Biases in pharmacovigilance reporting
The FDA Adverse Event Reporting System (FAERS) database is often taken as the “gold standard’ to which data gathered from social media is compared. This practice, although useful in a practical sense, has some limitations that need to be considered. Disproportionality calculations rely on comparator incidence as a valid and reliable control.
The use of FAERS as the gold standard in social media monitoring
The database is meant to support the post-market surveillance efforts and contains information on adverse events, medication errors and product quality complaints about drugs and biologics. The event terms contained within the database are coded in MedDRA. The FAERS database currently holds 17,530,164 reports in total, of which 9,304,998 are expedited reports, meaning the events are both serious and unexpected (not listed in the current product labeling). This fact brings us to the most important limitation of the FAERS database: the reports contained within are not representative of the actual rate of occurrence. The system’s focus is on qualitatively new signals, not on trends and epidemiological indicators. Moreover, the majority of the reports come from the industry. The number of direct reports submitted by healthcare professionals and consumers is minimal compared to the total volume. Additional limitations include uncertain causal relationship, varying quality of reports and duplication of reports resulting from a myriad of safety data exchange agreements (SDEAs) between market authorization holders that often lead to case resubmission.
The ADE National Action Plan is the work of The Office of Disease Prevention and Health Promotion (ODPHP), on behalf of the Federal Interagency Steering Committee for Adverse Drug Events. The ADE Action plan addresses ADEs that result from high-priority drug classes and have the highest impact in the health systems. The focus ADEs are common, clinically significant, preventable and measurable and occur in high-risk populations. The initial targets of the ADE National Action Plan include bleeding associated with anticoagulants, hypoglycemia with diabetes agents and accidental overdose, oversedation and respiratory depression associated with opioids .
None of the three focus areas mentioned above would appear in FAERS as a dominant problem due to the system’s focus on unexpected events and its inability to detect epidemiological indicators. The situation with opioids is further complicated by the involvement of other reporting systems (CDC and State) and the fact that many overdoses are the result of abuse.
Reporting bias. Even in serious cases, many ADEs are never reported. Under-reporting is especially prevalent in oncology where adverse events relating to treatment with cytotoxic drugs are considered “normal”. Moreover, ADEs are often difficult to distinguish from symptoms of underlying cancer. Intolerable side effects and treatment failure are the key concerns in oncology pharmacotherapy . The reporting medium and context of occurrence have an effect on the overall result. It is safe to assume that certain adverse events are unlikely to be reported on social media, e.g. erectile dysfunction linked to antidepressants.
Notoriety bias may occur with newly introduced drugs or if a specific adverse event is given notable publicity. Sertindole, an atypical neuroleptic was withdrawn in the EU because of fatal reports of prolongation of QT interval. The number of reports was ten times higher than for other atypical neuroleptics marketed in the UK. Studies utilizing data from Prescription-Event Monitoring (PEM) and a large retrospective cohort of patients on atypical neuroleptics enabled comparison of actual death rates on atypical neuroleptics. The results were substantially similar for sertindole and other neuroleptics, only cardiac deaths were about 3 times more likely to be reported than any other cause of death. In October 2001, the Committee on Proprietary Medicinal Products (CPMP) concluded that the difference was caused by notoriety bias rather than increased risk . Similarly, osteonecrosis of the jaw has been linked to bisphosphonates due to selective reporting of the problem and disregarding significant comedication such as chemotherapy . Social media are especially vulnerable to amplifying viral content, so media coverage of a specific drug-related problem is likely to be mirrored in the number of ADEs detected in social media.
Biases in pharmacovigilance reporting have serious consequences as amplification of rare occurrences can lead to drug withdrawal while under-reporting results in the continued use of drugs with unfavorable benefit:risk profile. More representative reporting would produce outputs that correspond with actual rates of occurrence and provide stakeholders with valuable actionable intelligence.
 ODPHP. (2019). ADE Action Plan – Adverse Drug Events – health.gov. Retrieved from https://health.gov/hcq/ade-action-plan.asp
 Baldo, P., Fornasier, G., Ciolfi, L., Sartor, I., & Francescon, S. (2018). Pharmacovigilance in oncology. International Journal Of Clinical Pharmacy, 40(4), 832-841. doi: 10.1007/s11096-018-0706-9
 Moore, N., Hall, G., Sturkenboom, M., Mann, R., Lagnaoui, R., & Begaud, B. (2003). Biases affecting the proportional reporting ratio (PRR) in spontaneous reports pharmacovigilance databases: the example of sertindole. Pharmacoepidemiology And Drug Safety, 12(4), 271-281. doi: 10.1002/pds.848
 de Boissieu, P., Kanagaratnam, L., Abou Taam, M., Roux, M., Dramé, M., & Trenque, T. (2014). Notoriety bias in a database of spontaneous reports: the example of osteonecrosis of the jaw under bisphosphonate therapy in the French national pharmacovigilance database. Pharmacoepidemiology And Drug Safety, 23(9), 989-992. doi: 10.1002/pds.3622