Healthcare Informatics:
 Harvard Medical School: Instructor of Medicine (Jan 12-)
 Mass. General Hospital Lab of Computer Science (MGH LCS): Assistant in Computer Science (Jan 12-).
 Regenstrief Institute: NLM Informatics Fellow (Aug 08-Aug 11), Student Researcher (Aug 07-08, Sep-Oct 11)
 Indiana University: PhD Student in Healthcare Informatics (Aug 07-Sep 11).

Interoperability (software, queries, and data): A recent RAND study suggested that lack of interoperacility is one of the major reasons that the promise of electronic medical records has not met 2005 predictions. I am engaged in several projects to increase interoperability in healthcare systems. One is the SMART Platform, which is developing an iPhone-inspired 'app-store' platform for medical apps, reusable across systems. This is creating an ecosystem for medical app innovation. Another is the Query Health initiative, which is developing a distributed-query platform for population health queries. This platform allows population surveillance by 'sending questions to the data', therefore avoiding many of the data-sharing privacy concerns present in health information exchange. The data formats used by Query Health are also being adopted by the Meaningful Use incentive program. Third, I am involved in the Informatics for Integrating Biology and the Bedside (i2b2) clinical data warehouse and analytics platform, which is an open-source data warehouse in use at over 100 locations worldwide. I believe that all of these tools and more are needed to realize the promise of electronic health records.

Wisdom of the crowd for clinical decision support: Although ample research has shown that individual physciains make errors, collaborative filtering experiments like Netflix and qualitiative and quantitative research into crowd wisdom have demonstrated there is truth to be extracted from heterogeneous groups of decision makers. How can systems utilize the growing mass of healthcare data to: assist physicians directly, to aid in the development of localized guidelines, and to detect changes in practice patterns over time?

Healthcare information systems: Healthcare presents a uniquely complex and time-constrained environment in which massive amounts of information must be presented to clinicians, and interactions with the computer must be fast and efficient. Yet most interfaces are cumbersome and do not utilize data intelligence. Google uses the search interface for applications such as a calculator, and Amazon uses past purchases to customize the display. The EHR could benefit from applications of this kind.


 MIT CS and AI Laboratory (CSAIL): Clinical Decision Making Group (Fall 2006-Summer 2007)

Intelligent listening project: Medical note writing is cumbersome and time-consuming, and documentation requirements are increasing with the advent of EHRs. Additionally, notes are written without patient input, who could sometimes provide valuable insights into their disease which were missed during the interview. One desirable solution could be an 'intelligent office' which would listen to doctor-patient interactions (through microphones and voice recognition software) and interpret these into an editable, interactive draft of the note in real time (through medical natural language processing).


Laboratory informatics:
 MIT Microsystems Technology Laboratory (MTL): Operations Dept. Research Assistant (2002)

Research and design of a module for multi-institution extensible laboratory management software, using Java, SQL, XML, CORBA: "run manager" module, which collects machine-usage data in a flexible, generic, configurable way.


Social informatics:
 MIT Media Laboratory: Epistemology and Learning Group (Fall 1999-Spring 2001)

Utilizing portable electronic devices to produce and analyze social behavior in a playful environment. Design and implementation of kid-friendly language and GUI; collection, analysis, and visualization of social-interaction data.


Software Projects:

  • PeBL exporter: PeBL is a simple, elegant, open-source Bayesian structure learning toolkit written in Python and Numpy. I have written a .DNET exporter, which will save a PeBL network in a format compatible with Netica, Genie, and several other Bayesian inference packages. A conditional-independence chi-squared test (not well tested) is also available here.
  • More to come as I clean up some code from my dissertation.


Design and implementation by Jeff Klann in HTML/CSS/PHP, except the tooltip script and tooltip stylesheet, which are under copyright by The Nucleus Group. Thanks also to Nucleus CMS for inspiring this site's look-and-feel.