TCP retransmission takes care of resending the dropped TCP segments. This is why the BGP peer remains up despite intermittent packet loss.
If you want to troubleshoot this issue, you have to find the common point of all the observed failures, i.e. how do BGP sessions where the issue is reported differ from those where the issue is not reported: are they all established through the same line card; same interface; if the affected BGP peer have peering with other routers, do they also report invalid md5 digest; etc.
If the common point of all failures is the router on which the error is reported, you can always open a service request with Cisco TAC.