By Aaron Krol

April 13, 2016 | As the final keynote session of last week’s Bio-IT World Conference & Expo drew to a close, a member of the audience that filled the amphitheater of the Seaport World Trade Center in downtown Boston asked the three panelists onstage a pointed question. “How do you see what you’re doing right now… eventually translating into community care?” she said. “How is it going to get to the community hospitals? How is that knowledge and analytics actually going to trickle down to the healthcare system that we have now?”

For the past hour, the keynote speakers―Yaron Turpaz, Chief Information Officer of Human Longevity, Inc.; Catherine Brownstein, Manager of the Molecular Genomics Core Facility at Boston Children’s Hospital; and Bill Evans, Chief Marketing Officer at IBM Watson Health—had described ambitious projects to bring together genomic and clinical data on the scale of whole populations. Yet only Brownstein had done this in the context of general patient care. As extraordinary efforts get underway to link up vast, distributed datasets and to profile patients’ health more deeply than ever before, the sponsors of this work are also calling on extraordinary resources: perhaps putting the results out of reach of the average physician or patient.

A defining challenge of precision medicine will be to vault this gap, and all three speakers had subtly different takes on how it will be done.

The Multi-Omics Titan

Turpaz, who opened the Thursday morning session, represents a grandiose vision for personalized healthcare. His employer, the J. Craig Venter-led Human Longevity, Inc. (HLI), operates the largest genome sequencing center in the world, gearing up to produce 40,000 whole human genomes a year; Turpaz revealed that they’ve cranked through roughly 22,000 already. Even more ambitiously, the company also absorbs huge volumes of non-genomic data on the patients it enrolls: medical records from hospital partnerships, personal health data provided by participants, and new experimental data including MRIs, microbiome surveys, and “metabolomics” profiles.

As CIO, Turpaz oversees the HLI Knowledgebase, a monumental R&D platform built around this data that received the Judges’ Prize in the Bio-IT World Best Practices Awards at the expo. (See also his interview with Bio-IT World from July.) Yet, Turpaz said in his presentation, HLI did not originally set out to build a custom operation on this scale.

“We were really trying, two years ago when we were established, not to reinvent the wheel,” he said. At times, HLI considered licensing or outsourcing everything from analytics, to storage and data management, to the sequencing itself. Ultimately, potential partners proved not to have systems in place to deal with multi-omics data collation on the scale of tens of thousands of patients a year. In designing the HLI Knowledgebase, Turpaz’s team had to not only reconcile and label this data to keep uniform records on individual patients, but also set up systems for mass, organization-wide upgrades―such as a recent one that realigned all HLI’s sequencing data to the latest build of the human reference genome.

The achievement is a serious one, but it does run headlong into the problem of access. “Once we have this information,” Turpaz asked, “how many clinics are there who can actually use this information to make a review of your health and take action on it? Just from the informatics side, not thinking about the element of training and education of physicians?”

HLI is working with data that would be exotic even in isolation: there are vanishingly few standards, for instance, around using a patient’s microbiome to make medical decisions. The volume of data HLI brings together is even more problematic. While physicians can rely on a number of professional guidelines to make good use of limited types of genetic data, such as risk factors in cancer or drug-gene relationships, almost no one in clinical practice has access to systems for parsing a whole genome.

HLI now runs its own clinic, inside its San Diego headquarters, where it gathers comprehensive “Health Nucleus” profiles and offers health assessments under a research protocol. This clinic, according to Turpaz, can now process an astonishing six to eight new participants a day (although he did not say whether it was consistently keeping up that volume). In addition to fueling R&D efforts, the Health Nucleus has evidently produced some remarkable medical results. Turpaz described one patient with colon cancer whose doctor gave her three weeks to live. “When something like this comes in, everybody at the company focuses on what we need to do to generate this data in no time,” he said. Sequencing the patient’s tumor uncovered a plausible but non-standard treatment pathway that had not been tried, which HLI suggested to the patient’s oncologist.

