Invisible by Design

The Data Imbalance Problem

AI systems in healthcare are trained on datasets that systematically underrepresent women's health conditions. This creates models that perform well on average but fail catastrophically for the very populations that need accurate diagnosis the most.

When evaluation datasets share the same imbalance, the problem is hidden: the model looks "high-performing" overall because its weakest cases are rare in both training and testing. This is known as hidden stratification— strong average scores that mask poor subgroup performance.

The False Negative Disparity

In liver disease detection, models show 21 percentage points higher false-negative rates for women. This means women with liver disease are significantly more likely to be told they're healthy when they're not—delaying treatment and worsening outcomes.

The Solution: Sex-Aware AI

The path forward requires building AI systems that are explicitly designed to account for sex-specific differences. This means:

• Training on sex-balanced datasets
• Evaluating performance across sex-specific subgroups
• Developing sex-aware algorithms that account for biological differences
• Capturing the "Barrier Data" that traditional systems miss

How ORI Addresses This

ORI is designed from the ground up to capture the data that has been historically missing. By generating continuous, structured health data that accounts for sex-specific presentations and outcomes, ORI enables:

• Sex-aware AI models that reduce diagnostic bias
• Better clinical decision-making through comprehensive data
• Coverage and benefits design informed by real-world patterns
• Policy development based on complete, not partial, data