crt2power is an R package that allows users to calculate
the statistical power or sample size of their cluster randomized trials
(CRTs) with two continuous co-primary outcomes, given a set of input
parameters. The motivation for this package is to aid in the design of
hybrid 2 studies. Hybrid 2 studies are studies where there are two
co-primary outcomes, namely an implementation outcome (such as fidelity
or reach) and a health outcome (such as infection rates, or change from
baseline health scores). When powering these studies, cluster
correlations and the inflation of the Type I error rate must be
accounted for.
The five key study design approaches are included in this package that can be used to power hybrid 2 CRTs. 1. P-Value Adjustments for Multiple Testing 2. Combined Outcomes Approach 3. Single 1-Degree of Freedom (DF) Combined Test for Two Outcomes 4. Disjunctive 2-DF Test for Two Outcomes 5. Conjunctive Intersection-Union Test for Two outcomes
For details on the methods listed above, please refer to the publication that discusses these methods by Owen et al., available here.
This package is available on CRAN, so it is recommended to run the following code:
install.packages("crt2power")
require(crt2power)
If you wish to directly install it from the GitHub repository instead, you can run the following code:
install.packages("devtools")
require(devtools)
install_github("https://github.com/melodyaowen/crt2power")
require(crt2power)
Table of Key Required Input Parameters: | Parameter |
Statistical Notation | Variable Name | Description | | — | — | — | — | |
Statistical power | \(\pi\) |
power | Probability of detecting a true effect under \(H_A\) | | Number of clusters | \(K\) | K | Number of clusters
in each treatment arm | | Cluster size | \(m\) | m | Number of
individuals in each cluster | | Family-wise false positive rate | \(\alpha\) | alpha | Probability
of one or more Type I error(s) | | Effect for \(Y_1\) | \(\beta_1^*\) | beta1 |
Estimated intervention effect on the first outcome (\(Y_1\)) | | Effect for \(Y_2\) | \(\beta_2^*\) | beta2 |
Estimated intervention effect on the second outcome (\(Y_2\)) | | Total variance of \(Y_1\) | \(\sigma_1^2\) | varY1 | Total
variance of the first outcome, \(Y_1\)
| | Total variance of \(Y_2\) | \(\sigma_2^2\) | varY2 | Total
variance of the second outcome, \(Y_2\)
| | Endpoint-specific ICC for \(Y_1\) |
\(\rho_0^{(1)}\) | rho01 |
Correlation for \(Y_1\) for two
different individuals in the same cluster | | Endpoint-specific ICC for
\(Y_2\) | \(\rho_0^{(2)}\) | rho02 |
Correlation for \(Y_2\) for two
different individuals in the same cluster | | Inter-subject
between-endpoint ICC | \(\rho_1^{(1,2)}\) | rho1 |
Correlation between \(Y_1\) and \(Y_2\) for two different individuals in the
same cluster | | Intra-subject between-endpoint ICC | \(\rho_2^{(1,2)}\) | rho2 |
Correlation between \(Y_1\) and \(Y_2\) for the same individual | | Treatment
allocation ratio | \(r\) |
r | Treatment allocation ratio; \(K_2 = rK_1\) where \(K_1\) is number of clusters in experimental
group | | Statistical distribution | – | dist |
Specification of which distribution to base calculation on, either the
\(\chi^2\)-distribution or \(F\)-distribution1 1. When
selecting the \(\chi^2\)-distribution,
all methods will use this distribution with the exception of the
conjunctive IU test, which will use the multivariate normal (MVN)
distribution; when selecting the \(F\)-distribution, all methods will use this
distribution with the exception of the conjunctive IU test, which will
use the \(t\)-distribution.
Each method has a set of functions for calculating the statistical power (\(\pi\)), required number of clusters per treatment group (\(K\)), or cluster size (\(m\)) given a set of input parameters. The names of all functions offered in this package are listed below, organized by study design method.
calc_pwr_pval_adj() calculates power for this
methodcalc_K_pval_adj() calculates number of clusters per
treatment group for this methodcalc_m_pval_adj() calculates cluster size for this
methodcalc_pwr_comb_outcome() calculates power for this
methodcalc_K_comb_outcome() calculates number of clusters per
treatment group for this methodcalc_m_comb_outcome() calculates cluster size for this
methodcalc_pwr_single_1dftest() calculates power for this
methodcalc_K_single_1dftest() calculates number of clusters
per treatment group for this methodcalc_m_single_1dftest() calculates cluster size for
this methodcalc_pwr_disj_2dftest() calculates power for this
methodcalc_K_disj_2dftest() calculates number of clusters per
treatment group for this methodcalc_m_disj_2dftest() calculates cluster size for this
methodcalc_pwr_conj_test() calculates power for this
methodcalc_K_conj_test() calculates number of clusters per
treatment group for this methodcalc_m_conj_test() calculates cluster size for this
methodrun_crt2_design(output = "power", ...) calculates power
for all 5 methodsrun_crt2_design(output = "K", ...) calculates number of
clusters per treatment group for all 5 methodsrun_crt2_design(output = "m",...) calculates cluster
size for all 5 methods# Example of using the combined outcomes approach for calculating power
calc_pwr_comb_outcome(dist = "Chi2", K = 8, m = 50, alpha = 0.05,
beta1 = 0.2, beta2 = 0.4, varY1 = 0.5, varY2 = 1,
rho01 = 0.05, rho02 = 0.1, rho1 = 0.01, rho2 = 0.1,
r = 1)
# Example of using the single weighted 1-DF test for calculating K
calc_K_single_1dftest(dist = "F", power = 0.9, m = 70, alpha = 0.05,
beta1 = 0.4, beta2 = 0.3, varY1 = 1.5, varY2 = 0.5,
rho01 = 0.1, rho02 = 0.07, rho1 = 0.05, rho2 = 0.3,
r = 2)
# Example of using conjunctive IU test for m calculation
calc_m_conj_test(dist = "MVN", power = 0.8, K = 10, alpha = 0.05,
beta1 = 0.4, beta2 = 0.4, varY1 = 0.5, varY2 = 1,
rho01 = 0.05, rho02 = 0.1, rho1 = 0.07, rho2 = 0.9,
r = 1, two_sided = TRUE)
# Example of calculating power based on all five methods
run_crt2_design(output = "power", K = 6, m = 70, alpha = 0.05,
beta1 = 0.4, beta2 = 0.4, varY1 = 0.5, varY2 = 0.5,
rho01 = 0.1, rho02 = 0.1, rho1 = 0.07, rho2 = 0.9, r = 1)
For questions or comments, please email Melody Owen at melody.owen@yale.edu, or submit an issue to this repository.