

Small groups give wide intervals. We say when the data can't tell.

Quick answers

What does the equity auditor do?

Subgroup error and denial analysis; language and dialect parity.

How is the equity auditor checked against clinicians?

Clinician review of flagged groups; small-subgroup CI discipline.

Which groups do you check?

Language, accent and dialect first, then groups relevant to your population and legal duties.

What if we lack demographic data?

We can use simulated cases with controlled differences, and say that's what we did.

What can the equity auditor not do?

Findings reviewed by governance committee. No agent makes a clinical decision.

Use the equity auditor on your product