How AI for Deterministic Approaches Can Fix PA Denials
A recent survey by the American Medical Association (AMA) revealed a growing concern among physicians: over 60% believe that unregulated AI tools are systematically denying necessary patient care. “Emerging evidence shows that insurers use automated decision-making systems to create systematic batch denials with little or no human review, placing barriers between patients and necessary medical care,” said AMA President Bruce A. Scott, MD, reacting to the survey results.
This alarming trend exacerbates the already burdensome prior authorization (PA) process in the US, delaying essential treatments and placing undue strain on providers, payers and patients alike.
At Smile Digital Health, we believe that AI should support—not override—good medical judgment. That’s why we developed KALM Services (Knowledge Authoring Lifecycle Management), with AI capabilities designed to exponentially accelerate the creation of deterministic computable policies for prior authorization workflows and other use cases like clinical reasoning and quality measure evaluations.
Why the Deterministic Approach Beats Probabilistic
A Deterministic approach is based on published medical research and established clinical policies. Leveraging Smile’s custom AI tooling to transform policy documents (such as PDFs) are transformed into FHIR® and CQL. Since the workflows are based on established guidelines, the process is transparent, reviewable and auditable.
Relying on probabilistic AI models introduces a level of randomness and uncertainty into the decision-making process, rendering those decisions opaque and difficult to trust. Since probabilistic models assign likelihood to potential outcomes, even when provided with the same inputs, the outputs (or in the case of Prior Authorization, the adjudication) can be unpredictable.
Deterministic AI is like a calculator, which given the same inputs, arrives at the same answer. Probabilistic AI is like a weather forecast, giving a probability of rain, based on patterns and data. They are both useful in different ways but for healthcare and prior authorization, Smile's approach is to leverage deterministic AI along with human intervention to deliver Clinical Intelligence.
Smile’s KALM Services—which use AI—are powerful accelerators in assisting the human authoring of computable clinical content, reducing the process from over 6 weeks to 4 days; computable content is a human-readable decision tree, and ultimately, the review and publishing of these guidelines are done by human experts.
Balancing AI with Clinical Judgement
The AI-supported intermediary documents are reviewed and refined by a Knowledge Expert (KE) to ensure human oversight before deployment. They are also validated and codified by a Clinical SME using additional modules of KALM Services. This approach reduces the heavy labour, cost and time associated with the generation of computable clinician knowledge, and allows sound clinical judgement to validate the clinical decision pathways as they are coded.
When policies are updated, changes can be made easily and reflected in real-time in any use-case, including but not limited to:
- Real-time Prior Authorization adjudication
- Automated identification of care gaps
- Determination of next-best clinical actions (i.e. clinical practice guidelines)
- Real-time quality measure evaluations
Towards a Transparent Future for Prior Authorization
Smile’s KALM Services run on our FHIR®-native Health Data Platform and Clinical Quality Intelligence (CQI) Suite, enabling real-time automated prior authorization workflows. By ensuring compliance while supporting data modernization, we provide payers and providers with a transparent, repeatable solution to prior authorization challenges.
