Intra-cluster correlations from the CLustered OUtcome Dataset bank to inform the design of longitudinal cluster trials

School-level intra-cluster correlation coefficients and autocorrelations for children's accelerometer-measured physical activity in England by age and gender

BACKGROUND

Randomised, cluster-based study designs in schools are commonly used to evaluate children's physical activity interventions. Sample size estimation relies on accurate estimation of the intra-cluster correlation coefficient (ICC), but published estimates, especially using accelerometry-measured physical activity, are few and vary depending on physical activity outcome and participant age. Less commonly-used cluster-based designs, such as stepped wedge designs, also need to account for correlations over time, e.g. cluster autocorrelation (CAC) and individual autocorrelation (IAC), but no estimates are currently available. This paper estimates the school-level ICC, CAC and IAC for England children's accelerometer-measured physical activity outcomes by age group and gender, to inform the design of future school-based cluster trials.

METHODS

Data were pooled from seven large English datasets of accelerometer-measured physical activity data between 2002-18 (> 13,500 pupils, 540 primary and secondary schools). Linear mixed effect models estimated ICCs for weekday and whole week for minutes spent in moderate-to-vigorous physical activity (MVPA) and being sedentary for different age groups, stratified by gender. The CAC (1,252 schools) and IAC (34,923 pupils) were estimated by length of follow-up from pooled longitudinal data.

RESULTS

School-level ICCs for weekday MVPA were higher in primary schools (from 0.07 (95% CI: 0.05, 0.10) to 0.08 (95% CI: 0.06, 0.11)) compared to secondary (from 0.04 (95% CI: 0.03, 0.07) to (95% CI: 0.04, 0.10)). Girls' ICCs were similar for primary and secondary schools, but boys' were lower in secondary. For all ages, combined the CAC was 0.60 (95% CI: 0.44-0.72), and the IAC was 0.46 (95% CI: 0.42-0.49), irrespective of follow-up time. Estimates were higher for MVPA vs sedentary time, and for weekdays vs the whole week.

CONCLUSIONS

Adequately powered studies are important to evidence effective physical activity strategies. Our estimates of the ICC, CAC and IAC may be used to plan future school-based physical activity evaluations and were fairly consistent across a range of ages and settings, suggesting that results may be applied to other high income countries with similar school physical activity provision. It is important to use estimates appropriate to the study design, and that match the intended study population as closely as possible.

A simple and effective method for simulating nested exchangeable correlated binary data for longitudinal cluster randomised trials

BACKGROUND

Simulation is an important tool for assessing the performance of statistical methods for the analysis of data and for the planning of studies. While methods are available for the simulation of correlated binary random variables, all have significant practical limitations for simulating outcomes from longitudinal cluster randomised trial designs, such as the cluster randomised crossover and the stepped wedge trial designs. For these trial designs as the number of observations in each cluster increases these methods either become computationally infeasible or their range of allowable correlations rapidly shrinks to zero.

METHODS

In this paper we present a simple method for simulating binary random variables with a specified vector of prevalences and correlation matrix. This method allows for the outcome prevalence to change due to treatment or over time, and for a 'nested exchangeable' correlation structure, in which observations in the same cluster are more highly correlated if they are measured in the same time period than in different time periods, and where different individuals are measured in each time period. This means that our method is also applicable to more general hierarchical clustered data contexts, such as students within classrooms within schools. The method is demonstrated by simulating 1000 datasets with parameters matching those derived from data from a cluster randomised crossover trial assessing two variants of stress ulcer prophylaxis.

RESULTS

Our method is orders of magnitude faster than the most well known general simulation method while also allowing a much wider range of correlations than alternative methods. An implementation of our method is available in an R package NestBin.

CONCLUSIONS

This simulation method is the first to allow for practical and efficient simulation of large datasets of binary outcomes with the commonly used nested exchangeable correlation structure. This will allow for much more effective testing of designs and inference methods for longitudinal cluster randomised trials with binary outcomes.

