geeCRT: a package for implementing the bias-corrected generalized estimating equations in analyzing cluster randomized trials
Hengshi Yu, Fan Li, Paul Rathouz, Elizabeth L. Turner, John Preisser
[paper] | [arXiv] | [R package] | [example code]
Maintainer: Hengshi Yu ([email protected])
geeCRT is an R package for implementing the bias-corrected generalized estimating equations in analyzing cluster randomized trials.
Population-averaged models have been increasingly used in the design and analysis of cluster randomized trials (CRTs). To facilitate the applications of population-averaged models in CRTs, we implement the generalized estimating equations (GEE) and matrix-adjusted estimating equations (MAEE) approaches to jointly estimate the marginal mean models correlation models both for general CRTs and stepped wedge CRTs.
Despite the general GEE/MAEE approach, we also implement a fast cluster-period GEE method specifically for stepped wedge CRTs with large and variable cluster-period sizes. The individual-level GEE/MAEE approach becomes computationally infeasible in this setting due to inversion of high-dimensional covariance matrices and the enumeration of a high-dimensional design matrix for the correlation estimation equations. The package gives a simple and efficient estimating equations approach based on the cluster-period means to estimate the intervention effects as well as correlation parameters.
In addition, the package also provides functions for generating correlated binary data with specific mean vector and correlation matrix based on the multivariate probit method (Emrich and Piedmonte, 1991) or the conditional linear family method (Qaqish, 2003). These two functions facilitate generating correlated binary data in future simulation studies.
The geeCRT package constains four main functions. In the analysis of individual-level CRT data, users can use the geemaee() function to perform joint estimation of the marginal mean model and intraclass correlation parameters. In the analysis of cross-sectional stepped wedge CRTs, users can use the cpgeeSWD() function to perform computationally efficient joint estimation of the marginal mean and intraclass correlation parameters simply based on the cluster-period means. For generating binary data with specified mean and correlation structures, user can use the simbinPROBIT() function with the multivariate probit method or the simbinCLF() function with the conditional linear family method.
-
geemaee function: GEE and MAEE for estimating the marginal mean and correlation parameters in CRTs
- Liang, K. Y., Zeger, S. L. (1986). Longitudinal data analysis using generalized linear models. Biometrika, 73(1), 13-22.
- Prentice, R. L. (1988). Correlated binary regression with covariates specific to each binary observation. Biometrics, 1033-1048.
- Zhao, L. P., Prentice, R. L. (1990). Correlated binary regression using a quadratic exponential model. Biometrika, 77(3), 642-648.
- Prentice, R. L., & Zhao, L. P. (1991). Estimating equations for parameters in means and covariances of multivariate discrete and continuous responses. Biometrics, 825-839.
- Sharples, K., & Breslow, N. (1992). Regression analysis of correlated binary data: some small sample results for the estimating equation approach. Journal of Statistical Computation and Simulation, 42(1-2), 1-20.
- Mancl, L. A., DeRouen, T. A. (2001). A covariance estimator for GEE with improved small sample properties. Biometrics, 57(1), 126-134.
- Kauermann, G., Carroll, R. J. (2001). A note on the efficiency of sandwich covariance matrix estimation. Journal of the American Statistical Association, 96(456), 1387-1396.
- Fay, M. P., Graubard, B. I. (2001). Small sample adjustments for Wald type tests using sandwich estimators. Biometrics, 57(4), 1198-1206.
- Lu, B., Preisser, J. S., Qaqish, B. F., Suchindran, C., Bangdiwala, S. I., Wolfson, M. (2007). A comparison of two bias corrected covariance estimators for generalized estimating equations. Biometrics, 63(3), 935-941.
- Preisser, J. S., Lu, B., Qaqish, B. F. (2008). Finite sample adjustments in estimating equations and covariance estimators for intracluster correlations. Statistics in Medicine, 27(27), 5764-5785.
- Li, F., Turner, E. L., & Preisser, J. S. (2018). Sample size determination for GEE analyses of stepped wedge cluster randomized trials. Biometrics, 74(4), 1450-1458.
- Li, F., Forbes, A. B., Turner, E. L., Preisser, J. S. (2019). Power and sample size requirements for GEE analyses of cluster randomized crossover trials. Statistics in Medicine, 38(4), 636-649.
- Li, F. (2020). Design and analysis considerations for cohort stepped wedge cluster randomized trials with a decay correlation structure. Statistics in medicine, 39(4), 438-455.
- Li, F., Yu, H., Rathouz, P., Turner, E. L., Preisser, J. S. (2020+). Marginal modeling of cluster period means and intraclass correlations in stepped wedge designs with binary outcomes. Under Revision at Biostatistics.
-
cpgeeSWD function: cluster-period generalized estimating equations for estimating the marginal mean and correlation parameters in cross-sectional stepped wedge CRTs
- Zhao, L. P., Prentice, R. L. (1990). Correlated binary regression using a quadratic exponential model. Biometrika, 77(3), 642-648.
- Mancl, L. A., DeRouen, T. A. (2001). A covariance estimator for GEE with improved small sample properties. Biometrics, 57(1), 126-134.
