eocusum_arl_sim: Compute ARLs of EO-CUSUM control charts using simulation

In vlad: Variable Life Adjusted Display and Other Risk-Adjusted Quality Control Charts

Description

Compute ARLs of EO-CUSUM control charts using simulation.

Usage

eocusum_arl_sim(r, pmix, k, h, RQ = 1, yemp = FALSE, side = "low")

Arguments

r	Integer. Number of of simulation runs.
pmix	Data Frame. A three column data frame. First column is the operation outcome. Second column are the predicted probabilities from the risk model. Third column can be either the predicted probabilities from the risk model or average outcome.
k	Double. Reference value of the CUSUM control chart. Either 0 or a positive value. Can be determined with function optimal_k.
h	Double. Decision interval (alarm limit, threshold) of the CUSUM control chart.
RQ	Double. Defines the true performance of a surgeon with the odds ratio ratio of death RQ. Use RQ = 1 to compute the in-control ARL and other values to compute the out-of-control ARL.
yemp	Logical. If TRUE use observed outcome value, if FALSE use estimated binary logistc regression model.
side	Character. Default is "low" to calculate ARL for the upper arm of the V-mask. If side = "up", calculate the lower arm of the V-mask.

Value

Returns a single value which is the Run Length.

Author(s)

Philipp Wittenberg

References

Wittenberg P, Gan FF, Knoth S (2018). A simple signaling rule for variable life-adjusted display derived from an equivalent risk-adjusted CUSUM chart. Statistics in Medicine, 37(16), pp 2455–2473.

Examples

## Not run: 
library("dplyr")
library("tidyr")
library(ggplot2)

## Datasets
data("cardiacsurgery", package = "spcadjust")
cardiacsurgery <- cardiacsurgery %>% rename(s = Parsonnet) %>%
  mutate(y = ifelse(status == 1 & time <= 30, 1, 0))
s5000 <- sample_n(cardiacsurgery, size = 5000, replace = TRUE)
df1 <- select(cardiacsurgery, s, y)
df2 <- select(s5000, s, y)

## estimate coefficients from logit model
coeff1 <- round(coef(glm(y ~ s, data = df1, family = "binomial")), 3)
coeff2 <- round(coef(glm(y ~ s, data = df2, family = "binomial")), 3)

## set up
RNGkind("L'Ecuyer-CMRG")
m <- 10^3
kopt <- optimal_k(QA = 2, df = S2I, coeff = coeff1, yemp = FALSE)
h <- eocusum_arloc_h_sim(L0 = 370, df = df1, k = kopt, m = m, side = "low", coeff = coeff1,
                         coeff2 = coeff2, nc = 4)

## Serial simulation
RLS <- do.call(c, lapply(1:m, eocusum_arloc_sim, h = h, k = kopt, df = df1, side = "low",
                         coeff = coeff1, coeff2 = coeff2))
data.frame(cbind(ARL = mean(RLS), ARLSE = sd(RLS)/sqrt(m)))

## Parallel simulation (FORK)
RLS <- simplify2array(parallel::mclapply(1:m, eocusum_arloc_sim, h = h, k = kopt, df = df1,
                                         side = "low", coeff = coeff1, coeff2 = coeff2,
                                         mc.cores = parallel::detectCores()))
data.frame(cbind(ARL = mean(RLS), ARLSE = sd(RLS)/sqrt(m)))

## Parallel simulation (PSOCK)
no_cores <- parallel::detectCores()
cl <- parallel::makeCluster(no_cores)
side <- "low"
h_vec <- h
QS_vec <- 1
k <- kopt
parallel::clusterExport(cl, c("h_vec", "eocusum_arloc_sim", "df1", "coeff1", "coeff2",
                              "QS_vec", "side", "k"))
time <- system.time( {
  RLS <- array(NA, dim = c( length(QS_vec), length(h_vec), m))
  for (h in h_vec) {
    for (QS in QS_vec) {
      cat(h, " ", QS, "\n")
      RLS[which(QS_vec==QS), which(h==h_vec), ] <- parallel::parSapply(cl, 1:m, eocusum_arloc_sim,
                                                                       side = side, QS = QS, h = h,
                                                                       k = k, df = df1,
                                                                       coeff = coeff1,
                                                                       coeff2 = coeff2,
                                                                       USE.NAMES = FALSE)
    }
  }
} )
ARL <- apply(RLS, c(1, 2), mean)
ARLSE <- sqrt(apply(RLS, c(1, 2), var)/m)
print(list(ARL, ARLSE, time))
parallel::stopCluster(cl)

## End(Not run)

vlad documentation built on Feb. 15, 2021, 5:12 p.m.