Faculty Bibliography

Introduction: Disproportionality analysis in pharmacovigilance entails a large number of analytical choices [1], including database background [2]. These would be expected to result in changed signal/noise ratios, sometimes favorably and sometimes unfavorably. Restricting the database to specific subsets more reflective of background diseases may improve noise/noise ratios [3], but may be associated with an unacceptable loss of signals depending on the nature of the restriction. Pharmacovigilance signal detection in oncology is not straightforward for numerous reasons [4] including polydrug regimens, complex patient histories that result in confounding and effect modification (i.e. drug-disease interactions) and unique benefit/risk considerations resulting in higher thresholds for recognizing/ reporting adverse drug events (ADEs). New drugs involving novel mechanisms may present difficult to anticipate/rationalize ADEs. Aim: To explore the effect of an oncology drug restriction (i.e. analyzing the subset of the database consisting only of oncology drugs) on the performance of a data mining analysis using a defined reference set of oncology drug-event pairs. Methods: We used the FAERS database. Positive control (PC) drugmedical concept (MC) pairs were selected from safety information not included in the product's first label but subsequently added as label changes [5]. These MCs were mapped to the Medical Dictionary for Regulatory Activities (MedDRA) preferred terms (PTs) used in FAERS to code adverse events. Negative controls (NC) were MCs with circumscribed PTs not in the corresponding Unites States Package Insert. We calculated shrinkage-adjusted observed-to-expected reporting frequencies for the aforementioned drug-PT pairs. We formulated an adjudication framework to calculate performance at the MC level. Performance metrics [sensitivity, specificity, positive and negative predictive value (PPV, NPV), signal/noise (S/N), F and Matthews correlation coefficient (MCC)] were calculated. Results: The PC reference set consisted of 11 drugs, 487 PTs, 27 MCs, 37 Drug-MC combinations and 638 drug-event combinations (DECs). The NC reference set consisted of 11 drugs, 9 PTs, 5 MCs, 40 Drug-MC combinations and 67 DECs. Most drug-event pairs were not highlighted by either analysis. A small percentage of signals of disproportionate reporting were lost, more noise than signal; little or no gains. Specificity, PPV, NPV, F and MCC improved or showed no visible change while sensitivity declined substantially. Overall S/N improved. Conclusion: Oncology drug restriction substantially improved the S/N ratio, but with significant credible signal loss. Without broader experience and a calculus of costs and utilities of correct versus incorrect classifications in oncology pharmacovigilance such restricted analyses should be optional rather than a default analysis

BACKGROUND: Many illnesses demonstrate seasonal and geographic variations. Pharmacovigilance is unique among public health surveillance systems in terms of the clinical diversity of the events under surveillance. Since many pharmacovigilance signal detection methodologies are geared towards looking for increased frequency of spontaneous adverse drug event (ADE) reporting over variable time frames, seasonality of ADEs may have implications for signal detection. OBJECTIVE: The aim of this study was to investigate whether a set of illnesses that might be expected to display seasonality in general, did so when spontaneously reported as ADEs. METHODS: We performed our analysis with the publically available US FDA Adverse Event Reporting System (FAERS) data. We selected a convenience sample of events possibly triggered by seasonal factors (hypothermia, Raynaud's phenomenon, photosensitivity reaction, heat exhaustion, heat stroke, and sunburn) and events for which previous literature experience suggests seasonality (anencephaly and interstitial lung disease). Our statistical procedures can be explained in terms of a simple physicogeometric setting: the unit circle divided into 6 (semiannual sinusoidal) or 12 (annual sinusoidal) arcs. When reporting frequencies (weights) are more or less evenly distributed across months, the center of mass of the circle would not be significantly displaced from the origin (0, 0). Distinct seasonal patterns will significantly displace the center of mass of the circle. RESULTS: Various patterns of seasonality were identified for some, but not all, events and region-event pairs. USA displayed the most instances of seasonality. Scandinavia did not display seasonality for any events. Seasonality was usually annual sinusoidal. Possible explanations for failure to observe seasonality are briefly considered. CONCLUSIONS: Understanding seasonality of spontaneous ADE reporting may have public health policy and research implications and may mitigate false-positive and missed true-positive pharmacovigilance signals. More systematic study of seasonality of spontaneous ADE reporting, including additional events with more or less biological rationale for seasonality, is a logical extension of this analysis.

