Abstract

Background: The traditional approach for assessing drug safety is to compute the baseline to end of treatment change in a laboratory measure. This strategy may miss important time-dependent value trends in patient subsets during or after treatment. We describe a strategy for evaluating time trends in laboratory parameters and applied the approach to platelet (plt) count analyses during linezolid (lzd) therapy.

Methods: Plt counts obtained before, during and after treatment were available from 2789 patients enrolled in 7 Phase III trials comparing linezolid to comparator antibiotics. Box plots (25th to 75th percentiles) and whiskers (5th to 95th percentiles) of platelet counts were superimposed on scattergrams (< 5th and >95th percentiles) and normalized to start of treatment for each group. Additional plots were further developed to investigate patients with substantially low values defined as < 75% of the lower limit of normal or < 75% of baseline if abnormal at baseline.

Results: The median plt count rose from baseline but remained within normal limits during treatment; this trend was similar in comparator patients. Twenty-seven (1.9%) patients on lzd had at least one substantially low value compared to 11 (0.8%) in the comparator group. These patients were more likely to have had a low baseline plt count, regardless of treatment group. The mean (SD) time to first occurrence of a substantially low plt count was 12.0 + 6.2 days for the lzd patients and 11.3 +10.1 for the comparator group.

Conclusions: Examination of the time course of laboratory values with boxplots allows for detection of trends and development of risk assessment models for laboratory indices. This comparative method also allows for the identification and tracking of outlier performance.

Interscience Conference on Antimicrobial Agents and Chemotherapy (ICAAC); Toronto, Ontario, Canada; September 2000

By Jon Bruss, Ed Antal, B Hafkin, B Cirincione, S Sardella, M Redman, Elizabeth A Ludwig, Thaddeus H.  Grasela