Monitoring antibiotic usage: an update

Abstract

Within a hospital, antibiotic usage can be affected by clustering of infections. In addition, pharmacy imprest systems may deliver stock in one time period that are used in a subsequent time period. As a result, hospital antibiotic usage data can be unpredictable and highly variable. Usage of an antibiotic can be conveniently displayed in a time-series chart with monthly defined daily doses (DDDs) per 1000 bed-days on the vertical axis and months on the horizontal axis. To account for random variation, the chart should ideally have control limits, for example a Shewhart chart. However, conventional Shewhart control charts rely on the availability of a run of predictable data values so that the average and its variability can be determined. Since this may be difficult to achieve with hospital antibiotic usage data, a conventional control chart may give misleading information. A modified control chart based on a generalised additive model can overcome the difficulty in the analysis of these data. This chart is not difficult to employ or interpret.