Aug 15, 2005 | How do you conduct clinical research in the genomic era? A team of Harvard scientists is building an answer from the ground up. A $20-million project, dubbed i2b2 — Informatics for Integrating Biology and the Bedside — will extract information from the private medical files of some 2.5 million people at Boston hospitals in the Partners HealthCare System.
Other research centers are trying to harness their information assets, but Isaac Kohane, one of i2b2’s directors, believes the project is in a unique position to succeed. Partners has invested in IT for two decades — likely the reason i2b2 won the NIH grant — and also has extensive experience in handling ethical concerns surrounding patient information, he says.
Kohane is a believer in clinical data. He points out that there have been no studies comparing the effectiveness of gene expression data and existing clinical data at predicting patient response to various treatments. “The full benefit of either kind of data does not happen until you combine them,” he says.
While microarrays produce vast quantities of data, collecting a comparative amount of clinical information is much more labor intensive. The process of conducting a classic clinical study is both expensive and inefficient because most of the desired data already exists in the patients’ medical records. The i2b2 team is creating a platform that processes information in clinical records and makes it usable in a research context.
Information passes through three steps. After it has been extracted from clinical records, the data must be validated. For example, if researchers want to know a patient’s smoking status, they could call up the patient, compare several clinical documents, or use Bayesian logic algorithms to extract the information from several different patient records. Next the information is organized in a database and finally displayed in a way that the investigator can interpret.
Research centers around the country have specialized in extracting different kinds of information. For example, Columbia University is very good at pulling
information from radiology reports. Recognizing this specialization, the i2b2 team has created what is called a “hive” infrastructure in which many different Web services can be linked together to create workflows. i2b2 has the potential to link research centers around the country so that they can share data-mining tools.
The project has focused on four disease targets including asthma, hypertension, type II diabetes, and Huntington’s disease. Researchers are forming plans to contact asthma patients for a possible clinical study. “What we want to do is turn the medical record world into a clinical research machine,” Kohane says.
The five-year project, which began in the fall of 2004, is an NIH-funded National Center for Biomedical Computing and was established in response to the NIH Roadmap Initiative.