INTRODUCTION: There is an interest in methodologies to expeditiously detect credible signals of drug-induced pancreatitis. An example is the reported signal of pancreatitis with rasburicase emerging from a study [the 'index publication' (IP)] combining quantitative signal detection findings from a spontaneous reporting system (SRS) and electronic health records (EHRs). The signal was reportedly supported by a clinical review with a case series manuscript in progress. The reported signal is noteworthy, being initially classified as a false-positive finding for the chosen reference standard, but reclassified as a 'clinically supported' signal. OBJECTIVE: This paper has dual objectives: to revisit the signal of rasburicase and acute pancreatitis and extend the original analysis via reexamination of its findings, in light of more contemporary data; and to motivate discussions on key issues in signal detection and evaluation, including recent findings from a major international pharmacovigilance research initiative. METHODOLOGY: We used the same methodology as the IP, including the same disproportionality analysis software/dataset for calculating observed to expected reporting frequencies (O/Es), Medical Dictionary for Regulatory Activities Preferred Term, and O/E metric/threshold combination defining a signal of disproportionate reporting. Baseline analysis results prompted supplementary analyses using alternative analytical choices. We performed a comprehensive literature search to identify additional published case reports of rasburicase and pancreatitis. RESULTS: We could not replicate positive findings (e.g. a signal or statistic of disproportionate reporting) from the SRS data using the same algorithm, software, dataset and vendor specified in the IP. The reporting association was statistically highlighted in default and supplemental analysis when more sensitive forms of disproportionality analysis were used. Two of three reports in the FAERS database were assessed as likely duplicate reports. We did not identify any additional reports in the FAERS corresponding to the three cases identified in the IP using EHRs. We did not identify additional published reports of pancreatitis associated with rasburicase. DISCUSSION: Our exercise stimulated interesting discussions of key points in signal detection and evaluation, including causality assessment, signal detection algorithm performance, pharmacovigilance terminology, duplicate reporting, mechanisms for communicating signals, the structure of the FAERs database, and recent results from a major international pharmacovigilance research initiative.

INTRODUCTION: Pharmacovigilance includes analysis of large databases of information on drugs and events using algorithms that detect disproportional frequencies of associations. In order to test such algorithms, attempts have been made to provide canonical reference lists of so-called 'positive controls' and 'negative controls'. Reference sets with even modest levels of misclassification may result in under- or overstatement of the performance of algorithms. AIM: We sought to determine the extent to which 'negative control' drug-event pairs in the Observational Medical Outcomes Partnership (OMOP) database are misclassified METHODS: We searched the medical literature for evidence of associations between drugs and events listed by OMOP as negative controls. RESULTS: The criteria used in OMOP to classify positive and negative controls are asymmetric; drug-event associations published only as case series or case reports are classified as positive controls if they are cited in Drug-Induced Diseases by Tisdale and Miller, but as negative controls if case series or case reports exist but are not cited in Tisdale and Miller. Of 233 drug-event pairs classified in the 2013 version of OMOP as negative controls, 21 failed to meet pre-specified OMOP adjudication criteria; in another 19 cases we found case reports, case series, or observational evidence that the drug and event are associated. Overall, OMOP misclassified, or may have misclassified, 40 (17 %) of all 'negative controls.' CONCLUSIONS: Results from studies of the performance of signal-detection algorithms based on the OMOP gold standard should be viewed with circumspection, because imperfect gold standards may lead to under/overstatement of absolute and relative signal detection algorithm performance. Improvements to OMOP would include omitting misclassified drug-event pairs, assigning more specific event labels, and using more extensive sources of information.

INTRODUCTION: Perflutren microbubble/microsphere ultrasound contrast agents have a black-box warning based on case reports of serious cardiopulmonary events. There have been several subsequent observational safety studies. Large spontaneous reporting databases may help detect/refine signals of rare adverse events that elude other data sources/study designs. OBJECTIVE: The objective of this study was to supplement existing knowledge of the reported safety of perflutren using statistical analysis of spontaneous reports. METHODS: We analyzed information from the US Food and Drug Administration Adverse Event Reporting System using a disproportionality analysis. Analysis of overall reporting for perflutren was supplemented by subset (age, indication) analysis. A signal of disproportionate reporting (SDR) was defined as EB05 >2. RESULTS: Overall, 18/380 Preferred Terms and 1/83 Standardized Medical Queries had SDRs. Most were small (EB05 = 2-4). Back pain and flank pain were the largest SDRs followed by events compatible with signs/symptoms of hypersensitivity. The general pattern of SDRs in the subset analysis was consistent with the overall analysis. Almost all events with SDRs were literally or conceptually labeled. Except for chest pain (higher in the age <65 years subgroup) and back pain (higher in the age >/=65 years subgroup), there were no statistically significant differences between age subsets. Except for the Preferred Terms Pruritus and Urticaria and the narrow Standardized Medical Queries Ventricular tachyarrhythmia, Angioedema, Oropharyngeal allergic conditions, and Hypersensitivity (higher in the stress test subgroup), there were no statistically significant reporting differences between indication subsets. There were no SDRs associated with the major cardiovascular events of death, myocardial infarction/ischemia, angina, arrhythmias, or convulsions in any analysis. CONCLUSIONS: Our combined signal detection/evaluation analysis did not identify SDRs of novel adverse events or major cardiovascular events associated with perflutren ultrasound contrast agents. The negative results for major cardiovascular events extend previous signal evaluation exercises supporting the relative cardiovascular safety of these agents.

