~ Abstracts

Cathepsin D, the urokinase-type plasmogene activator (uPA) and its inhibitor (PAl- 1) in tomour tissue extracts for prognostic significance in relation to relapse and survival.

For this screening procedure a new statistical method, the CART-technique (= Classification _and Regression Tree) has been applied. The idea of this technique is to split the whole-sample into two different groups with greatest difference in prognosis. For this split all variables with all possible divisions into two groups are considered. The split-criteria is the maximum log-rank test. After a split is performed, both groups are analyzed independently in the same way until the size of a group becomes too small or the test is no longer significant.

The major new findings are that node-negative breast cancer patients with low values in both parameters do have a good prognosis (no relapse in 30 patients).

Only those patients with high value in at least one of both parameters are considered as potential candidates for an adjuvant therapy.

I'41 EQUIVALENCE TESTING USING AVAILABLE

STATISTICAL SOFTWARE

Jose Ma. J. Alvir Hillside Hospital

Glen Oaks, New York

Testing whether a new drug is 'at least as good as' a standard comparison drug is not readily addressed by statistical procedures implemented in popular softwarepackages such as SAS, SPSS, and BMDP. Procedures in these packages are generally desigued to test for differences and not similarities between treatment groups. It is well known that failure to reject the null hypothesis of no differences between groups does not provide evidence that two treatment are indeed equivalent. Determining equivalence typically entails arbitrary specification of a difference (m) between the new and standard

drugs. A represents the maximum loss of efficacy allowed for the new drug, beyond which the new drug is deemed inferior. A one-sided eordidenee interval for the observed difference between treatments provides a clear picture of whether the true difference between new and standard treatments can be expected to exceed A . Note that while lack of statistical power can lead to failure to reject the null hypothesis of no treatment difference, this same lack of power will only serve to widen the confidence interval for the observed difference and thus make it more difficult to infer equivalence. Confidence intervals, while generally not provided in the standard output of procedures in commonly used statistical software, are easily computed using the results of these procedures. Samples of program code which compute eordidenee intervals for differences between means and proportions as well as odds and hazards ratios will be demonstrated. These samples, written in SAS ®, attempt to take full advantage of output data sets produced by SAS so as to avoid the need for manual entry of parameters necessary for computing confidence intervals. Examples will also be provided for cases wherein two-sided confidence intervals are used to determine whether a new treatment is 'similar to' a standard treatment. In such cases, one specifies A 1 as the maximal loss and A 2 as the maximal

gain of efficacy, with A 1 and/~2 not necessarily equal.

1'42 THE EFFECT OF DRUG EXPOSURE ON INCIDENCE

RATES OF ADVERSE CLINICAL EVENTS WITHIN ANTIBIOTIC TRIALS

Victoria Rutkiewicz and Barbara Conetta Bristol-Myers Squibb

WaUingford, Connecticut

Integrated safety reports within the antibiotic New Drug Application (NDA) include the investigation of on-study adverse experiences: adverse clinical events (ACE), laboratory abnormalities, drug intolerance, and deaths. Most frequently, the cumulative incidence rates based on the tallies of ACE within the treated population are reported for short-term acute treatment trials. While this information is a necessary component of the NDA from a labelling and regulatory perspective, it may not be a complete representation of the safety profile of a new drug. Appropriate clinical perspective for events with increased incidence necessitates analysis of a model that includes length of exposure of the drug; an exposure that varies among patient groups.

Kaplan-Meier cumulative incidence estimates and curves using life-table methods permit an appropriate framework to account for variability in treatment duration. A related logrank test enables