Healthcare AI Engineer
Diego Muñoz Casinos
Healthcare AI Engineer
We audit bioinformatics pipelines and define concordance thresholds defensible before regulatory bodies.
Get in touchFig. 1 — Simulated data. Each column represents one subject; colored dots show each rater's assigned score. Vertical lines indicate inter-rater range.
Pairwise Inter-Rater Agreement: Cohen’s Kappa Matrix
Mean κ = 0.87 · N = 230 variants · 5 raters
| Rater A | Rater B | Rater C | Rater D | Rater E | |
|---|---|---|---|---|---|
| Rater A | |||||
| Rater B | |||||
| Rater C | |||||
| Rater D | |||||
| Rater E |
Landis & Koch Scale
Fleiss' κ (5 raters)
0.87 [CI 95%: 0.83–0.91]
Fig. 2 — Simulated data. Pairwise Cohen's Kappa for variant pathogenicity classification (5 ACMG categories). Interpretation scale per Landis & Koch (1977).
Fig. 3 — Simulated data. Score scatter: pipeline vs established method. Solid line = regression; dashed = perfect agreement (y=x). The gap between them reveals systematic bias that ρ alone doesn't detect.