By Deborah Borfitz 

March 19, 2012 | Hypotheses culled from real-world outcomes data will be getting large-scale testing by Harvard-affiliated Brigham and Women’s Hospital. The advanced analytics of GNS Healthcare are being applied to de-identified data from electronic health records (EHRs), pharmacy data, and administrative claims information to determine what factors contribute to adverse drug reactions and hospital readmissions in patients with congestive heart failure (CHF), says David Bates, MD, director of the hospital’s Center for Patient Safety Research and Practice. 

David Bates

Hospital readmission consequently represents the single biggest expenditure born by the Centers for Medicare & Medicaid Services (CMS) for CHF, as well as a host of other conditions like asthma, chronic obstructive pulmonary disease, and coronary artery disease. CMS now financially penalizes hospitals for readmissions above a certain threshold, says Bates, giving hospitals everywhere a “strong incentive” to tease out causal factors. 

The collaboration with GNS Healthcare (formerly Gene Network Sciences) began last October, with Bates serving as the lead investigator on the Brigham and Women’s side. The reverse engineering forward simulation (REFS) platform of GNS Healthcare is tasked with looking through a subset of the hospital’s dataset specific to CHF patients for variables associated with adverse drug events and readmissions. The hope is that novel hypotheses, “ones we wouldn’t have made ourselves,” will emerge, says Bates. 

Discovered connections will be evaluated by a research team that includes a programmer, research assistant, and a couple of pharmacists, says Bates. “Pages and pages of associations” have by now been produced by GNS Healthcare. Unexpected associations that make sense clinically will be investigated in the traditional fashion. “We’d also like to see [the dozens of] expected associations, because it validates the tool and current [medical] knowledge.”  

The plan is to use this approach subsequently on other Brigham and Women’s datasets “to predict readmissions more broadly,” says Bates. It is not yet known if one generic tool, or a series of diagnosis-specific ones, will be most effective. “The intent is to understand what the opportunities are to use data to identify people at risk of having problems and intervene earlier, which should both improve safety and reduce costs. The overall quest is to improve the value of care.” 

The most significant risk of this “hypotheses-free” approach to clinical research is idiosyncrasies in the dataset being mined, says Bates. If the computer model gets “over-fitted” to those quirks, it won’t work in independent datasets. “We need to create rules, models, and associations that are generally valid.” 

Although the current focus is on identifying medication-related safety issues, says Bates, the search could be broadened to include other potential causes of hospital readmissions like blood clots.  

Pharmaceutical companies will join health insurance companies and pharmacy benefit managers as licensees of software that emerges from the Harvard project, predicts Colin Hill, CEO of GNS Healthcare. Results of the CHF inquiry will be useful in pharmaceutical marketing, medical affairs, and late-stage drug development in terms of labeling and endpoints that “differentiate over competitors.” 

The Pursuit in Other Haystacks 

Brigham and Women’s Hospital is one of a number of health care institutions working with GNS Healthcare to test outcomes-based hypotheses. Most of the top pharmaceutical firms have already tapped GNS Healthcare to look for needles in haystacks of data, says Hill. Tasks assigned by the research and development (R&D) wing have included drug target discovery, mechanisms of drug efficacy, and especially biomarkers for patient selection in clinical trials. Interest in comparative effectiveness research is also particularly acute. 

The REFS “sifts through the hay…like a Hoover magnet” accomplishing the equivalent of 10,000 years of human labor, says Hill. IBM supercomputers power the hay sorting, which started in 2000, a full year before sequencing of the human genome added huge amounts of complex data to the stack.  

REFS crunches vast amounts of data from sources—ranging from gene-expression assays and next-generation DNA sequencers to EHRs and patient outcomes databases—to churn out novel hypotheses about not only drugs but also patients and the impact of providers and care paths on treatment results. Bringing down the cost of care by focusing on what works best is the overarching issue, says Hill. Health care in the U.S. now costs $2.7 trillion annually and drugs account for only 12% of that expenditure. 

On the R&D side, GNS Healthcare helps drug companies accelerate discovery by making their wet lab experiments more productive. It also facilitates patient-to-drug matching for heterogeneous oncology, autoimmune and neurological diseases that require a more personalized treatment approach. “Our niche with biomarkers is when the data has multiple layers to it,” says Hill. Identifying subsets of patients for whom a drug works not only reduces the size and duration of clinical trials; it “increases their probability of success.” 

Drug researchers are keen on the technology because it quickly generates hypotheses without relying on the comparatively limited universe of biological knowledge available in scientific literature, says Hill. Several other companies offer bio-simulation discovery and development services, but most of these ply the existing knowledge base for leads. 

The REFS approach has resulted in a number of potentially very important R&D discoveries, notably novel therapeutic targets and markers for the one-third of rheumatoid arthritis patients who are unresponsive to anti-TNF drugs, says Hill. Study partner Biogen Idec now hopes GNS Healthcare can help it better determine the parameters of efficacy for a multiple sclerosis drug it already has on the market. 

Separately, GNS Healthcare aided a leading global pharmaceutical company in the discovery of a diagnostic marker of patient response to an oncology drug, says Hill. Presuming the marker gets validated in a year-long study now underway, it will be used to prospectively stratify patients into phase III trials. Johnson & Johnson is likewise working with GNS Healthcare to optimize its oncology program using analytics to help unravel how its anti-cancer drugs work. 

The National Cancer Institute has GNS Healthcare analyzing gene expression and growth inhibition data for well-known anti-cancer drugs in hopes of understanding how the drugs work and identifying potential targets for combination therapies. In collaboration with Bristol-Myers-Squibb, GNS Healthcare is also working on computer models of inflammatory disease. 

GNS Healthcare additionally has the capability to address the troubling problem of drug toxicity, particularly in cases where a drug is toxic to a certain sub-population or in combination with other therapies, says Hill. Patients routinely ingest powerful drugs with mysterious mechanisms of action. In some infamous cases, blockbuster medicines therapeutically beneficial to multitudes of patients turn out to be deadly for a genetically unfortunate few.