Test on device-events using a one-layer BCPNN as proposed by Bate et al (1998), which assumes beta-distributed probabilities and a normal approximation of the variance of the information component (IC). From the family of disproportionality analyses (DPA) used to generate signals of disproportionate reporting (SDRs).
bcpnn(df, ...)# S3 method for mds_ts
bcpnn(df, ts_event = c(Count = "nA"), analysis_of = NA, ...)
# S3 method for default
bcpnn(
df,
analysis_of = NA,
eval_period = 1,
null_ratio = 1,
conf_interval = 0.9,
quantiles = c(0.05, 0.95),
cont_adj = 0,
...
)
Required input data frame of class mds_ts or, for generic
usage, any data frame with the following columns:
Unique times of class Date
Cell A count (class numeric) of the 2x2 table:
device/event of interest.
Cell B count (class numeric) of the 2x2 table:
device/non-event of interest.
Cell C count (class numeric) of the 2x2 table:
non-device/event of interest.
Cell D count (class numeric) of the 2x2 table:
non-device/non-event of interest.
Further arguments passed onto bcpnn methods
Required if df is of class mds_ts. Named string
indicating the variable corresponding to the event count (cell A in the 2x2
contingency table). In most cases, the default is the appropriate setting.
See details for alternative options.
Default: c("Count"="nA") corresponding to the event count column in
mds_ts objects. Name is generated from mds_ts metadata.
Optional string indicating the English description of what
was analyzed. If specified, this will override the name of the
ts_event string parameter.
Default: NA indicates no English description for plain df
data frames, or ts_event English description for df data frames
of class mds_ts.
Example: "Count of bone cement leakages"
Required positive integer indicating the number of unique times counting in reverse chronological order to sum over to create the 2x2 contingency table.
Default: 1 considers only the most recent time in df.
Example: 12 sums over the last 12 time periods to create the 2x2
contingency table.
Numeric value representing the null relative reporting
ratio (RR), used with conf_interval to establish the signal status.
This null_ratio is saved in the output as the signal threshold. See
details for more.
Default: 1 indicates a null RR of 1 and tests if the lower bound of
the IC conf_interval exceeds 1.
Numeric value between 0 and 1 representing the width of
the IC confidence interval, where the lower bound of the
interval is assessed against the null_ratio. The interval bounds are
returned as the lcl and ucl. See details for more.
Default: 0.90 indicates a 90% confidence interval with bounds at 5%
and 95%. The signal test is against the lower 5% bound, effectively
creating a one-sided test at the conventional 0.05 alpha level.
Vector of quantiles between 0 and 1. bcpnn() will
return an equal length vector of Gaussian-approximated quantiles from the
posterior distribution of the IC. Specify quantiles=NULL if no
quantiles are desired.
Default: c(.05, .95) corresponds to the 5% and 95% quantiles of the
IC.
Non-negative number representing the continuity adjustment to be added to each cell of the 2x2 contingency table. A value greater than 0 allows for contingency tables with a 0 in the D cell to run the algorithm. Adding a continuity adjustment will adversely affect the algorithm estimates, user discretion is advised. See details for more.
Default: 0 adds zero to each cell, thus an unadjusted table.
A named list of class mdsstat_test object, as follows:
Name of the test run
English description of what was analyzed
Named boolean of whether the test was run. The name contains the run status.
A standardized list of test run results: statistic
for the test statistic, lcl and ucl for the set
confidence bounds, p for the p-value, signal status, and
signal_threshold.
The test parameters
The data on which the test was run
mds_ts: BCPNN on mds_ts data
default: BCPNN on general data
null_ratio and conf_interval are used together to establish the
signal criteria. The null_ratio is conceptually similar to the
relative reporting ratio under a null hypothesis of no signal. Common values
are 1 and, more conservatively (fewer false signals), 2. The
conf_interval is the IC confidence interval used to test for a
signal. A value of 0.90 returns the 5
tests if the lower bound exceeds null_ratio. Effectively,
conf_interval=0.90 conducts a one-sided test at the conventional 0.05
alpha level.
cont_adj provides the option to allow bcpnn() to proceed running,
however this is done at the user's discretion because there are adverse
effects of adding a positive number to every cell of the contingency table.
By default, bcpnn() allows 0 in all cells except D.
It has been suggested that 0.5 may be an appropriate value. However, the
user is cautioned that interpretation may be compromised by adding continuity
adjustments.
For parameter ts_event, in the uncommon case where the
device-event count (Cell A) variable is not "nA", the name of the
variable may be specified here. Note that the remaining 3 cells of the 2x2
contingency table (Cells B, C, D) must be the variables "nB",
"nC", and "nD" respectively in df. A named character
vector may be used where the name is the English description of what was
analyzed. Note that if the parameter analysis_of is specified, it will
override this name. Example: ts_event=c("Count of Bone Cement
Leakages"="event_count")
Bate A, Lindquist M, et al. A Bayesian Neural Network Method for Adverse Drug Reaction Signal Generation. European Journal of Clinical Pharmacology, 1998, 54, 315-321.
Ahmed I, Poncet A. PhViD: PharmacoVigilance Signal Detection, 2016. R package version 1.0.8.
Lansner A, Ekeberg <U+00D6>. A one-layer feedback artificial neural network with a bayesian learning rule. Int. J. Neural Syst., 1989, 1, 77-87.
# NOT RUN {
# Basic Example
data <- data.frame(time=c(1:25),
nA=as.integer(stats::rnorm(25, 25, 5)),
nB=as.integer(stats::rnorm(25, 50, 5)),
nC=as.integer(stats::rnorm(25, 100, 25)),
nD=as.integer(stats::rnorm(25, 200, 25)))
a1 <- bcpnn(data)
# Example using an mds_ts object
a2 <- bcpnn(mds_ts[[3]])
# }
Run the code above in your browser using DataLab