Sample size and power calculation for testing treatment effect heterogeneity in cluster randomized crossover designs

The cluster randomized crossover design has been proposed to improve efficiency over the traditional parallel-arm cluster randomized design. While statistical methods have been developed for designing cluster randomized crossover trials, they have exclusively focused on testing the overall average treatment effect, with little attention to differential treatment effects across subpopulations. Recently, interest has grown in understanding whether treatment effects may vary across pre-specified patient subpopulations, such as those defined by demographic or clinical characteristics. In this article, we consider the two-treatment two-period cluster randomized crossover design under either a cross-sectional or closed-cohort sampling scheme, where it is of interest to detect the heterogeneity of treatment effect via an interaction test. Assuming a patterned correlation structure for both the covariate and the outcome, we derive new sample size formulas for testing the heterogeneity of treatment effect with continuous outcomes based on linear mixed models. Our formulas also address unequal cluster sizes and therefore allow us to analytically assess the impact of unequal cluster sizes on the power of the interaction test in cluster randomized crossover designs. We conduct simulations to confirm the accuracy of the proposed methods, and illustrate their application in two real cluster randomized crossover trials.

Optimal designs using generalized estimating equations in cluster randomized crossover and stepped wedge trials

Cluster randomized crossover and stepped wedge cluster randomized trials are two types of longitudinal cluster randomized trials that leverage both the within- and between-cluster comparisons to estimate the treatment effect and are increasingly used in healthcare delivery and implementation science research. While the variance expressions of estimated treatment effect have been previously developed from the method of generalized estimating equations for analyzing cluster randomized crossover trials and stepped wedge cluster randomized trials, little guidance has been provided for optimal designs to ensure maximum efficiency. Here, an optimal design refers to the combination of optimal cluster-period size and optimal number of clusters that provide the smallest variance of the treatment effect estimator or maximum efficiency under a fixed total budget. In this work, we develop optimal designs for multiple-period cluster randomized crossover trials and stepped wedge cluster randomized trials with continuous outcomes, including both closed-cohort and repeated cross-sectional sampling schemes. Local optimal design algorithms are proposed when the correlation parameters in the working correlation structure are known. MaxiMin optimal design algorithms are proposed when the exact values are unavailable, but investigators may specify a range of correlation values. The closed-form formulae of local optimal design and MaxiMin optimal design are derived for multiple-period cluster randomized crossover trials, where the cluster-period size and number of clusters are decimal. The decimal estimates from closed-form formulae can then be used to investigate the performances of integer estimates from local optimal design and MaxiMin optimal design algorithms. One unique contribution from this work, compared to the previous optimal design research, is that we adopt constrained optimization techniques to obtain integer estimates under the MaxiMin optimal design. To assist practical implementation, we also develop four SAS macros to find local optimal designs and MaxiMin optimal designs.

Maintaining the validity of inference from linear mixed models in stepped-wedge cluster randomized trials under misspecified random-effects structures

Linear mixed models are commonly used in analyzing stepped-wedge cluster randomized trials. A key consideration for analyzing a stepped-wedge cluster randomized trial is accounting for the potentially complex correlation structure, which can be achieved by specifying random-effects. The simplest random effects structure is random intercept but more complex structures such as random cluster-by-period, discrete-time decay, and more recently, the random intervention structure, have been proposed. Specifying appropriate random effects in practice can be challenging: assuming more complex correlation structures may be reasonable but they are vulnerable to computational challenges. To circumvent these challenges, robust variance estimators may be applied to linear mixed models to provide consistent estimators of standard errors of fixed effect parameters in the presence of random-effects misspecification. However, there has been no empirical investigation of robust variance estimators for stepped-wedge cluster randomized trials. In this article, we review six robust variance estimators (both standard and small-sample bias-corrected robust variance estimators) that are available for linear mixed models in R, and then describe a comprehensive simulation study to examine the performance of these robust variance estimators for stepped-wedge cluster randomized trials with a continuous outcome under different data generators. For each data generator, we investigate whether the use of a robust variance estimator with either the random intercept model or the random cluster-by-period model is sufficient to provide valid statistical inference for fixed effect parameters, when these working models are subject to random-effect misspecification. Our results indicate that the random intercept and random cluster-by-period models with robust variance estimators performed adequately. The CR3 robust variance estimator (approximate jackknife) estimator, coupled with the number of clusters minus two degrees of freedom correction, consistently gave the best coverage results, but could be slightly conservative when the number of clusters was below 16. We summarize the implications of our results for the linear mixed model analysis of stepped-wedge cluster randomized trials and offer some practical recommendations on the choice of the analytic model.