- Kauermann, G., Carroll, R. J. (2001). A note on the efficiency of sandwich covariance matrix estimation. Journal of the American Statistical Association, 96(456), 1387-1396.
- Fay, M. P., Graubard, B. I. (2001). Small sample adjustments for Wald type tests using sandwich estimators. Biometrics, 57(4), 1198-1206.
- Lu, B., Preisser, J. S., Qaqish, B. F., Suchindran, C., Bangdiwala, S. I., Wolfson, M. (2007). A comparison of two bias corrected covariance estimators for generalized estimating equations. Biometrics, 63(3), 935-941.
- Preisser, J. S., Lu, B., Qaqish, B. F. (2008). Finite sample adjustments in estimating equations and covariance estimators for intracluster correlations. Statistics in Medicine, 27(27), 5764-5785.
- Li, F., Turner, E. L., & Preisser, J. S. (2018). Sample size determination for GEE analyses of stepped wedge cluster randomized trials. Biometrics, 74(4), 1450-1458.
- Li, F. (2020). Design and analysis considerations for cohort stepped wedge cluster randomized trials with a decay correlation structure. Statistics in medicine, 39(4), 438-455.
- Li, F., Yu, H., Rathouz, P. J., Turner, E. L., & Preisser, J. S. (2022). Marginal modeling of cluster-period means and intraclass correlations in stepped wedge designs with binary outcomes. Biostatistics, 23(3), 772-788.
-
simbinPROBIT function: generating correlated binary data using the multivariate probit method
- Emrich, L. J., & Piedmonte, M. R. (1991). A method for generating high-dimensional multivariate binary variates. The American Statistician, 45(4), 302-304.
- Preisser, J. S., Qaqish, B. F. (2014). A comparison of methods for simulating correlated binary variables with specified marginal means and correlations. Journal of Statistical Computation and Simulation, 84(11), 2441-2452.
-
simbinCLF function: generating correlated binary data using the conditional linear family method
- Qaqish, B. F. (2003). A family of multivariate binary distributions for simulating correlated binary variables with specified marginal means and correlations. Biometrika, 90(2), 455-463.
- Preisser, J. S., Qaqish, B. F. (2014). A comparison of methods for simulating correlated binary variables with specified marginal means and correlations. Journal of Statistical Computation and Simulation, 84(11), 2441-2452.
The geeCRT R package is available on CRAN.
install.packages('geeCRT')
The geemaee() function implements the matrix-adjusted GEE or regular GEE developed for analyzing cluster randomized trials (CRTs). It provides valid estimation and inference for the treatment effect and intraclass correlation parameters within the population-averaged modeling framework. The program allows for flexible marginal mean model specifications. The program also offers bias-corrected intraclass correlation coefficient (ICC) estimates as well as bias-corrected sandwich variances for both the treatment effect parameter and the ICC parameters. The technical details of the matrix-adjusted GEE approach are provided in Preisser et al. (2008) and Li et al. (2018).
For the individual-level data, we use the geemaee() function to estimate the marginal mean and correlation parameters in CRTs. We use two simulated stepped wedge CRT datasets with true nested exchangeable correlation structure to illustrate the geemaee() function examples. We first create an auxiliary function createzCrossSec() to help create the design matrix for the estimating equations of the correlation parameters. We then collect design matrix X for the mean parameters with five period indicators and the treatment indicator.
We implement the geemaee() function on both the continuous, binary and count outcomes using the nested exchangeable correlation structure, and consider both matrix-adjusted estimating equations (MAEE) with alpadj = TRUE and uncorrected generalized estimating equations (GEE) with alpadj = FALSE. We use the binary outcome for the count outcome type of the geemaee() function, and consider both poisson and quasipoisson distributions for the count outcome. For the shrink argument, we use the "ALPHA" method to tune step sizes and focus on using estimated variances in the correlation estimating equations rather than using unit variances by specifying makevone = FALSE.
