OBJECTIVES: Statin-induced adverse events may prompt additional pharmacotherapy resulting in prescribing cascades. We previously performed untargeted prescribing cascade signal detection using high throughput sequence symmetry analysis (HTSSA), identifying 57 plausible prescribing cascades. However, this endeavor requires access to big data sources and significant computational efforts; false positives are common. We evaluated whether a targeted approach – using Side Effect Resource (SIDER), a public database containing pharmaceutical package inserts data – captures similar findings more efficiently.

METHODS: Anatomical Therapeutic Chemical level 4 codes, representing drug classes, and Medical Dictionary for Regulatory Activities (MeDRA) codes, representing statin-related adverse events were identified from SIDER and linked so that an adverse event for statin was an indication for another medication class. MeDRA codes were then dropped, leaving ‘statin-other medication class’ potential prescribing cascade signals. These signals were compared to empirically-derived signals from prior claims-based HTSSA (gold standard) to calculate sensitivity and specificity for SIDER signal detection.

RESULTS: We detected 432 potential signals using SIDER, compared to 160 signals using HTSSA screening, with a sensitivity of 78.1% (125 of 160 signals) and specificity of 33.8% (128 of 379 signals). To assess predictive ability of SIDER, signals were screened to capture the 57 empirically-identified plausible statin-related prescribing cascades. Out of these, 46 were predicted using SIDER with a sensitivity of 80.7% and specificity of 31.5%. Conversely, SIDER predicted 79 signals that represented therapeutic escalation or clinically implausible prescribing cascades for statins.

CONCLUSIONS: SIDER predicted plausible statin-related prescribing cascades, empirically identified by HTSSA, with high sensitivity; however, it had low specificity and 18% of the signals were implausible prescribing cascades. Our overall findings suggest that medication package inserts are not more efficient to identify potential statin-related prescribing cascades but may be useful combined with expert review to classify drugs and adverse events in distinguishing true from false positive signals.