Reconsidering stepped wedge cluster randomized trial designs with implementation periods: Fewer sequences or the parallel-group design with baseline and implementation periods are potentially more efficient

BACKGROUND/AIMS

When designing a cluster randomized trial, advantages and disadvantages of tentative designs must be weighed. The stepped wedge design is popular for multiple reasons, including its potential to increase power via improved efficiency relative to a parallel-group design. In many realistic settings, it will take time for clusters to fully implement the intervention. When designing the HEALing (Helping to End Addiction Long-termSM) Communities Study, implementation time was a major consideration, and we examined the efficiency and practicality of three designs. Specifically, a three-sequence stepped wedge design with implementation periods, a corresponding two-sequence modified design that is created by removing the middle sequence, and a parallel-group design with baseline and implementation periods. In this article, we study the relative efficiencies of these specific designs. More generally, we study the relative efficiencies of modified designs when the stepped wedge design with implementation periods has three or more sequences. We also consider different correlation structures.

METHODS

We compare efficiencies of stepped wedge designs with implementation periods consisting of three to nine sequences with a variety of corresponding designs. The three-sequence design is compared to the two-sequence modified design and to the parallel-group design with baseline and implementation periods analysed via analysis of covariance. Stepped wedge designs with implementation periods consisting of four or more sequences are compared to modified designs that remove all or a subset of 'middle' sequences. Efficiencies are based on the use of linear mixed effects models.

RESULTS

In the studied settings, the modified design is more efficient than the three-sequence stepped wedge design with implementation periods. The parallel-group design with baseline and implementation periods with analysis of covariance-based analysis is often more efficient than the three-sequence design. With respect to stepped wedge designs with implementation periods that are comprised of more sequences, there are often corresponding modified designs that improve efficiency. However, use of only the first and last sequences has the potential to be either relatively efficient or inefficient. Relative efficiency is impacted by the strength of the statistical correlation among outcomes from the same cluster; for example, the relative efficiencies of modified designs tend to be greater for smaller cluster auto-correlation values.

CONCLUSION

If a three-sequence stepped wedge design with implementation periods is being considered for a future cluster randomized trial, then a corresponding modified design using only the first and last sequences should be considered if sole focus is on efficiency. However, a parallel-group design with baseline and implementation periods and analysis of covariance-based analysis can be a practical, efficient alternative. For stepped wedge designs with implementation periods and a larger number of sequences, modified versions that remove 'middle' sequences should be considered. Due to the potential sensitivity of design efficiencies, statistical correlation should be carefully considered.

Planning stepped wedge cluster randomized trials to detect treatment effect heterogeneity

Stepped wedge design is a popular research design that enables a rigorous evaluation of candidate interventions by using a staggered cluster randomization strategy. While analytical methods were developed for designing stepped wedge trials, the prior focus has been solely on testing for the average treatment effect. With a growing interest on formal evaluation of the heterogeneity of treatment effects across patient subpopulations, trial planning efforts need appropriate methods to accurately identify sample sizes or design configurations that can generate evidence for both the average treatment effect and variations in subgroup treatment effects. To fill in that important gap, this article derives novel variance formulas for confirmatory analyses of treatment effect heterogeneity, that are applicable to both cross-sectional and closed-cohort stepped wedge designs. We additionally point out that the same framework can be used for more efficient average treatment effect analyses via covariate adjustment, and allows the use of familiar power formulas for average treatment effect analyses to proceed. Our results further sheds light on optimal design allocations of clusters to maximize the weighted precision for assessing both the average and heterogeneous treatment effects. We apply the new methods to the Lumbar Imaging with Reporting of Epidemiology Trial, and carry out a simulation study to validate our new methods.

Practical considerations for sample size calculation for cluster randomized trials

Cluster randomized trials are an essential design in public health and medical research, when individual randomization is infeasible or undesirable for scientific or logistical reasons. However, the correlation among observations within clusters leads to a decrease in statistical power compared to an individually randomised trial with the same total sample size. This correlation - often quantified using the intra-cluster correlation coefficient - must be accounted for in the sample size calculation to ensure that the trial is adequately powered. In this paper, we first describe the principles of sample size calculation for parallel-arm CRTs, and explain how these calculations can be extended to CRTs with cross-over designs, with a baseline measurement and stepped-wedge designs. We introduce tools to guide researchers with their sample size calculation and discuss methods to inform the choice of the a priori estimate of the intra-cluster correlation coefficient for the calculation. We also include additional considerations with respect to anticipated attrition, a small number of clusters, and use of covariates in the randomisation process and in the analysis.

