Scores on the five measures of balance and mobility were
obtained before and after participation in the exercise
program. Fall rirk, a surrogated measure defined as the
predicted probability for falling, was calculated for each
patient based on a logistic regression model that was
developed and tested on a different cohort of 44 subjects
(22 older adults with a history of falls and 22 older adults
without a history of falls) (unpublished research). In the
development of the fall-risk model, univariate analyses
were first applied to select variables that could potentially
be used to identify older adults with a high risk for
falling. Potential predictors included demographic variables;
variables related to the subjects' medical and
balance history; current balance and mobility status as
determined by five different clinical tests; and impairments
in specific sensory and motor systems such as
visual, vibratory or touch/pressure sense, range of
motion, strength, and static alignment. Six variables that
emerged to be individually and importantly associated
with faller and nonfaller status were then used in a
stepwise logistic regression analysis. Two variables, selfreported
history of imbalance (IMBALANCE) and performance
on the Berg Balance Scale (BERG), were
identified by the stepwise logistic regression analysis as
highly predictive of falling. The resulting logistic model