Abstracts 621

protocol and the revised protocol. For example, the original data verification programs continued to process the data, even though the database structure had been modified to accommodate the new protocol.

The second application was to create several levels of views to provide a consistent definition of data that had been collected on multiple revisions of study forms. For example, views were created to combine data from several revisions of multiple forms to merge the data for analysis. We conclude that selective use of views can contribute to the efficient management of study data.

A8 LOW COST DATABASE MANAGEMENT IN CLINICAL TRIALS-RIDING THE TIDAL WAVE

OF TECHNOLOGY

David Duncan end lan Ford Glasgow University Glasgow, Scotland

In the spring of 1988 we were presented with the prospect of managing the data for a screening program and clinical trial (the West of Scotland Coronary Prevention Study) which would generate data on approximately 100,000 screenees and 6,000 trial subjects followed up for a period 5-6 years. The number of records generated would be in excess of 10 e. Available expertise was with the database management package SIR on a busy mainframe with an uncertain future. Decisions were required on a computing and database platform for the project. Factors affecting the decisions included budget restrictions and the need to get a system running within 6 months.

A decision was made to buy the best PC's available and to utilize cheap and easily accessible PC software to create a base system for data entry and database accrual. The theory was that, given the rapid developments in hardware and software, we would optimize our budget by buying into technology only when necessary.

We are now 3 years into the project and are in a position to discuss the advantages and dangers of this approach. We are also about to upgrade our system and will review the options available for anyone starting out from the same position in 1991.

A9 A NEW CLINICAL TRIALS DESIGN FOR AIDS RESEARCH: THE "MEASURED-CHOICE" DESIGN

James Rochon University of Western Ontario

London, Ontario, Canada

The AIDS epidemic has placed a severe strain not only on the delivery of health care but also on the process by which experimental therapies are tested and approved for license. AIDS patients benefit from organized patient advocacy groups who have challenged many of the methodological underpinnings of randomized controlled tdals. As a result, RCTs conducted in this area have suffered from high withdrawal rates, pooling of study drugs, systematic patient unblinding, and the widespread use of proscribed medications. This raises the possibility of selection and withdrawal biases entering the analysis and casts doubt on the validity of the results.

In this paper, a new methodology called the "Measured-Choice" design is described. Under this model, patients are entered into the study in matched pairs, with matching performed according to relevant prognostic factors. Randomization is then performed within each matched pair, and beth patients and investigators are blinded to the treatment allocation and to the identities of the subjects comprising the matched pairs. At every follow-up, the differences in measured clinical outcomes between the patient and his pair-mate are evaluated, and the patient can then choose to switch to the other treatment arm if he is dissatisfied with his progress vis-a-vis his pair-mate. The patient can switch only once and is then followed as usual thereafter.

The practicability of this design and the anticipated benefits in terms of its acceptability to the AIDS community and enhanced protocol compliance are presented. As well, accompanying statistical issues are considered, and an approach for analyzing the ensuing data is outlined.