Generated Narrative: Evidence

url: https://fevir.net/resources/Evidence/179690

identifier: FEvIR Object Identifier/179690

version: 1.0.0-ballot

name: CIBIC_EndpointAnalysisPlan_from_PHUSE_Lilly_Redacted_Protocol_EBMonFHIR_IG_Version

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

status: active

date: 2024-04-27 20:42:39+0000

publisher: HL7 International / Clinical Decision Support

contact: HL7 International / Clinical Decision Support: http://www.hl7.org/Special/committees/dss

author: Brian S. Alper:

UseContexts

copyright: https://creativecommons.org/licenses/by-nc-sa/4.0/

RelatedArtifacts

description: An example of an EndpointAnalysisPlan Profile which uses intended='true' and include-if extensions within Evidence.statistic.modelCharacteristic elements.

note: Approximately 100 patients will be randomized to each of the 3 treatment groups (high dose, low dose, and placebo). Previous experience with the oral formulation of xanomeline suggests that this sample size has 90% power to detect a 3.0 mean treatment difference in ADAS-Cog (p<.05, twosided), based on a standard deviation of 6.5. Furthermore, this sample size has 80% power to detect a 0.36 mean treatment difference in CIBIC+ (p<.05, two-sided), based on a standard deviation of 0.9 Group mean changes from baseline in the primary efficacy parameters will serve as efficacy criteria. The primary analysis of efficacy will include only the data obtained up to and including the visit of discontinuation of study drug. Furthermore, the primary analysis will not include efficacy data obtained at any visit where the study drug was not administered in the preceding three days. Analyses that include the retrieved dropouts are considered secondary. In general, all patients will be included in all analyses of efficacy if they have a baseline measurement and at least one postrandomization measurement. The primary analysis of ADAS-Cog (11) and CIBIC+ will be the 24-week endpoint, which is defined for each patient and variable as the last measurement obtained postrandomization (prior to protocol defined reduction in dose). a last-observation-carried-forward (LOCF). Note that the LOCF analysis at 24 weeks is the same as the endpoint analysis described previously. CIBIC+ will be analyzed by analysis of variance (ANOVA), with effects in the model being investigator and treatment. Investigator-by-treatment interaction will be tested in a full model prior to conducting the primary ANOVA (see description below). Because 3 treatment groups are involved, the primary analysis will be the test for linear dose response in the ANCOVA model described in the preceding paragraph. The result is then a single p-value for CIBIC+. Investigator-by-treatment interaction will be tested in a full ANOVA model, which takes the models described above, and adds the interaction term to the model. Interaction will be tested at α = .10 level. When the interaction is significant at this level, the data will be examined for each individual investigator to attempt to identify the source of the significant interaction. When the interaction is not significant, this term will be dropped from the model as described above, to test for investigator and treatment main effects. By doing so, all ANCOVA and ANOVA models will be able to validly test for treatment differences without weighting each investigator equally, which is what occurs when using Type III sums of squares (cell means model) with the interaction term present in the model. This equal weighting of investigators can become a serious problem when sample sizes are dramatically different between investigators. For all ANOVA and ANCOVA models, data collected from investigators who enrolled fewer than 3 patients in any one treatment group will be combined prior to analysis. If this combination still results in a treatment group having fewer than 3 patients in any one treatment group, then this group of patients will be combined with the next fewestenrolling investigator. In the event that there is a tie for fewest-enrolling investigator, one of these will be chosen at random by a random-number generator. The inherent assumption of normally distributed data will be evaluated by generating output for the residuals from the full ANCOVA and ANOVA models, which include the interaction term, and by testing for normality using the Shapiro-Wilk test from PROC UNIVARIATE. In the event that the data are predominantly nonnormally distributed, analyses will also be conducted on the ranked data. This rank transformation will be applied by ranking all the data for a particular variable, across all investigators and treatments, from lowest to highest. Integer ranks will be assigned starting at 1; mean ranks will be assigned when ties occur. All comparisons between xanomeline and placebo with respect to efficacy variables should be one-sided. The null hypothesis is that the drug is equal or worse than placebo. The alternative hypothesis is that the drug has greater efficacy than placebo. Different regulatory agencies require different type I error rates. Treatment differences that are significant at the .025 α-level will be declared to be “statistically significant.” When a computed p-value falls between .025 and .05, the differences will be described as “marginally statistically significant.” This approach satisfies regulatory agencies who have accepted a one-sided test at the .05 level, and other regulatory agencies who have requested a two-sided test at the .05 level, or equivalently, a one-sided test at the .025 level. In order to facilitate the review of the final study report, two-sided p-values will be presented in addition to the one-sided p-values.