### function to create the design matrix for correlation parameters
### under the nested exchangeable correlation structure of SW-CRTs
createzCrossSec = function (m) {
Z = NULL
n = dim(m)[1]
for (i in 1:n) {
alpha_0 = 1; alpha_1 = 2; n_i = c(m[i, ]); n_length = length(n_i)
POS = matrix(alpha_1, sum(n_i), sum(n_i))
loc1 = 0; loc2 = 0
for (s in 1:n_length) {
n_t = n_i[s]; loc1 = loc2 + 1; loc2 = loc1 + n_t - 1
for (k in loc1:loc2) {
for (j in loc1:loc2) {
if (k != j) { POS[k, j] = alpha_0 } else { POS[k, j] = 0 }}}}
zrow = diag(2); z_c = NULL
for (j in 1:(sum(n_i) - 1)) {
for (k in (j + 1):sum(n_i)) {z_c = rbind(z_c, zrow[POS[j,k],])}}
Z = rbind(Z, z_c) }
return(Z)}
########################################################################
### Example 1): simulated SW-CRT with smaller cluster-period sizes (5~10)
########################################################################
sampleSWCRT = sampleSWCRTSmall
### Individual-level id, period, outcome, and design matrix
id = sampleSWCRT$id; period = sampleSWCRT$period;
X = as.matrix(sampleSWCRT[, c('period1', 'period2', 'period3', 'period4', 'treatment')])
m = as.matrix(table(id, period)); n = dim(m)[1]; t = dim(m)[2]
### design matrix for correlation parameters
Z = createzCrossSec(m)
### (1) Matrix-adjusted estimating equations and GEE
### on continous outcome with nested exchangeable correlation structure
### MAEE
est_maee_ind_con = geemaee(y = sampleSWCRT$y_con,
X = X, id = id, Z = Z,
family = "continuous",
maxiter = 500, epsilon = 0.001,
printrange = TRUE, alpadj = TRUE,
shrink = "ALPHA", makevone = FALSE)
print(est_maee_ind_con)
### GEE
est_uee_ind_con = geemaee(y = sampleSWCRT$y_con,
X = X, id = id, Z = Z,
family = "continuous",
maxiter = 500, epsilon = 0.001,
printrange = TRUE, alpadj = FALSE,
shrink = "ALPHA", makevone = FALSE)
print(est_uee_ind_con)
### (2) Matrix-adjusted estimating equations and GEE
### on binary outcome with nested exchangeable correlation structure
### MAEE
est_maee_ind_bin = geemaee(y = sampleSWCRT$y_bin,
X = X, id = id, Z = Z,
family = "binomial",
maxiter = 500, epsilon = 0.001,
printrange = TRUE, alpadj = TRUE,
shrink = "ALPHA", makevone = FALSE)
print(est_maee_ind_bin)
### GEE
est_uee_ind_bin = geemaee(y = sampleSWCRT$y_bin,
X = X, id = id, Z = Z,
family = "binomial",
maxiter = 500, epsilon = 0.001,
printrange = TRUE, alpadj = FALSE,
shrink = "ALPHA", makevone = FALSE)
print(est_uee_ind_bin)
### (3) Matrix-adjusted estimating equations and GEE
### on count outcome with nested exchangeable correlation structure
### using Poisson distribution
### MAEE
est_maee_ind_cnt_poisson = geemaee(y = sampleSWCRT$y_bin,
X = X, id = id, Z = Z,
family = "poisson",
maxiter = 500, epsilon = 0.001,
printrange = TRUE, alpadj = TRUE,
shrink = "ALPHA", makevone = FALSE)
print(est_maee_ind_cnt_poisson)
### GEE
est_uee_ind_cnt_poisson = geemaee(y = sampleSWCRT$y_bin,
X = X, id = id, Z = Z,
family = "poisson",
maxiter = 500, epsilon = 0.001,
printrange = TRUE, alpadj = FALSE,
shrink = "ALPHA", makevone = FALSE)
print(est_uee_ind_cnt_poisson)
### (4) Matrix-adjusted estimating equations and GEE
### on count outcome with nested exchangeable correlation structure
### using Quasi-Poisson distribution
### MAEE
est_maee_ind_cnt_quasipoisson = geemaee(y = sampleSWCRT$y_bin,
X = X, id = id, Z = Z,
family = "quasipoisson",
maxiter = 500, epsilon = 0.001,
printrange = TRUE, alpadj = TRUE,
shrink = "ALPHA", makevone = FALSE)
print(est_maee_ind_cnt_quasipoisson)
### GEE
est_uee_ind_cnt_quasipoisson = geemaee(y = sampleSWCRT$y_bin,
X = X, id = id, Z = Z,
family = "quasipoisson",
maxiter = 500, epsilon = 0.001,
printrange = TRUE, alpadj = FALSE,
shrink = "ALPHA", makevone = FALSE)
print(est_uee_ind_cnt_quasipoisson)
########################################################################
### Example 2): simulated SW-CRT with larger cluster-period sizes (20~30)
########################################################################
## This will elapse longer.
sampleSWCRT = sampleSWCRTLarge
### Individual-level id, period, outcome, and design matrix
id = sampleSWCRT$id; period = sampleSWCRT$period;
X = as.matrix(sampleSWCRT[, c('period1', 'period2', 'period3', 'period4', 'period5', 'treatment')])
m = as.matrix(table(id, period)); n = dim(m)[1]; t = dim(m)[2]
### design matrix for correlation parameters
Z = createzCrossSec(m)
### (1) Matrix-adjusted estimating equations and GEE
### on continous outcome with nested exchangeable correlation structure
### MAEE
est_maee_ind_con = geemaee(y = sampleSWCRT$y_con,
X = X, id = id, Z = Z,
family = "continuous",
maxiter = 500, epsilon = 0.001,
printrange = TRUE, alpadj = TRUE,
shrink = "ALPHA", makevone = FALSE)
print(est_maee_ind_con)
### GEE
est_uee_ind_con = geemaee(y = sampleSWCRT$y_con,
X = X, id = id, Z = Z,
family = "continuous",
maxiter = 500, epsilon = 0.001,
printrange = TRUE, alpadj = FALSE,
shrink = "ALPHA", makevone = FALSE)
print(est_uee_ind_con)
### (2) Matrix-adjusted estimating equations and GEE
### on binary outcome with nested exchangeable correlation structure
### MAEE
est_maee_ind_bin = geemaee(y = sampleSWCRT$y_bin,
X = X, id = id, Z = Z,
family = "binomial",
maxiter = 500, epsilon = 0.001,
printrange = TRUE, alpadj = TRUE,
shrink = "ALPHA", makevone = FALSE)
print(est_maee_ind_bin)
### GEE
est_uee_ind_bin = geemaee(y = sampleSWCRT$y_bin,
X = X, id = id, Z = Z,
family = "binomial",
maxiter = 500, epsilon = 0.001,
printrange = TRUE, alpadj = FALSE,
shrink = "ALPHA", makevone = FALSE)
print(est_uee_ind_bin)
We can also use the simple exchangeable correlation structure to fit geemaee().
