How AI is Predicting Catastrophic Claims (Before They Happen) | with Sasha Gribov

This video provides an in-depth exploration of how artificial intelligence and real-time data analysis are being leveraged by Milu Health to create a proactive "early warning system" within the employer-sponsored healthcare ecosystem. Sasha Gribov, the co-founder of Milu Health, discusses the necessity of shifting the healthcare model from reactive claim processing to proactive intervention. The core mission is to use technology layered on comprehensive data to save patient lives and dramatically reduce costs for self-insured employers by identifying potential catastrophic health events months before they materialize.

The progression of the discussion emphasizes the overwhelming volume of data generated in healthcare—far exceeding terabytes—which is currently siloed across various systems (EHRs, claims, labs, genetic tests). Milu Health's unique value proposition lies in aggregating this disparate data, particularly real-time health records (doctor's notes, pathology, radiology reports), which they access as a provider organization (employing nurses and pharmacists for medication review and care gap analysis). This comprehensive data view, combined with AI algorithms, allows them to identify care gaps, potential medication conflicts, and the progression of chronic conditions that often lead to expensive surgeries or emergency visits.

A key methodology highlighted is the use of AI to augment, not replace, human clinicians. Milu's system identifies potential issues, which are then reviewed by human nurses and pharmacists in a "human in the loop" system before any outreach occurs. Gribov shared a compelling case study showing that their system catches indications of future surgeries (like orthopedic procedures) an average of 90 days in advance, achieving an 80% catch rate—significantly higher than the 30% expectation set by industry partners. This early identification allows Milu's team to proactively reach out to members via simple text messages (avoiding app fatigue) to guide them toward high-value, often free, solutions already available in their health plan, or to schedule necessary appointments, thereby preventing costly escalation.

The conversation also touches on the strategic application of this technology for various stakeholders. For Third-Party Administrators (TPAs) and stop-loss carriers, Milu offers a crucial differentiator: the ability to intervene nine months before an emergency room visit, optimizing existing case management efforts. For employers, the system maximizes the utilization of existing, often underutilized, point solutions (e.g., Centers of Excellence, specific disease management programs). Gribov asserts that while high savings (like 30%) require disruptive plan changes (e.g., RBP), Milu provides real, measurable savings in the single digits by optimizing the existing plan structure and ensuring members receive the right care at the right time.

Detailed Key Takeaways

• Proactive Intervention is the Future of Cost Management: The current reactive model, where intervention only occurs after a catastrophic claim is filed, is fundamentally broken. AI-driven systems like Milu Health enable a forward-looking approach by identifying worsening chronic conditions (e.g., escalating knee pain) months in advance, allowing for timely, less expensive interventions like physical therapy instead of emergency surgery.
• The Power of Integrated Health Data: The true "magic" of AI in healthcare is unlocked when siloed data—including claims, lab results, genetic tests, and especially real-time Electronic Health Records (EHRs)—are aggregated. EHR data provides rich context (doctor's notes, pathology reports) that claims data alone cannot offer, leading to superior predictive accuracy.
• AI Augments, It Doesn't Replace, Clinicians: Milu utilizes a "human in the loop" system where AI agents identify potential care gaps or risks, but licensed nurses and pharmacists review and validate every finding before communicating with the patient. This ensures safety, security, and trust, preventing the delivery of "asinine stuff" sometimes generated by raw LLMs.
• Significant Predictive Lead Time: Milu's AI models demonstrated the ability to catch indications of future surgeries (e.g., orthopedic procedures) with an 80% success rate, providing an average lead time of 90 days. This lead time is critical for effective care coordination, specialist referrals, and cost-saving negotiations.
• Simplicity Drives Engagement: Patient outreach is conducted primarily via simple text messages, avoiding the need for members to download new apps or navigate complex portals. This low-friction communication strategy is highly effective, leading to high engagement rates (90%+ consent rates when the value proposition is clearly explained).
• Provider Status is Key to Data Access: Milu Health gains access to comprehensive, real-time health records by operating as a provider organization, employing pharmacists (for medication review/reconciliation) and nurses (for care gap review). This status allows them to access data that typical vendors or consultants cannot.
• Strategic Differentiation for TPAs and Stop-Loss: For TPAs, Milu offers a way to differentiate by providing proactive identification services that augment their existing case management teams. For stop-loss carriers, the technology offers a real-time understanding of patient health status beyond outdated case notes, though it is currently used for optimization post-enrollment, not initial underwriting.
• Implementation is Low-Friction: Implementation does not require ripping out existing systems, changing the plan design, or setting up complex claims feeds initially. Milu only requires an enrollment file to begin and can go live mid-year, often preferred by consultants to avoid the confusion of open enrollment.
• Focus on Optimization, Not Disruption: Milu's expected ROI is typically 3x to 5x their cost, resulting in real, single-digit percentage savings achieved by optimizing the utilization of existing high-value plan benefits (e.g., directing members to contracted Centers of Excellence or high-quality doctors).

Key Concepts

• Early Warning System: A proactive technology platform that analyzes real-time health data to predict the escalation of chronic conditions or the need for high-cost procedures (like surgery or hospitalization) months in advance.
• Human in the Loop (HITL): An AI system design where human experts (nurses, pharmacists) review, validate, and act upon the insights generated by the AI before they are deployed or communicated to the end-user (patient).
• Medication Review and Reconciliation: A service performed by Milu's pharmacists to ensure prescribed medications do not conflict with existing prescriptions and that patients are adhering to necessary treatments.
• Care Gaps: Identified instances where a patient has a documented need for follow-up care (e.g., an MRI ordered eight months ago, a specialist referral mentioned in notes) that has not yet been acted upon.

Examples/Case Studies

• Predicting Surgeries: Milu ran a study demonstrating that their AI could identify indications of future surgeries (e.g., orthopedic procedures stemming from chronic pain) an average of 90 days before the procedure, achieving an 80% catch rate.
• Effective Communication: Early attempts at automated, branded outreach ("Milu alert with a cute emoji") failed. Success was achieved when nurses reached out simply as human beings ("Hi, I'm Brooke, I'm a registered nurse. I saw this thing. Let me know if you need some help."), leading to dramatically higher response rates.
• Optimizing Point Solution Adoption: Milu is used by consultants to drive proactive adoption of other high-value, zero-cost solutions (e.g., specialized doctor networks) that are typically underutilized (sometimes only 2% utilization) because members are unaware of them or confused by the rollout.