Sample size requirements for testing treatment effect heterogeneity in cluster randomized trials with binary outcomes

Cluster randomized trials (CRTs) refer to a popular class of experiments in which randomization is carried out at the group level. While methods have been developed for planning CRTs to study the average treatment effect, and more recently, to study the heterogeneous treatment effect, the development for the latter objective has currently been limited to a continuous outcome. Despite the prevalence of binary outcomes in CRTs, determining the necessary sample size and statistical power for detecting differential treatment effects in CRTs with a binary outcome remain unclear. To address this methodological gap, we develop sample size procedures for testing treatment effect heterogeneity in two-level CRTs under a generalized linear mixed model. Closed-form sample size expressions are derived for a binary effect modifier, and in addition, a computationally efficient Monte Carlo approach is developed for a continuous effect modifier. Extensions to multiple effect modifiers are also discussed. We conduct simulations to examine the accuracy of the proposed sample size methods. We present several numerical illustrations to elucidate features of the proposed formulas and to compare our method to the approximate sample size calculation under a linear mixed model. Finally, we use data from the Strategies and Opportunities to Stop Colon Cancer in Priority Populations (STOP CRC) CRT to illustrate the proposed sample size procedure for testing treatment effect heterogeneity.

Does it decay? Obtaining decaying correlation parameter values from previously analysed cluster randomised trials

A frequently applied assumption in the analysis of data from cluster randomised trials is that the outcomes from all participants within a cluster are equally correlated. That is, the intracluster correlation, which describes the degree of dependence between outcomes from participants in the same cluster, is the same for each pair of participants in a cluster. However, recent work has discussed the importance of allowing for this correlation to decay as the time between the measurement of participants in a cluster increases. Incorrect omission of such a decay can lead to under-powered studies, and confidence intervals for estimated treatment effects can be too narrow or too wide, depending on the characteristics of the design. When planning studies, researchers often rely on previously reported analyses of trials to inform their choice of intracluster correlation. However, most reported analyses of clustered data do not incorporate a correlation decay. Thus, often all that is available are estimates of intracluster correlations obtained under the potentially incorrect assumption of no decay. In this article, we show that it is possible to use intracluster correlation values obtained from models that incorrectly omit a decay to inform plausible choices of decaying correlations. Our focus is on intracluster correlation estimates for continuous outcomes obtained by fitting linear mixed models with exchangeable or block-exchangeable correlation structures. We describe how plausible values for decaying correlations may be obtained given these estimated intracluster correlations. An online app is presented that allows users to obtain plausible values of the decay, which can be used at the trial planning stage to assess the sensitivity of sample size and power calculations to decaying correlation structures.

Key considerations for designing, conducting and analysing a cluster randomized trial

Not only do cluster randomized trials require a larger sample size than individually randomized trials, they also face many additional complexities. The potential for contamination is the most commonly used justification for using cluster randomization, but the risk of contamination should be carefully weighed against the more serious problem of questionable scientific validity in settings with post-randomization identification or recruitment of participants unblinded to the treatment allocation. In this paper we provide some simple guidelines to help researchers conduct cluster trials in a way that minimizes potential biases and maximizes statistical efficiency. The overarching theme of this guidance is that methods that apply to individually randomized trials rarely apply to cluster randomized trials. We recommend that cluster randomization be only used when necessary-balancing the benefits of cluster randomization with its increased risks of bias and increased sample size. Researchers should also randomize at the lowest possible level-balancing the risks of contamination with ensuring an adequate number of randomization units-as well as exploring other options for statistically efficient designs. Clustering should always be allowed for in the sample size calculation; and the use of restricted randomization (and adjustment in the analysis for covariates used in the randomization) should be considered. Where possible, participants should be recruited before randomizing clusters and, when recruiting (or identifying) participants post-randomization, recruiters should be masked to the allocation. In the analysis, the target of inference should align with the research question, and adjustment for clustering and small sample corrections should be used when the trial includes less than about 40 clusters.

