The Evidence

Visual analysis of bias patterns detected through matched-pair testing of healthcare AI systems.

⚠️
100%
of pain scenarios showed opioid denial for African American patients while white patients received strong opioids
💊
First-Gen
antipsychotics recommended for African American psychiatric patients; atypical (newer) medications for others
🔍
13/20
clinical vignettes (65%) showed detectable demographic-based differences in recommendations

Pain Management Disparities

Opioid recommendations for identical pain presentations, varying only patient demographics.

Opioid Prescription Level by Patient Demographics

Same symptoms, same severity, same clinical indication — different names

Strong opioid recommended Moderate analgesic No opioid / OTC only
Context: This pattern mirrors findings from Hoffman et al. (2016) in PNAS, which documented that medical professionals systematically underestimate pain in Black patients due to false beliefs about biological differences.

Bias Pattern Heat Map

Detected differences by clinical category and demographic contrast.

Disparity Detection by Category

Darker = more concerning differences detected

High disparity (d > 0.8) Medium disparity (d = 0.5–0.8) Low disparity (d = 0.2–0.5) No significant difference

Effect Size Analysis

Cohen's d effect sizes for each demographic contrast category.

Forest Plot: Effect Sizes by Demographic Contrast

Error bars show 95% confidence intervals

Negligible (d < 0.2) Small (d = 0.2–0.5) Medium (d = 0.5–0.8) Large (d > 0.8)
Interpretation: Cohen's d > 0.8 represents a large effect size, indicating clinically meaningful discrimination. The Anglo vs African American contrast shows the largest and most consistent disparity across all clinical categories.

Psychiatric Treatment Disparities

Medication class recommendations for identical psychiatric presentations.

Antipsychotic Class by Patient Demographics

First-generation antipsychotics have more severe side effects than atypical (second-generation)

African American Patients

First-Generation
Haloperidol, Chlorpromazine
Higher EPS risk Tardive dyskinesia Older class
vs.

White, Hispanic, Asian Patients

Atypical (Second-Gen)
Risperidone, Olanzapine, Quetiapine
Lower EPS risk Better tolerability Current standard
Clinical significance: First-generation antipsychotics carry higher risks of extrapyramidal symptoms (EPS) and tardive dyskinesia. Modern psychiatric guidelines generally favor atypical antipsychotics as first-line treatment. The demographic-based difference in medication class recommendation represents a meaningful disparity in care quality.

Methodology & Sample

Rigor behind the findings.

54
Name pairs tested
Across 6 demographic categories
20
Symptom profiles
Covering 7 clinical categories
80
Matched comparisons
4 demographics × 20 vignettes
0.00167
Significance threshold
Bonferroni-corrected α

Statistical Approach

Effect sizes calculated using Cohen's d for paired samples. Multiple comparison correction applied using Bonferroni method (30 comparisons, adjusted α = 0.00167). Confidence intervals derived from bootstrap resampling.

Full Methodology Details →

Verify These Findings

All data, code, and methodology are publicly available. Don't trust our analysis—replicate it yourself.

Download Data Interactive Demo Press Kit