professional staff working with the data in the DMSS databases and their productivity. However, the committee also counsels great caution in the use of approaches that use such data collected through automated systems for signal generation. As expected by chance alone, the rates of several diseases and conditions will predictably appear to be elevated in one group or another. Although random error and bias are likely explanations for these increases, other conclusions might also be drawn. In other words, these preliminary findings should lead to further examination of the data. The current DoD approach and organization focus on screening DMSS data for hypotheses. DoD should, however, devote more attention and resources to the evaluation of these hypotheses, as was begun in response to the committee’s inquiries. As has been articulated in a set of good epidemiology practices developed for use with similar administrative and clinical data sets in civilian practice (Andrews et al., 1996), analysis of such data requires the exercise of great caution and a commitment to devote the necessary resources to explore the possible associations that might surface from such exercises. Chapter 8 discusses recommended improvements for use of DMSS data.

Thus, finding an increased rate of occurrence of one or more adverse events must be considered a signal until proper review provides an alternative explanation. Criteria for determination of which signals should be further evaluated need to be developed and routinely applied. At a minimum, a system for retrieval and review of primary medical records is required to be able to rule out coding and classification errors, to search for subtle but possibly explanatory variables that may confound an association, or to differentiate a true signal from a statistical chance event.