openEBGM (version 0.9.1)

processRaw: Process raw data

Description

processRaw finds the actual and expected counts using the methodology described by DuMouchel (1999); however, an adjustment is applied to expected counts to prevent double-counting (i.e., using unique marginal ID counts instead of contingency table marginal totals). Also calculates the relative reporting ratio (RR) and the proportional reporting ratio (PRR).

Usage

processRaw(
  data,
  stratify = FALSE,
  zeroes = FALSE,
  digits = 2,
  max_cats = 10,
  list_ids = FALSE
)

Value

A data frame with actual counts (N), expected counts (E), relative reporting ratio (RR), and proportional reporting ratio (PRR) for var1-var2 pairs. Also includes a column for IDs (ids) if list_ids = TRUE.

Arguments

data

A data frame containing columns named: id, var1, and var2. Possibly includes columns for stratification variables identified by the substring 'strat' (e.g. strat_age). Other columns will be ignored.

stratify

A logical scalar (TRUE or FALSE) specifying if stratification is to be used.

zeroes

A logical scalar specifying if zero counts should be included. Using zero counts is only recommended for small data sets because it will dramatically increase run time.

digits

A whole number scalar specifying how many decimal places to use for rounding RR and PRR.

max_cats

A whole number scalar specifying the maximum number of categories to allow in any given stratification variable. Used to help prevent a situation where the user forgets to categorize a continuous variable, such as age.

list_ids

A logical scalar specifying if a column for pipe-concatenated IDs should be returned.

Warnings

Use of the zeroes = TRUE option will result in a considerable increase in runtime. Using zero counts is not recommended if the contingency table is moderate or large in size (~500K cells or larger). However, using zeroes could be useful if the optimization algorithm fails to converge when estimating hyperparameters.

Any columns in data containing the substring 'strat' in the column name will be considered stratification variables, so verify that you do not have any extraneous columns with that substring.

Details

An id column must be included in data. If your data set does not include IDs, make a column of unique IDs using df$id <- 1:nrow(df). However, unique IDs should only be constructed if the cells in the contingency table are mutually exclusive. For instance, unique IDs for each row in data are not appropriate with CAERS data since a given report can include multiple products and/or adverse events.

Stratification variables are identified by searching for the substring 'strat'. Only variables containing 'strat' (case sensitive) will be used as stratification variables. PRR calculations ignore stratification, but E and RR calculations are affected. A warning will be displayed if any stratum contains less than 50 unique IDs.

If a PRR calculation results in division by zero, Inf is returned.

References

DuMouchel W (1999). "Bayesian Data Mining in Large Frequency Tables, With an Application to the FDA Spontaneous Reporting System." The American Statistician, 53(3), 177-190.

Examples

data.table::setDTthreads(2)  #only needed for CRAN checks
var1 <- c("product_A", rep("product_B", 3), "product_C",
          rep("product_A", 2), rep("product_B", 2), "product_C")
var2 <- c("event_1", rep("event_2", 2), rep("event_3", 2),
          "event_2", rep("event_3", 3), "event_1")
strat1 <- c(rep("Male", 5), rep("Female", 3), rep("Male", 2))
strat2 <- c(rep("age_cat1", 5), rep("age_cat1", 3), rep("age_cat2", 2))
dat <- data.frame(
  var1 = var1, var2 = var2, strat1 = strat1, strat2 = strat2,
  stringsAsFactors = FALSE
)
dat$id <- 1:nrow(dat)
processRaw(dat)
suppressWarnings(
  processRaw(dat, stratify = TRUE)
)
processRaw(dat, zeroes = TRUE)
suppressWarnings(
  processRaw(dat, stratify = TRUE, zeroes = TRUE)
)
processRaw(dat, list_ids = TRUE)