Physician building open-source tools for health. Santiago, Chile.
A label-choice bias audit toolkit for healthcare risk-stratification models.
Most audits ask how accurate is the model?
riskaudit asks a different question: who becomes invisible because of
the label we chose?
When a risk model is trained on a proxy for the outcome that actually matters — cost instead of need, utilization instead of illness — the bias sits in the target variable, not in the features. Standard fairness metrics do not see it.
→ github.com/conradoproromant/riskaudit
MD, Pontificia Universidad Católica de Chile MSc Data Science, [institución] — in progress
Interests: clinical prediction models · algorithmic fairness · computational psychiatry
Contact: cproromant@gmail.com · LinkedIn · ORCID