From Signal to Confirmed Risk — RWE and the Triangulation Principle

This lesson is where Module 7’s signal detection content and this entire module’s RWE content converge directly, closing a loop this course has been building toward since Module 7 first established that a disproportionality flag is a candidate signal, not a validated finding. When a disproportionality analysis in VigiBase, FAERS, or EudraVigilance flags a drug-event combination above threshold, it provides a genuine hypothesis worth investigating — but it doesn’t provide an incidence rate, it doesn’t control for confounding, and it doesn’t tell anyone whether the association actually holds consistently across different populations. RWE is specifically what supplies each of those missing pieces.

The full workflow connecting a spontaneous-database signal to a regulatory decision runs through six concrete steps, each one drawing on a lesson this module has already built. It starts with the signal itself, already validated through Module 7’s human review process. From there, a precise research question gets defined — which drug, which event coded to which specific outcome definition, which population, which comparison group, what time horizon — with the general principle that the more specific the question, the more useful the eventual answer. Data source selection follows Lesson 15.3’s discipline directly: does the outcome need lab confirmation, is the population size adequate for the expected incidence, is follow-up sufficient. Study design selection draws on Lessons 15.4 and 15.5’s full toolkit — cohort, case-control, SCCS, nested case-control, or sequence symmetry analysis, chosen to fit the specific question, with the protocol registered on ENCEPP or ClinicalTrials.gov before data access begins, exactly the transparency discipline Lesson 15.2 flagged as part of what makes RWE regulatory-grade. Analysis then applies Lesson 15.6’s confounding control methods, calculating a risk ratio or odds ratio with a 95% confidence interval and running sensitivity analyses on the key assumptions. And finally, results get interpreted alongside the original ICSR signal strength and clinical plausibility, feeding into the signal assessment report and, if confirmed, into an RMP safety specification update (Module 9), the relevant PBRER sections (Module 8), and formal regulatory communication.

The principle tying this entire workflow together, worth naming explicitly, is triangulation: regulatory confidence in a safety signal is highest when multiple independent evidence streams converge on the same conclusion. A signal showing strong disproportionality in VigiBase, independently supported by a published case series with a plausible biological mechanism, and then confirmed by a properly designed PASS study showing a genuinely elevated risk after appropriate confounding adjustment — that combination is a signal that has earned regulatory action, because three structurally different sources of potential error (reporting bias, publication bias, and confounding) have all independently pointed to the same conclusion, and it’s statistically implausible for all three to be wrong in exactly the same direction by coincidence.

This directly extends the Bradford Hill consistency criterion Module 7 introduced when it covered causality assessment more broadly: the more independent data sources agree, the stronger the causal case becomes. RWE from a federated network like DARWIN EU or Sentinel adds a genuinely independent line of confirmation specifically because it draws on data with entirely different generation mechanisms and entirely different potential biases than spontaneous reporting — a systematic reporting bias that might inflate a VigiBase signal has no reason to also appear in an EHR-based cohort study’s incidence calculation, which is exactly why agreement between the two is so much more persuasive than either alone.