Home

Current Issue

Archives

Guidelines for Authors

Classified Ads

Links

Search PubMed

Subscriptions

Subscriber Registration

Guidelines for Website Users

JRheum Update Service

Contact Info

Dr. Genovese, et al reply

To the Editor:

We thank the Drs. Yazici for their response to our article¹ and believe that it offers an excellent forum for understanding the appropriate use of statistical analysis in randomized clinical trials that cross over into longterm observational studies.

In the 2-year article² patients had been randomized and remained on their original treatment assignments for the duration of the 2 years. The blind for the study was broken during year 2, but on average patients remained blinded for 18.4 months and this study had very low attrition. Safety and efficacy outcomes were compared between the 3 groups utilizing statistical tests for significance. This was feasible and appropriate because all groups continued to receive treatment according to random assignment, and, more importantly, all subjects who received any active drug were included in the analysis. The analysis in this study used the last observation carried forward (LOCF) technique to account for missing data and early drop-outs.

The formal longterm extension study began when each patient completed year 2. In this observational study, all interested subjects were allowed to receive etanercept at the 25 mg biweekly dose, with the use of methotrexate (MTX) and corticosteroids determined by the investigator based on disease activity. The decision to enter the extension was certainly subject to a number of biases. Formal statistical significance testing was not performed in this study for several reasons: (1) differential drop-out rates were likely to bias results due to confounding by indication; (2) there was an increasing amount of heterogeneity of treatment within each group as other medication changes were made according to disease activity; and (3) the safety analysis focusing on the use of etanercept examined all patients who received at least 1 dose of etanercept, and excluded patients from the MTX arm who did not enter into the etanercept only phase of the study.

Table 2¹ shows the safety data for all patients who received at least one dose of etanercept. Careful examination of subjects included in this portion of safety analysis shows that only 143 of the original 217 subjects (66%) randomized to the MTX arm received any etanercept due to early termination from the original phase of the study or elected not to participate in the longterm extension. In comparison, all subjects originally assigned to receive either dose of etanercept (208/208 of 25 mg etanercept and 207/207 of 10 mg etanercept) are included in the safety analysis. So although it appears that there may be fewer adverse events in the original MTX arm, fewer subjects were included. Many of the adverse events seen in the early phases of the study in subjects receiving MTX without etanercept were not evaluated in the analysis of safety in subjects receiving etanercept. Inferential statistics in this instance would be inappropriate because one arm is deliberately left-censored. The 3 groups in comparison were not equal with respect to length of time receiving etanercept or with period of time under study. We were concerned that the inclusion of formal statistics would not only be unable to uncover any inherent biases, but would lead to misinterpretations of the data. In fact the application of statistical analysis in this situation would ignore or obscure the fact these significant biases exist and would legitimize this type of analysis in the eyes of many readers. While testing for statistical significance is very important in evaluating study results, careful attention to the composition of each group and the clinical significance of the measured outcomes should not be ignored in the absence of p values.

MARK C. GENOVESE, MD, Division of Rheumatology, Stanford University Medical Center, Palo Alto, California; JOAN M. BATHON, MD, Division of Rheumatology, Johns Hopkins University, Baltimore, Maryland; ROY M. FLEISCHMANN, MD, Division of Rheumatology, University of Texas Southwestern Medical Center at Dallas, Dallas, Texas; LARRY W. MORELAND, MD, Clinical Immunology and Rheumatology, University of Alabama at Birmingham, Birmingham, Alabama; RICHARD W. MARTIN, MD, College of Human Medicine, Michigan State University, Grand Rapids, Michigan; JONATHAN A. LEFF, MD, Amgen Inc., Thousand Oaks, CA, USA.

REFERENCES

Search PubMed for: