Hypothesis-free Screening of Nationwide Administrative Health Data Sources for Unknown Beneficial Drug Effects: An Application of Sequence Symmetry Analysis

Drugs are developed and prescribed for specific, approved indications, yet many act through only partially understood mechanisms and interact with multiple biological pathways. As a result, their effects can extend beyond the original indication. Some additional effects are adverse and systematically captured through pharmacovigilance and spontaneous reporting. Beneficial effects, however, are rarely detected. Unlike adverse reactions, benefits often appear as the absence or prevention of events, which is inherently less observable and seldom reported by patients or clinicians. This asymmetry makes protective effects difficult to recognize and argues for systematic approaches can supplement serendipitous approaches.

This thesis evaluates a systematic screening framework to detect beneficial drug effects in nationwide Danish health registries. Within this framework,large-scale pharmacoepidemiologic data analyzed with self-controlled designscan reveal protective associations that would otherwise go unnoticed. By screening drug-outcome pairs, these approaches complement traditional drug development and pharmacovigilance and may reveal potential candidates for subsequent repurposing.

The thesis comprises three studies. Study I is a proof-of-concept, showing that sequence symmetry analysis (SSA) can identify a beneficial effect unknown at the time of drug approval but later verified in a randomized controlled trial. The same signal is detectable in registry data from periods preceding that trial, indicating that real-world data can reflect early protective signals even before clinical confirmation in trials. Study II scales this principle to a nation wide screening across the entire Danish population where it detects potential beneficial drug-outcome associations across all prescriptions and secondary care diagnoses. We use SSA as the primary method for its efficiency and inherent control for time-invariant confounding and supplement it with a casecrossover design to provide a complementary, unidirectional estimate to help reduce spurious findings and give direction to estimates. Study III validates one of the high-priority associations from Study II, namely acetylcholine esterase inhibitor treatment reduces the risk of delirium. Using a cohort design with high-dimensional propensity scores to strengthen confounder control, the study assessed whether the association reflects a true beneficial effect and whether the Study-2 association replicates under a more elaborate design suitable for evaluating repurposing candidates.