tystan / pharmsignal

Safety signal detection in pharmacological data (pharmacovigilance)

pharmsignal

Safety signal detection in pharmacological data (pharmacovigilance).

This package contains implementations of

• Reporting Odds Ratio (ROR),
• Proportional Reporting Ratio (PRR),
• Bayesian Confidence Propagation Neural Network Information Component (BCPNN IC, both MCMC and normal approximation), and
• Observed-to-expected ratio with simple shrinkage.

To be implemented

• Empirical Bayes Geometric Mean (EBGM), and
• Sequence Symmetry Analysis (SSA).

Background

Signal detection of adverse drug reaction (ADR) events often requires analysis of tabulated count data as seen in the below table.

	ADR(s) X	ADR(s) Y
Drug(s) A	a	b
Drug(s) B	c	d

The package's functions to calculate signal detection measures are:

• ror_signal(): ROR with 100(1-alpha/2)% confidence intervals
• prr_signal(): PRR with 100(1-alpha/2)% confidence intervals
• bcpnn_norm_signal(): BCPNN IC using the 'exact' expectation and variance of the the IC posterior distribution of Gould (2003) using a normal approximation to construct the 100(1-alpha/2)% confidence interval on the log_2 scale (see equations (A3) and (A4) of the Appendix)
• bcpnn_mcmc_signal(): BCPNN IC using the maximum a posteriori (m.a.p.) central estimate of the IC with MCMC simulation of the exact empirical distribution for 100(1-alpha/2)% confidence (credible) regions of Noren (2006)
• obsexp_shrink_signal(): Observed-to-expected ratio with simple shrinkage transformation with 100(1-alpha/2)% confidence intervals of Noren (2013)
• ebgm_signal() [to be implemented]
• seq_sym_signal() [to be implemented]

Example usage

library(devtools) # see https://www.r-project.org/nosvn/pandoc/devtools.html
devtools::install_github('tystan/pharmsignal')
library(pharmsignal)

# The example values are taken from Table III (page 3749) of:
# Noren et al. (2006). Stats in Med, 25:3740–3757 (DOI: 10.1002/sim.2473).

n11 <- 25
n1. <- 1126
n.1 <- 87
n.. <- 572573

# convert to the corresponding a, b, c, d values for a 2x2 table
a <- n11
b <- n1. - n11
c <- n.1 - n11
d <- n.. - a - b - c

# perform calculations for values
set.seed(123456)
results_tab <- bcpnn_mcmc_signal(a, b, c, d)

# print in two parts as is a wide results table
n_cols <- ncol(results_tab)
half_cols <- ceiling(n_cols / 2)
results_tab[, 1:half_cols]

# analysis n11 drug margin event margin    n..     E_n11   est_name est_scale
#        1  25        1126           87 572573 0.1710908 bcpnn_mcmc      log2

results_tab[, (half_cols + 1):n_cols]

#      est var_scale var_est sd_est alpha   ci_lo   ci_hi
# 5.247846      <NA>      NA     NA  0.05 4.64927 5.73115

Acknowledgements

The BCPNN calculations have been altered from that found in the PhViD package.

I have attempted to contact the authors about the incorrect use of log() instead of log2() in the code here without luck.