# simulated SW-CRT with smaller cluster-period sizes (5~10)
sampleSWCRT = sampleSWCRTSmall
### Individual-level id, period, outcome, and design matrix
id = sampleSWCRT$id; period = sampleSWCRT$period;
X = as.matrix(sampleSWCRT[, c('period1', 'period2', 'period3', 'period4', 'treatment')])
m = as.matrix(table(id, period)); n = dim(m)[1]; t = dim(m)[2]
### design matrix for correlation parameters (simple exchangeable correlation structure)
Z = createzCrossSec(m)
Z = matrix(1, dim(Z)[1], 1)
# Z for simple exchangeable correlation structure
### (1) Matrix-adjusted estimating equations and GEE
### on continuous outcome with nested exchangeable correlation structure
### MAEE
est_maee_ind_con = geemaee(y = sampleSWCRT$y_con,
X = X, id = id, Z = Z,
family = "continuous",
maxiter = 500, epsilon = 0.001,
printrange = TRUE, alpadj = TRUE,
shrink = "ALPHA", makevone = FALSE)
print(est_maee_ind_con)
### GEE
est_uee_ind_con = geemaee(y = sampleSWCRT$y_con,
X = X, id = id, Z = Z,
family = "continuous",
maxiter = 500, epsilon = 0.001,
printrange = TRUE, alpadj = FALSE,
shrink = "ALPHA", makevone = FALSE)
print(est_uee_ind_con)
### (2) Matrix-adjusted estimating equations and GEE
### on binary outcome with nested exchangeable correlation structure
### MAEE
est_maee_ind_bin = geemaee(y = sampleSWCRT$y_bin,
X = X, id = id, Z = Z,
family = "binomial",
maxiter = 500, epsilon = 0.001,
printrange = TRUE, alpadj = TRUE,
shrink = "ALPHA", makevone = FALSE)
print(est_maee_ind_bin)
### GEE
est_uee_ind_bin = geemaee(y = sampleSWCRT$y_bin,
X = X, id = id, Z = Z,
family = "binomial",
maxiter = 500, epsilon = 0.001,
printrange = TRUE, alpadj = FALSE,
shrink = "ALPHA", makevone = FALSE)
print(est_uee_ind_bin)geemaee() examples: matrix-adjusted GEE for estimating the mean and correlation parameters in SW-CRTs when some clusters had only 1 observation
We can use the geemaee() function for SW-CRTs when some clusters had only one observation.
# simulated SW-CRT with smaller cluster-period sizes (5~10)
# let the cluster 5 and cluster 10 be with only 1 observation
sampleSWCRT = sampleSWCRTSmall[sampleSWCRTSmall$id != 5 & sampleSWCRTSmall$id != 10, ]
sampleSWCRT5 = sampleSWCRTSmall[sampleSWCRTSmall$id == 5 & sampleSWCRTSmall$period == 2, ][1, ]
sampleSWCRT10 = sampleSWCRTSmall[sampleSWCRTSmall$id == 10 & sampleSWCRTSmall$period == 3, ][1, ]
sampleSWCRT = rbind(sampleSWCRT, sampleSWCRT5)
sampleSWCRT = rbind(sampleSWCRT, sampleSWCRT10)
sampleSWCRT = sampleSWCRT[order(sampleSWCRT$id, sampleSWCRT$period), ]
id = sampleSWCRT$id; period = sampleSWCRT$period;
X = as.matrix(sampleSWCRT[, c('period1', 'period2', 'period3', 'period4', 'treatment')])
m = as.matrix(table(id, period)); n = dim(m)[1]; t = dim(m)[2]
## function to generate Z matrix (ignore the clusters with only 1 observation)
createzCrossSecIgnoreOneObsCount = function (m) {
Z = NULL
n = dim(m)[1]
for (i in 1:n) {
alpha_0 = 1; alpha_1 = 2;
n_i = c(m[i, ]);
n_i = c(n_i[n_i>0])
n_length = length(n_i)
POS = matrix(alpha_1, sum(n_i), sum(n_i))
loc1 = 0; loc2 = 0
for (s in 1:n_length) {
n_t = n_i[s]; loc1 = loc2 + 1; loc2 = loc1 + n_t - 1
for (k in loc1:loc2) {
for (j in loc1:loc2) {
if (k != j) { POS[k, j] = alpha_0 } else { POS[k, j] = 0 }}}}
zrow = diag(2); z_c = NULL
if (sum(n_i) > 1) {
for (j in 1:(sum(n_i) - 1)) {
for (k in (j + 1):sum(n_i)) {
z_c = rbind(z_c, zrow[POS[j,k],])
}}
}
Z = rbind(Z, z_c) }
return(Z)}
