What is the Information Component–Bicluster–Bayesian Hierarchical Model?
The Information Component–Bicluster–Bayesian Hierarchical Model (IC-Bic-BHM) is a relatively new signal detection and prioritization framework for pharmacovigilance. Rather than being a single statistical model, it combines three complementary methods:
Information Component (IC): Initial signal detection
The IC is a Bayesian disproportionality measure that compares the observed number of reports for a medication–event pair with the expected number under independence. Bayesian shrinkage stabilizes estimates for rare medication–event combinations.
Biclustering (Bic): Finding patterns
Biclustering identifies subsets of medications that share subsets of adverse events simultaneously. This is valuable because adverse reactions often occur in biologically meaningful groups rather than as isolated associations.
Bayesian Hierarchical Model (BHM): Borrowing strength
Rather than estimating every medication–event association independently, the model allows related associations to “borrow strength” from one another, providing more stable estimates for rare events, less random variation, and quantification of uncertainty.
Why is this useful in pregnancy pharmacovigilance?
IC-Bic-BHM could be particularly useful when screening many medication–congenital malformation combinations. The biclustering step may reveal groups of medications associated with similar patterns of congenital anomalies, while the hierarchical model provides more stable estimates for very rare outcomes– an important advantage in pregnancy pharmacovigilance.
Most pharmacovigilance methods stop after disproportionality analysis. IC-Bic-BHM goes further: IC identifies candidate signals, biclustering discovers related groups of medication–event associations, and Bayesian hierarchical modelling refines the estimates by borrowing information across these related groups.
Study Results
A total of 1,611 medication–outcome pairs with at least three observations were analysed, and 153 signals involving 63 different medications were identified.
Of the 20 highest-ranked signals, 10 medication–congenital anomaly pairs were prioritized for independent evaluation: beclometasone–bilateral renal agenesis, levothyroxine–oesophageal atresia, levothyroxine–tetralogy of Fallot, paracetamol–gastroschisis, salbutamol–gastroschisis, nifedipine–oesophageal atresia, betamethasone–atrial septal defect, lamotrigine–small intestinal atresia/stenosis, ampicillin–oesophageal atresia, and iodine therapy–severe microcephaly.
IC-Bic-BHM is still a signal detection method, and any identified associations require confirmation using pharmacoepidemiologic study designs such as cohort or case-control analyses.
Johnson, H., Dolk, H., Loane, M., Damase‐Michel, C., Given, J., Nordeng, H., Bergman, J. E. H., Carey, I., Den Hond, E., Garne, E., Rouget, F., Bruneau, L., Monier, I., Rissmann, A., O’Mahony, M., Gatt, M., Lutke, R., Sichitiu, J., Ballardini, E., … Morris, J. (2026). Case‐malformed signal detection and prioritisation using EUROmediCAT data for pharmacovigilance in pregnancy. British Journal of Clinical Pharmacology, e70645. https://doi.org/10.1002/bcp.70645