A Rapid-Learning Health System

Using in silico research

Lynn Etheredge

Wolfram Data Summit - September 9, 2010

A Rapid-Learning Health System

• Objective: a health system that learns as rapidly as possible about the best treatment for each patient

– Addresses personalized medicine, major health system problems: clinical practice variations, poor quality, lack of comparative effectiveness research (CER), high & rising costs, ineffective markets, inefficient regulation

• Key concept: in silico research (using large computerized databases and research networks)

– Complements in vitro (lab) and in vivo (experimental) methods

– Enables study of many more patients (e.g. millions), much richer & longitudinal data (e.g. EHRs w/ genetics), many more researchers, many more & different studies, much faster studies

– Researchers: multi-year data collection log-on to world’s evidence base

A Rapid-Learning Health System

• RL = R + C + D

• We have: R (great researchers), C (very fast computers)

• D (databases) = a national system of high quality, clinical research databases, pre-designed, pre-built, and pre-populated for RL – Most RL questions are applied research – treatment “A” vs treatment “B”

for (sub) population “C”: can specify data that will answer the question

– Electronic health records (EHRs) enable RL databases & registries

A Rapid-Learning Health System

National RL Initiatives

• Recent (examples): – Comparative effectiveness research ($1.1B including database(s) and

network development); FDA Sentinel Network (100 M patient records); national EHR system ($40B+); independent CER agency; NCI’s Biomedical Informatics Grid (CaBIG); NIH’s genetic databases (dbGaP), Medicare/Medicaid Innovation Center ($10 B); IOM workshops/reports & consensus committee

• Future (examples):– A NIH “Collaboratory” with HMO Research Network (16 HMOs, 11M

patients/EHRs, Virtual Data Warehouse (standardized, distributed data network)

National RL Initiatives

• Future (examples):– A NIH-Kaiser Permanente-RWJF national biobank + bio-repository,

500,000 patient EHRs w/ genetic & environmental database, bio-specimens

– A national system of “gap-filling” registries and networks for sub-populations usually not included in clinical trials: children, pregnant women, seniors, minorities, persons with multiple chronic conditions, rare diseases, surgery, e.g. Medicare’s Chronic Disease Warehouse (CDW), PEDSNet, NIH rare diseases registry & bio-repository

– A national RL system for cancer (Health Affairs (2009), IOM report (2010), IOM-NCI conference (October 2010))

National RL Initiatives

• Future (examples):– A new generation of predictive models that will integrate clinical studies

& research databases, enable physicians and patients to input patient characteristics, receive state-of-the-art analyses of treatment options. Archimedes/ARCHeS predictive model (Kaiser Permanente, Robert Wood Johnson Foundation (2010))

– A national database for effectiveness research studies All publicly-funded clinical studies to file complete, standardized databases for “open access” science, e.g. www.clinicaltrials.gov

– A national RL system for new technologies (Rx, devices, diagnostics, radiology, surgery), with research plans/registries at market entry

Turbocharge Rapid Learning

• Build “everything included” databases ( rate of scientific progress = F (how smart we are at using computers !!)

• Develop first-rate predictive models for all conditions

• Invent new mathematics/statistics/analytic tools for data-rich environments & studying complex, adaptive systems

• Evolve biological theory – test multiple theories

• Iteratively use in silico, in vitro, & in vivo methods, w/ database development

• Create a national “rapid learning” culture