Many folks in this situation would use simple thresholds for "enrichment" since the background (null hypothesis) is not well understood. In your data, for example, one might argue that even two samples sharing a position would constitute "enrichment", such as position 300. In the context of NGS data, many fusion event detection algorithms have problems with a high false-positive-rate, so even after filtering for recurrence, there may be further filters that need to be applied (uniqueness of region, mapping quality, coverage of individual fusion events, split-read support, etc.).