into account rapid reporting increase, seriousness, time on the market for the involved drug and if the ADR was of special interest. It was mentioned that these criteria proved to be successful in selecting true signals, however the exact testing method was not provided.
Another decision support framework [11] took into account the strength of evidence and the potential public health impact. The included components of strength of evidence were: disproportionality score, quality of reports and biological plausibility. The components of public health score were: number of reports, seriousness and reporting rate. Using an empirical cut-off point for both scales, four priority categories for a signal were obtained, each having a different course of regulatory action. All subjective variables were quantified using graded scales. This tool was validated [25] and is currently used for prioritization in a regulatory setting at Medicines and Healthcare Products Regulatory Agency (MHRA) in the UK.
Seabroke et al. [15] updated the framework mentioned above by adding two categories: agency regulatory obligations and public perceptions. This updated tool was designed to be used in a later stage of signal management. The tool was piloted and validated against expert group opinion and is also routinely used in the same organization.
A multi criteria decision analysis (MCDA) weighted framework was developed by Levitan et al. [16] based on 11 criteria related to novelty of event (10% weight), strength of evidence (40% weight) and medical impact (50% weight). In addition, two extra criteria were used for pre-selection: evaluation of novelty of the event and of confounding by indication. Each criterion had an assigned weight and there were graded scales for each attribute. The model was tested against expert group judgment and the agreement was found to be moderate.
FDA drafted a prioritization guidance aimed to classify post-marketing drug safety issues [18]. This guidance recommends to estimate the hazard posed by a safety issue, based on three criteria: (1) the seriousness of the issue; (2) the estimated size of the population exposed to the drug; and (3) the suspected frequency of harm for exposed patients. The combination of factors 2 and 3 provides an estimate of population risk, while the combination of factors 1 and 3 provides an estimate of individual risk.
The vigiRank predictive model developed by Caster et al. [10] is an algorithm for emerging safety signals that accounts mainly for reports quality and content. The following criteria were included: disproportionate reporting, number of informative and recent reports, number of reports with a narrative and multinational reporting. The advantages of this method are that it is automated in VigiBase® and was tested in a comprehensive manner, by means of multiple logistic regression, and against a reasonably large reference set. Public health impact was not considered within this algorithm.
519883-L-sub01-bw-Pacurariu 519883-L-sub01-bw-Pacurariu 519883-L-sub01-bw-Pacurariu 519883-L-sub01-bw-Pacurariu Processed on: 5-6-2018 Processed on: 5-6-2018 Processed on: 5-6-2018
Processed on: 5-6-2018 PDF page: 98PDF page: 98PDF page: 98PDF page: 98
Chapter 5
98
Last but not least, Strandell et al. [21] propose two prioritization frameworks specific to drug-drug interactions. This was the first application of predictive regression models for first-pass screening of large collections of spontaneous reports, when looking for drug interactions. Due to its specificity for drug interaction, this was not described further.
DISCUSSION
Prioritization decisions are typically complex and resource intensive, as they blend the numerical information with scientific knowledge and judgment [8]. In this paper, we conducted a review of signal prioritization criteria and associated decision support frameworks that were built upon those, in order to increase awareness and facilitate the process.
A total of 48 criteria were identified in the literature for signal prioritization, and they were categorized according to the following key concepts: novelty, strength of evidence, public health impact and general public and media attention.
One important distinction should be made early on regarding the criteria found in the studies: some of them were used for prioritization based on their predictive value, while others were used independent of this property. Usually, criteria related to strength of evidence would fall in the first category, whereas criteria related to public health impact would be included in the second one. The novelty concept is intrinsic to signal definition and six studies mentioned it, albeit this might be an underestimate of its actual use in signal prioritisation process. Novelty could related to the drug event association or just to the drug. Although ‘Weber effect’ [26] (i.e., AE reporting peaks at the end of the second year after approval) was not reproduced, [27,28] two studies [9,17] showed that new drugs are more likely to have more safety signals.
Another key concept, strength of evidence, was at the core of prioritization algorithms, being considered by 10 out of 11 studies and by all six decision support frameworks. This was expected, since it is logical to focus the resources, even from a very early stage, on those signals which have a high probability to be true. From the strength of evidence related criteria, the multi-national reporting was repeatedly demonstrated to have predictive value [9,10]. In addition, the quality of reports predicted a true signal and, therefore, it is worthwhile to consider this when prioritizing. The fact that the report quality/completeness is associated with true signals might seem counter- intuitive at first, since an increase in the amount of information should not necessarily mean an increase in likelihood of a causal association. A potential explanation might be that the reporter is more likely to provide more complete information about a report once he genuinely believes that the drug is the real culprit. An alternative possibility is that only complete reports can provide the necessary information for a causality assessment that can give rise to a true signal.
519883-L-sub01-bw-Pacurariu 519883-L-sub01-bw-Pacurariu 519883-L-sub01-bw-Pacurariu 519883-L-sub01-bw-Pacurariu Processed on: 5-6-2018 Processed on: 5-6-2018 Processed on: 5-6-2018
Processed on: 5-6-2018 PDF page: 99PDF page: 99PDF page: 99PDF page: 99