basket {basket} | R Documentation |

## Create a Basket Trial Analysis

### Description

This function creates an analysis modeling the exchangeability and distribution characteristics of cohorts in a basket trial, where a single therapy is given to multiple cohorts. The model is based on the multi-source exchangeability model. This is a generalization of the Bayesian hierarchical model and it accomodates multiple sets of latent factors shared combinations of cohorts.

### Usage

```
basket(
responses,
size,
name,
p0 = 0.15,
shape1 = 0.5,
shape2 = 0.5,
prior = diag(length(responses))/2 + matrix(0.5, nrow = length(responses), ncol =
length(responses)),
hpd_alpha = 0.05,
alternative = "greater",
call = NULL,
cluster_function = cluster_membership,
method = c("mcmc", "lmem", "exact"),
mcmc_iter = 2e+05,
mcmc_burnin = 50000,
initial_mem = round(prior - 0.001),
cluster_analysis = FALSE,
seed = get_seed()
)
```

### Arguments

`responses` |
the number of responses in each basket. |

`size` |
the size of each basket. |

`name` |
the name of each basket. |

`p0` |
the null response rate for the poster probability calculation. (default 0.15) |

`shape1` |
the first shape parameter(s) for the prior of each basket. (default 0.5) |

`shape2` |
the second shape parameter(s) for the prior of each basket. (default 0.5) |

`prior` |
the matrix giving the prior inclusion probability for each pair of baskets. The default is on on the main diagonal and 0.5 elsewhere. |

`hpd_alpha` |
the highest posterior density trial significance. (default 0.05) |

`alternative` |
the alternative case definition (default "greater"). |

`call` |
the call of the function. (default NULL) |

`cluster_function` |
a function to cluster baskets. |

`method` |
"mcmc", "lmem" (local MEM), or "exact". See details for an explanation. (default "mcmc") |

`mcmc_iter` |
if the method is "mcmc" then this spcifies the number of mcmc iterations. Otherwise, it is ignored. (default 200000) |

`mcmc_burnin` |
if the method is "mcmc" then this specifies the number of burn-in iterations. (default 50000) |

`initial_mem` |
if the method is "mcmc" then this spcifies the initial MEM matrix. Otherwise, it is ignored. |

`cluster_analysis` |
if the cluster analysis is conducted. |

`seed` |
the random seed for the mcmc calculations. By default this is .Random.seed. If this value is not initialized, then it is first initialized with Sys.time() and then returned. |

### Details

The model may be fit using an exact calculation via mcmc, or a local version of MEM. The first conducts posterior inference through the entire set of exchangeability relationships in the sample domain. This approach is computationally feasible only when the number of cohorts is relatively small. As a rule of thumb this option should be used with a maximum of 20 cohorts. By default, the second(mcmc) is used and it is based on the Metropolis algorithm and it extends the model's implementation to larger collections of subpopulations. The algorithm initiates with a burn-in period (see mcmc_burnin), which are discarded from the analysis. The third is based on a recent paper by Wei et al. The calculation is computationally efficient and tends to borrow to a lesser extent than other implemented methods.

### Examples

```
# 3 baskets, each with enrollement size 5
trial_sizes <- rep(5, 3)
# The response rates for the baskets.
resp_rate <- 0.15
# The trials: a column of the number of responses and a column of the
# the size of each trial.
trials <- data.frame(
responses = rbinom(trial_sizes, trial_sizes, resp_rate),
size = trial_sizes,
name = letters[1:3]
)
summary(basket(trials$responses, trials$size, trials$name))
```

*basket*version 0.10.11 Index]