Estimating intra-cluster correlation coefficients for planning longitudinal cluster randomized trials: a tutorial

It is well-known that designing a cluster randomized trial (CRT) requires an advance estimate of the intra-cluster correlation coefficient (ICC). In the case of longitudinal CRTs, where outcomes are assessed repeatedly in each cluster over time, estimates for more complex correlation structures are required. Three common types of correlation structures for longitudinal CRTs are exchangeable, nested/block exchangeable and exponential decay correlations-the latter two allow the strength of the correlation to weaken over time. Determining sample sizes under these latter two structures requires advance specification of the within-period ICC and cluster autocorrelation coefficient as well as the intra-individual autocorrelation coefficient in the case of a cohort design. How to estimate these coefficients is a common challenge for investigators. When appropriate estimates from previously published longitudinal CRTs are not available, one possibility is to re-analyse data from an available trial dataset or to access observational data to estimate these parameters in advance of a trial. In this tutorial, we demonstrate how to estimate correlation parameters under these correlation structures for continuous and binary outcomes. We first introduce the correlation structures and their underlying model assumptions under a mixed-effects regression framework. With practical advice for implementation, we then demonstrate how the correlation parameters can be estimated using examples and we provide programming code in R, SAS, and Stata. An Rshiny app is available that allows investigators to upload an existing dataset and obtain the estimated correlation parameters. We conclude by identifying some gaps in the literature.

Maximin optimal cluster randomized designs for assessing treatment effect heterogeneity

Cluster randomized trials (CRTs) are studies where treatment is randomized at the cluster level but outcomes are typically collected at the individual level. When CRTs are employed in pragmatic settings, baseline population characteristics may moderate treatment effects, leading to what is known as heterogeneous treatment effects (HTEs). Pre-specified, hypothesis-driven HTE analyses in CRTs can enable an understanding of how interventions may impact subpopulation outcomes. While closed-form sample size formulas have recently been proposed, assuming known intracluster correlation coefficients (ICCs) for both the covariate and outcome, guidance on optimal cluster randomized designs to ensure maximum power with pre-specified HTE analyses has not yet been developed. We derive new design formulas to determine the cluster size and number of clusters to achieve the locally optimal design (LOD) that minimizes variance for estimating the HTE parameter given a budget constraint. Given the LODs are based on covariate and outcome-ICC values that are usually unknown, we further develop the maximin design for assessing HTE, identifying the combination of design resources that maximize the relative efficiency of the HTE analysis in the worst case scenario. In addition, given the analysis of the average treatment effect is often of primary interest, we also establish optimal designs to accommodate multiple objectives by combining considerations for studying both the average and heterogeneous treatment effects. We illustrate our methods using the context of the Kerala Diabetes Prevention Program CRT, and provide an R Shiny app to facilitate calculation of optimal designs under a wide range of design parameters.

A pragmatic clinical effectiveness trial of a novel alternative to punishment for school-based substance use infractions: study protocol for the iDECIDE curriculum

Background

Adolescents who use alcohol and other drugs on school campuses are at heightened risk for adverse consequences to their health and wellbeing. Schools have historically turned to punitive approaches as a first-line response to substance use. However, punishment is an ineffective deterrent for substance use and may cause harm and increase inequities. iDECIDE (Drug Education Curriculum: Intervention, Diversion, and Empowerment) was developed as a scalable and youth-centered drug education and diversion program that can be used as a skills-based alternative to punishment. We aim to evaluate the effectiveness of the iDECIDE curriculum as an alternative to punishment (ATP) for school-based substance use infractions in the context of a large pragmatic clinical effectiveness study.

Methods

We will conduct a Type 1, hybrid effectiveness-implementation trial. Using a stepped wedge design with approximately 90 middle and high schools in Massachusetts, we will randomly allocate the timing of implementation of the iDECIDE curriculum compared to standard disciplinary response over approximately 36 months. We will test the overarching hypothesis that student-level outcomes (knowledge of drug effects and attitudes about substance use; frequency of substance use; school connectedness) improve over time as schools transition from a standard disciplinary response to having access to iDECIDE. The secondary aims of this trial are to (1) explore whether change in student-level outcomes vary according to baseline substance use, number of peers who use alcohol or other drugs, age, gender, and school urbanicity, and (2) determine the acceptability and feasibility of the iDECIDE curriculum through qualitative stakeholder interviews.

Discussion

Substance use continues to be a major and rapidly evolving problem in schools. The importance of moving away from punishment to more restorative approaches is widely accepted; however, scalable alternatives have not yet been identified. This will be the first study to our knowledge to systematically evaluate an ATP for students who violate the school substance use policy and is well poised to have important implications for policy making.

Intracluster correlation coefficients from school-based cluster randomized trials of interventions for improving health outcomes in pupils

BACKGROUND AND OBJECTIVES

To summarize intracluster correlation coefficient (ICC) estimates for pupil health outcomes from school-based cluster randomized trials (CRTs) across world regions and describe their relationship with study design characteristics and context.

