Study design, statistical analyses, and results reporting in the bone marrow transplantation literature

To assess the statistical methods in the bone marrow transplantation (BMT) literature, we reviewed 255 articles from eight major journals (Annals of Internal Medicine, Blood, British Journal of Cancer, Cancer, Lancet, Journal of Clinical Oncology, New England Journal of Medicine, and Transplantation) between 1990 and 1992. Study designs, quality of statistical methods, and types of statistical procedures employed were summarized. There were similar percentages of case series (24%) and prospective trials (28%); however, the former design is often lacking in rigor of patient selection and treatment, and may therefore be less convincing. The reporting of eligibility criteria, treatment complications, and losses to follow-up was complete in the majority of articles (72-94%). Deficits in reporting were seen in power and sample size justifications (86%) and in stating whether the alternative hypothesis of the study was one-sided or two-sided (75%), information without which the magnitude of the p value cannot be interpreted. Although there has been recent emphasis on the evaluation of cost-effectiveness and quality of life as important endpoints, only 4-5% of the articles included these topics. Survival analysis was the most frequently employed technique (69%), and regression modeling (Cox and logistic) was also among the more common statistical techniques. The methods reported, however, often were inadequate to demonstrate that these techniques had been correctly applied and therefore that the results were correctly interpreted. Guidelines to further strengthen the design, analysis, and reporting of future BMT research are presented.