| SP01 |
Sample Size Estimation for Trials of Recurrent Events |
Kuolung Hu (Amgen) |
In some clinical trials, repeated occurrences of the same type of event are the study endpoint. For example, numbers of adverse event of interest occurred during the study period. There are several standard statistical methods for the analysis of such data, including the Anderson-Gill model (Anderson and Gill, 1982) and the Poisson regression model (Frome et. al. 1973). Signorini (1991) has proposed a method to estimate power for the Poisson model in the similar situation, but no approaches to power calculations are available for the Anderson-Gill model. We illustrate SAS macro to implement Signorini's method and extend it to incorporate unequal randomization between the treatment and control groups. Furthermore, using simulations, we show that the power for both the Poisson model and the Anderson-Gill model are similar under a variety of scenarios, so that this approach to sample size calculations can be used for either method of analysis in a study. |
| SP02 |
Sample size estimation for (Bio)equivalence Testing between two treatments |
Madan G. Kundu (i3 Statprobe) |
Standard 2X2 and replicated 2X2m crossover designs are recommended in the regulatory guidelines to establish bioequivalence of generic drug with off patent brand-name drug. Estimation of sample size, in any clinical trial, targets to optimize the resource usage with assurance of having adequate probability (Power) to get significant result. This paper discusses the statistical concept behind sample size estimation for 2X2 and 2X2m bioequivalence trials along with its implementation in SAS. Procedure of sample size estimation in SAS has been discussed through some simple datastep functions (%BESS, %BESSm and %BESSg) and PROC POWER, a procedure for calculation of sample size. To assess the accuracy of the macro (%BESS, %BESSm and %BESSg) and PROC POWER, estimated sample size obtained through these two procedures have been compared with those estimated by Diletti et. al.(1991). It has been observed that sample size obtained through macros and PROC POWER is matching almost perfectly with those obtained by Diletti et. al. |
| SP03 |
Sample Size Estimation Through Simulation of a Random Coefficient Model by Using SAS |
Junliang Chen (Talecris Biotherapeutics, Inc) |
In chronic pulmonary diseases, the development of emphysema is a slow progress over many years and the assessment of drug effects on this process requires the observation of large number of patients for a long period of time. Recently, lung densitometry (measuring the lung density through CT scan) has been studied as a potential clinical endpoint for assessing the lung tissue loss over time in patients with emphysema. The clinical trial with lung densitometry as an endpoint is typically designed as a longitudinal study with repeated measured at fixed intervals. Also, considering that the lung density measurement is closely correlated with lung volume measurement, the lung volume has to be included in the statistical model as a time-dependent covariate. The clinical benefits can be assessed by comparing the difference in slow progression of lung density loss between two treatment groups through random coefficient model – a longitudinal linear mixed model with random intercept and slope. However, in planning the clinical trial with such complex statistical analyses, estimating the sample size required for a given power to detect a specified treatment difference is an important issue. In this article, an empirical approach is proposed to estimate the sample size by simulating trajectories of lung density and lung volume using SAS. We present the details step-by-step for sample size estimation through simulation and discuss the pros and cons of this approach. SAS Macros will be provided. |
| SP04 |
Statistical Approach to Establishing Bioequivalence |
William F. McCarthy (MMRI), Nan Guo (MMRI) |
This paper presents the statistical approaches outlined in the FDA guidance for industry regarding bioequivalence studies for orally administered drug products. Three forms of bioequivalence are covered: average bioequivalence (ABE), population bioequivalence (PBE) and individual bioequivalence (IBE). The method of analysis for each of the three forms is based on Jones and Kenward (2003) and a modification of their SAS Macro is provided.
Jones B and Kenward MG (2003). Design and Analysis of Cross-Over Trials, Second Edition. Chapman & Hall/CRC, New York. |
| SP05 |
Approximate Simultaneous Nonparametric Confidence Intervals for All-Pairwise Comparisons and Comparisons with a Designated (Control) Group in a One-Way Layout Using the SAS System |
Michelle M. Gayari (MMS Holdings), Paul Juneau (MMS Holdings, Inc), Aditya Sikka (MMS Holdings, Inc.), Jiawei Qiao (MMS Holdings, Inc.) |
The SAS System (Version 9) presents users with the ability to perform standard parametric multiple comparisons for a one-way layout with PROC GLM and PROC MIXED using the means and lsmeans statements, respectively. Options in these statements exist to construct simultaneous confidence intervals for all pair-wise comparisons or all pair-wise comparisons with a designated (control) group and may be used under circumstances were measurement error is approximately Gaussian (normal).
It is quite common in the early phases of medical research for investigators to have limited knowledge regarding the distributional properties of their measurements. Moreover, in later phases of clinical research, previous experience with important efficacy measurements suggests that the distributional nature of the error associated with some is non-Gaussian. To date, SAS code exists to perform simultaneous nonparametric inference for all pair-wise comparisons and all pair-wise comparisons with a control group, but not to construct the corresponding interval estimates.
In this presentation, the speaker will introduce some original SAS code to construct approximate simultaneous confidence intervals for all pair-wise comparisons and all pair-wise treatment comparisons with a designated control group from measurements occurring in a one-way layout with errors that are non-Gaussian. After illustrating a statistical solution for this setting, the speaker will discuss the flow of this macro code, illustrate its usage and output and demonstrate the application of the program to the analysis of a real data set from a medical research investigation in cancer research.
|
| SP06 |
Use SAS Programming to Calculate AUC in Pharmacokinetic Studies |
John He (Barr Laboratories) |
The AUC (the area under curve) is an important parameter in PK (pharmacokinetics) analysis. There are different ways to calculate AUC. Sometime, the statistician or pharmacokineticist have to choose one particular method to calculate AUC depending on the concentration data.
