BCPNN: Bayesian Confidence Propagation Neural Network (BCPNN)
In bips-hb/pvm: PharmacoVigilance Methods

Description Usage Arguments Details Value References Examples

Applies the BCPNN to a collection of 2 x 2 tables of the form

	event	not event
drug	`a`	`c`
not drug	`b`	`d`

There are two versions of the BCPNN:

'original' - The original version proposed by Bate et al. (1998)
'alternative' - The BCPNN as proposed by Norén et al. (2006)

1 2	BCPNN(a, b, c, d, alpha = NULL, version = "original", mc_estimate = FALSE, mc_runs = 1000)

`a`	A vector with the counts of the upper left corner of the tables
`b`	A vector with the counts of the lower left corner of the tables
`c`	A vector with the counts of the upper right corner of the tables
`d`	A vector with the counts of the lower right corner of the tables
`alpha`	Value between (0,1). If set, the lower endpoint the confidence (or credible) interval is returned. (Default = `NULL`)
`version`	Version of the BCPNN used. Can either be `'original'` (Default) for the BCPNN as proposed orignally by Bate et al. (1998), or `'alternative'` for the BCPNN as proposed by Norén et al. (2006).
`mc_estimate`	The value is estimated using Monte Carlo runs (Default = `FALSE`). Only used when `version = 'alternative'`.
`mc_runs`	The number of Monte Carlo runs used to estimate the credible interval. (Default: 1000). Only used when `version = 'alternative'`.

The implementation of this function is based on the implementation in the PhViD package.

The maximum aposteriori estimate of the information component (IC) or the lower endpoint of the approximate credible interval

Bate, A., Lindquist, M., Edwards, I. R., Olsson, S., Orre, R., Lansner, A., & De Freitas, R. M. (1998). A Bayesian neural network method for adverse drug reaction signal generation. European Journal of Clinical Pharmacology, 54(4), 315–321. http://doi.org/10.1007/s002280050466

Norén, G. N., Bate, A., Orre, R., & Edwards, I. R. (2006). Extending the methods used to screen the WHO drug safety database towards analysis of complex associations and improved accuracy for rare events. Statistics in Medicine, 25(21), 3740–3757. http://doi.org/10.1002/sim.2473

# get the tables
a <- srdata$tables$a
b <- srdata$tables$b
c <- srdata$tables$c
d <- srdata$tables$d

# Applying the original BCPNN: 
BCPNN(a, b, c, d)
# [1]  0.349783103 -0.609077730 -0.168446711 -0.277981964 ...

# Getting the lower end point of the 95% confidence intervaL: 
BCPNN(a, b, c, d, alpha = 0.05) 
# [1]  0.280077253 -0.994960076 -0.293624528 -0.408661852 ...

# Using the alternative version: 
BCPNN(a, b, c, d, version = 'alternative')
# [1]  0.350235800 -0.595807902 -0.166901050 -0.276387348 ...

# Getting the lower end points of the 95% confidence interval 
# using the alternative version. The estimates are based on 
# 10,000 Monte Carlo samples:
BCPNN(a, b, c, d, version = 'alternative', 
      alpha = 0.05, mc_estimate = TRUE, mc_runs = 10^4)
# [1]  [1]  0.31621489 -0.92490130 -0.25601307 -0.37040303 ...