(348) Multimodal surveillance and the importance of enabling gender-based risk stratification capabilities in real world data

Background: Spontaneous reports data are routinely used for drug and vaccine safety surveillance and have many known limitations, including the non-random nature of how reports are received and content contained within them. Surveillance strategies that rely primarily on spontaneous reports data may therefore not be well-adapted to identify signals that are specific to subgroups (e.g., safety issues that differ by gender or other demographic or clinical factors).

Objectives: We aimed to evaluate the ability to identify a drug-outcome association that is known to differ by gender (i.e., a higher risk of anaphylaxis among antibiotic-exposed adult females compared to males) using a multimodal surveillance approach (spontaneous reports and real-world data (RWD)).

Methods: Among patients > 18 years of age, associations between anaphylaxis and antibiotics were assessed using the Food and Drug Administration’s Adverse Event Reporting System (FAERS), a spontaneous reports database and separately in the TriNetX Global database derived from de-identified electronic health records (EHR) of a federated network with natural language processing (NLP). Disproportionality analyses with Multi-item Gamma-Poisson Shrinker (MGPS) model were used to evaluate associations in FAERS, overall, and by gender. In TriNetX, risk ratio (RR) estimates with 95% confidence intervals (CI) were calculated for females compared to males within a 0-14 day surveillance window post index date. Significant associations and differences by gender were based on the following criteria: for MGPS, if the lower limit of the 90%CI is greater than 2, overall and by gender; for RRs, the 95%CI not including 1.

Results: Although gender is captured in FAERS, the association was identified overall (MGPS=3.9, 90%CI(3.8-4.0)) and, in both genders (females: MGPS=4.3, 90%CI(4.1-4.4); males: MGPS=3.4, 90%CI(3.3-3.6)), and in RWD, the signal was higher among females compared to males (RR = 1.3, 95%CI(1.2-1.3)) within 14 days of treatment initiation. Although comparisons between disproportionality analysis effect estimates are fraught with limitations and can be grossly misleading, we do note that the ratio of females to males is similar to that of RRs from RWD.

Conclusions: This type of analyses supports enabling the rapid examination of safety outcomes using a multimodal surveillance approach. RWD, such as de-identified EHR in TriNetX’s Global Network, with a dashboard that supports rapid data query analyses and risk estimation by gender, may provide additional insights into risk subgroups for drug- or vaccine-outcome pairs.