|
ABSTRACT Jeff Blume, PhD Candidate Biostatistics
Throughout an entire clinical trial, study investigators are
ethically obligated to monitor participant safety, as well as
accumulating evidence concerning the effectiveness of treatments.
However, current statistical tools discourage continuous monitoring of
study data. For example, it is well know that when conducting repeated
significance tests on accumulating data the probability of a type one
error quickly approaches unity (Armitage's repeated significance testing
paradox). Sequential, Group Sequential, and Bayesian methods have
failed to fill this void in current practice for a variety of reasons.
An Evidential Paradigm, based on the Law of Likelihood, is examined
in the context of constant monitoring. This paradigm uses (1)
likelihood ratios, not p-values, to measure the strength of statistical
evidence and (2) provides a bound on and control over the frequency of
both misleading and weak evidence. Instead of representing evidence
against a null hypothesis, the likelihood function measures relative
evidence supporting one simple hypothesis over another. Re-examination
of accumulating evidence does not diminish its strength, because the
likelihood function is unaffected by the number of examinations.
A procedure fashioned after the Law of Likelihood is proposed to
accommodate composite hypotheses. Brownian Motion techniques (Siegmund,
1985) can be used to approximate and demonstrate control over the
probability of misleading evidence. This procedure allows constant
monitoring for evidence of a clinically significant treatment effect
over no treatment effect, while maintaining a low probability of
misleading evidence.
Return to Longitudinal/Survival Working Group List | Return to Home Page |
|