ABSTRACT
Although methods used for data collection and quality assurance for large-scale clinical trials are important to critical reading of trial results and have been published, such reporting is the exception rather than the rule. In the MRFIT, systematic methods for processing large volumes of data over a long period of time were developed. The methods were designed to detect and control a variety of errors and to leave a complete audit trial of the processing of forms and corrections to forms. Many of these methods evolved and were refined during the course of the study as a result of trial and error. If one were to start over, the methods described herein would be modified. The field of data processing is evolving, and it is important for statistical and data processing staff of coordinating centers to recognize this and continually evaluate and update their methods. For example, the simultaneous entry and computer editing of forms is becoming more feasible with time. Also, more sophisticated intelligent data entry equipment is available for central use. Near the end of MRFIT, some data received at the Coordinating Center were entered and edited on a minicomputer. The parameter-driven edits described previously were performed at the time of data entry. Additional modifications to the content of the data dictionary for future studies are also being considered. The incorporation into the data dictionary of consistency checks (both deterministic and probabilistic) between fields on different forms would facilitate the specification of complex edit checks and would provide better documentation of the edit checks actually performed. Incorporating definitions of the numeric codes for each field would improve the documentation and facilitate reporting using statistical packages. Dedicated computer hardware should also be a major consideration of coordinating centers in future clinical trials. For MRFIT, a dedicated system was used from 1978 to the end of the trial. With the continued decline in hardware costs, dedicated systems can and should be considered, even for trials much smaller than MRFIT. We believe the system developed for processing data in the MRFIT has several advantages. It satisfies the requirements identified by Karrison or a system of data editing and control, it is largely self-documenting as a result of the data dictionary approach taken, and it is easily adaptable to other clinical studies.
