Background: Spontaneous reporting of suspected adverse drug reactions has been the traditional signaling system in pharmacovigilance. The increased availability of longitudinal Electronic Healthcare Record (EHR) databases offers new opportunities for drug safety signal detection using observational data from daily practice. Objectives: To describe the preliminary results of signal detection using different methods applied to longitudinal EHR data from the EU-ADR Project. Methods: The evaluation dataset consisted of clinical and drug use information from 9,202,393 individuals with 47,881,511 person-years (PYs) of follow-up representing six databases of the EU-ADR network for the period 1995–2010. Disproportionality-based methods (Reporting Odds Ratio (ROR), Proportional Reporting Ratio (PRR), Gamma Poisson Shrinker (GPS), Bayesian Confidence Propagation Neural Network (BCPNN)) and exposure-based methods (Cohort and Longitudinal GPS) at varying thresholds were compared using a verification set of true positive (i.e., known ‘signals’) and true negative drug-event associations for six events considered important in pharmacovigilance. Only the first occurrence of an event in a patient was counted. Data processing was performed in distributed data architecture using custom-built software Jerboa# and statistical analyses were performed in R. Results: Disproportionality-based analyses gave an area under the receiver operating characteristic curve (AUC) of 0.68–0.71, while exposure-based methods gave an AUC of 0.66–0.73. There was variability in the performance of the methods with respect to individual events, suggesting that some methods may be more appropriate for particular event types. Conclusions: Disproportionality-based and exposure-based signal detection methods were broadly comparable when applied to longitudinal data. There appears to be a need to tailor methods to specific adverse event types to improvement method performance.
Coloma, P., Trifiro, G., Gini, R., Herings, R., Mazzaglia, G., Giaquinto, C., et al. (2011). Comparison of Methods for Drug Safety Signal Detection Using Electronic Healthcare Record (EHR) Databases: The Added Value of Longitudinal, Time-Stamped Patient Information. Intervento presentato a: International Conference on Pharmacoepidemiology & Therapeutic Risk Management, Chicago, Illinois, USA.
Comparison of Methods for Drug Safety Signal Detection Using Electronic Healthcare Record (EHR) Databases: The Added Value of Longitudinal, Time-Stamped Patient Information
Mazzaglia, G;
2011
Abstract
Background: Spontaneous reporting of suspected adverse drug reactions has been the traditional signaling system in pharmacovigilance. The increased availability of longitudinal Electronic Healthcare Record (EHR) databases offers new opportunities for drug safety signal detection using observational data from daily practice. Objectives: To describe the preliminary results of signal detection using different methods applied to longitudinal EHR data from the EU-ADR Project. Methods: The evaluation dataset consisted of clinical and drug use information from 9,202,393 individuals with 47,881,511 person-years (PYs) of follow-up representing six databases of the EU-ADR network for the period 1995–2010. Disproportionality-based methods (Reporting Odds Ratio (ROR), Proportional Reporting Ratio (PRR), Gamma Poisson Shrinker (GPS), Bayesian Confidence Propagation Neural Network (BCPNN)) and exposure-based methods (Cohort and Longitudinal GPS) at varying thresholds were compared using a verification set of true positive (i.e., known ‘signals’) and true negative drug-event associations for six events considered important in pharmacovigilance. Only the first occurrence of an event in a patient was counted. Data processing was performed in distributed data architecture using custom-built software Jerboa# and statistical analyses were performed in R. Results: Disproportionality-based analyses gave an area under the receiver operating characteristic curve (AUC) of 0.68–0.71, while exposure-based methods gave an AUC of 0.66–0.73. There was variability in the performance of the methods with respect to individual events, suggesting that some methods may be more appropriate for particular event types. Conclusions: Disproportionality-based and exposure-based signal detection methods were broadly comparable when applied to longitudinal data. There appears to be a need to tailor methods to specific adverse event types to improvement method performance.File | Dimensione | Formato | |
---|---|---|---|
ICPE 2011.pdf
Solo gestori archivio
Dimensione
3.1 MB
Formato
Adobe PDF
|
3.1 MB | Adobe PDF | Visualizza/Apri Richiedi una copia |
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.