Affordable Access

Publisher Website

Central statistical monitoring in clinical trials

Springer (Biomed Central Ltd.)
Publication Date
DOI: 10.1186/1745-6215-12-s1-a55
  • Poster Presentation
  • Design


Central statistical monitoring in clinical trials POSTER PRESENTATION Open Access Central statistical monitoring in clinical trials Amy A Kirkwood*, Allan Hackshaw From Clinical Trials Methodology Conference 2011 Bristol, UK. 4-5 October 2011 Background On-site monitoring is a common but time-consuming and expensive activity, with little evidence that it is worthwhile. Centralised statistical monitoring (CSM) is a much cheaper alternative, where data checks are per- formed by the co-ordinating centre, reducing the need to visit every site. Although some publications have out- lined possible methods, few have applied them to data from real clinical trials. Methods R-programs were developed to check data at either the patient or site level, for fraud or data errors. These included finding anomalous data patterns, digit prefer- ence, rounding, incorrect dates (eg weekends/holidays), values of variables too close or too far from the means, odd correlation structures and extreme values or var- iances. We applied these to 3 trials: (i) where data had already been checked, (ii) an ongoing trial where our findings could be checked in real-time, and (iii) where data errors and fake patients were created. Findings The programs were designed to be run automatically and produce simple tables or figures. Few errors were detected in the trial where data had already been checked (as expected). Most data errors were found in the two other trials. The programs were able to detect data errors, as well as fabricated patients that we gener- ated to have values that were too close to the multivari- ate mean (fig. 1). They also detected centres that had too few or too many serious adverse events (fig. 2). It might be difficult to reliably apply some of the programs to centres with few patients. Several patients that were fabricated were not detected because the data did not follow the assumptions used by the R-programs, or the number of fabricated patients within a centre was too small. Examples

There are no comments yet on this publication. Be the first to share your thoughts.


Seen <100 times