Implementing expert systems into ongoing data analyses

The POSCH Study is a recently completed national multiclinic clinical trial (Buchwald et al., 1990). POSCH collected about 1400 variables at each annual visit on 838 participants, producing a database of about 300-million characters. The size inspired the use of innovative ways to automate the analyses (Long, 1987; Long at al., Long et al., 1991a). We previously reported the development of two expert systems that automate certain steps of data analyses not previously possible because they required clinical judgment (Slage at al., 1986a; Long et al., 1988a; Long et al., 1987; Long et al., 1991b;). This report is an expanded and updated of our experience implementing these systems into our day-to-day operations (Long et al., 1988b). Issues discussed include hardware selection, use of LISP versus FORTRAN, other software considerations, statistical concerns, database access and several other mundane issues, the resolution of which were essential to the actual use of these expert systems.