Statistical techniques for signal generation : The Australian experience
- Authors: Purcell, Patrick , Barty, Simon
- Date: 2002
- Type: Text , Journal article
- Relation: Drug safety Vol. 25, no. 6 (2002), p. 415-421
- Full Text: false
- Reviewed:
- Description: National voluntary reporting systems generate large volumes of clinical data pertinent to drug safety. Currently descriptive statistical techniques are used to assist in the detection of drug safety 'signals'. Australian data have been coded according to guidelines formulated almost 30 years ago and which have resulted in many drugs which are not associated with an adverse drug reaction or 'innocent bystander' drugs being recorded as 'suspected' in individual reports. In this paper we explore the application of an iterative probability filtering algorithm titled 'PROFILE'. This serves to identify the 'signals' and remove the 'innocent bystander' drugs, thus providing a clearer view of the drugs most likely to have caused the reactions. Reaction terms analysed include neutropenia, agranulocytosis, hypotension, hypertension, myocardial infarction, neuroleptic malignant syndrome, and rectal haemorrhage. In this version of PROFILE, Fishers exact test has been used as the statistical tool but other methods could be used in future. Advantages and limitations of the method and its assumptions are discussed together with the rationale underlying the method and suggestions for further enhancements.
- Description: 2003000065
Data mining of the Australian adverse drug reactions database : A comparison of Bayesian and other statistical indicators
- Authors: Harvey, Jack , Turville, Christopher , Barty, Simon
- Date: 2004
- Type: Text , Journal article
- Relation: International Transactions in Operational Research Vol. 11, no. (2004), p. 419-433
- Full Text: false
- Reviewed:
- Description: The Australian adverse drug reactions database is derived from 140,000 reports over 30 years, including many instances of multiple drugs and multiple reactions. There are several thousand different drugs and reactions, and so the drug-reaction table is large and sparse. To aid rapid expert assessment of new reports, Bayesian approaches are being compared with other statistical methods for the re-evaluation of historical data and to provide early indications of emerging trends. Bayesian methods provide more balanced detection criteria than either descriptive statistics such as relative risks, which are subject to large sampling variation for rare co-occurrences, or statistical significance levels which are conversely weighted towards the most common co-occurrences. In this paper the various methods are reviewed and some indicative early results are presented.
- Description: C1
- Description: 2003000879
Statistical detection techniques to reduce confounding by association (co-medication) and confounding by indication in the Australian spontaneous reporting system
- Authors: Barty, Simon
- Date: 2005
- Type: Text , Thesis , PhD
- Full Text: false
- Description: This research is the incorporation and melding of classical and Bayesian statistical techniques into an iterative methodology aimed at reducing two major confounding issues in Adverse Drug Reactions (ADRs), confounding by association (drug therapy) and confounding by indication (medical condition) to assist in the detection of signals. [...] This study highlights the ability of the STATFILE algorithm to detect drugs that are potential signals. More importantly, it also flags those drugs that are considered to be bystander drugs or noise, consequently reducing confounding by association. [...] This work highlights the significance and viability of an automated signal detection system and its practical application for the Australian spontaneous reporting of ADRs scene and potentially the international scene.
- Description: Doctor of Philosophy