Bayesian confidence propagation neural network in pharmacovigilance

Quantitative and data mining approach has been created to enhance the emphasis of the clinical signal detection procedure. The process employed, is known as the BCPNN (Bayesian Confidence Propagation Neural Network). This not just aids in the early detection of adverse drug reactions (ADRs) but in addition further evaluation of such signals. The process makes use of Bayesian statistical principles to quantify apparent dependencies in the data set. This quantifies the degree to which a certain drug- ADR mixture is distinct from a background (in this case the WHO database). The measure of disproportionality utilized, is termed as the Information Component (IC) due to its’ origins in Information Theory. A confidence interval is computed for the IC of each and every pairing. A neural network method enables all drug-ADR combinations in the database to be evaluated in an automatic manner. Evaluations of the effectiveness of the BCPNN in signal detection are detailed.To examine how a drug association compares in unexpectedness to associated drugs, which may be utilised for precisely the same clinical indication, the approach is extended to consideration of groups of drugs.

Contents

Part I Finding Signals
Chapter 1. Pharmacovigilance

1.1 Risk benefit assessment
1.2 The role of WHO in Pharmacovigilance: past and present
1.3 The current work of WHO in signal detection
1.4 Signal detection and analysis
Chapter 2. Approaches to Quantitative Signal detection
2.1 Data mining: knowledge finding or discovery
2.2 Approaches to quantitative signal detection in pharmacovigilance
2.2.1 Drug associated event systems
i) Intensive hospital based systems Cohort based
ii) Outpatient cohort studies
iii) Case Control Surveillance
2.2.2 Spontaneous reporting systems
i) The use of spontaneous reporting combined with an estimate of utilisation data such as sales/prescription data
ii) Changes in reporting rates: trends
2.2.3 Comparison to database background
Chapter 3. Measures of disproportionality compared to database
background
3.1 What they are in general
3.2 Similarities and differences
3.2.1 Multiple testing
3.2.2 ADRs are also hard to detect clinically
Chapter 4. Bayesian Statistics
4.1 History
4.2 Bayesian statistical principles
4.3 Comparison between classical and Bayesian statistics
4.4 Distinction between using Bayes Law and using Bayesian statistics
4.5 Priors and Bayesian controversy
4.5.1 Prior and Posterior probabilities
4.5.2 Standard deviation v Standard Error
4.5.3 Bayesian approach advantages
4.5.4 Disadvantages of Bayesian approach
4.6 Applications
4.6.1 Benefits of using Bayesian statistics when data mining
Chapter 5. Neural Networks
5.1 The use of a neural network
5.2 Applications
5.3 Implementation in the WHO database
5.3.1 Feed forward network
5.3.2 The recurrent neural network
Chapter 6. BCPNN
6.1 History
6.2 IC analysis in general
6.2.1 Choice of measure of disproportionality
6.3 Information theory
6.4 IC as a distribution
6.5 Choice of priors for BCPNN on WHO database
6.6 Limitations of current implementation of IC
6.6.1 Specific Limitations
6.7 Benefits of a Bayesian approach in data mining and use ofstandard deviation measure
6.8 IC properties and interpretation
6.9 Other application of IC like analysis
6.10 Other issues and how the BCPNN approaches them
Chapter 7. Implementation of quantitative signal detection in signalling strategy
7.1 The problem of data mining in general
7.2 Computational architecture
7.3 Choice of precision estimate
7.4 Stratification
7.5 Variation of background for comparison (role of stratification)
7.6 Good drug, Bad drug
7.7 Choice of background for comparison and calculation ofexpected and observed numbers
7.8 Suspected only or concomitant drugs used counted?
7.9 Combinations or Cases?
7.10 Other exclusion criteria for cases
7.11 Level of drug and ADR classification
7.12 Other Implementation issues
7.12.1 Measuring Impact of approach
7.12.2 Feedback mechanisms
7.12.3 Use of ad hoc and routine runs
7.12.4 Frequency of data scanning
7.12.5 Combination with other signal detection strategies
7.12.6 Output information
i) UMC routine output: Combinations database
ii) Trend analysis
iii) Triage – use and need
iv) Best-case-worse-case scenarios
v) Non-highlighted combinations interesting?
vi) False positives and false negatives
vii) The implication of missing data and how to approach it
viii) Implications of data set used
ix) Data quality
7.13 A summary on the use and implementation of the BCPNN
Chapter 8. Method evaluation
8.1 Introduction and initial UMC tests
8.2 Problems of validation
8.3 Retrospective evaluation
8.3.1 Limitations of the retrospective study
8.3.2 Commentary on the results of the retrospective study
8.4 Comparison to other methods
8.5 Further tests needed
Part II Refining signals
Chapter 9. Group effect detection
9.1 Grouping of drugs
9.1.1 A clinically interesting example
9.1.2 Refinements
9.2 Strengths and weaknesses
9.3 Implementation into routine work
9.4 Further work
9.5 Other methods
Chapter 10. Complex dependencies
10.1 The need for searching for complex relations in generaland the WHO database
10.2 Pairs and triplets search in WHO database
10.3 Pattern recognition
10.4 Recurrent neural network
10.5 Bayesian Classifier
10.6 Artificial test set…………..

Source: Umea University

Download URL 2: Visit Now

Leave a Comment