variableDefinition

VariableDefinitionVariableRoleCode: exposure

VariableDefinitionComparatorCategory: placebo

description: high dose xanomeline vs. low dose xanomeline vs. placebo

note: exposure

variableRole: Use extension:variableRoleCode instead. (EvidenceVariableRole#exposure "exposure")

observed: EvidenceVariable/179689: GroupAssignment: high dose xanomeline vs. low dose xanomeline vs. placebo "GroupAssignment_high_dose_xanomeline_vs_low_dose_xanomeline_vs_placebo"

variableDefinition

VariableDefinitionVariableRoleCode: outcome

description: CIBIC+ at 24 weeks

note: outcome

variableRole: Use extension:variableRoleCode instead. (EvidenceVariableRole#measuredVariable "measured variable")

statistic

statisticType: (mean treatment difference) (StatisticStatisticType#0000457 "Mean Difference")

attributeEstimate

description: p value for one-sided test

type: p value for one-sided test (Scientific Evidence Code System (SEVCO) -- EXAMPLE VERSION for EBMonFHIR Implementation Guide#TBD:p-value-one-sided)

attributeEstimate

description: p value for two-sided test

type: p value for two-sided test (Scientific Evidence Code System (SEVCO) -- EXAMPLE VERSION for EBMonFHIR Implementation Guide#TBD:p-value-two-sided)

modelCharacteristic

StatisticModelIntended: true

StatisticModelValueCodeableConcept: In general, all patients will be included in all analyses of efficacy if they have a baseline measurement and at least one postrandomization measurement. ()

code: participant inclusion criteria for analysis (Scientific Evidence Code System (SEVCO) -- EXAMPLE VERSION for EBMonFHIR Implementation Guide#TBD:participant-inclusion-criteria-for-analysis)

modelCharacteristic

StatisticModelIntended: true

StatisticModelValueCodeableConcept: The primary analysis of efficacy will include only the data obtained up to and including the visit of discontinuation of study drug. Furthermore, the primary analysis will not include efficacy data obtained at any visit where the study drug was not administered in the preceding three days. ()

code: data inclusion criteria for analysis (Scientific Evidence Code System (SEVCO) -- EXAMPLE VERSION for EBMonFHIR Implementation Guide#TBD:data-inclusion-criteria-for-analysis)

modelCharacteristic

StatisticModelIntended: true

StatisticModelValueCodeableConcept: single imputation by last-observation-carried-forward (LOCF) (Scientific Evidence Code System (SEVCO) -- EXAMPLE VERSION for EBMonFHIR Implementation Guide#TBD:single-imputation-by-LOCF)

code: handling of missing endpoint data (Scientific Evidence Code System (SEVCO) -- EXAMPLE VERSION for EBMonFHIR Implementation Guide#TBD:handling-of-missing-endpoint-data)

modelCharacteristic

StatisticModelIntended: true

StatisticModelValueCodeableConcept: The primary analysis of efficacy will include only the data obtained up to and including the visit of discontinuation of study drug. Furthermore, the primary analysis will not include efficacy data obtained at any visit where the study drug was not administered in the preceding three days. Analyses that include the retrieved dropouts are considered secondary. ()

code: data inclusion criteria for secondary analysis (Scientific Evidence Code System (SEVCO) -- EXAMPLE VERSION for EBMonFHIR Implementation Guide#TBD:data-inclusion-criteria-for-secondary-analysis)

modelCharacteristic

StatisticModelIntended: true

code: one-tailed test (Scientific Evidence Code System (SEVCO) -- EXAMPLE VERSION for EBMonFHIR Implementation Guide#STATO:0000286)

modelCharacteristic

StatisticModelIntended: true

StatisticModelValueQuantity: 0.025

code: alpha setting (Scientific Evidence Code System (SEVCO) -- EXAMPLE VERSION for EBMonFHIR Implementation Guide#TBD:0000081)

modelCharacteristic

StatisticModelIntended: true

StatisticModelValueQuantity: 0.05

code: threshold for marginal statistical significance (Evidence Based Medicine on FHIR Implementation Guide Code System#defined-in-text "defined in text")

modelCharacteristic

StatisticModelIntended: true

StatisticModelValueCodeableConcept: xanomeline is equal or worse than placebo ()

code: null hypothesis (Scientific Evidence Code System (SEVCO) -- EXAMPLE VERSION for EBMonFHIR Implementation Guide#TBD:null-hypothesis)

modelCharacteristic

StatisticModelIntended: true

StatisticModelValueCodeableConcept: xanomeline has greater efficacy than placebo ()

code: alternative hypothesis (Scientific Evidence Code System (SEVCO) -- EXAMPLE VERSION for EBMonFHIR Implementation Guide#TBD:alternative-hypothesis)

modelCharacteristic

StatisticModelIntended: true

StatisticModelValueCodeableConcept: SAS ()

code: statistical software package (Scientific Evidence Code System (SEVCO) -- EXAMPLE VERSION for EBMonFHIR Implementation Guide#TBD:statistical-software-package)

modelCharacteristic

StatisticModelIntended: true

code: Sample Size/Power Calculation ~ 80% power to detect a 0.36 mean treatment difference in CIBIC+ (p<.05, two-sided), based on a standard deviation of 0.9 and sample size of 100 patients in each of 3 groups (Scientific Evidence Code System (SEVCO) -- EXAMPLE VERSION for EBMonFHIR Implementation Guide#TBD:sample-size "Sample size estimation")

