Design and Analysis of Pragmatic Trials
本书讨论了整群随机试验中实用性试验的设计问题，涵盖缺失数据处理、样本量计算、分层设计、配对设计及阶梯楔形设计等，适合研究人员和生物统计学研究生参考。
Cluster randomized trials (CRTs) serve as a key component of health promotion and prevention strategies.Under this, interventions are carried out at community level, in a routine clinical practice.They remain pragmatic trials (PCT) involving various pragmatic issues to be dealt with at the design stage.The present book has dealt with pragmatic issues under CRTs at design stage.In the first chapter, PCTs and involved statistical issues like missing data mechanisms and required imputation methods; within cluster similarity measures; complicated correlation structure; and limited number of clusters are discussed.Further, under CRTs, total variance, within and between cluster variance; multiple-period CRTs including longitudinal, crossover and stepped-wedge (SW); and related sample size explorations and analytical methods are discussed.In the second chapter, considering hierarchical structure of data and continuous outcome, analytical methods including generalized estimating equation (GEE) method, and mixed-effects linear regression models are described.Likewise, in case of binary outcomes, GEE approach, and generalized linear mixed model approach are discussed.Also, in case of count outcomes, GEE approach is discussed.Cluster size determination for a fixed number of clusters is also discussed.The third chapter on matched-pair CRT consists of associated impact of correlation and missing data.In presence of missing continuous outcomes, sample size estimation based on GEE approach, adjustment of inflated type I error, and sensitivity analysis is discussed.Similar discussion is also presented in case of missing binary outcomes.The authors have opined to take into account related impact of correlation structures and missing data on sample size exploration.While discussing stratified cluster randomized design in the fourth chapter, they have discussed about its due considerations.In case of continuous outcomes, sample size estimation based on GEE approach, and relative sample size changes due to varying cluster size are described.For binary outcomes, they have reported sample size estimation based on Cochran-Mantel-Haenszel statistic, relative sample size change due to varying cluster size, and estimation of clustering parameter.The GEE approach for SW trial design has been discussed in fifth chapter covering its brief review; applications with a continuous outcome accounting for missing data, simulation research, adjusting for underestimated variances for small sample sizes, and considerations for efficiency and robustness.For binary outcomes, they have accounted for extension to outcomes from the exponential family.They have investigated remedy for known limitation of GEE approach as underestimation of variance with limited numbers of clusters.Finally, the mixed-effect model approach and adaptive strategies for SW trial design are covered in the sixth chapter.This covers sample size calculation based on cluster-step means considering commonly used correlation structures.Also, adaptive strategies for SW trials investigating frequentist group sequential design and Bayesian predictive probability approach are discussed.It is correctly pointed out that adaptive strategies enable researchers to modify design in the middle of a clinical trial, without destroying the validity and integrity of intended studies.This book may be helpful to the researchers in the design and implementation of PCTs; and serve as text book for the graduate level biostatistics students.