Application of biclustering algorithm in adverse drug reaction monitoring system of China.

PURPOSE: Signal evaluation is considered to be a tedious process owing to the large number of disproportional signals detected. This study aimed to apply a biclustering algorithm in the spontaneous reporting system of China and to obtain the optimal parameters. The biclustering algorithm is expected to improve the efficiency of signal evaluation by identifying similar signal groups.

METHODS: Information component (IC) was the method used for disproportionality analysis. By using IC thresholds of various strengths (0.05-4.00), the original quantitative data matrix was transformed into 80 different binary data matrices, where each cell contained either a 1 or 0. The biclustering results were obtained using a total of 720 Bimax algorithm parameters (minimal number of columns and rows was 3, 4, or 5). Next, the optimal parameters were determined through the comprehensive evaluation of the rank sum ration. Finally, we examined the biclustering results under the optimal parameters and evaluated the effect of biclustering analysis on adverse drug reaction (ADR) data in China.

RESULTS: The optimal strength of the IC threshold was 0.80, and the minimum number of rows and columns was 3. After taxonomic evaluation, we also found that 1836 biclusters (42.8%) contained similar drugs or similar ADRs, which accounted for 72.3% of signals unevaluated.

CONCLUSIONS: Applying biclustering analysis in spontaneous reporting system could provide support in confirming unrecognized ADRs, identifying rare ADRs, and screening drug-ADR pairs, which need more attention. Biclustering algorithm could improve the efficiency of signal detection and evaluation in China.