attributeEstimate

description: 80% power

type: power (Scientific Evidence Code System (SEVCO) -- EXAMPLE VERSION for EBMonFHIR Implementation Guide#TBD:power "Power")

quantity: 80 % (Details: UCUM code % = '%')

attributeEstimate

description: minimally detectable threshold value 0.36

type: minimally detectable threshold value (Scientific Evidence Code System (SEVCO) -- EXAMPLE VERSION for EBMonFHIR Implementation Guide#TBD:checkIfInSTATOtesting-margin "Hypothesis testing margin")

quantity: 0.36

attributeEstimate

description: p<.05

type: alpha setting (Scientific Evidence Code System (SEVCO) -- EXAMPLE VERSION for EBMonFHIR Implementation Guide#TBD:0000081)

quantity: 0.05

attributeEstimate

description: standard deviation of 0.9

type: Standard deviation (Scientific Evidence Code System (SEVCO) -- EXAMPLE VERSION for EBMonFHIR Implementation Guide#STATO:0000237)

quantity: 0.9

attributeEstimate

description: sample size of 100 patients in each of 3 groups

type: sample size per group (Scientific Evidence Code System (SEVCO) -- EXAMPLE VERSION for EBMonFHIR Implementation Guide#TBD:sample-size-per-group)

quantity: 100

attributeEstimate

description: two-tailed test

type: two-tailed test (Scientific Evidence Code System (SEVCO) -- EXAMPLE VERSION for EBMonFHIR Implementation Guide#STATO:0000287)

attributeEstimate

description: sample permutation testing - Because there will be roughly 300!/(3 * 100!) possible permutations of the data, random data permutations will be sampled (10,000 random permutations). -- distribution assumption for comparison if random

type: prospective sample permutation testing (Scientific Evidence Code System (SEVCO) -- EXAMPLE VERSION for EBMonFHIR Implementation Guide#TBD:prospective-sample-permutation-testing)

attributeEstimate

attributeEstimate

modelCharacteristic

StatisticModelIntended: true

StatisticModelValueCodeableConcept: analysis of variance (ANOVA) (Scientific Evidence Code System (SEVCO) -- EXAMPLE VERSION for EBMonFHIR Implementation Guide#TBD:0000146 "ANOVA")

code: Primary analytic method (Scientific Evidence Code System (SEVCO) -- EXAMPLE VERSION for EBMonFHIR Implementation Guide#TBD:primary-analytic-method "primary analytic method")

variable

variableDefinition: EvidenceVariable/159673: investigator "Investigator"

handling: polychotomous

variable

variableDefinition: : treatment

handling: ordinal

valueCategory: high dose xanomeline (), low dose xanomeline (), placebo ()

variable

StatisticModelIncludeIf

url

attribute

value: p value for F test ()

url

value

value: <0.1

variableDefinition: EvidenceVariable/156673: Investigator-by-treatment interaction "Investigator_by_treatment_interaction_for_ADAS_Cog11_ANCOVA"

handling: polychotomous

modelCharacteristic

StatisticModelIncludeIf

url

attribute

value: Defined by Expression ()

url

value

value: Evidence[id='171171'].statistic[statisticType='F test'].attributeEstimate[type='p value'].quantity < 0.1("When the Investigator-by-treatment interaction (p value for F test) is significant at the 0.1 level")

StatisticModelIntended: true

StatisticModelValueCodeableConcept: the data will be examined for each individual investigator [EvidenceVariable/159673] to attempt to identify the source of the significant interaction. (Scientific Evidence Code System (SEVCO) -- EXAMPLE VERSION for EBMonFHIR Implementation Guide#TBD:identify-source-of-interaction "identify source(s) of significant interaction")

code: additional investigation (Evidence Based Medicine on FHIR Implementation Guide Code System#defined-in-text "defined in text")

Variables

-	VariableDefinition	Handling
*	EvidenceVariable/159673: investigator "Investigator"	polychotomous

modelCharacteristic

StatisticModelIncludeIf

url

attribute

value: Defined by Expression ()

url

value

value: (Evidence[id='171172'].statistic[statisticType='Shapiro-Wilk test'].attributeEstimate[type='p value'].quantity < 0.05), using SAS PROC UNIVARIATE("residuals are nonnormally distributed using the Shapiro-Wilk test from PROC UNIVARIATE")

StatisticModelIntended: true

StatisticModelValueCodeableConcept: Analyses will also be conducted on the ranked data. This rank transformation will be applied by ranking all the data for a particular variable, across all investigators and treatments, from lowest to highest. Integer ranks will be assigned starting at 1; mean ranks will be assigned when ties occur.. ()

code: rank-based analytic method (Scientific Evidence Code System (SEVCO) -- EXAMPLE VERSION for EBMonFHIR Implementation Guide#TBD:rank-based-analytic-method)