Central data monitoring in the multicentre randomised SafeBoosC-III trial - a pragmatic approach

Abstract

Background: Data monitoring of clinical trials is a tool aimed at reducing the risks of random errors (e.g. clerical errors) and systematic errors, which include misinterpretation, misunderstandings, and fabrication. Traditional 'good clinical practice data monitoring' with on-site monitors increases trial costs and is time consuming for the local investigators. This paper aims to outline our approach of time-effective central data monitoring for the SafeBoosC-III multicentre randomised clinical trial and present the results from the first three central data monitoring meetings.

Methods: The present approach to central data monitoring was implemented for the SafeBoosC-III trial, a large, pragmatic, multicentre, randomised clinical trial evaluating the benefits and harms of treatment based on cerebral oxygenation monitoring in preterm infants during the first days of life versus monitoring and treatment as usual. We aimed to optimise completeness and quality and to minimise deviations, thereby limiting random and systematic errors. We designed an automated report which was blinded to group allocation, to ease the work of data monitoring. The central data monitoring group first reviewed the data using summary plots only, and thereafter included the results of the multivariate Mahalanobis distance of each centre from the common mean. The decisions of the group were manually added to the reports for dissemination, information, correcting errors, preventing furture errors and documentation.

Results: The first three central monitoring meetings identified 156 entries of interest, decided upon contacting the local investigators for 146 of these, which resulted in correction of 53 entries. Multiple systematic errors and protocol violations were identified, one of these included 103/818 randomised participants. Accordingly, the electronic participant record form (ePRF) was improved to reduce ambiguity.

Discussion: We present a methodology for central data monitoring to optimise quality control and quality development. The initial results included identification of random errors in data entries leading to correction of the ePRF, systematic protocol violations, and potential protocol adherence issues. Central data monitoring may optimise concurrent data completeness and may help timely detection of data deviations due to misunderstandings or fabricated data.

Keywords: Central monitoring; Clinical trials; Data deviations; Data quality; Mahalanobis distance; Missing data.

Fig. 1

Generation of the central data monitoring reports utilises R and markdown. The full reports, i.e. A full data completeness report and C full data quality and deficiencies report are utilised by the data monitoring group, while the short reports, i.e. B short data completeness report and D short data quality and deficiencies report, are used for newsletters and benchmarking. ePRF: electronic Participant Report Forms

Fig. 2

This is an example of how the Mahalanobis distance is presented in the full data quality and deficiencies reports. The centres are presented with blinded acronyms, which are used throughout the central data monitoring meetings

Fig. 3

This is a part of the first central data monitoring log which exemplify the flags, and the course of action for two of the flags

Fig. 4

These diagrams show the results from the first three meeting. The first column shows how many participants were included, and the second row show the number of entries which were flagged, and furthermore, in how many where an action was not deemed necessary. The investigators contacted received information about the entries which were flagged and an explanation of standard operating procedure. The last column present if an entry was correct or incorrect after response from the local investigator. ePRF: electronic participation report form; SOP: standard operating procedure