Z = createzCrossSecIgnoreOneObsCount(m)
### (1) Matrix-adjusted estimating equations and GEE
### on continuous outcome with nested exchangeable correlation structure
### MAEE
est_maee_ind_con = geemaee(y = sampleSWCRT$y_con,
X = X, id = id, Z = Z,
family = "continuous",
maxiter = 500, epsilon = 0.001,
printrange = TRUE, alpadj = TRUE,
shrink = "ALPHA", makevone = FALSE)
print(est_maee_ind_con)
### GEE
est_uee_ind_con = geemaee(y = sampleSWCRT$y_con,
X = X, id = id, Z = Z,
family = "continuous",
maxiter = 500, epsilon = 0.001,
printrange = TRUE, alpadj = FALSE,
shrink = "ALPHA", makevone = FALSE)
print(est_uee_ind_con)
### (2) Matrix-adjusted estimating equations and GEE
### on binary outcome with nested exchangeable correlation structure
### MAEE
est_maee_ind_bin = geemaee(y = sampleSWCRT$y_bin,
X = X, id = id, Z = Z,
family = "binomial",
maxiter = 500, epsilon = 0.001,
printrange = TRUE, alpadj = TRUE,
shrink = "ALPHA", makevone = FALSE)
print(est_maee_ind_bin)
### GEE
est_uee_ind_bin = geemaee(y = sampleSWCRT$y_bin,
X = X, id = id, Z = Z,
family = "binomial",
maxiter = 500, epsilon = 0.001,
printrange = TRUE, alpadj = FALSE,
shrink = "ALPHA", makevone = FALSE)
print(est_uee_ind_bin)
geemaee() examples: matrix-adjusted GEE for estimating the mean and correlation parameters in general CRTs
We can also use the geemaee() function for general CRTs with only one time period. We need to specify a single column of 1's for the Z matrix.
# simulated SW-CRT with smaller cluster-period sizes (5~10)
# use only the second period data to get a general CRT data
sampleSWCRT = sampleSWCRTSmall[sampleSWCRTSmall$period == 2, ]
### Individual-level id, period, outcome, and design matrix
id = sampleSWCRT$id; period = sampleSWCRT$period;
X = as.matrix(sampleSWCRT[, c('period2', 'treatment')])
m = as.matrix(table(id, period)); n = dim(m)[1]; t = dim(m)[2]
### design matrix for correlation parameters
Z = createzCrossSec(m)
Z = matrix(1, dim(Z)[1], 1)
### (1) Matrix-adjusted estimating equations and GEE
### on continuous outcome with nested exchangeable correlation structure
### MAEE
est_maee_ind_con = geemaee(y = sampleSWCRT$y_con,
X = X, id = id, Z = Z,
family = "continuous",
maxiter = 500, epsilon = 0.001,
printrange = TRUE, alpadj = TRUE,
shrink = "ALPHA", makevone = FALSE)
print(est_maee_ind_con)
### GEE
est_uee_ind_con = geemaee(y = sampleSWCRT$y_con,
X = X, id = id, Z = Z,
family = "continuous",
maxiter = 500, epsilon = 0.001,
printrange = TRUE, alpadj = FALSE,
shrink = "ALPHA", makevone = FALSE)
print(est_uee_ind_con)
### (2) Matrix-adjusted estimating equations and GEE
### on binary outcome with nested exchangeable correlation structure
### MAEE
est_maee_ind_bin = geemaee(y = sampleSWCRT$y_bin,
X = X, id = id, Z = Z,
family = "binomial",
maxiter = 500, epsilon = 0.001,
printrange = TRUE, alpadj = TRUE,
shrink = "ALPHA", makevone = FALSE)
print(est_maee_ind_bin)
### GEE
est_uee_ind_bin = geemaee(y = sampleSWCRT$y_bin,
X = X, id = id, Z = Z,
family = "binomial",
maxiter = 500, epsilon = 0.001,
printrange = TRUE, alpadj = FALSE,
shrink = "ALPHA", makevone = FALSE)
print(est_uee_ind_bin)geemaee() examples: matrix-adjusted GEE for estimating the mean and correlation parameters in SW-CRTs with all cluster-period sizes being 1
We consider an edge case when all cluster period sizes are 1, the geemaee() function can still estimate the intra-period correlation parameter.
