BACKGROUND: Childhood cancer is an important contributor to childhood mortality in high-income countries. Information on associations between childhood cancer and in-utero exposure is absent or limited for most medications. Signal detection methods identify medications where research should be focused but have not been applied to datasets containing prenatal medication exposures and childhood cancers. RESEARCH DESIGN AND METHODS: The aim of this study was to apply and evaluate four signal detection methods - odds ratios (OR), the information component (IC), sequential probability ratio testing (SPRT), and Bayesian hierarchical models (BHM) - for identification of associations between medications dispensed during pregnancy and subsequent, incident diagnosis of childhood cancer <
10 years, using linked Nordic registry data. Signal detection results were compared to propensity score adjusted odds ratios from generalized linear models. RESULTS: Analysis was performed for 117 medication-cancer pairs with 5 or more observations. The OR had the greatest sensitivity (0.75). The IC had a greater specificity (0.98) than the OR (0.95). CONCLUSIONS: The IC may be the most appropriate method for identifying signals within this type of data. Reported signals should not be considered sufficient evidence of causal association and must be followed-up by tailored investigations that consider confounding by indication.