"Everything should be made as simple as possible, but not simpler." - Albert Einstein

Seminar Announcement

VA Opportunities and Resources

Robert A. Lew, MAVERIC Research, Department of Veterans Affairs, Boston, MA

Monday, May 2, 2011

3:45 p.m., room 223, Weber Bldg

ABSTRACT

The VA has built a 10-million person cohort with detailed medical histories on veterans over the last decade.   Besides the basic details of diagnoses, outcomes, and procedures, the VA computer record includes prescriptions, laboratory tests, radiology results, and death records.   Recently, programmers have applied natural language processing techniques to extract data from patient charts.   Also, the VA has initiated a project to collect a blood sample from each of a million veterans and thereby add genetic data to the database.   To make the most of this resource, the VA needs statisticians to apply state-of-the-art methods to cohort data that minimize bias and adjust results for non-random missing data.
The same issues arise in clinical trials.    Trials run by the VA can address issues that large pharmaceutical companies might not choose to study.    Therefore, VA statisticians have begun intensive validation of the primary results of VA trials to explore and potentially strengthen the capacity to generalize study results to non-VA populations.

The VA has initiated a novel form of cohort analysis, point-of-care research.   All records at the VA are computerized, including the physician notes on each patient visit.    For a common disease such as diabetes, many physicians are indifferent to two forms of insulin regimen that control blood glucose levels.   Thus, adding only one step to routine care, the option to randomize treatment at the point of care when the physician assigns a treatment, has virtually no cost.   This inexpensive option inserts a scientific comparison into routine practice and potentially creates large cohorts useful for comparing treatments.   The highly integrated computerized VA patient databases and VA software have largely automated the task of creating an appropriate analytic database.   Nevertheless, designing viable point-of-care studies remains a challenge.