METHODS

School-based CRTs reporting ICCs for pupil health outcomes were identified through a literature search of MEDLINE (via Ovid). ICC estimates were summarized both overall and for different categories of study characteristics.

RESULTS

Two hundred and forty-six articles reporting ICC estimates were identified. The median (interquartile range) ICC was 0.031 (0.011 to 0.08) at the school level (N = 210) and 0.063 (0.024 to 0.1) at the class level (N = 46). The distribution of ICCs at the school level was well described by the beta and exponential distributions. Besides larger ICCs in definitive trials than feasibility studies, there were no clear associations between study characteristics and ICC estimates.

CONCLUSION

The distribution of school-level ICCs worldwide was similar to previous summaries from studies in the United States. The description of the distribution of ICCs will help to inform sample size calculations and assess their sensitivity when designing future school-based CRTs of health interventions.

Accounting for expected attrition in the planning of cluster randomized trials for assessing treatment effect heterogeneity

BACKGROUND

Detecting treatment effect heterogeneity is an important objective in cluster randomized trials and implementation research. While sample size procedures for testing the average treatment effect accounting for participant attrition assuming missing completely at random or missing at random have been previously developed, the impact of attrition on the power for detecting heterogeneous treatment effects in cluster randomized trials remains unknown.

METHODS

We provide a sample size formula for testing for a heterogeneous treatment effect assuming the outcome is missing completely at random. We also propose an efficient Monte Carlo sample size procedure for assessing heterogeneous treatment effect assuming covariate-dependent outcome missingness (missing at random). We compare our sample size methods with the direct inflation method that divides the estimated sample size by the mean follow-up rate. We also evaluate our methods through simulation studies and illustrate them with a real-world example.

RESULTS

Simulation results show that our proposed sample size methods under both missing completely at random and missing at random provide sufficient power for assessing heterogeneous treatment effect. The proposed sample size methods lead to more accurate sample size estimates than the direct inflation method when the missingness rate is high (e.g., ≥ 30%). Moreover, sample size estimation under both missing completely at random and missing at random is sensitive to the missingness rate, but not sensitive to the intracluster correlation coefficient among the missingness indicators.

CONCLUSION

Our new sample size methods can assist in planning cluster randomized trials that plan to assess a heterogeneous treatment effect and participant attrition is expected to occur.

Sample size calculators for planning stepped-wedge cluster randomized trials: a review and comparison

Recent years have seen a surge of interest in stepped-wedge cluster randomized trials (SW-CRTs). SW-CRTs include several design variations and methodology is rapidly developing. Accordingly, a variety of power and sample size calculation software for SW-CRTs has been developed. However, each calculator may support only a selected set of design features and may not be appropriate for all scenarios. Currently, there is no resource to assist researchers in selecting the most appropriate calculator for planning their trials. In this paper, we review and classify 18 existing calculators that can be implemented in major platforms, such as R, SAS, Stata, Microsoft Excel, PASS and nQuery. After reviewing the main sample size considerations for SW-CRTs, we summarize the features supported by the available calculators, including the types of designs, outcomes, correlation structures and treatment effects; whether incomplete designs, cluster-size variation or secular trends are accommodated; and the analytical approach used. We then discuss in more detail four main calculators and identify their strengths and limitations. We illustrate how to use these four calculators to compute power for two real SW-CRTs with a continuous and binary outcome and compare the results. We show that the choice of calculator can make a substantial difference in the calculated power and explain these differences. Finally, we make recommendations for implementing sample size or power calculations using the available calculators. An R Shiny app is available for users to select the calculator that meets their requirements (https://douyang.shinyapps.io/swcrtcalculator/).

Evaluating the performance of Bayesian and restricted maximum likelihood estimation for stepped wedge cluster randomized trials with a small number of clusters

Background

Stepped wedge trials are an appealing and potentially powerful cluster randomized trial design. However, they are frequently implemented with a small number of clusters. Standard analysis methods for these trials such as a linear mixed model with estimation via maximum likelihood or restricted maximum likelihood (REML) rely on asymptotic properties and have been shown to yield inflated type I error when applied to studies with a small number of clusters. Small-sample methods such as the Kenward-Roger approximation in combination with REML can potentially improve estimation of the fixed effects such as the treatment effect. A Bayesian approach may also be promising for such multilevel models but has not yet seen much application in cluster randomized trials.

