In the context of public procurement, several indicators, called red flags, are used to estimate fraud risk. These red flags are calculated according to certain contract attributes and are therefore dependant on the proper filling of the award notices. In this paper, we propose a general framework based on pattern extraction to detect anomalous graphs. It aims to identify subgraph patterns associated with the presence of red flags, in order to construct a set of new red flag indicators. These patterns can then be used in cases where red flags information is missing.