greedy.fi: Calculate a fragility index using a greedy algorithm
In brb225/FragilityTools: Toolbox For Patient Count Fragility Measures

Description Usage Arguments Details Value Examples

This is a very general function which approximately calculates a fragility index using covariates, response(s), and any p value function. It's described in detail in the article Generalized fragility index.

greedy.fi(
  X,
  Y,
  get.replacements,
  get.p.val,
  alpha = 0.05,
  verbose = FALSE,
  cl = NULL,
  only.consider = NULL,
  dont.consider = c()
)

`X`	a data frame of covariates which are not subject to modification.
`Y`	a data frame of responses which are subject to modification.
`get.replacements`	a function which outputs a data frame containing all possible row replacements for Y which are to be considered. The functions inputs the row of Y under consideration, the row of X under consideration, the row name, and the full original data frames X and Y.
`get.p.val`	a function that inputs X and Y and returns a p value
`alpha`	a numeric for the significance cutoff
`verbose`	a logical value for whether to print status updates while running
`cl`	a cluster from the `parallel` package, used to compute fragility index over each modified observation at each stage of the greedy algorithm
`only.consider`	a vector of row names to only consider, by default NULL which considers all patients
`dont.consider`	a vector of row names to not consider

This is a general function which uses a greedy algorithm to compute a fragility index.The function arguments accept data from two sources: covariates in X and responses in Y. Covariates are unchanged by the algorithm, while responses are iteratively changed. The type of each response is specified to determine which substitute outcomes to consider.

a list containing a fragility index and related information

FI: A fragility index, positive if the initial $p$-value was < alpha and negative otherwise
p_val_sequence: An atomic vectors of the p-values at each step of the iteratative algorithm, where the first is the starting p value and the last is the p value which crossed the alpha threshold
reverse: A boolean indicating whether the original p value is larger than alpha
num_patients: The number of patients whose responses were subject to change
patients: An atomic vector with the rownames of the rows which had their responses changed
old_responses: A list of dataframes of the original responses of each changed patient
new_responses: A list of dataframes of the new responses of each changed patient
Zmod: A data frame of the modified cbind(X,Y)

n <- 100
X <- data.frame("tr_group" = sample(c("treated", "not treated"), n, TRUE))
Y <- data.frame("outcome" = sample(c("sick", "healthy"), n, TRUE))
get.p.val <- function(X, Y) fisher.test(table(X[[1]], Y[[1]]))$p.value
get.replacements <- function(y, x, rn, Y, X) data.frame(Y=setdiff(unique(Y[[1]]), y))
greedy.fi(X, Y, get.p.val = get.p.val, get.replacements = get.replacements)$FI