“She is now in full remission,” Turpaz said. “Now we take, on a weekly basis, a liquid biopsy to make sure nothing relapses.”

This is the kind of result the national drive toward precision medicine is designed to make routine. For now, however, admission to HLI’s Health Nucleus runs customers a tidy $25,000, which is almost certain not to be reimbursed by insurance. As tailored health profiles with actionable information emerge, they seem destined to be geared toward the interests of the wealthiest. And to the extent that this fraction of the public differs, in its demographics and clinical history, from the general population, that also has an impact on research based on their data.

Turpaz believes the gap is only temporary. “Absolutely, the price is high,” he said in answer to an audience question to this point. “For us, it’s a proof of concept. What can be done? We want the Health Nucleus as a demonstration of how important is this concept of preventive health… The assumption is that once people realize the value, including the cost-saving elements of this preventive medicine, then insurance companies will pick it up.”

He also offered praise for Discovery Health, an insurer in South Africa and the UK, which is beginning to offer exome and genome sequencing to its members—in partnership, naturally, with HLI.

Genetics in the Trenches

Catherine Brownstein, from Boston Children’s, doesn’t have the resources of HLI at her disposal; taking the stage after Turpaz, she joked about “taking it down a notch.” Yet the project that formed the core of her talk at the Bio-IT World Conference illustrates how, with dedication and some savvy about population health, geneticists working on individual cases can provide benefits that go far beyond their own patients.

Working in pediatric genetics, Brownstein often sees children with mysterious disorders who have gone through multiple specialists and misdiagnoses before being referred for some form of genome-wide testing. In her presentation, she recalled a particular pair of cases that, at first, were not obviously connected.

The first patient, a nine-year-old boy, had been suffering from hallucinations since age six, along with moderate developmental delays. The second, a five-year-old girl, had developed hallucinations even earlier, but also dealt with seizures and more severe developmental delays. Sequencing revealed that both had copy number variations in the same region of the genome, 16p13.11, although the boy had a copy gain and the girl a copy loss.

It was at this point that the population health-based approach at Boston Children’s distinguished itself from a standard diagnostic process. “The electronic medical record isn’t designed for genomic queries,” said Brownstein. “At Children’s we’re lucky, in that we spun out our clinical laboratory, and in conjunction with Life Technologies and now WuXi NextCODE, formed Claritas Genomics, a provider of genomic services. So we’re able to ask Claritas… how many children have aberrations at 16p13.11, and get the list of their MRNs [medical record numbers].”

The immediate result was that Brownstein’s team found three more families, who had undergone genetic testing for a variety of reasons, carrying copy number variants in this region—two of which had family members with suggestive symptoms. “Mental health disorders are quite unique,” she said, “because the symptoms can often go unnoticed,” as physicians struggle to deal with more acute problems like epilepsy. System-wide checks like the one Claritas conducted can sweep in patients whose underlying genetic conditions are missed in the broader context of their care. (In this case, unfortunately, the diagnoses were made too late to forestall a pair of psychotic breaks in a father and son with relevant mutations.)

In the longer term, too, a population health approach to genetics can inform the standard of care. “For those symptoms that could indicate a 16p aberration, we can do mental health screening,” said Brownstein. “In epilepsy or some of the other clinics, we can ask questions that are actually quite sensitive for determining whether a mental health screening is warranted, and put that on the intake form.” Measures like this are now being implemented at Boston Children’s as a result of finding these patients, several of whom have responded well to medication now that their underlying disorder has been diagnosed.

During audience questions, Brownstein offered a big technological reason for optimism that efforts like hers will have broader impact in the future. “Precompetitive sharing has gotten much better in the past five years where I’ve been working,” she said. “Being able to have genomes on the cloud is huge, and for hospitals to embrace that finally, not having to ship hard drives, is amazing.”

“There’s also just a culture shift,” she added. “People are beginning to understand that you can’t hoard your data. You might hoard and have nothing, while if you share it you will find something. It’s beginning to sink in.”

