Adverse drug reactions (ADRs) are one of the leading causes of mortality and thus should be detected as early as possible to reduce health risks of patients. Data mining approach using large-scale medical record might be a useful method for early detection of ADRs. There have been many researches analyzing medical record to detect ADRs; however most of the researches focused only on a narrow range of late-onset ADRs, limiting its usefulness. In this study, we developed an early and simple detection scheme of ADRs based on association rule mining (ARM) of JMDC insurance claims data. The assessment of its ability in detecting a broader range of ADRs was performed using a global gold standard of ADRs consisting of 92 positive control ADR-drug pairs and 88 negative control pairs rationally selected from the statistical analyses of large-scale ADR self-reports. In this assessment of ARM, the areas under the receiver operating characteristic curve and the precision-recall curve were 0.80 and 0.83, respectively. Moreover, when capability of ARM to detect ADRs with varying short periods of accumulating data was tested, our method detected much more positive controls (65 pairs) than conventional sequence symmetry analysis (9 pairs) frequently-used for ADR detection. These results suggest that ARM is effective in the early and versatile detection of ADRs, complementarily to the existing pharmacovigilance strategies.