### Simulated dataset with 12 clusters and large cluster sizes
# use the first observation for each cluster-period combination
sampleSWCRT = NULL
for (i in 1:12) {
for (j in 1:5) {
row = sampleSWCRTLarge[sampleSWCRTLarge$id == i &
sampleSWCRTLarge$period == j, ][1, , drop = FALSE]
sampleSWCRT = rbind(sampleSWCRT, row)
}
}
### Individual-level id, period, outcome, and design matrix
id = sampleSWCRT$id; period = sampleSWCRT$period;
X = as.matrix(sampleSWCRT[, c('period2', 'treatment')])
m = as.matrix(table(id, period)); n = dim(m)[1]; t = dim(m)[2]
### design matrix for correlation parameters
Z = createzCrossSec(m)
Z = matrix(1, dim(Z)[1], 1)
### (1) Matrix-adjusted estimating equations and GEE
### on continous outcome with nested exchangeable correlation structure
### MAEE
est_maee_ind_con = geemaee(y = sampleSWCRT$y_con,
X = X, id = id, Z = Z,
family = "continuous",
maxiter = 500, epsilon = 0.001,
printrange = TRUE, alpadj = TRUE,
shrink = "ALPHA", makevone = FALSE)
print(est_maee_ind_con)
### GEE
est_uee_ind_con = geemaee(y = sampleSWCRT$y_con,
X = X, id = id, Z = Z,
family = "continuous",
maxiter = 500, epsilon = 0.001,
printrange = TRUE, alpadj = FALSE,
shrink = "ALPHA", makevone = FALSE)
print(est_uee_ind_con)
### (2) Matrix-adjusted estimating equations and GEE
### on binary outcome with nested exchangeable correlation structure
### MAEE
est_maee_ind_bin = geemaee(y = sampleSWCRT$y_bin,
X = X, id = id, Z = Z,
family = "binomial",
maxiter = 500, epsilon = 0.001,
printrange = TRUE, alpadj = TRUE,
shrink = "ALPHA", makevone = FALSE)
print(est_maee_ind_bin)
### GEE
est_uee_ind_bin = geemaee(y = sampleSWCRT$y_bin,
X = X, id = id, Z = Z,
family = "binomial",
maxiter = 500, epsilon = 0.001,
printrange = TRUE, alpadj = FALSE,
shrink = "ALPHA", makevone = FALSE)
print(est_uee_ind_bin)cpgeeSWD() examples: cluster-period GEE for estimating the marginal mean and correlation parameters in cross-sectional SW-CRTs
The cpgeeSWD() function implements the cluster-period GEE developed for cross-sectional stepped wedge cluster randomized trials (SW-CRTs). It provides valid estimation and inference for the treatment effect and intraclass correlation parameters within the GEE framework, and is computationally efficient for analyzing SW-CRTs with large cluster sizes. The program currently only allows for a marginal mean model with discrete period effects and the intervention indicator without additional covariates. The program offers bias-corrected ICC estimates as well as bias-corrected sandwich variances for both the treatment effect parameter and the ICC parameters. The technical details of the cluster-period GEE approach are provided in Li et al. (2021).
We summarize the individual-level simulated SW-CRT data to cluster-period data and use the cpgeeSWD() function to estimate the marginal mean and correlation parameters on cluster-period means of binary outcome. We first transform the variables to get the cluster-period mean outcome y_cp, mean parameters' design matrix X_cp as well as other arguments.
We implement the cpgeeSWD() function on all the three choices of the correlation structure including "exchangeable", "nest_exch" and "exp_decay". We consider both matrix-adjusted estimating equations (MAEE) with alpadj = TRUE and uncorrected generalized estimating equations (GEE) with alpadj = FALSE.
# Simulated SW-CRT example with binary outcome
########################################################################
### Example 1): simulated SW-CRT with smaller cluster-period sizes (5~10)
########################################################################
sampleSWCRT = sampleSWCRTSmall
### cluster-period id, period, outcome, and design matrix
### id, period, outcome
id = sampleSWCRT$id; period = sampleSWCRT$period; y = sampleSWCRT$y_bin
X = as.matrix(sampleSWCRT[, c('period1', 'period2', 'period3', 'period4', 'treatment')])
m = as.matrix(table(id, period)); n = dim(m)[1]; t = dim(m)[2]
clp_mu = tapply(y,list(id,period), FUN=mean)
y_cp = c(t(clp_mu))
### design matrix for correlation parameters
trt = tapply(X[, t + 1], list(id, period), FUN=mean)
trt = c(t(trt))
time = tapply(period,list(id, period), FUN = mean); time = c(t(time))
X_cp = matrix(0, n * t, t)
s = 1
for (i in 1:n) { for (j in 1:t) { X_cp[s, time[s]] = 1; s = s + 1 }}
