Creating Clinical Trial Summary Tables Containing P …

1 1 Creating Clinical Trial Summary Tables Containing P-Values:A practical approach using Standard SAS MacrosJun Zuo, , STATPROBE, Inc., Ann Arbor, MICarl R. Haske, , STATPROBE, Inc., Ann Arbor, MIABSTRACTP-value is a key criterion for evaluating the effectiveness andsafety of new drugs in Clinical trials, particularly in comparativestudies. However, p-values are generally not presented in datasummary Tables generated with SAS software, because of thecomplexity of incorporating p-values into a formatted table thatcontains Summary statistics, such as mean, proportion, or stan-dard purpose of this paper is to present a practical approach tocreate Summary Tables Containing p-values which are calculatedusing customized SAS macros developed at STATPROBE, major types of SAS macros are presented in this paper tocalculate p-value, according to whether variables are categorical(Fisher's exact test, Cochran-Mantel-Haenzel procedure, or logis-tic regression)

2 , quantitative (t-test, Wilcoxon rank-sum test, orgeneral linear models), or survival (parametric or nonparametricestimates). SAS statistical procedures, such as PROC FREQ,PROC NPAR1 WAY, PROC CATMOD, PROC GLM, and PROCLIFETEST, provide the function part of the macros. The outputlocations of p-values from SAS statistical procedures are alsoconsidered for the purpose of effective creation of the macrosand generation of the Summary Tables . The programming tech-niques to prepare the input data set and to merge data sets con-taining p-values and other Summary statistics are discussed is a probability value commonly used in comparativeclinical trials to test whether there exists a significant differencefor a specific Summary measure ( , mean, proportion, re-sponse, or survival time) among different treatment groups.

3 It isa key criterion used in Clinical trials to evaluate the effectivenessand safety of a new drug, a new medical product, or a new medi-cal intervention. However, in general, most Summary Tables gen-erated with SAS software contain only such common summarystatistics as proportion, mean, median, or standard are not presented in these Summary Tables because ofthe theoretical and technical complexity of generating andincorporating them into the Tables , and also because of theseparation of responsibilities between statisticians andprogrammers. Although a word processor can be used tocombine the common Summary statistics, computed mostly byprogrammers using Base SAS, and the p-values, calculatedmainly by statisticians using SAS/STAT or SAS statistical macros( , Ward and Haske 1996), it is still desirable for bothprogrammers and statisticians to pursue a practical approach togenerate Summary Tables Containing paper presents a practical approach to create integratedsummary Tables Containing p-values.

4 Specifically, it discussesvarious standard and customized SAS macros developed to cal-culate different types of p-values, as well as some strategies andtechniques for incorporating p-values into Summary Tables . Thepaper focuses on generating Summary Tables for the Clinical trialswith two independent treatment groups. However, the methodsdiscussed here can be extended and applied to Clinical trials withmultiple dosing regimen or paired treatment groups by extendingthe corresponding SAS macros. table 1 illustrates the specifica-tion of the Summary Tables that are the focus of this 1. Specification of Summary table with P-ValuesReport X(Page x of y)Title of Clinical TrialTitle of TableInvestigator: abcVariableTreatment A (N = xx)Treatment B ( N= xx)P-ValueVar1, N (%)na(%)nb(%) STDstdastdb MEDIAN medamedb MIN-MAXmina-maxaminb-maxbThe authors first discuss the classification of various p-values interms of methodological considerations and output locations of p-values from SAS statistical procedures.

5 The methodological con-siderations include: Type (categorical, quantitative, or survival) of key variable tobe analyzed. Number of other variables (class variables, independentvariables, or controlling variables) to be included. Concern about statistical distributions (parametric or non-parametric).Output locations include three possible sources of the p-valuesfrom SAS statistical procedures: SAS data set created by procedures. Log file printed after using procedures. Calculation based on estimated parameters and classification provides an adequate framework for effectivedevelopment of various standard and customized SAS macrosfor generating groups of SAS macros are developed in the next section:function macros, procedure macros, and utility macros.

