Clinical Trial Design, Monitoring, and Analysis

This task view gathers information on specific R packages for design, monitoring and analysis of data from clinical trials. It focuses on including packages for clinical trial design and monitoring in general plus data analysis packages for a specific type of design. Also, it gives a brief introduction to important packages for analyzing clinical trial data. Please refer to task views ExperimentalDesign, Survival, Pharmacokinetics for more details on these topics. Please feel free to e-mail me regarding new packages or major package updates.

Design and Monitoring

  • TrialSize This package has more than 80 functions from the book Sample Size Calculations in Clinical Research (Chow & Wang & Shao, 2007, 2nd ed., Chapman &Hall/CRC).
  • asd This Package runs simulations for adaptive seamless designs using early outcomes for treatment selection.
  • bcrm This package implements a wide variety of one and two-parameter Bayesian CRM designs. The program can run interactively, allowing the user to enter outcomes after each cohort has been recruited, or via simulation to assess operating characteristics.
  • blockrand creates randomizations for block random clinical trials. It can also produce a PDF file of randomization cards.
  • This small package contains a series of simple tools for constructing and manipulating confounded and fractional factorial designs.
  • CRTSize This package contains basic tools for the purpose of sample size estimation in cluster (group) randomized trials. The package contains traditional power-based methods, empirical smoothing (Rotondi and Donner, 2009), and updated meta-analysis techniques (Rotondi and Donner, 2011).
  • dfcrm This package provides functions to run the CRM and TITE-CRM in phase I trials and calibration tools for trial planning purposes.
  • experiment contains tools for clinical experiments, e.g., a randomization tool, and it provides a few special analysis options for clinical trials.
  • FrF2 This package creates regular and non-regular Fractional Factorial designs. Furthermore, analysis tools for Fractional Factorial designs with 2-level factors are offered (main effects and interaction plots for all factors simultaneously, cube plot for looking at the simultaneous effects of three factors, full or half normal plot, alias structure in a more readable format than with the built-in function alias). The package is currently subject to intensive development. While much of the intended functionality is already available, some changes and improvements are still to be expected.
  • GroupSeq performs computations related to group sequential designs via the alpha spending approach, i.e., interim analyses need not be equally spaced, and their number need not be specified in advance.
  • gsDesign derives group sequential designs and describes their properties.
  • ldBand from Hmisc computes and plots group sequential stopping boundaries from the Lan-DeMets method with a variety of a-spending functions using the ld98 program from the Department of Biostatistics, University of Wisconsin written by DM Reboussin, DL DeMets, KM Kim, and KKG Lan.
  • ldbounds uses Lan-DeMets Method for group sequential trial; its functions calculate bounds and probabilities of a group sequential trial.
  • longpower The longpower package contains functions for computing power and sample size for linear models of longitudinal data based on the formula due to Liu and Liang (1997) and Diggle et al (2002). Either formula is expressed in terms of marginal model or Generalized Estimating Equations (GEE) parameters. This package contains functions which translate pilot mixed effect model parameters (e.g. random intercept and/or slope) into marginal model parameters so that the formulas of Diggle et al or Liu and Liang formula can be applied to produce sample size calculations for two sample longitudinal designs assuming known variance.
  • PIPS generates predicted interval plots, simulates and plots confidence intervals of an effect estimate given observed data and a hypothesis about the distribution of future data.
  • PowerTOST contains functions to calculate power and sample size for various study designs used for bioequivalence studies. See function known.designs() for study designs covered. Moreover the package contains functions for power and sample size based on 'expected' power in case of uncertain (estimated) variability. Added are functions for the power and sample size for the ratio of two means with normally distributed data on the original scale (based on Fieller's confidence ('fiducial') interval).
  • pwr has power analysis functions along the lines of Cohen (1988).
  • PwrGSD is a set of tools to compute power in a group sequential design.
  • qtlDesignprovides tools for the design of QTL experiments.
  • samplesize computes sample size for Student's t-test with equal and nonequal variances and for the Wilcoxon-Mann-Whitney test for categorical data with and without ties.
  • seqmon is computes the probability of crossing sequential efficacy and futility boundaries in a clinical trial. It implements the Armitage-McPherson and Rowe Algorithm using the method described in Schoenfeld (2001).

