Big Data for Better Cures: Where to Invest?

Marty Cohn (Jointly Health) defines big data in healthcare through the "four Vs," with variety (structured, unstructured, genomic, clinical) and veracity/value (handling inconsistent or conflicting data) being the primary technical challenges.

• He cites an IBM estimate that 80% of future data will be inconsistent or in conflict.
• Cohn references a JAMA study by Don Berwick estimating 21% to 47% of U.S. healthcare spending is wasteful, positioning big data as a mechanism to reduce waste and personalize care.
• His startup, Jointly Health, focuses on chronic disease management to prevent avoidable hospital admissions by risk-categorizing patients using longitudinal home monitoring data.
• He notes that clinical incentives must align with value; currently, invasive cardiologists may resist tools that reduce procedural volume because their revenue models are fee-for-service.

Joel Dudley (Mount Sinai) argues that while the "hype cycle" for big data is high, the technology exists to enable a learning health system where data exhaust from monitors and EHRs is actively mined rather than discarded.

• He challenges the concept of disease as discrete buckets (ICD-9 codes), asserting that molecular and clinical data reveal disease as a continuous spectrum that requires new definitions.
• Dudley highlights a critical research gap: the lack of a "model of the normal" human, noting that current medical knowledge is derived almost exclusively from sick populations.
• He describes a project to build a multiscale model of a normal average human to track minute-by-minute changes in DNA, immune response, and metabolites.
• Mount Sinai has a $250 million investment in biomedical informatics and is testing broad consent models where patients consent to the reuse of their genetic and EMR data for any disease research.
• He cites a specific finding where blood potassium levels predicted patient satisfaction scores, demonstrating how clinical data can objectively explain subjective feedback.

Magali Haas (Orion BioNetworks) states that brain disorders affect over 2 billion people globally and cost the U.S. alone over $2 trillion, yet lack definitive diagnostics or cures due to insufficient data scale.

• Her organization builds predictive models (comparable to weather forecasting) for brain disorders by aggregating data from partners to close the loop from basic science to clinical application.
• She emphasizes the need for common data standards and open systems to allow different organizations to analyze data for distinct purposes without silos.
• Haas identifies the lack of a mechanism for cross-organizational data sharing as a primary barrier to accelerating discovery in neurology.

Risa Sack (GE Healthcare) notes that while nearly $800 million was invested in digital health and big data IT in the last year, business models remain challenging regarding reimbursement and payer incentives.

• She identifies Clinical Decision Support (CDS) and Population Health as key areas where NLP and patient-centered data can drive value, specifically by integrating social determinants (e.g., income) into care decisions.
• Sack points out that data connectivity remains a major hurdle, with much hospital data unutilized or disconnected from external devices like Fitbits.
• She observes a macroeconomic shift toward consumer-directed health, driven by high-deductible plans, which increases patient demand for data-driven utility similar to "digital hooks" used by Google or Apple.
• Sack highlights the difficulty in selling health IT to hospitals due to a checkered history of IT failures and the fact that wellness technologies are rarely reimbursed compared to sick-care interventions.

Andy von Eschenbach (Samaritan Health Initiatives) argues that the core barrier to big data adoption is cultural, not technological, specifically within regulatory frameworks.

• He advocates for shifting the FDA regulatory model from Phase I-III trials to Phase IV (post-market surveillance) using real-world data to validate drug performance on a larger scale.
• Von Eschenbach asserts that data possession (who controls the data repository) is more critical than data ownership (which patients should legally own).
• He warns against "legislating ignorance," citing a failed Virginia bill that would have prohibited statistical analysis of medical records for population health management.

Panel Consensus on Implementation Barriers:

• Incentive Misalignment: The transition from fee-for-service to value-based care (ACOs) is essential to align financial incentives with data-driven quality improvements; currently, less than 10% of healthcare is delivered in such models.
• Physician Adoption: Even with good tools, physician compliance with guidelines and decision support averages only 30% to 40%, requiring a cultural shift in how clinicians view data usage.
• Consent and Trust: Patients are generally willing to share broad data for research if they trust the institution, as seen in the Cystic Fibrosis Foundation's registry, but adherence monitoring faces patient resistance due to fears of being "scolded."
• Data Utility: A panelist from PCORnet distinguished between hypothesis-seeking research (clinical trials) and Bayesian decision-making used in daily practice, noting that patients care more about individual outcomes than mechanistic understanding.
• Error Reduction: With 20% of diagnoses being incorrect or incomplete and 1.5 million medication errors annually, big data is seen as critical for closing these safety gaps through real-time analytics.

Forward-Looking Requests from the Panel:

• Andy von Eschenbach: Calls for formalized post-market surveillance systems for medical products.
• Marty Cohn: Urges the creation of incentive structures that reward the use of information rather than just its collection.
• Joel Dudley: Requests relaxed regulations on diagnostics to facilitate faster adoption of data-driven tools.
• Risa Sack: Emphasizes the need for better payer piloting processes for new technologies.