6 A func-tion macro corresponds to a specific statistical test to generate aspecified p-value; it contains a SAS statistical procedure that maybe integrated in a procedure macro and several data manipula-tion steps that may be captured by the utility macros. The thirdsection discusses some programming techniques and examplesfor efficient and accurate use of those function macros, such aspreparing input data sets, manipulating output data sets, andutilizing options in the macros. The concluding section providessome directions for building a complete standardized system togenerate most Summary Tables for any Clinical OF STATISTICAL TESTS (P-Values)In this section, methodological aspects and output locations re-lated to corresponding statistical tests and SAS statistical proce-dures are discussed and used to classify statistical tests, with aview to establish an adequate framework for effective devel-opment of various standard and customized SAS macros forgenerating p-values.

7 The methodological aspects focus on differ-ences in the statistical tests used to generate p-values in terms ofdifferences in the variables analyzed and distributions assumed;2the output locations refer to the differences in output of p-valuesfrom various SAS statistical those statistical tests dealing with comparative Clinical trialstudies for two independent treatment groups are considered inthis paper. These statistical tests, or corresponding p-values,can be classified by three factors: types of key variables to betested, number of other variables to be included, and thestatistical distribution to be assumed. table 2 displays theclassification of statistical tests for generating p-values and thecorresponding SAS of Statistical Tests (or P-Values) forClinical Trials with Two Independent Treatment GroupsType ofOtherStatistical Tests (SAS Procedures)KeyVariableVariablesIncludedP arametricNonparametricCategoryTreatmentC hi-square(Pearson orcontinuity adj.)

8 (PROC FREQ)Fisher s exacttest(PROC FREQ)Treatment+ OthersCMH with tablescore(PROC FREQ)LogisticRegression(PROCLOGISTIC orPROCCATMOD)CMH with rankscore(PROC FREQ)QuantitativeTreatmentUnpaired t test(PROC TTEST)Wilcoxon rank-sum test(PROCNPAR1 WAY)Treatment+ OthersGLM(PROC GLM)GLM with rank-transformeddata(PROC RANK+ PROC GLM)SurvivalTreatmentorTreatment+ OthersChi-square(PROCLIFEREG)Log-rank test(PROCLIFETEST)Cox proportional hazards model(Semiparametric) (PROC PHREG)Three types of variables are regularly treated in statistical analy-sis for Clinical trials: categorical, quantitative, and variables are characterized as having a limited num-ber of discrete values that can be nominal, ordinal, or examples of categorical variables in Clinical trials aremeasures of treatment outcomes ( , failure or success, NIHstroke scale score, Glasgow outcome score), investigator num-ber, treatment group, race, sex, age group, and variables captur-ing whether a patient took a specific medication or had a specificadverse event.

9 Chi-square statistic, Fisher s exact test, Cochran-Mantel-Haenzel (CMH) procedure, and logistic regression arecommonly used to deal with such categorical , or continuous, variables are defined as those thatcan be put into an infinite number of continuous values. In clini-cal trials, some treatment outcomes ( , rate of new episodesover a specific period) and most laboratory test parameters ( ,RBC, WBC, SGOT, SGPT, and pH) are quantitative statistical tests used for quantitative variables includeunpaired t-test, Wilcoxon rank-sum test, and general linear mod-els (GLM).Survival variables are used to measure the duration of time forthe occurrence of a specific event. Some common examples ofsurvival variables are time to failure, time to a new episode, ortime to withdraw from treatment.

10 Survival data can be analyzedby Chi-square statistic, log-rank test, or the Cox proportionalhazards major task of statistical analysis in a comparative Clinical trialis to compare the difference of a specific measure ( , mean orproportion) between two treatment groups. The treatment vari-able alone may enter statistical models as an independent vari-able or a class variable. In this case, simple statistical tests andcorresponding SAS statistical procedures can be applied to gen-erate a single p-value. Unpaired t-test, Fisher s exact test, orWilcoxon rank-sum test is usually used to deal with this variables (either categorical or quantitative) often need tobe included into the statistical models as additional independentvariables, stratified variables, or controlling variables.