The Intelligence Machine

The final speaker, Bill Evans, began his presentation with a personal story. Evans—not to be confused with either the inimitable jazz pianist or the Boston Police Commissioner—had a health scare about five years ago, experiencing frequent heart palpitations and occasional fainting spells. Swiftly working his way through more than a dozen specialists and racking up both diagnoses and medications, he had accumulated a large amount of health data on himself, as his symptoms worsened over seven months.

“Being a technology nerd, I collected all my data,” Evans said. “I had all my radiology scans on my iPad… I had every single blood test and workup printed out.” Eventually, a friend referred Evans to a gastroenterologist, whom he visited with his reams of personal information. “I sat there with him for 45 minutes, giving him every single record, every single scan, every single piece of paper. And in two minutes he said, ‘Here’s your problem. You have irritable bowel syndrome, and I’m going to put you on a low dose of an anti-anxiety mediation.’ And I’ve never felt better.”

“I didn’t have a big data problem,” Evans concluded. “I had a small data problem. I had plenty of analytics, and I had no wisdom.”

Today, Evans works with IBM’s Watson Health division, perhaps the hottest application of Big Blue’s famous artificial intelligence platform. Watson Health blends natural language processing, training in “grammars” like gene pathways and drug mechanisms of action, and the same kinds of massive datasets HLI is eagerly ingesting, to explore areas of medicine where machine learning might lead health practitioners to new approaches. Watson has been deployed in oncology, in pediatrics and rare disease, and in clinical trial matching, all in pilot programs that IBM hopes will eventually advance to active aids in the practice of healthcare.

Evans stressed that the promise of Watson is not so much that it will collate data for population-scale insights, but that it will develop some “intuition” for what information is truly relevant to a particular problem or a particular patient—the “wisdom” that was missing from Evans’ own diagnostic journey.

“If you’re a radiologist,” he said, “on an average day you read 100,000 scans. There’s a high margin of error, a high fatigue rate. Watson will analyze that data, mark what’s significant, prioritize those patients, and get that clinician back to consulting with his patients.” (IBM last year acquired the medical imaging company Merge Healthcare to train Watson on exactly these types of applications.)

Evans also wants Watson Health to tackle problems caused by misaligned incentives and information gaps. One frontier for the technology is drug repurposing: learning whether cheap, already approved generic drugs could be usefully applied in new disease areas. Generic drug manufacturers don’t have access to data on the patients they serve, their outcomes, or their comorbidities, Evans pointed out, but in collaboration with pharmacists and hospitals, Watson Health might.

From Evans’ perspective, the challenge of making precision medicine an everyday and not a boutique service is less about mustering the resources to exhaustively profile every patient, and more about moving machine learning to places where useful data already exists, but may not be well-structured or easily available for analysis. The integration of Watson with the major electronic medical record vendors is one way IBM could rapidly expand into community health networks. By releasing free APIs to developers working with Watson services, IBM also intends to regularly open up its cognitive computing capabilities to new applications—although implementing these programs at scale is likely to remain a barrier for small clients.

Whatever the long-term prospects for precision medicine, all three speakers shared a conviction that patients want to contribute to building a more intelligent, personalized healthcare system. At one point, an audience member asked about privacy concerns around collecting and sharing genomic data, and whether patients should worry about how their data could be misused—a major topic at the Bio-IT World Conference two years ago.

Brownstein, however, said her own patients are eager to see their data used to its full potential. “The interventions that are going to be benefiting from these new types of technologies are going to benefit the very sickest patients first,” she said. “I abstractly understand the privacy concerns, but in the hospital setting, I don’t think we’re going to deal with them for quite a while, because the benefits so greatly outweigh the risks.”

Turpaz agreed, and added that it is the personal drive of patients to see the full picture of their health that provides researchers with the fuel for broader insights. “The sicker the individual or the loved one, the more they’re willing or their family is willing to share and contribute to research,” he said. “We have to have this vision of how valuable it is to have knowledge integrated. Large-scale knowledge, as well as the n-of-one, the individual.”