(BSPOT42) Signal Detection using Real World Data - Anticipating Serious Outcomes in Cancer Patients through Thresholds Ranges derived from US-based Claims Data

Background: The detection of adverse events (AEs) is often difficult to achieve during clinical trials (CT) given the restrictive enrollment criteria and small sample sizes. Additionally, signal detection requires reliable thresholds, yet the literature only partially covers reference values for expected rates. This study uses administrative claims data (RWD) to anticipate potential serious adverse events (PSAEs) for NHL (non-Hodgkin’s Lymphoma) and estimate threshold values (expected rates) for different population settings.
Identifying PSAEs early and understanding the patient's characteristics better may facilitate signal monitoring and timely interventions to mitigate associated risks.

Objectives: Building RWD cohorts to emulate two populations: general NHL population, synthetic NHL CT patients.
Estimating the incidence of anticipated serious AEs of interest and deriving corresponding thresholds for signal detection against CTs rates.

Methods:

Design: Two retrospective NHL cohorts (general vs synthetic CT patients) were designed using relevant ICD codes from Jan 2017 to Mar 2023. Applicable exclusion criteria were derived from NHL-relevant CT
Setting: IQVIA PharMetrics Claims database, which includes reviewed medical and pharmacy claims for over 115 million patients across various care settings (hospitals, specialist facilities, pharmacies).
Exposures: includes all eligible therapies currently used for NHL treatment.
Main outcome measures & Statistical analysis: Post NHL diagnosis Incidence proportion of 35 AEs, with 1 to 3 years follow-up times.

Results: The study identified Sepsis and Cytopenia as two PSAEs for further evaluation. The 1y cumulative incidence proportion was elevated in both cohorts (N=118’890 and 92’037), ranging from 4.64% (4.5-4.77, Sepsis) to 13.73% (15.53-15.93, Cytopenia). The incidence of Sepsis increased monotonically over time, gaining 1% and 0.5% in both cohorts during year 2 and year 3, respectively.
The synthetic NHL cohort thresholds were higher than the proportions obtained in a comparative NHL CT under monotherapy (N=798), except for pneumonias related to Covid-19: 0.48% (0.43-0.52) vs 1.13% (0.52-2.13) and P. jirovecii: 0.17% (0.15-0.2) vs. 0.75% (0.28-1.63).

Conclusions: Claims databases can be used to generate different population profiles, which may enable the identification, monitoring, and characterization of PSAEs, along with thresholds usable for signal detection purposes against CTs and additional safety datasets. Limitations include the high missingness of mortality outcomes, event underreporting, and the lack of granularity of ICD to identify the main diagnoses in the dataset. Further identification of confounders and risk factors related to AEs is necessary to refine findings.