WHITEHOUSE STATION, N.J., May 30, 2006 -
Merck & Co., Inc. is correcting its prior description of one of the statistical methods used to analyze certain data in the APPROVe study published in 2005, and has notified the study authors, the New England Journal of Medicine (NEJM) and regulatory authorities. Merck recently discovered the need for this correction while reviewing the preliminary analyses of the off-drug extension data for the APPROVe study. Merck believes that this correction does not change the results of the APPROVe study, in which an increased relative risk for confirmed thrombotic cardiovascular events for VIOXX compared to placebo was observed beginning after 18 months of continuous daily treatment. This correction is unrelated to the recently announced preliminary analysis of the off-drug extension of the APPROVe study.
The VIOXX cardiovascular data analysis plan called for numerous statistical and graphical methods to be used to assess whether the relative risk of VIOXX compared to placebo was constant over time or if it changed over time (see attached). The use of the variable, logarithm of time, was an element in the primary method specified. The reference to logarithm of time in the description of methods published in the NEJM and submitted to regulatory agencies was in error. The reported result (p-value = 0.01) came from a statistical model using linear time, not logarithm of time. Recent tests show that the result using logarithm of time has a p-value = 0.07. The results of diagnostic steps specified in the data analysis plan indicate that the linear test is an appropriate method to assess changes in relative risk over time.
As specified in the analysis plan, Merck used additional analytical and graphical methods to evaluate whether the relative risk changed over time. These methods included a Kaplan-Meier plot that showed similar curves for placebo and VIOXX during the first 18 months, which then began to separate at approximately 18 months. Relative risks were also calculated over successive six-month intervals in the study. Over the 36-month period of the study, the relative risk was lowest in the first three sets of six-month intervals and highest in the last three sets of six-month intervals, again demonstrating changing relative risk over time. The results of the linear time analysis, the Kaplan-Meier plot, and the interval relative risks together confirm that the relative risk changes over time.
The linear time and logarithm of time analyses only test whether relative risk changes over time, they do not test the overall magnitude of relative risk. The overall magnitude of the relative risks and their associated p-values were described correctly.
Merck believes that today's correction does not change the results of the APPROVe study. Nonetheless, Merck intends to retain an independent statistical expert to review data and analyses from the APPROVe study. The study's authors advised Merck that they intend to submit a correction to the NEJM.
About the APPROVe Study
APPROVe (Adenomatous Polyp Prevention on VIOXX) was a multi-center, randomized, placebo-controlled, double-blind study designed to evaluate the efficacy of 156 weeks (three years) of treatment with VIOXX 25 mg in preventing recurrence of colorectal polyps in patients with a history of colorectal adenomas. There was no Statistical Analysis Plan (SAP) for the cardiovascular data from APPROVe alone. Merck planned to combine the cardiovascular data from APPROVe with data from two other placebo-controlled studies, VICTOR and ViP. Given the decision to stop the study early, the APPROVe data were analyzed separately.
About Merck
Merck & Co., Inc. is a global research-driven pharmaceutical company dedicated to putting patients first. Established in 1891, Merck currently discovers, develops, manufactures and markets vaccines and medicines to address unmet medical needs. The Company devotes extensive efforts to increase access to medicines through far-reaching programs that not only donate Merck medicines but help deliver them to the people who need them. Merck also publishes unbiased health information as a not-for-profit service. For more information, visit www.merck.com.
Forward-Looking Statement
This press release (including the attachment) contains "forward-looking statements" as that term is defined in the Private Securities Litigation Reform Act of 1995. These statements are based on management's current expectations and involve risks and uncertainties, which may cause results to differ materially from those set forth in the statements. The forward-looking statements may include statements regarding product development, product potential or financial performance. No forward-looking statement can be guaranteed, and actual results may differ materially from those projected. Merck undertakes no obligation to publicly update any forward-looking statement, whether as a result of new information, future events, or otherwise. Forward-looking statements in this press release should be evaluated together with the many uncertainties that affect Merck's business, particularly those mentioned in the cautionary statements in Item 1 of Merck's Form 10-K for the year ended Dec. 31, 2005, and in its periodic reports on Form 10-Q and Form 8-K, which the Company incorporates by reference.
# # #
Excerpted from the MK-0966 CV Outcomes Data Analysis Plan for Protocol 203 Combining APPROVe, ViP, and VICTOR (Jan. 9, 2004)
Check of Model Assumptions
Analytical and graphical methods will be employed to verify the proportional hazards assumption. The primary method for testing the proportional hazards assumption will be by including the factor treatment*log(time) in the model; nonsignificance (p>0.050) of this factor is not inconsistent with proportionality, i.e., constancy of treatment effect over time. The semi-parametric nature of the proportional hazards model does not require other distributional assumptions.
Further, the log HR will be plotted over time by stratifying time into intervals containing approximately the same number of events within each interval. The log HR within each successive 6-month time interval with confidence limits will also be calculated and plotted. Such plots will provide an indication of any time effect on the HR.
Additional diagnostic steps based on residuals and HR will be performed to examine model fitting if the previously described analyses indicate issues that warrant further assessment. For instance, assessment of the log HR over time may also be made using Schoenfeld's partial residuals [10] with respect to each covariate in the fitted model. Visual and test diagnostics of scaled Schoenfeld residuals [11; 12; 13] may be used here to study the PH assumption for the model covariates. Schoenfeld residuals are defined at event times and compute the difference between the observed covariate (treatment group indicator) and its conditional expectation from the fitted Cox PH model. If the PH assumption holds for a covariate in the Cox model, these scaled residuals when plotted over time should be randomly distributed to either side of "zero" line and should not have any discernible trend/pattern over time. Constancy of relative treatment effect over time may also be assessed for each covariate (discrete or continuous) by using the zph test [13]. A zph test of treatment effect (based on log-transformation) is essentially the same as testing the significance of adding treatment*log(time) as a time-dependent covariate ([13], pp. 144). Thus, a nonsignificant zph test may also be used to validate proportionality assumption for associated covariate.
If the nonproportionality is large and real, various approaches may be explored to investigate the causes of nonproportionality: stratification of covariates, partition of the time axis [14; 15; 16], adding a time-dependent covariate, or using a different model. Some of these approaches are discussed with illustrations in the book by Therneau and Grambsch [13] (Sections 6.5 and Chapter 7). The choice of a non-proportional hazard model, such as a piecewise exponential model, that will be a good fit to the data may be assessed from some of the above diagnostic measures.
References
10. Schoenfeld D. Partial residuals for the proportional hazard regression model. Biometrika 1982;69:239-41.
11. Venables WN and Ripley BD. Modern applied statistics with S-Plus, Springer-Verlag Inc (Berlin; New York). 1994.
12. Hess K. Graphical methods for assessing violations of the proportional hazards assumption in Cox regression. Statistics in Medicine 1995;14:1707-23.
13. Therneau TM and Grambsch PM Modeling survival data: extending the cox model. Springer, New York 2000.
14. Martink DO and Austin H. Exact estimates of a rate ratio. Epidemiology 1996;7:29-33.
15. Martin DO and Austin H. An exact method for meta-analysis of case-control and follow-up studies. Epidemiology 2000;11:255-60.
16. Zelen M. The analysis of several 2x2 contingency tables. Biometrika 1971;58:129-37.
|
Printer Friendly Page
