non-inferiority trials

superiority trial null hypothesis

• for the outcome of interest, treatment x is no more effective than treatment y

• rejecting means there is a difference

superiority trials alternative hypothesis

• for the outcome of interest, treatment is more more effective than treatment y

disprove/reject

• for superiority trials, the aim is to ___ the null hypothesis and show that the 2 treatment differ in effecgt

non-inferiority trial null hypothesis

• for the outcome of interest, treatment x is worse than treatment y by more than a certain margin

• rejecting at a prespecfifed level of statistical significance supports a claim that permits conclusion

non-inferiority trial alternative hypothesis

• for the outcome of interest, treatment x is not worse than treatment y by more than a certain margin

unconcerned

• a non-inferiority trial is ))) if the novel treatment is better so long as it is “not much worse” than the comparator

equivalence trial null hypothesis

• for the outcome of interest, treatment x is not worse nor better than treatment y by more than a certain margin

why use a non-inferiority trial

• designed to demonstrate that an experimental therapy is not worse than an active control

• can use this trial instead of conventional superiority because the therapeutic strategies may have characteristics beyond their overall treament effect that make them attractive to clincians and patients

foundations of non-inferiority trials

• one or more prior RCTs evaluating the superiority of the active control over placebo. entriely based on having relevant historical evidence that establishes substantial effect of the ocntrol drug

• an endpoint is selected, and on the basis of prior experience, the expected performance of the active control is derived

• an acceptable margin is defined during the design phase, which preserves a minumum clinically acceptable proportion of the historical effect of the active treatment as compared with placebo

• the trial must be designed to adequately distinguish between effective and ineffective therapues

• preserves the conditions of the trial in which the active control was shown to be effective

important assumptions in non-inferiority trials

• new treatment is not worse than standard treatment

• new treatment would be better than placebo

• new treatment has other benefits over standard rx

non-inferiority margin

• investigators set their own margins

• there is no universally accepted method for defining the threshold

• the margin depends on the eye of the beholder, what is sound reasoning for one clinician is misguided in another

non-inferiory margin guidelines

• often based on historical placebo controlled trials

• may be determined by: differences in patient population, different dosage regimens, different endpoint definitions or different concomitant treatments, active treatment is historical placebo controlled trials having an inconsistent effect

choose a stringent NI margin

• if choose insuffieiently stringent thresholds, NI may be concluded while many patients might be unwilling to accept the novel tx if they were informed of the largest possible increased risk associated with its use

• wide uptake of novel txs could result if the choices of thresholds go uncontested

• use your own judgement on what is reasonable threshold

• the NI margin is usually chosen with some buffering to take into account the current trial may not have full predicted effect of the active control

• the NI margin is often based on the lower bound of a CI for that effect, or on retaining at least a certain percentage of the effect of the active control

• the value of the NI margin should be chosen as the maximum acceptable difference between X and Y

superiority trials

• to be successful in disproving Ho, it should maximally demonstrate the difference in efficacy between the experimental and active control

• any methodological issues minimize differences will weaken the evidence for the experimental drug

• any that maximizes the differences between groups’ effects will make it harder to see the benefit of one drug over the other

non-inferiority trials

• by contrast for a NI trial to disprove H0, it should maximally demonstrate similarity between the 2 groups

• any methodolofic issues that minimize differences between the 2 groups will falsely strengthen the evidence for the experimental therapy

• if you do not look very hard for a difference, you won’t find one

• the implication is deeming the drug non-inferior to the active control when it is actually inferior

blinding in NI trial

• does not protect against bias

• researchers are slightly more generous when judging outcomes

• bias makes both treatment look more similar- similarity is exactly what the investigator wants

method of analysis in NI trials

• ITT is usually conservative

• exception: PPA becomes equally important

• ideally, both ITT and PPA will be used in NI trials

• in the case both analyses agree, the clinical decision is easy

• in the case the 2 analyses don’t agree, then the inferences may be drawn into question

superiority

• Noninferioty trial can show ___

NOT

• superiority trial that fails to show superiority does ___ default to showing noninferiority

maximum acceptable

• the NI margin should be chosen as the ___ difference between drugs x and y