History of BMIR

During the 2010s some impressive faculty joined Stanford Center for Biomedical Informatics Research.  In 2011, Nigam Shah was recruited; he subsequently became Associate Division Chief.   Manisha Desai, Section Chief for BioStatistics, joined BMIR in 2013 to facilitate the mission of the Quantitative Sciences Unite for collaborative consultation in both biostatistics and biomedical informatics. Purvesh Khatri also joined BMIR that year, in conjunction with his appointment in the Stanford Institute for Immunity, Transplantation, and Infection.  Tina Hernandez-Boussard joined BMIR the following year, exploring work at the intersection of machine learning, natural language processing, and health-services research.  During 2015, BMIR greeted Olivier Gevaert, who brought expertise in multi-scale modeling to BMIR. In 2016, we welcomed Andrew Gentles and Lance Downing.  In 2017, Jonathan Chen came aboard.

BMIR began some important initiatives during this decade.  The Center for Expanded Data Annotation and Retrieval (CEDAR) was established in 2014 to create a computational ecosystem for development, evaluation, use, and refinement of biomedical metadata so that experimental data can be understood, are reusable, and accessible to humans and machines for analysis.  With CEDAR, it becomes possible to learn from online data in new ways because the data are more understandable and more searchable.

Since 2017, Nigam Shah and his team at BMIR have been offering the "Green Button" Clinical Informatics Consultation Service that helps clinicians learn about patients who have similarities to a patient whom they are treating, allowing the clinicians to inform their decisions with knowledge of the outcomes of patients whose data are available from electronic health record (EHR) systems.

Analysis of EHR data emerged as a major theme broadly at this time.  Tina Hernandez-Boussard, for example, used EHR data to drive policy decisions (for example, by elucidating the effects of commonly prescribed antidepressants on patients’ response to treatment with opioids).  Jonathan Chen gleaned patterns in physician prescribing behavior from archived EHR data in order to discern possible inadvertent omissions in future sets of physician orders.

Olivier Gevaert explored news computational techniques for characterizing tumors by combining data from many different information sources.  Andrew Gentles also addressed the challenges of integrating diverse types of data at different scales, focusing on high-throughput ’omics data, as well as more traditional clinical data sources.

In 2005, BMIR became home to the National Center for Biomedical Ontology (NCBO).  The NCBO is a project that was supported by the NIH program of National Centers for Biomedical Computing.  The NCBO developed a repository of most of the world’s publicly accessible biomedical ontologies and terminologies, and built tools and Web services that enable biomedical scientists and clinicians to put those ontologies to use.  Among the most widely used services is the NCBO Annotator, which Nigam Shah conceived as a mechanism for the automated association of textual data with appropriate controlled terms.  The NCBO Annotator has become the foundation for ongoing work to perform text mining of patient-record data to identify evidence of adverse drug events and of other indicators reflecting the quality of medical care.

The group continued to build on the successful research in clinical decision support systems that had emerged during the SUMEX-AIM years.  The 1990s saw the development and deployment of a major decision-support system for the management of patients living with HIV infection known as T-Helper—a project that spawned Amar Das’ development of the Chronus temporal data management system and Yuval Shahar’s work on Résumé, a system that used application-specific knowledge to abstract patient data into meaningful temporal intervals.  Résumé allowed the computer to process medical record data to identify intervals of disease and episodes of toxicity in a manner that had not been possible previously. T-Helper provided a domain-independent infrastructure for offering clinical decision-support for protocol-based therapy known as EON.  EON became the foundation of the ATHENA suite of decision-support tools for guideline-based care of patients with chronic diseases.  Developed in collaboration with Mary Goldstein, EON has been installed at several VA medical centers.  More than two decades later, ATHENA is probably the clinical decision-support system based on AI techniques that has been in continuous operation in a clinical environment for the longest period of time in the world.

During the 1980s, Ted Shortliffe’s work on ONCOCIN led to an intelligent system to provide decision support to clinicians caring for patients who had breast cancer or lymphoma. The problem of building knowledge bases for ONCOCIN to encode the procedures for applying cancer chemotherapy protocols led to Mark Musen’s early work on the Protégé system, which originally was developed to acquire knowledge about cancer chemotherapy.  The challenge of reasoning about the temporal relationships among patient data values and therapeutic interventions led to Michael Kahn’s work on a system known as TOPAZ.

During this time, Greg Cooper had identified ways to overcome limitations of naïve Bayesian reasoning, and led a group of students who investigated new approaches to building and applying of Bayesian belief networks. This pioneering research offered some of the first practical demonstrations of belief-network technology, which is widely used to perform reasoning under uncertainty.

At this time, the Medical Computer Science group was supported by the Stanford University Medical Experimental Computer for Artificial Intelligence in Medicine (SUMEX-AIM), the brainchild of Joshua Lederberg and Ed Feigenbaum in the 1970s, which had become the major national resource that sustained research on intelligent systems in biomedicine for more than 15 years.

In 1988, the Medical Computer Science Group reorganized to become a section of the Division of General Internal Medicine of the Department of Medicine. In 1992, it became its own freestanding division. The Stanford Center for Biomedical Informatics Research (BMIR) adopted its current name in 2007 to emphasize its broad scope and its emphasis on integrating the breadth of biomedical informatics research.

Stanford was the origin of much of the early excitement about AI in biomedicine in the 1960s and 1970s.  In the 1960s, computer scientist Edward Feigenbaum, biochemist Carl Djerassi, and Nobel laureate Joshua Lederberg joined forces to create Heuristic Dendral, a computer system that could interpret data from mass spectrometry.  The remarkable success of the Dendral project led to other, foundational work in AI technology, including the MYCIN system for recommending empirical treatment of patients with presumed bacteremia and meningitis, and E-MYCIN, a rule-based “shell” that spawned a whole industry devoted to the creation of intelligent computer systems.  Work in the 1970s on computer systems such as MYCIN provided strong evidence that biomedical knowledge could be modeled computationally and processed to inform clinical decision making.