Yubin Park

• Yubin Park, co-founder and CTO of Aquarian Health, states the current U.S. healthcare system is unsustainable due to excessive waste and rapidly growing patient burdens.
• The company is a startup comprised of data scientists and medical professionals aiming to increase system efficiency through personalized healthcare, an evolution beyond the narrower concept of "precision medicine."
• Precision medicine is defined as customizing prevention and treatment based on individual genes, lifestyle, environment, and medical conditions.
• Personalized healthcare expands this scope to include personalization of every axis: service bundles, benefit plans, financial incentives, and care trajectories, not just medication or treatment.

Challenges in Implementation

• Data Bias: A CMS proposed bundled payment model for joint replacements (covering 90 days post-surgery) highlighted that machine learning models fail without accounting for selection bias, where sicker patients are disproportionately sent to skilled nursing facilities (SNFs) while healthier patients go home.
• Industry Distrust: Healthcare professionals often mistrust machine learning models built without industry insight, viewing them as products of academics or "young nerd geeks" unaware of clinical realities.
• Non-Randomized Data: Healthcare data is rarely generated from randomized trials or A/B testing, making traditional classification or regression frameworks insufficient.
• Operational Constraints: Predictive models must account for local realities, such as rural hospitals lacking SNF availability or facilities being at full capacity.
• Legal and Ethical Complexity: Solutions must address worst-case scenarios and legal implications before deployment.

Required Paradigm Shift

• Effective healthcare AI requires deep collaboration with doctors, nurses, and administrators to understand undocumented incentives and operational workflows.
• The field needs a new machine learning paradigm designed from scratch to handle multidimensional, temporal tracking of cohorts and phenotypes, rather than applying traditional frameworks.
• Predictive outputs must be scientifically grounded and explainable to medical professionals to be actionable.

Data Ecosystem for Personalization

• The strategy relies on aggregating vast "floods" of data, including:
  • Traditional Sources: Claims data, Electronic Health Records (EHR), and prescription data (structured and unstructured).
  • New Sources: Social media (Facebook, Twitter), mobile devices (Apple Watch, Fitbit), and crawled scientific literature.
• These data sources are intended to answer "what-if" questions and simulate patient trajectories.

Aquarian Health's Specific Applications

• Post-Discharge Placement Prediction: Developed a tool for a specific hospital to predict optimal post-op care (home health vs. SNF) for knee replacement patients, considering:
  • Patient factors: Age, gender, co-morbidities (e.g., congestive heart failure), and prior surgery history.
  • Facility factors: Local SNF and home health agency quality ratings and capacity.
  • Outcome simulation: Modeling readmission rates and costs for different trajectories.
• Surgery Route Personalization: Working with clients to personalize surgical approaches (e.g., robotic, arthroscopic, laparoscopic) based on individual patient suitability rather than standard protocols.
• Medication Adherence Programs: Designing personalized adherence plans for chronic conditions that address:
  • Complexity of regimens (10–20+ daily medications).
  • Financial implications (generic vs. brand costs).
  • Incentive structures for both providers and patients.

Future Outlook

• The healthcare system is not yet fully equipped for personalized healthcare, requiring significant time and development beyond current machine learning capabilities.
• Early attempts to deploy traditional algorithms often yielded more questions than answers, necessitating a shift toward designing problems that are meaningful to medical practice.
• Aquarian Health was founded one year ago by Park, Dr. Sriyam Vishwanath (UT Austin Professor), and Dr. Joyce Ho to reduce costs and improve outcomes through these efficiencies.
• The company has faced initial pushback regarding industry understanding but has secured opportunities with progressive customers willing to share data for pilot testing.
• A specific CMS bundled payment model for joint replacements is scheduled to be instituted the following year, driving immediate demand for such decision-support tools.