X_cp = cbind(X_cp, trt); id_cp = rep(1:n, each= t); m_cp = c(t(m))
### cluster-period matrix-adjusted estimating equations (MAEE)
### with exchangeable, nested exchangeable and exponential decay correlation structures
# exponential
est_maee_exc = cpgeeSWD(y = y_cp, X = X_cp, id = id_cp,
m = m_cp, corstr = "exchangeable",
alpadj = TRUE)
print(est_maee_exc)
# nested exchangeable
est_maee_nex = cpgeeSWD(y = y_cp, X = X_cp, id = id_cp,
m = m_cp, corstr = "nest_exch",
alpadj = TRUE)
print(est_maee_nex)
# exponential decay
est_maee_ed = cpgeeSWD(y = y_cp, X = X_cp, id = id_cp,
m = m_cp, corstr = "exp_decay",
alpadj = TRUE)
print(est_maee_ed)
### cluster-period GEE
### with exchangeable, nested exchangeable and exponential decay correlation structures
# exchangeable
est_uee_exc <- cpgeeSWD(y = y_cp, X = X_cp, id = id_cp,
m = m_cp, corstr = "exchangeable",
alpadj = FALSE)
print(est_uee_exc)
# nested exchangeable
est_uee_nex <- cpgeeSWD(y = y_cp, X = X_cp, id = id_cp,
m = m_cp, corstr = "nest_exch",
alpadj = FALSE)
print(est_uee_nex)
# exponential decay
est_uee_ed <- cpgeeSWD(y = y_cp, X = X_cp, id = id_cp,
m = m_cp, corstr = 'exp_decay',
alpadj = FALSE)
print(est_uee_ed)
########################################################################
### Example 2): simulated SW-CRT with larger cluster-period sizes (20~30)
########################################################################
sampleSWCRT = sampleSWCRTLarge
### cluster-period id, period, outcome, and design matrix
### id, period, outcome
id = sampleSWCRT$id; period = sampleSWCRT$period; y = sampleSWCRT$y_bin
X = as.matrix(sampleSWCRT[, c('period1', 'period2', 'period3', 'period4', 'period5', 'treatment')])
m = as.matrix(table(id, period)); n = dim(m)[1]; t = dim(m)[2]
clp_mu<-tapply(y,list(id,period), FUN=mean)
y_cp <- c(t(clp_mu))
### design matrix for correlation parameters
trt <- tapply(X[, t + 1], list(id, period), FUN=mean)
trt <- c(t(trt))
time <- tapply(period,list(id, period), FUN = mean); time <- c(t(time))
X_cp <- matrix(0, n * t, t)
s = 1
for(i in 1:n){for(j in 1:t){X_cp[s, time[s]] <- 1; s = s + 1}}
X_cp <- cbind(X_cp, trt); id_cp <- rep(1:n, each= t); m_cp <- c(t(m))
### cluster-period matrix-adjusted estimating equations (MAEE)
### with exchangeable, nested exchangeable and exponential decay correlation structures
# exponential
est_maee_exc <- cpgeeSWD(y = y_cp, X = X_cp, id = id_cp,
m = m_cp, corstr = "exchangeable",
alpadj = TRUE)
print(est_maee_exc)
# nested exchangeable
est_maee_nex <- cpgeeSWD(y = y_cp, X = X_cp, id = id_cp,
m = m_cp, corstr = "nest_exch",
alpadj = TRUE)
print(est_maee_nex)
# exponential decay
est_maee_ed <- cpgeeSWD(y = y_cp, X = X_cp, id = id_cp,
m = m_cp, corstr = "exp_decay",
alpadj = TRUE)
print(est_maee_ed)
### cluster-period GEE
### with exchangeable, nested exchangeable and exponential decay correlation structures
# exchangeable
est_uee_exc <- cpgeeSWD(y = y_cp, X = X_cp, id = id_cp,
m = m_cp, corstr = "exchangeable",
alpadj = FALSE)
print(est_uee_exc)
# nested exchangeable
est_uee_nex <- cpgeeSWD(y = y_cp, X = X_cp, id = id_cp,
m = m_cp, corstr = "nest_exch",
alpadj = FALSE)
print(est_uee_nex)
# exponential decay
est_uee_ed <- cpgeeSWD(y = y_cp, X = X_cp, id = id_cp,
m = m_cp, corstr = 'exp_decay',
alpadj = FALSE)
print(est_uee_ed)
The simbinPROBIT() function generates correlated binary data using the multivariate Probit method (Emrich and Piedmonte, 1991). It simulates a vector of binary outcomes according the specified marginal mean vector and correlation structure. Constraints and compatibility between the marginal mean and correlation matrix are checked.
We use the simbinPROBIT() function to generate correlated binary data with different correlation structures. We consider simulating a cross-sectional SW-CRT dataset with 2 clusters, 3 periods with the same cluster-period size of 5. We use two mean vectors for the two clusters and specify the mu argument.
For the exchangeable correlation structure, we specify both the within-period and inter-period correlation parameters to be 0.015. We use 0.03 and 0.015 for the within-period and inter-period correlations, respectively. The exponential decay correlation structure has an decay parameter 0.8 with the within-period correlation parameter 0.03.