Design and Analysis

  • Package AGSDest This package provides tools and functions for parameter estimation in adaptive group sequential trials.
  • Package clinfun has functions for both design and analysis of clinical trials. For phase II trials, it has functions to calculate sample size, effect size, and power based on Fisher's exact test, the operating characteristics of a two-stage boundary, Optimal and Minimax 2-stage Phase II designs given by Richard Simon, the exact 1-stage Phase II design and can compute a stopping rule and its operating characteristics for toxicity monitoring based repeated significance testing. For phase III trials, it can calculate sample size for group sequential designs.
  • Package CRM Continual Reassessment Method (CRM) simulator for Phase I Clinical Trials.
  • Package DoseFinding provides functions for the design and analysis of dose-finding experiments (for example pharmaceutical Phase II clinical trials). It provides functions for: multiple contrast tests, fitting non-linear dose-response models, calculating optimal designs and an implementation of the MCPMod methodology. Currently only normally distributed homoscedastic endpoints are supported.
  • MCPMod This package implements a methodology for the design and analysis of dose-response studies that combines aspects of multiple comparison procedures and modeling approaches (Bretz, Pinheiro and Branson, 2005, Biometrics 61, 738-748). The package provides tools for the analysis of dose finding trials as well as a variety of tools necessary to plan a trial to be conducted with the MCPMod methodology.
  • Package TEQR The target equivalence range (TEQR) design is a frequentist implementation of the modified toxicity probability interval (mTPI) design and a competitor to the standard 3+3 design (3+3). The 3+3 is the work horse design in Phase I. It is good at determining if a safe dose exits, but provides poor accuracy and precision in estimating the level of toxicity at the maximum tolerated dose (MTD). The TEQR is better than the 3+3 when compared on: 1) the number of times the dose at or nearest the target toxicity level was selected as the MTD, 2) the number of subjects assigned to doses levels, at or nearest the MTD, and 3) the overall trial DLT rate. TEQR more accurately and more precisely estimates the rate of toxicity at the MTD because a larger number of subjects are studied at the MTD dose. The TEQR on average uses fewer subjects and provide reasonably comparable results to the continual reassessment method (CRM) in the number of times the dose at or nearest the target toxicity level was selected as the MTD and the number of subjects assigned doses, at, or nearest the target and in overall DLT rate.

Analysis for Specific Designs

  • adaptTest The functions defined in this program serve for implementing adaptive two-stage tests. Currently, four tests are included: Bauer and Koehne (1994), Lehmacher and Wassmer (1999), Vandemeulebroecke (2006), and the horizontal conditional error function. User-defined tests can also be implemented. Reference: Vandemeulebroecke, An investigation of two-stage tests, Statistica Sinica 2006.
  • bifactorial makes global and multiple inferences for given bi- and trifactorial clinical trial designs using bootstrap methods and a classical approach.
  • clinsig This function calculates both parametric and non-parametric versions of the Jacobson-Truax estimates of clinical significance.
  • MChtest performs Monte Carlo hypothesis tests. It allows a couple of different sequential stopping boundaries (a truncated sequential probability ratio test boundary and a boundary proposed by Besag and Clifford (1991). It gives valid p-values and confidence intervals on p-values.
  • nppbibimplements a nonparametric statistical test for rank or score data from partially-balanced incomplete block-design experiments.
  • speff2trial, the package performs estimation and testing of the treatment effect in a 2-group randomized clinical trial with a quantitative or dichotomous endpoint.
  • ThreeGroups This package implements the Maximum Likelihood estimator for three-group designs proposed by Gerber, Green, Kaplan, and Kern (2010).

Analysis in General

  • Base R, especially the stats package, has a lot of functionality useful for design and analysis of clinical trials. For example, chisq.test, prop.test, binom.test, t.test, wilcox.test, kruskal.test, mcnemar.test, cor.test, power.t.test, power.prop.test, power.anova.test, lm, glm, nls, anova (and its lm and glm methods) among many others.
  • asypow has a set of routines for calculating power and related quantities utilizing asymptotic likelihood ratio methods.
  • binomSamSize is a suite of functions for computing confidence intervals and necessary sample sizes for the success probability parameter Bernoulli distribution under simple random sampling or under pooled sampling.
  • coin offers conditional inference procedures for the general independence problem including two-sample, K-sample (non-parametric ANOVA), correlation, censored, ordered and multivariate problems.
  • epibasix has functions such as diffdetect, n4means for continuous outcome and n4props and functions for matched pairs analysis in randomized trials.
  • ae.dotplot from HH shows a two-panel display of the most frequently occurring adverse events in the active arm of a clinical study.
  • The Hmisc package contains around 200 miscellaneous functions useful for such things as data analysis, high-level graphics, utility operations, functions for computing sample size and power, translating SAS datasets into S, imputing missing values, advanced table making, variable clustering, character string manipulation, conversion of S objects to LaTeX code, recoding variables, and bootstrap repeated measures analysis.
  • multcomp covers simultaneous tests and confidence intervals for general linear hypotheses in parametric models, including linear, generalized linear, linear mixed effects, and survival models.
  • survival contains descriptive statistics, two-sample tests, parametric accelerated failure models, Cox model. Delayed entry (truncation) allowed for all models; interval censoring for parametric models. Case-cohort designs.
  • ssanv is a set of functions to calculate sample size for two-sample difference in means tests. Does adjustments for either nonadherence or variability that comes from using data to estimate parameters.


  • metasens is a package for statistical methods to model and adjust for bias in meta-analysis
  • meta is for fixed and random effects meta-analysis. It has Functions for tests of bias, forest and funnel plot.
  • metafor consists of a collection of functions for conducting meta-analyses. Fixed- and random-effects models (with and without moderators) can be fitted via the general linear (mixed-effects) model. For 2x2 table data, the Mantel-Haenszel and Peto's method are also implemented.
  • metaLik Likelihood inference in meta-analysis and meta-regression models.
  • rmeta has functions for simple fixed and random effects meta-analysis for two-sample comparisons and cumulative meta-analyses. Draws standard summary plots, funnel plots, and computes summaries and tests for association and heterogeneity.

View on CRAN

10 months ago

Ed Zhang and Harry G. Zhang