Leveraging the Variability of Pharmacovigilance Disproportionality Analyses to Improve Signal Detection Performances

Background: A plethora of methods and models of disproportionality analyses for safety surveillance have been developed to date without consensus nor a gold standard, leading to methodological heterogeneity and substantial variability in results. We hypothesized that this variability is inversely correlated to the robustness of a signal of disproportionate reporting (SDR) and could be used to improve signal detection performances.Methods: We used a validated reference set containing 399 true and false drug-event pairs and performed, with a frequentist and a Bayesian disproportionality method, seven types of analyses (model) for which the results were very unlikely to be related to actual differences in absolute risks of ADR. We calculated sensitivity, specificity and plotted ROC curves for each model. We then evaluated the predictive capacities of all models and assessed the impact of combining such models with the number of positive SDR for a given drug-event pair through binomial regression models.Results: We found considerable variability in disproportionality analysis results, both positive and negative SDR could be generated for 60% of all drug-event pairs depending on the model used whatever their truthfulness. Furthermore, using the number of positive SDR for a given drug-event pair largely improved the signal detection performances of all models.Conclusion: We therefore advocate for the pre-registration of protocols and the presentation of a set of secondary and sensitivity analyses instead of a unique result to avoid selective outcome reporting and because variability in the results may reflect the likelihood of a signal being a true adverse drug reaction.

Statistical learning reflects inferences about unique predictive relations

The mind adeptly registers statistical regularities in experience, often incidentally and implicitly. We used a visual statistical learning paradigm to study what kinds of statistics it spontaneously computes in such conditions. We found that participants’ learning of pairwise predictive relations was best explained by an inferentially sophisticated quantity, deltaP, which reflects whether a high conditional probability between an event pair is unique. We showed that uniqueness can be reduced by either a strong competing predictor or an overall high base rate of the outcome. Both can result from normalization: if predictors of the same effect trade off, a predictor must raise the probability of the effect more than the others to be effective. Adding normalization to the Rescorla-Wagner learning model captures these results. We argue that the uniqueness of a relation is an intrinsically important statistical property that governs learning without incentive or deliberation.

Desogestrel and severe psychiatric disorders, a retrospective analysis of Federal Adverse Event Reporting System and Eudravigilance

Background: Several studies have been published which stated that there is some connection between severe psychiatric disorders and contraceptive drug “desogestrel”. However, nothing in the summary of product characteristics (SmPC) or patient information leaflets of desogestrel about anxiety, more severe anxiety leading to panic attacks, or about risks of severe depression leading to suicidal thoughts or suicide attempts. Objective: To examine the safety and risk association between hormonal contraceptive desogestrel among women with psychiatric disorders using adverse drug reaction database of FDA Adverse Events Reporting System (FAERS) and Eudravigilance (EV). Methods: Individual case safety reports (ICSRs) of only female patients from Jan 1999 to Nov 2019 and Jan 2004 to Nov 2019 were downloaded from FAERS and EV database respectively. Reports of drug desogestrel, dienogest, norgestimate, cyproterone acetate and drospirenone were downloaded. Disproportionality method of data mining was used to calculate the risk association. Results and Discusion: The lower limit of 95 % CI of PRR is -0.28 and 2.02, PRR was 1.08 and 9.18, ROR is 1.09 (95%CI: 0.74, 1.59) and 9.26 (95% CI: 7.21, 11.89), Chi square value is 1.21 and 433.68, and IC-2SD is -0.27 and 2.60 respectively for data obtained from FAERS and EV. Conclusion: From this study, we conclude that there is no new emerging signal for the drug-event pair studied. Further study and continuous monitoring is required in future to know more about this drug-event pair association, as severe psychiatric disorders is not yet mentioned or included in SmPC and patient leaflet of desogestrel.

A mathematical framework to quantify the masking effect associated with the confidence intervals of measures of disproportionality

Background: The lower bound of the 95% confidence interval of measures of disproportionality (Lower95CI) is widely used in signal detection. Masking is a statistical issue by which true signals of disproportionate reporting are hidden by the presence of other medicines. The primary objective of our study is to develop and validate a mathematical framework for assessing the masking effect of Lower95CI. Methods: We have developed our new algorithm based on the masking ratio (MR) developed for the measures of disproportionality. A MR for the Lower95CI (MRCI) is proposed. A simulation study to validate this algorithm was also conducted. Results: We have established the existence of a very close mathematical relation between MR and MRCI. For a given drug–event pair, the same product will be responsible for the highest masking effect with the measure of disproportionality and its Lower95CI. The extent of masking is likely to be very similar across the two methods. An important proportion of identical drug–event associations affected by the presence of an important masking effect is revealed by the unmasking exercise, whether the proportional reporting ratio (PRR) or its confidence interval are used. Conclusion: The detection of the masking effect of Lower95CI can be automated. The real benefits of this unmasking in terms of new true-positive signals (rate of true-positive/false-positive) or time gained by the revealing of signals using this method have not been fully assessed. These benefits should be demonstrated in the context of prospective studies.

Distant Supervision with Transductive Learning for Adverse Drug Reaction Identification from Electronic Medical Records

Information extraction and knowledge discovery regarding adverse drug reaction (ADR) from large-scale clinical texts are very useful and needy processes. Two major difficulties of this task are the lack of domain experts for labeling examples and intractable processing of unstructured clinical texts. Even though most previous works have been conducted on these issues by applying semisupervised learning for the former and a word-based approach for the latter, they face with complexity in an acquisition of initial labeled data and ignorance of structured sequence of natural language. In this study, we propose automatic data labeling by distant supervision where knowledge bases are exploited to assign anentity-levelrelation label for each drug-event pair in texts, and then, we use patterns for characterizing ADR relation. The multiple-instance learning with expectation-maximization method is employed to estimate model parameters. The method applies transductive learning to iteratively reassign a probability of unknown drug-event pair at the training time. By investigating experiments with 50,998 discharge summaries, we evaluate our method by varying large number of parameters, that is, pattern types, pattern-weighting models, and initial and iterative weightings of relations for unlabeled data. Based on evaluations, our proposed method outperforms the word-based feature for NB-EM (iEM), MILR, and TSVM with F1 score of 11.3%, 9.3%, and 6.5% improvement, respectively.

The focal mechanisms and depths of small earthquakes as determined from Rayleigh-wave radiation patterns

abstract Many earthquakes of body-wave magnitude less than about 5.5 are located in key positions for defining tectonic processes at plate boundaries and in the interiors of plates. Due to the sparsity of observed body-wave data from these events, their focal mechanisms and depths cannot be well determined from body-wave data alone. Earthquakes with mb as low as 4.5 radiate surface-wave trains which can be observed at teleseismic distances. In this study, Rayleigh waves observed at WWSSN stations have been inverted to obtain focal mechanisms and depths of two small earthquakes (mb = 4.7 and mb = 5.1) in the Caribbean plate region. An event pair inversion patterned after Weidner and Aki (1973) is contrasted with a single event inversion which makes use of the seismic moment tensor formalism (Mendiguren, 1977; Gilbert, 1970) to linearize the relationship between observed signal and model. The event pair approach requires master events whose focal parameters are well known and which are located near the smaller events. The single event approach, on the other hand, does not require master events, but is limited to amplitude data only. Phase information can be utilized by considering a pair of events located near each other and recovering the difference in phase for the event pair. For the events studied, the single event analysis yields the same solution as the event pair analysis. In general, amplitude is more sensitive to mechanism than phase. However, noise in amplitude increases more rapidly than noise in phase as the size of the event is reduced.