Abstract:Introduction: Several quantitative methods have been established, in pharmacovigilance, to detect signals of disproportionate reporting (SDRs) from databases containing reports of adverse drug reactions (ADRs). The signal detection algorithms (SDAs) and the source of the reporting per product vary, but it is unclear whether any algorithm can provide satisfactory performance using data with such large variance factors. Objective: Determine the appropriate SDA for Biogen's internal Global Safety Database (GSD) given the characteristics of the database including frequencies of events, data skewness, outliers, and missing information. Compare performance of standard approaches (EBGM, EB05, PRR, and ROR), well accepted by industry, to a Biogen-developed Machine Learning (ML) Regression Decision Tree (RDT) model, across several Biogen products, to determine a champion SDA. Methods: All data associated with seven marketed Biogen products were chosen and a historical subset of reported ADRs were considered. Six SDAs (five common industry disproportionality methods) and RDT were evaluated. The SDRs were calculated on training and test data composed of quarterly reporting intervals from 2004-2019. The performance measures used were sensitivity, precision, time to detect new events, and frequency of detected cases for each algorithm for each product. Outcomes in the test data are known a priori and easily compared to predicted outcomes. Validation was performed via rates of misclassification. This work solely represents Biogen's internal information, intentionally chosen to serve the performance review of its signal detection systems, and results will not necessarily be generalizable to other external sources. Results: Several algorithms performed differently among products, but no one method dominated any other. Performance was dependent on the thresholds used to define a signal according to different criteria. However, those different statistics subtly influenced the achievable performance. The relative performance of RDT and Medicines and Healthcare products Regulatory Agency (MHRA) algorithms were superior and paired across products. A reduction in precision for all methods spanning the products was present. Hence, companies evaluating signal detection approaches, search for innovative methods to minimize this effect. Conclusions: In designing signal detection systems, careful consideration should be given to the criteria that are used to define SDRs. The choice of disproportionality statistics does not affect the achievable range of signal detection performance. These choices should consider mainly ease of implementation and interpretation. The implementation of a method is specific to its accuracy. The RDT attempted to take advantage of known methods and compare results on a per-product basis. Many factors influencing ADRs may improve RDT in future efforts. In this experiment, RDT demonstrated superiority in terms of quickest time to detect and capturing of the highest number of ADRs. Next steps include expansion of data for products representing other indications and testing models in external databases to investigate generalizability of estimates when comparing SDAs.

Evaluation of Statistical Methods for Safety Signal Detection: a Simulation Study

An evaluation of statistical approaches to postmarketing surveillance

Drug Safety Signal Detection in a Regional Healthcare Database Using the Tree-Based Scan Statistic and Comparison to 3 Other Mining Methods

Quantitative Safety Monitoring in Clinical Trials: Application of Multiple Statistical Methodologies for Infrequent Events

Methods for drug safety signal detection using routinely collected observational electronic health care data: a systematic review

Simultaneous Confidence Intervals for Signal Detection and Ascertaining Precision of Adverse Event Rates in Clinical Trials

Statistical Techniques for Signal Generation

Statistical Signal Detection Algorithm in Safety Data: A Proprietary Method Compared to Industry Standard Methods

Detection algorithms and attentive points of safety signal using spontaneous reporting systems as a clinical data source

A Bayesian Group Sequential Approach to Safety Signal Detection.

Comparison of Statistical Signal Detection Methods in Adverse Events Following Immunization — China, 2011–2015

Robust safety monitoring and signal detection using alternatives to the standard poisson distribution

A simulation-based comparison of drug-drug interaction signal detection methods

Comparison of Data Mining Methods for the Signal Detection of Adverse Drug Events with a Hierarchical Structure in Postmarketing Surveillance

A Bayesian method for safety signal detection in ongoing blinded randomised controlled trials

A Computational Method to Quantitatively Measure Pediatric Drug Safety Using Electronic Medical Records

Signal Detection Methods Based on False Discovery Rate of Adverse Drug Reaction

Present Status and Future Prospect of Signal Detection Methods for Adverse Drug Reaction

Tree‐temporal scan statistics for safety signal detection in vaccine clinical trials

Combining a Pharmacological Network Model with a Bayesian Signal Detection Algorithm to Improve the Detection of Adverse Drug Events

The Application of Tree-based Scan Statistic in Drug Safety Surveillance