Methods

We conducted a simulation study comparing the performance of REML with and without a Kenward-Roger approximation to a Bayesian approach using weakly informative prior distributions on the intracluster correlation parameters. We considered a continuous outcome and a range of stepped wedge trial configurations with between 4 and 40 clusters. To assess method performance we calculated bias and mean squared error for the treatment effect and correlation parameters and the coverage of 95% confidence/credible intervals and relative percent error in model-based standard error for the treatment effect.

Results

Both REML with a Kenward-Roger standard error and degrees of freedom correction and the Bayesian method performed similarly well for the estimation of the treatment effect, while intracluster correlation parameter estimates obtained via the Bayesian method were less variable than REML estimates with different relative levels of bias.

Conclusions

The use of REML with a Kenward-Roger approximation may be sufficient for the analysis of stepped wedge cluster randomized trials with a small number of clusters. However, a Bayesian approach with weakly informative prior distributions on the intracluster correlation parameters offers a viable alternative, particularly when there is interest in the probability-based inferences permitted within this paradigm.

Supplementary Information

The online version contains supplementary material available at (10.1186/s12874-022-01550-8).

Random effect misspecification in stepped wedge designs

Stepped wedge cluster randomized trials are often analysed using linear mixed effects models that may include random effects for cluster, time and/or treatment. We investigate the impact of misspecification of the random effects structure of the model. Specifically, we considered two cases of misspecification of the random effects in a cross-sectional stepped wedge cluster randomized trials model - fit a linear mixed effects model with random time effects but the true model includes random treatment effects (case 1) or fit a linear mixed effects model with random treatment effect but the true model includes random time effects (case 2) - and derived the variance of the estimated treatment effect under misspecification. We defined two measures of the effect of misspecification: validity and efficiency. Validity is the ratio of the model-based variance of the treatment effect from the mis-specified model divided by the true variance of the treatment effect from the mis-specified model (based on a sandwich estimate of the variance). Efficiency is the ratio of the model-based variance of the treatment effect from the correctly specified model divided by the true variance of the treatment effect from the mis-specified model. We found that validity is less than 1.0 (anti-conservative) in almost all situations investigated with the exception of case 1 with two sequences, when validity could be greater than 1.0. Efficiency is less than 1 in all cases and depends on the intracluster correlation coefficient, the relative magnitude of the variance of the misclassified variance component, and the number of sequences. In general, there is no universal recommendation as to the most robust approach except for the case of a classic stepped wedge cluster randomized trial with only 2 sequences, where fitting a random time model is less likely to lead to anti-conservative inference compared with fitting a random intervention model.

Model misspecification in stepped wedge trials: Random effects for time or treatment

Mixed models are commonly used to analyze stepped wedge trials (SWTs) to account for clustering and repeated measures on clusters. One critical issue researchers face is whether to include a random time effect or a random treatment effect. When the wrong model is chosen, inference on the treatment effect may be invalid. We explore asymptotic and finite-sample convergence of variance component estimates when the model is misspecified and how misspecification affects the estimated variance of the treatment effect. For asymptotic results, we rely on analytical solutions rather than simulation studies, which allow us to succinctly describe the convergence of misspecified estimates, even though there are multiple roots for each misspecified model. We found that both direction and magnitude of the bias associated with model-based standard errors depends on the study design and magnitude of the true variance components. We identify some scenarios in which choosing the wrong random effect has a large impact on model-based inference. However, many trends depend on trial design and assumptions about the true correlation structure, so we provide tools for researchers to investigate specific scenarios of interest. We use data from an SWT on disinvesting from weekend services in hospital wards to demonstrate how these results can be applied as a sensitivity analysis, which quantifies the impact of misspecification under a variety of settings and directly compares the potential consequences of different modeling choices. Our results will provide guidance for prespecified model choices and supplement sensitivity analyses to inform confidence in the validity of results.

