As part of a planned $25 million investment over the next 10 years, Hewlett-Packard will support the development of bioinformatics tools to help analyze clinical data stored in electronic medical records at the Lucille Packard Children's Hospital at Stanford University.
The investment, announced last week, will support treatment and intervention efforts and expand the hospital's facilities in Palo Alto, Calif.
It will also provide resources to explore different kinds of analysis that can be done on clinical data stored in EMRs, Atul Butte, director of the Center for Pediatric Bioinformatics at LPCH, told BioInform.
Harry Kim, senior director of healthcare at HP enterprise business, told BioInform that HP will provide the hospital with a mix of hardware, financial support, and informatics expertise and capabilities.
He added that the partners are still discussing how the funds will be split between the expansion and research efforts.
The grant will help sustain and advance interdisciplinary research collaborations between physicians at LPCH and a team of six to eight scientists at HP Labs – the company’s central research arm.
With the aid of bioinformatics tools, the project team will investigate issues such as patient safety and personalized care, HP said.
This project builds on a prior working relationship between HP and LPCH that began in 2007, in which HP donated more than $1 million worth of technology and funding to the hospital. The partnership is aimed at "driving innovation that matters" and could potentially have a larger economic benefit for the company, Kim said.
"We fundamentally believe that if we drive and demonstrate innovation, there is a business economic return that follows both in the context of our business relationship with LPCH and Stanford as well as taking that innovation and capability and bringing it to the rest of the world," he said.
Previous investments from HP have enabled LPCH investigators to study environmental factors in disease development, personalized cancer therapy, and genome sequencing.
Improve Care; Do No Harm
The Obama administration has called for all doctors offices and hospitals to adopt EMRs by 2015, which means that there will soon be "tens of thousands of hospitals in the country with huge repositories [containing] millions up to hundreds of millions of measurements on human organisms," Butte said in a conversation with BioInform.
While some of the more obvious uses for that data would be to "improve the quality of care" and "to try to prevent harm [such as] medical errors" in hospital settings, Butte said, "a fun part is actually going to be thinking about what kind of science we can derive from these human organism measurements — new biomarkers, new ways to think about drugs, how drugs might be working on diseases that we don’t know about."
He added that "all of that is enabled by this HP donation."
Clinical information in these records includes details on biomarkers, cell counts, pathology and radiology reports, and second-by-second measurements from intensive care units, among other data.
Furthermore, the researchers have some genetic data stored in the EMRs that is currently being used in pharmacogenetics-based studies in oncology at Stanford.
Another consideration is the fact that next-generation sequence data is looming on the horizon. At nearby Stanford University School of Medicine, NGS technologies have already been used to sequence data from hundreds of individuals, Butte said. While NGS data will not likely be a part of the HP project, "we anticipate that kind of research getting connected" in the future, he said.
Meanwhile, there is a growing awareness of the role the environment plays in disease susceptibility, Butte said, adding that it may have an even larger role than genetics.
"I think the future is now to think about how we can figure out the environment that these patients are in, in an automated or in a computable way," Butte said. He added that this "kind of research is going to be brand new and enabled by [HP's contribution]."
The project will require the development of data-mining software to tap into the wealth of information in EMRs. Tool development for clinical data "is still in its infancy," Butte said, explaining that while the molecular world benefits from open source packages like R and Bioconductor, clinical data doesn’t have such tools readily on hand.
Furthermore, he noted that there aren’t many open source middleware options that would allow researchers to access information stored in databases via analytical programs. While there are standard ways to connect tools like R and Bioconductor, Butte said these methods don't work as well in the complex environment of hospital data centers.
As an illustration, Microsoft's Open Database Connectivity interface — which provides a standard software interface for accessing database management systems — offers a "pretty naïve connection between an analysis program and a database" if a user needs to go through a federated system or issue queries that get mapped to other queries, he said.
Other issues that the project will need to contend with include data de-identification as well as patients who don't want their information to be used, even if it is scrubbed of any data that could be traced back to them.
Compute, Data, and Healthcare Delivery
For this project, Butte said the team plans to tweak some open source resources to work in the clinical setting, as well as develop some new software.
For example, his lab plans to develop middleware, new packages for R, and some visualization schemas to help users analyze clinical data. Some other tools that will likely play a part include MySQL, PostgreSQL, and Bioconductor
On the data de-identification front, Butte and his colleagues will develop tools to anonymize the data, but will also use previously published methods of accomplishing the same task.
Although he could not give specifics, HP's Kim said the company will likely provide personal computers, laptops, and monitors for the project, as well as back-office infrastructure with an eye towards providing resources that can be used to create "intelligent workflows" and to "bring services capabilities to drive integration of information."
In addition, HP can provide security software that could be used to ensure data safety and maintain patient privacy, all of which are necessary considerations.
Sharad Singhal, a researcher and distinguished technologist at HP Labs, said that his team, which has expertise in data analytics, will work jointly with doctors to ascertain which algorithms are needed, optimize them, and in many cases implement the results in the hospital so that doctors only have access to patient data that is considered appropriate.
"In many cases what we [find] is that it's not that [researchers] need a large amount of computational power, it's simply that [they] need to understand how to go about doing things," he told BioInform. He added that the type of analysis required and the amount of data involved also determines the computational cost of a given research question.
Singhal highlighted an ongoing pilot project involving researchers at HP Labs and physicians at LPCH that aims to determine whether an effective marriage between analytics and medicine can lead to improved outcomes for patients at risk for heart attacks.
As part of the pilot, the teams developed an automatic system that was used to predict which patients were at risk of "decompensating" 24 hours ahead of time based on information stored in electronic documents such as physician orders and notes and nursing documentation.
Chris Longhurst, chief medical information officer at LPCH, told BioInform that the development team extracted laboratory and clinical event data from 27 patients whose conditions worsened during the study period and used machine-learning techniques to compare this data to data from 3,500 controls to obtain "some sensitivity and specificity for all the different pieces of data in the system in terms of a predictive value for decompensation."
For the study, the team collected data on 85,091 clinical laboratory tests of 1,105 types and 69,795 clinical events of 2,769 types.
Using funds from a separate HP grant received last year, the development team, comprised of Longhurst, Butte, and others, is working on a patient-centered EMR dashboard tab that provides critical clinical data and ensures compliance with evidence-based protocols for each patient.
The team has also developed a multi-patient dashboard that will provide a "unit-wide view" of critical patient information. These tools are expected to provide real-time information to clinicians about their patients.
Moving forward, Longhurst expects to continue working on the system as well as to participate in other projects under the new grant.
While these projects haven’t been fully defined yet, "I think it's fair to say that at a high level, [they] will involve leveraging HP expertise in analytics to drive real-time data-driven improvements in healthcare and healthcare delivery," he said.
Have topics you'd like to see covered in BioInform? Contact the editor at uthomas [at] genomeweb [.] com.