#### Simulate 2 clusters, 3 periods and cluster-period size of 5
t = 3; n = 2; m = 5
# means of cluster 1
u_c1 = c(0.4, 0.3, 0.2)
u1 <- rep(u_c1, c(rep(m, t)))
# means of cluster 2
u_c2 = c(0.35, 0.25, 0.2)
u2 <- rep(u_c2, c(rep(m, t)))
# List of mean vectors
mu = list(); mu[[1]] = u1; mu[[2]] = u2;
# List of correlation matrices
## correlation parameters
alpha0 = 0.03; alpha1 = 0.015; rho = 0.8
## (1) exchangeable
Sigma = list()
Sigma[[1]] = diag(m * t) * ( 1 - alpha1) + matrix(alpha1, m * t, m * t )
Sigma[[2]] = diag(m * t) * ( 1 - alpha1) + matrix(alpha1, m * t, m * t )
y_exc = simbinPROBIT(mu = mu, Sigma = Sigma, n = n)
## (2) nested exchangeable
Sigma = list()
cor_matrix = matrix(alpha1, m * t, m * t)
loc1 = 0; loc2 = 0
for(t in 1:t){loc1 = loc2 + 1; loc2 = loc1 + m - 1
for(i in loc1:loc2){for(j in loc1:loc2){
if(i != j){cor_matrix[i, j] = alpha0}else{cor_matrix[i, j] = 1}}}}
Sigma[[1]] = cor_matrix; Sigma[[2]] = cor_matrix
y_nex = simbinPROBIT(mu = mu, Sigma = Sigma, n = n)
## (3) exponential decay
Sigma = list()
### function to find the period of the ith index
region_ij<-function(points, i){diff = i - points
for(h in 1:(length(diff) - 1)){if(diff[h] > 0 & diff[h + 1] <= 0){find <- h}}
return(find)}
cor_matrix = matrix(0, m * t, m * t)
useage_m = cumsum(m * t); useage_m = c(0, useage_m)
for(i in 1:(m * t)){i_reg = region_ij(useage_m, i)
for(j in 1:(m * t)){j_reg = region_ij(useage_m, j)
if(i_reg == j_reg & i != j){
cor_matrix[i, j] = alpha0}else if(i == j){cor_matrix[i, j] = 1
}else if(i_reg != j_reg){cor_matrix[i,j] = alpha0 * (rho^(abs(i_reg - j_reg)))}}}
Sigma[[1]] = cor_matrix; Sigma[[2]] = cor_matrix
y_ed = simbinPROBIT(mu = mu, Sigma = Sigma, n = n)
The simbinCLF() function generates correlated binary data using the conditional linear family method (Qaqish, 2003). It simulates a vector of binary outcomes according the specified marginal mean vector and correlation structure. Natural constraints and compatibility between the marginal mean and correlation matrix are checked.
We use the simbinCLF() function to generate correlated binary data with different correlation structures. We consider simulating a cross-sectional SW-CRT dataset with 2 clusters, 3 periods with the same cluster-period size of 5. We use two mean vectors for the two clusters and specify the mu argument.
For the exchangeable correlation structure, we specify both the within-period and inter-period correlation parameters to be 0.015. We use 0.03 and 0.015 for the within-period and inter-period correlations, respectively. The exponential decay correlation structure has an decay parameter 0.8 with the within-period correlation parameter 0.03.
##### Simulate 2 clusters, 3 periods and cluster-period size of 5
t = 3; n = 2; m = 5
# means of cluster 1
u_c1 = c(0.4, 0.3, 0.2)
u1 <- rep(u_c1, c(rep(m, t)))
# means of cluster 2
u_c2 = c(0.35, 0.25, 0.2)
u2 <- rep(u_c2, c(rep(m, t)))
# List of mean vectors
mu = list()
mu[[1]] = u1; mu[[2]] = u2;
# List of correlation matrices
## correlation parameters
alpha0 = 0.03; alpha1 = 0.015; rho = 0.8
## (1) exchangeable
Sigma = list()
Sigma[[1]] = diag(m * t) * ( 1 - alpha1) + matrix(alpha1, m * t, m * t )
Sigma[[2]] = diag(m * t) * ( 1 - alpha1) + matrix(alpha1, m * t, m * t )
y_exc = simbinCLF(mu = mu, Sigma = Sigma, n = n)
## (2) nested exchangeable
Sigma = list()
cor_matrix = matrix(alpha1, m * t, m * t)
loc1 = 0; loc2 = 0
for(t in 1:t){loc1 = loc2 + 1; loc2 = loc1 + m - 1
for(i in loc1:loc2){for(j in loc1:loc2){
if(i != j){cor_matrix[i, j] = alpha0}else{cor_matrix[i, j] = 1}}}}
Sigma[[1]] = cor_matrix; Sigma[[2]] = cor_matrix
y_nex = simbinCLF(mu = mu, Sigma = Sigma, n = n)
## (3) exponential decay
Sigma = list()
### function to find the period of the ith index
region_ij<-function(points, i){diff = i - points
for(h in 1:(length(diff) - 1)){if(diff[h] > 0 & diff[h + 1] <= 0){find <- h}}
return(find)}
cor_matrix = matrix(0, m * t, m * t)
useage_m = cumsum(m * t); useage_m = c(0, useage_m)
for(i in 1:(m * t)){i_reg = region_ij(useage_m, i)
for(j in 1:(m * t)){j_reg = region_ij(useage_m, j)
if(i_reg == j_reg & i != j){
cor_matrix[i, j] = alpha0}else if(i == j){cor_matrix[i, j] = 1
}else if(i_reg != j_reg){cor_matrix[i,j] = alpha0 * (rho^(abs(i_reg - j_reg)))}}}
Sigma[[1]] = cor_matrix; Sigma[[2]] = cor_matrix
y_ed = simbinCLF(mu = mu, Sigma = Sigma, n = n)