Illusions of objectivity and a recommendation for reporting data mining results

Hauben, Manfred; Reich, Lester; Gerrits, Charles; Younus, Muhammad

doi:10.1007/s00228-007-0279-3

Cited by 25 publications

(18 citation statements)

References 14 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Bias exists due to subjective decisions in the selection, data mining procedure algorithm, and output selection, system deployment and interpretation, as discussed by Hauben et al,22 this study reconfirmed that the generalizability of results of statistical analyses on spontaneous reporting system data is affected by spontaneous reporting system data. Database profiles, such as reported substances and AEs, could be responsible for this difference.…”

Section: Discussionsupporting

confidence: 75%

Effect of database profile variation on drug safety assessment: an analysis of spontaneous adverse event reports of Japanese cases

Nomura¹,

Takahashi²,

Hinomura³

et al. 2015

DDDT

View full text Add to dashboard Cite

BackgroundThe use of a statistical approach to analyze cumulative adverse event (AE) reports has been encouraged by regulatory authorities. However, data variations affect statistical analyses (eg, signal detection). Further, differences in regulations, social issues, and health care systems can cause variations in AE data. The present study examined similarities and differences between two publicly available databases, ie, the Japanese Adverse Drug Event Report (JADER) database and the US Food and Drug Administration Adverse Event Reporting System (FAERS), and how they affect signal detection.MethodsTwo AE data sources from 2010 were examined, ie, JADER cases (JP) and Japanese cases extracted from the FAERS (FAERS-JP). Three methods for signals of disproportionate reporting, ie, the reporting odds ratio, Bayesian confidence propagation neural network, and Gamma Poisson Shrinker (GPS), were used on drug-event combinations for three substances frequently recorded in both systems.ResultsThe two databases showed similar elements of AE reports, but no option was provided for a shareable case identifier. The average number of AEs per case was 1.6±1.3 (maximum 37) in the JP and 3.3±3.5 (maximum 62) in the FAERS-JP. Between 5% and 57% of all AEs were signaled by three quantitative methods for etanercept, infliximab, and paroxetine. Signals identified by GPS for the JP and FAERS-JP, as referenced by Japanese labeling, showed higher positive sensitivity than was expected.ConclusionThe FAERS-JP was different from the JADER. Signals derived from both datasets identified different results, but shared certain signals. Discrepancies in type of AEs, drugs reported, and average number of AEs per case were potential contributing factors. This study will help those concerned with pharmacovigilance better understand the use and pitfalls of using spontaneous AE data.

show abstract

Section: Discussionsupporting

confidence: 75%

Effect of database profile variation on drug safety assessment: an analysis of spontaneous adverse event reports of Japanese cases

Nomura¹,

Takahashi²,

Hinomura³

et al. 2015

DDDT

View full text Add to dashboard Cite

show abstract

“…[45] For example, one author might specify a threshold of >2 for the test statistic and another might note their threshold as being ‡2. [45] For example, one author might specify a threshold of >2 for the test statistic and another might note their threshold as being ‡2.…”

Section: Literature Search Resultsmentioning

confidence: 99%

Data Mining in Drug Safety

Deshpande

Gogolak²,

Smith

2010

Pharm Med

View full text Add to dashboard Cite

“…Some have claimed that data mining is ''objective''. Realworld experience indicates that such views may be unrealistic and sensitive to specific software selection and implementation [49]. Our decision rule based on the presence of non-overlapping intervals and at least one SDR for purposes of subset reporting comparisons, is conservative, in that overlapping confidence or credibility intervals up to a point, may occur despite significant differences because the rule is based on linear addition of variance rather than addition in quadrature [50].…”

Section: Discussionmentioning

confidence: 98%