Bioinformatics, meet medical informatics. Medical informatics, meet bioinformatics.
Perhaps the introductions haven’t been quite so obvious, but the National Institutes of Health is playing the role of go-between in aligning two research communities that have much in common but often seem to exist on different planets. Informatics could be the driving force behind a new age of personalized medicine, according to NIH officials, but that will require integrated IT platforms that encompass everything from genotype to phenotype, including DNA sequences, clinical trials information, and patient records. If informatics is indeed the key to next-generation clinical care, IT professionals at either end of the spectrum will have to meet somewhere in the middle.
To speed up the process a bit, NIH last week hosted a symposium at its Bethesda, Md., campus called “Biomedical Informatics for Clinical Decision Support.” Notably, it was the first such meeting to be jointly hosted by NIH’s BISTIC (Biomedical Information Science and Technology Consortium) and BECON (Bioengineering Consortium). The meeting’s stated purpose was to focus “primarily on the software tools and approaches” that will enable point-of-care biomedical IT, but for many attendees, the task at hand was sociological, not technical. As Isaac Kohane, director of informatics at Children’s Hospital Boston, explained, most bioinformaticists view clinical informaticists as unimaginative record-keepers in charge of medical records and order-entry forms, while clinical informaticists see bioinformaticists as blue-sky researchers running “billion-dollar projects with very little to do with medicine.”
This “unnecessary separation” between bioinformatics and medical informatics has a real-world impact on the scientific community, said Ted Shortliffe, a professor of medicine and computer science in the department of biomedical inform-atics at Columbia University Medical Center, where he oversees a team with backgrounds in imaging, public health, and biology. Biologists, Shortliffe said, “feel uncomfortable” on a medical informatics team, while clinicians have the same feeling toward the term bioinformatics. His solution? A classic compromise: “We need to be inclusive by changing the name of the field to biomedical informatics.”
Sociology may be partially to blame for the gulf between the bio- and medical informatics communities, and if Shortliffe has the right idea, perhaps semantics will go part of the way toward narrowing it a bit. But there are also very real technical challenges in the way of bench-to-bedside informatics systems.
The primary hurdle comes from integrating several levels of phenotypic data with molecular-scale information. While the analysis and management of genomic data still presents a number of unmet challenges, sequence and expression information is not nearly as complex as image data gathered from cells, tissues, and organs. In addition, the ultimate goal of throwing data from clinical trials and patient records into the mix to enable real diagnostics throws up security and access hurdles that don’t exist with pure research data.
Image analysis in particular presents a thicket of thorny problems. Ricardo Avila, project manager for computer-aided diagnosis at General Electric Global Research, discussed the “software gap” that currently exists between imaging technology and the ability of clinicians to get the most out of it. In 1996, he said, a CT scanner took 26 seconds to produce 30 images. By 2004, it took only 2.6 seconds to produce 480 images at a much higher resolution. By the time the software developers create accurate decision support algorithms, he said, “the world has moved on.”
Another problem for image informatics is the lack of a reference database to train and verify diagnostic algorithms. Maryellen Giger, chief of radiological science at the University of Chicago, said that her team trains its diagnostic algorithms by comparing a new tumor image to a small internal reference library of images — an approach that many groups currently use, she said. However, the problem is that these data sets are usually too small for adequate training, and researchers are unwilling — or unable — to share their data. Giger and a number of other researchers at the conference urged NIH to support a national imaging archive that would serve this demand.
NIH is already taking a step in that direction. Officials from the Foundation for NIH said last week that $3.7 million has been earmarked for a 16-month pilot project that will collect between 1,000 and 2,000 CT lung studies in a single archive with the goal of “supporting the development, independent testing, and PMA [premarket approva]l of software products.” The demonstration project will build on an existing NCI project called the Lung Image Database Consortium (LIDC), which supports five academic centers that are creating a database of spiral CT lung images.
NIH is also supporting a number of other projects to bring together biological and clinical data. NECTAR (National Electronic Clinical Trials and Research System) — a component of the recently introduced NIH Roadmap — is a proposal to create a “network of networks” for clinical trials that will marry genomics data with biomedical imaging data and patient medical records.
The National Cancer Institute Center for Bioinformatics’ caBIG (Cancer Biomedical Informatics Grid) project is also tackling the challenge of integrating genotype and phenotype [BioInform 11-03-03] and some of the features the NCICB has developed for this framework are being put to broader use in another project led by Laura Esserman, director of the Carol Buck Breast Care Center at the University of California, San Francisco. The project, called I SPY TRIAL (Investigation of Serial Studies to Predict Your Therapeutic Response with Imaging and Molecular Analysis), is a multi-center trial of serial imaging and biopsy for women with tumors at least 3 cm in size who are undergoing neoadjuvant chemother-apy. Assays include immunohistochemical markers; genomic, expression, and proteomic arrays; and serum for proteomics and protein levels. The goal of the project, said Esserman, “is to identify women likely to have a poor outcome at the time of diagnosis in order to introduce novel therapeutics early in treatment.” As of June 18, 107 patients had registered for the trial, Esserman said.
The NIH meeting was timely. Earlier this month, it was reported that the International Council of Medical Journal Editors — a group of 12 journals that includes the Journal of the American Medical Association, the New England Journal of Medicine, The Lancet, and the Annals of Internal Medicine — is considering a proposal to require all clinical trials data to be deposited in a public database as a prerequisite for publication. Michael Vannier, a professor of radiology at the University of Chicago, drew a parallel between clinical and genomic data, noting that bioinformatics advanced rapidly once the major journals required submission of sequence data. If clinical data becomes as accessible as genomic data, it will certainly spur development of new analytical tools — although getting these developers to communicate with bioinformaticists, and vice versa, may take a bit longer to accomplish.