There are two major different approaches to calculation of AUC: one is compartmental modeling analysis, the other is model independent (non-compartmental) analysis. Among the non-compartmental approach, traditional trapezoidal, log-linear trapezoidal, Lagrange polynomial integration, and Purves method are currently being used. The linear method can be use alone or in conjunction with log-linear method if concentration data has a smooth "up and down" profiles. The linear trapezoidal can be used for both ascending and descending portion of the curve while descending portion is generally concaved by log-linear method. It is generally understood by PK analyzers that the linear method usually underestimates the AUC but overestimate it in descending phase. The Lagrange method can be used to estimate the AUC by interpolating values between consecutive data points for the data with multiple peaks.
A SAS program is developed to examine the data and determine the method to calculate the AUC.
1. Linear trapezoidal
&AUC=((&C0+&C1)*(.5))*(&T1-&T0);
2. Log-linear trapezoidal
&AUC=((&C0-&C1)*(&T1-&T0))/(LOG(&C0)-LOG(&C1));
3. Lagrange polynomial integration
4. Purves method
A PK concentration data examples will be used for analysis. The concentration graphs and tables will be constructed with SAS program as the linear trapezoidal and Lagrange integration require the examination of concentration data and selection of data points used in the analysis.
Finally, the comparison from all four methods will be given and discussed.
|
| SP07 |
AUTOMATED SAS® MACRO FOR ADVERSE EVENT LOGISTIC REGRESSION ANALYSIS |
Suhas R. Sanjee (MaxisIT Inc), Sheng Zhang (Merck Co. & Inc.) |
Logistic regression analysis can be used to determine whether demographic or other factors are associated with the occurrence of adverse events (AE) in clinical trials. In this paper we begin by introducing a few SAS procedures for performing logistic regression analyses, e.g., GENMOD, LOGISTIC, etc. We then present a SAS macro which generates multivariate and univariate models dynamically for adverse events satisfying prespecified conditions. In these models, the outcome variable is defined as whether the subject had an AE of interest and the covariates include age, sex, race, and weight. Both univariate and multivariate models were fit for each AE under consideration. Scenarios are also presented if an AE does not occur in the reference category for a categorical covariate. If any given AE did not occur in the reference category for a given categorical covariate, the univariate model was not run for that covariate, and the multivariate model did not include that covariate. This is because the odds ratio will become inestimable for that covariate. In multivariate analysis, a single model is fit for each AE that includes all of the covariates (unless a categorical covariate needs to be excluded, as described above). The estimate and the corresponding p-values for continuous variables and for the individual levels of the categorical covariates, are stored in the ODS dataset created by GENMOD procedure. The odds ratio can be computed as exp(ESTIMATE). |
| SP09 |
Using ESTIMATE and CONTRAST Statements for Customized Hypothesis Tests |
Hanyu Chen (CAZ Consulting Corporation) |
ESTIMATE and CONTRAST statements in a number of SAS procedures permit customized hypothesis tests and make different statistical comparisons easy to perform. The key to use these two statements is to determine the proper weights or coefficients, which can be very challenging in complicated cases. Using PROC GLM and ANOVA model, this paper discusses a process for obtaining the weights appropriate for the intended customized comparison. Examples using the process are also provided. |
| SP10 |
A SAS Macro for Single Imputation |
xingshu Zhu (Merck), Shuping Zhang (Merck), Jane Liao (Merck) |
Single imputation is often used to replace the missing value of a variable in a dataset because this approach is both simple and efficient. Single imputation is particularly useful when working with an extensive dataset containing millions of records and a large number of variables. In this case, it would be highly impractical, or maybe even impossible, to use the multiple imputation method, performed by the procedure PROC MI, because of the overwhelmingly large number of datasets that would have to be created. In this paper, we introduce a simple SAS macro that allows the user to create a “complete” SAS dataset through single imputation by selecting different statistical methods and applying them to data on a given patient or to information from the entire dataset. |
| SP12 |
Bayesian Data Analysis Using %WinBUGS |
Lei Zhang (Celgene) |
WinBUGS is a powerful statistical tool for Bayesian analysis using Markov chain Monte Carlo (McMC) methods. It has been used to construct and analyze a wide variety of Bayesian models in many application areas; however, it has very limited capabilities for data manipulation, graph customization, and comparison with other statistical methods. The SAS System is a dominant statistical platform with rich functions for data management, graph generation, and analytical statistical analyses. Therefore, making WinBUGS work with SAS will create much-wanted synergy. This paper introduces a macro called %WinBUGS that gives you the edge to perform a Bayesian analysis using WinBUGS from within the SAS System. With %WinBUGS, you can convert SAS datasets into WinBUGS data files, invoke WinBUGS to perform the intended Bayesian analysis, and then get back results into SAS for further analyses and reporting. All that you need to do is to create a WinBUGS analysis file, and submit it to %WinBUGS for execution. In this paper, I first describe the mechanism used in %WinBUGS, explain the syntax of %WinBUGS directives, and then give examples to demonstrate how to write WinBUGS analysis files to automate Bayesian analyses using features from both SAS and WinBUGS. |