Generated Narrative: Evidence 179691

url: https://fevir.net/resources/Evidence/179691

identifier: FEvIR Object Identifier/179691

name: Example_EndpointAnalysisPlan_from_PHUSE_Lilly_Redacted_Protocol_EBMonFHIR_IG_Version

title: Example EndpointAnalysisPlan from PHUSE Lilly Redacted Protocol - EBMonFHIR IG Version

status: Active

publisher: Computable Publishing LLC

contact: support@computablepublishing.com

author: Brian S. Alper:

UseContexts

copyright:

relatedArtifact

type: Cite As

citation:

Example EndpointAnalysisPlan from PHUSE Lilly Redacted Protocol - EBMonFHIR IG Version [Evidence]. Contributors: Brian S. Alper [Authors/Creators]. In: Fast Evidence Interoperability Resources (FEvIR) Platform, FOI 179691. Revised 2023-12-04. Available at: https://fevir.net/resources/Evidence/179691. Computable resource at: https://fevir.net/resources/Evidence/179691.

relatedArtifact

type: Specification Of

Documents

-	ContentType	Data	Size	Title
*	application/pdf	(base64 data - 321,048 base64 chars)	240785	PHUSE.Lilly.Redacted.Protocol

description:

note: The statistic element will show an example of an endpoint analysis plan. To determine if there is a statistically significant relationship (overall Type 1 error rate, α=.05) between the change in both ADAS-Cog (see Attachment LZZT.2) and CIBIC+ (see Attachment LZZT.3) scores, and drug dose (0, 50 cm2 [54 mg], and 75 cm2 [81 mg]). 4.3.1. Efficacy Variables to be Analyzed Efficacy measures are described in Section 3.9.1.1. As stated in Section 3.9.1.2, the primary outcome measures are the ADAS-Cog (11) and CIBIC+ instruments. Because both of these variables must reach statistical significance, an adjustment to the nominal p-values is necessary in order to maintain a .05 Type I error rate for this study. This adjustment is described in detail in Section 4.3.5. 4.3.5. Nominal P-value Adjustments When there are multiple outcomes, and the study drug is declared to be effective when at least one of these outcomes achieves statistical significance in comparison with a placebo control, a downward adjustment to the nominal α-level is necessary. A well-known simple method is the Bonferroni method, that divides the overall Type I error rate, usually .05, by the number of multiple outcomes. So, for example, if there are two multiple outcomes, the study drug is declared to be effective if at least one of the two outcomes is significant at the .05/2 or .025 level. However, when one has the situation that is present in this study, where there are 2 (or 3 for Europe) outcome variables, each of which must be statistically significant, then the adjustment of the nominal levels is in the opposite direction, that is upwards, in order to maintain an overall Type 1 error rate of .05. In the case of two outcomes, ADAS-Cog (11) and CIBIC+, if the two variables were completely independent, then each variable should be tested at the nominal α-level of .051/2 = .2236 level. So if both variables resulted in a nominal p-value less than or equal to .2236, then we would declare the study drug to be effective at the overall Type 1 error rate of .05. We expect these two outcome measures to be correlated. From the first large-scale efficacy study of oral xanomeline, Study MC-H2Q-LZZA, the correlation between CIBIC+ and the change in ADAS-Cog(11) from baseline was .252. Consequently, we plan to conduct a randomization test to combine these two dependent dose-response p values into a single test, which will then be at the .05 Type I error level. Because there will be roughly 300!/(3 * 100!) possible permutations of the data, random data permutations will be sampled (10,000 random permutations). Designate the dose response p-values as p1 and p2 (computed as one-sided p-values), for ADAS-Cog(11) and CIBIC+, respectively. The rejection region is defined as [ {p1 ≤ α and p2 ≤ α} ]. The critical value, α, will be determined from the 10,000 random permutations by choosing the value of α to be such that 2.5% of the 10,000 computed pairs of dose response p-values fall in the rejection region. This will correspond to a one-sided test at the .025 level, or equivalently a two-sided test at the .05 level. In addition, by determining the percentage of permuted samples that are more extreme than the observed data, a single p-value is obtained. Do 10,000 random permutations of 'treatment group' assigned to the observed values for all other variables. Alpha level for this single p-value is 0.025, 'marginally statistically significant' threshold for this single p-value is 0.05

variableDefinition

description:

random permutation of treatment assignments matching the actual trial distribution of treatment assignment counts and keeping all other observed variables fixed

variableRole: Population

variableDefinition

description:

high dose xanomeline vs. low dose xanomeline vs. placebo

variableRole: Exposure

comparatorCategory: placebo

observed: GroupAssignment: high dose xanomeline vs. low dose xanomeline vs. placebo

variableDefinition

description:

whether the permuted sample is more extreme than the observed data (ANCOVA-derived ADAS-Cog(11) effect estimate > observed effect estimate AND ANOVA-derived CIBIC+ effect estimate > observed effect estimate)

variableRole: Outcome

synthesisType: not applicable

studyDesign: randomized assignment

statistic

statisticType: reported as an empirical p-value

SampleSizes

-	NumberOfParticipants	KnownDataCount
*	10000	10000

modelCharacteristic

code: prospective sample permutation testing

intended: true

modelCharacteristic

code: nominal Type I error rate

value: 0.025

intended: true

modelCharacteristic

code: threshold for marginal statistical significance

value: 0.05

intended: true

modelCharacteristic

code: null hypothesis

value: xanomeline is equal or worse than placebo

intended: true

modelCharacteristic

code: alternative hypothesis

value: xanomeline has greater efficacy than placebo

intended: true

modelCharacteristic

code: statistical software package

value: SAS

intended: true

modelCharacteristic

code: Sample Size/Power Calculation

value: Rationale: Because there will be roughly 300!/(3 * 100!) possible permutations of the data, random data permutations will be sampled (10,000 random permutations).

intended: true