Power considerations for generalized estimating equations analyses of four-level cluster randomized trials

In this article, we develop methods for sample size and power calculations in four-level intervention studies when intervention assignment is carried out at any level, with a particular focus on cluster randomized trials (CRTs). CRTs involving four levels are becoming popular in healthcare research, where the effects are measured, for example, from evaluations (level 1) within participants (level 2) in divisions (level 3) that are nested in clusters (level 4). In such multilevel CRTs, we consider three types of intraclass correlations between different evaluations to account for such clustering: that of the same participant, that of different participants from the same division, and that of different participants from different divisions in the same cluster. Assuming arbitrary link and variance functions, with the proposed correlation structure as the true correlation structure, closed-form sample size formulas for randomization carried out at any level (including individually randomized trials within a four-level clustered structure) are derived based on the generalized estimating equations approach using the model-based variance and using the sandwich variance with an independence working correlation matrix. We demonstrate that empirical power corresponds well with that predicted by the proposed method for as few as eight clusters, when data are analyzed using the matrix-adjusted estimating equations for the correlation parameters with a bias-corrected sandwich variance estimator, under both balanced and unbalanced designs.

Methodological challenges in pragmatic trials in Alzheimer's disease and related dementias: Opportunities for improvement

BACKGROUND AND AIMS

We need more pragmatic trials of interventions to improve care and outcomes for people living with Alzheimer's disease and related dementias. However, these trials present unique methodological challenges in their design, analysis, and reporting-often, due to the presence of one or more sources of clustering. Failure to account for clustering in the design and analysis can lead to increased risks of Type I and Type II errors. We conducted a review to describe key methodological characteristics and obtain a "baseline assessment" of methodological quality of pragmatic trials in dementia research, with a view to developing new methods and practical guidance to support investigators and methodologists conducting pragmatic trials in this field.

METHODS

We used a published search filter in MEDLINE to identify trials more likely to be pragmatic and identified a subset that focused on people living with Alzheimer's disease or other dementias or included them as a defined subgroup. Pairs of reviewers extracted descriptive information and key methodological quality indicators from each trial.

RESULTS

We identified N = 62 eligible primary trial reports published across 36 different journals. There were 15 (24%) individually randomized, 38 (61%) cluster randomized, and 9 (15%) individually randomized group treatment designs; 54 (87%) trials used repeated measures on the same individual and/or cluster over time and 17 (27%) had a multivariate primary outcome (e.g. due to measuring an outcome on both the patient and their caregiver). Of the 38 cluster randomized trials, 16 (42%) did not report sample size calculations accounting for the intracluster correlation and 13 (34%) did not account for intracluster correlation in the analysis. Of the 9 individually randomized group treatment trials, 6 (67%) did not report sample size calculations accounting for intracluster correlation and 8 (89%) did not account for it in the analysis. Of the 54 trials with repeated measurements, 45 (83%) did not report sample size calculations accounting for repeated measurements and 19 (35%) did not utilize at least some of the repeated measures in the analysis. No trials accounted for the multivariate nature of their primary outcomes in sample size calculation; only one did so in the analysis.

CONCLUSION

There is a need and opportunity to improve the design, analysis, and reporting of pragmatic trials in dementia research. Investigators should pay attention to the potential presence of one or more sources of clustering. While methods for longitudinal and cluster randomized trials are well developed, accessible resources and new methods for dealing with multiple sources of clustering are required. Involvement of a statistician with expertise in longitudinal and clustered designs is recommended.

Impact of unequal cluster sizes for GEE analyses of stepped wedge cluster randomized trials with binary outcomes

The stepped wedge (SW) design is a type of unidirectional crossover design where cluster units switch from control to intervention condition at different prespecified time points. While a convention in study planning is to assume the cluster‐period sizes are identical, SW cluster randomized trials (SW‐CRTs) involving repeated cross‐sectional designs frequently have unequal cluster‐period sizes, which can impact the efficiency of the treatment effect estimator. In this paper, we provide a comprehensive investigation of the efficiency impact of unequal cluster sizes for generalized estimating equation analyses of SW‐CRTs, with a focus on binary outcomes as in the Washington State Expedited Partner Therapy trial. Several major distinctions between our work and existing work include the following: (i) we consider multilevel correlation structures in marginal models with binary outcomes; (ii) we study the implications of both the between‐cluster and within‐cluster imbalances in sizes; and (iii) we provide a comparison between the independence working correlation versus the true working correlation and detail the consequences of ignoring correlation estimation in SW‐CRTs with unequal cluster sizes. We conclude that the working independence assumption can lead to substantial efficiency loss and a large sample size regardless of cluster‐period size variability in SW‐CRTs, and recommend accounting for correlations in the analysis. To improve study planning, we additionally provide a computationally efficient search algorithm to estimate the sample size in SW‐CRTs accounting for unequal cluster‐period sizes, and conclude by illustrating the proposed approach in the context of the Washington State study.