Healthcare AI has a transparency problem — and we address it with interpretable, auditable approaches built into Glance.
Two regulatory shifts drive the need for trustworthy analytical tools: the transition from fee-for-service to value-based care, and the data interoperability requirements of the 21st Century Cures Act. Providers and payers need models they can trust for quality monitoring, risk adjustment, decision support, and patient engagement. Healthcare decision-making is high-stakes, requiring interpretable algorithms — not black boxes.
Our Approach
We use deterministic algorithms and well-structured Bayesian modeling frameworks that retain the interpretability of classical methods while gaining the expressiveness of modern techniques. In Glance, this means:
- Deterministic ETL mapping — the same source input always produces the same OMOP CDM output, which matters for validation and regulatory review
- HCC risk adjustment coding from normalized clinical and claims data, with transparent logic that clinicians can audit
- HEDIS quality measure computation with clear numerator/denominator logic running on deduplicated, standardized records
- Reference range evaluation using clinically validated ranges stratified by patient demographics — no opaque scoring
Our models never use LLMs for clinical transformations. Every calculation is reproducible, every transformation is traceable, and every output can be explained to a regulator, a clinician, or a patient.
The Innovation
We pioneer and productize statistical models that combine the capabilities of contemporary methods within well-structured Bayesian frameworks. This gives us: economical training, uncertainty estimation, explicit data privacy guarantees, multi-modal data handling including missing and incomplete records, continuous model expansion as new data arrives, and transferability across datasets. Combined, these attributes power diverse inter-related use-cases for different end-users — each with different data available — through a single analytical platform.