Background: Prescription drug abuse is a complex public safety problem. Addressing this problem requires new/newly applied methods for the detection and monitoring of abuse. Objectives: To explore the use of multivariate statistical techniques in detecting and monitoring drug abuse potential by evaluating the ability of such methods to appropriately compare/group/classify drugs based on patterns of spontaneously reported abuse-related adverse events (AEs). Methods: Multivariate statistical techniques including correspondence analysis, factor analysis, hierarchical cluster analysis and discriminant analysis were applied to AE data in the US FDA AERS database for scheduled and non-scheduled CNS drugs. AEs associated with drug abuse, including terms in the MedDRA Drug Abuse SMQ and other terms generally considered by those in the field as indicative of abuse, were prespecified. AEs were also clustered into medically meaningful categories to improve some of the analyses. Drugs were grouped based on US DEA scheduling status. Results: Discriminant analysis demonstrated good performance in predicting drug scheduling class based on the pattern of spontaneously reported abuse-related AEs. Correspondence analysis, factor analysis, and hierarchical cluster analysis provided some further insights on the similarity/differences/distances between drug scheduling classes. Conclusions: This preliminary evaluation of the use of multivariate statistical techniques for assessing drug abuse potential suggests that at least some of these techniques may be useful tools in determining the types of AEs to look for with a given drug scheduling classification and may be useful in monitoring postmarketing AE reports for indications of drug abuse

BACKGROUND: Masking is a statistical issue by which signals are hidden by the presence of other medicines in the database. In the absence algorithm, the impact of the masking effect has not been fully investigated. OBJECTIVE: Our study is aimed at assessing the extent and the impact of the masking effect on two large spontaneous reporting databases. STUDY DESIGN: Cross sectional study using a set of terms of importance for public health in two spontaneous reporting databases. SETTING: The analyses were performed on EudraVigilance (EV) and the Pfizer spontaneous reporting database (PfDB). MAIN OUTCOME MEASURE: Using the masking ratio, we have identified and removed the products inducing the highest masking effect. RESULTS: Studying a total of almost 50 000 drug-event combinations masking had an impact on approximately 60% of drug-event combinations were masked by another product with a masking ratio >1 in EV and 84% in PfDB. The prevalence of important masking was quite rare (0.003% of the DECs) and mainly affected events rarely reported in EV. The products involved in the highest masking effects are products known to induce the reaction. The removal of the masking effect of the highest masking product has revealed 974 signals of disproportionate reporting in EV including true signals. The study shows that the original ranking provided by the quantitative methods included in our study is marginally affected by the removal of the masking product. CONCLUSION: Our study suggests that significant masking is rare in large spontaneous databases and mostly affects events rarely reported in EV

BACKGROUND: Masking is a statistical issue by which true signals of disproportionate reporting are hidden by the presence of other products in the database. Masking is currently not perfectly understood. There is no algorithm to identify the potential masking drugs to remove them for subsequent analyses of disproportionality. OBJECTIVE: The primary objective of our study is to develop a mathematical framework for assessing the extent and impact of the masking effect of measures of disproportionality. METHOD: We have developed a masking ratio that quantifies the masking effect of a given product. We have conducted a simulation study to validate our algorithm. RESULTS: The masking ratio is a measure of the strength of the masking effect whether the analysis is performed at the report or event level, and the manner in which reports are allocated to cells in the contingency table significantly impact the masking mechanisms. The reports containing both the product of interest and the masking product need to be handled appropriately. The proposed algorithm can use simplified masking provided that underlying assumptions (in particular the size of the database) are verified. For any event, the strongest masking effect is associated with the drug with the highest number of records (reports excluding the product of interest). CONCLUSION: Our study provides significant insights with practical implications for real-world pharmacovigilance that are supported by both real and simulated data. The public health impact of masking is still unknown