Data Sources for Signal Management

To prove that the product has no safety risks To assess potential residual bias or systematic error in the analytical design To increase the number of exposed patients To replace sensitivity analyses

Accept the data because claims databases are routinely used by regulators Qualify the vendor and document data provenance, refresh characteristics, mapping, validation and limitations Remove all records with missing variables without documenting the effect Use the vendor's analytic code without version control to preserve consistency

Whether an event is reported in routine care across all demographic groups The frequency, timing, dose relationship or laboratory pattern of an event under controlled data collection Whether a product has a class effect in medicines never studied in the trial The population incidence of an extremely rare event with certainty

A spontaneous-report disproportionality score alone A comparative epidemiological study using defined exposure and outcome criteria and confounding control A single mechanistic animal experiment A literature search limited to positive case reports

Trial data are never suitable for safety analysis Clinical trials generally exclude all patients receiving active treatment Trial populations, follow-up duration and sample sizes may be insufficient to represent rare events or routine-care populations Randomization guarantees that every adverse event is causal

Reporting timeliness and completeness figures without reviewing them Tracking trends, investigating out-of-specification performance and documenting CAPA or root-cause actions Using only the highest-performing month in management reports Replacing scientific review with a single numerical target

A large sample size without any prespecified definitions A clear design rationale, validated exposure and outcome definitions, sensitivity analyses and transparent analytic methods Exclusion of all patients with incomplete records without assessing the resulting bias Use of a statistically significant result as the sole decision criterion

To ensure that every initial data entry remains unchangeable, even when corrected To permit reconstruction of what was done, by whom, using which version and rationale, supporting reproducibility and oversight To demonstrate that all signals must result in regulatory action To avoid the need for staff training and quality control

As direct proof of human causality As contextual evidence that can support or challenge biological plausibility and guide human hypothesis testing As a substitute for human safety data whenever an animal finding is statistically significant Only after a labeling change has been implemented

Change the product label immediately based on the statistical result Verify data integrity, assess duplicates, retrieve complete case information and perform preliminary clinical review Dismiss the alert because disproportionality does not provide an incidence estimate Commission a large epidemiological study before reviewing the underlying cases