Interpretation of statistical methodology associated with maintenance trials.

Maintenance trials in ulcer disease provide a scientific framework through which the prophylactic performances of two or more therapies can be compared with respect to ulcer recurrence. For this purpose, a primary source of data is the endoscopic classification of patients at protocol-scheduled visits as well as at unscheduled visits. Several statistical issues are relevant to the analysis of this endoscopic information. First, the extent to which the respective patients have different data patterns for their ulcer recurrence status must be taken into account by actuarial methods that adjust for the length of follow-up time. Second, the role of multiple investigators requires careful attention when information is combined from them. A third issue is concerned with the assessment of the variation of ulcer recurrence experience across subgroups with respect to patient demographics, medical history, and baseline characteristics. Statistical methods can effectively address these issues so that the results of maintenance studies can be interpreted through estimates for ulcer recurrence rates. In this way, they can meaningfully support clinical conclusions about the performance of alternative therapies for maintaining the absence of ulcer disease over time.

Ranitidine and cimetidine in the healing of duodenal ulcer: meta-analysis of comparative clinical trials.

All available ranitidine and cimetidine comparative trials in acute duodenal ulcer disease were examined: of the 44 trials, 36 favoured ranitidine, and there was an overall difference in ulcer healing of 7%. Further stratification enabled examination of trials with common attributes: the most frequent endoscopic assessment was at 4 weeks to compare ranitidine 150 mg twice daily with cimetidine 1 g day-1 or 400 mg b.d. Twenty of these trials had sufficient data to permit pooling. Ranitidine was favoured in 18/20 trials and in three the differences achieved statistical significance. Results of the trials were combined using meta-analysis to calculate differences in ulcer healing. Most studies had sample sizes that were insufficient to detect clinically-important differences; the power to detect a 20% difference was less than 80% in 15/20 trials, and for a 10% difference was less than 80% in all but one trial. Fifteen trials compared ranitidine 150 mg b.d. with cimetidine 1 g day-1: healing after 4 weeks therapy was overall 6% greater for ranitidine. This was statistically significant (P less than 0.05) and the combined total number of patients had a power of 83% to detect this difference. In five trials the dose of cimetidine used was 400 mg b.d.: the 12% difference in healing in favour of ranitidine 150 mg b.d. was statistically significant, and the combined trials had a power of 95% to detect this difference. Ranitidine 150 mg twice daily heals significantly more duodenal ulcers after 4 weeks of therapy than either cimetidine 400 mg b.d. or cimetidine 1 g day-1.

The conflict between relational databases and the hierarchical structure of clinical trials data.

Relational database software has become popular for the management of certain types of commercial data. Its use is being given serious consideration in the management of data from clinical trials. Relational systems have a number of advantages over hierarchical or network systems for some types of data. However, as illustrated by an example, the data from clinical trials typically have an inherent hierarchical structure. The incorporation of hierarchically structured data into a relational database raises difficult problems of data integrity versus the complexity of the database structure. These problems, together with the long execution times of many relational operations, indicate that relational systems are not necessarily well suited for clinical trials data management.