Signal detection using via variable selection in logistic regression. The Bayesian Information Criterion maximization is assessed using Metropolis-Hastings algorithm.

Description

Spontaneous adverse event reports have a high potential for detecting adverse drug reactions. However, due to their dimension, the analysis of such databases requires statistical methods. The MHTrajectoryR package propose to use a logistic regression whose sparsity is viewed as a model selection challenge. Since the model space is huge, a Metropolis-Hastings algorithm carries out the model selection by maximizing the BIC criterion.

Usage

1
Analyze_oneAE(ae, drug, maxit, alpha, nbinit)

Arguments

ae

The response binary vector which indicates if the adverse event is observed (value 1) and (value 0) otherwise. It must have the sime length with the number of rows of matrix of drugs consumption.

drug

The matrix of drugs consumption or the matrix of binary covariates. Each row corresponds to one individual drugs consumption. Each column corresponds to one drug.

maxit

Number of iteration of the Metropolis-Hastings algorithm. In other words, the length of one trajectory of the Metropolis-Hastings Markov Chain.

alpha

The parameter that define the neighbordhood.

nbinit

The number of random initialisations.

Value

list of (best) the best model that maximize the BIC. (all) all explored models through the trajectory. (signals) the detected signals (positive, negative or unknown) in OMOP reference set.

Author(s)

Mohammed Sedki and Matthieu Marbac

References

Matthieu Marbac, Pascale Tubert-Bitter, Mohammed Sedki: Bayesian model selection in logistic regression for the detection of adverse drug reactions. http://arxiv.org/abs/1505.03366 (accepted for publication in Biometrical Journal).

Examples

1
2
3
4
5
6
7
8
## Not run: 
  data(exampleAE)
  data(exampleDrugs)
  res <- Analyze_oneAE(exampleAE[,1], exampleDrugs, 10, 1, 10)
  # print signals (drugs relied to the adverse event)
  print(res$signal)

## End(Not run)