The last I heard (source) the current thinking was that simple repeat expansion was the source of most de novo mutations, leading to phenotypic changes and symptom severity. This isn't really my field so there might have been a more recent update, but I spend enough time dealing with structural changes to the genome that I read this one when it came out.

Repeat expansion doesn't really behave like a classical mutation, and is more a product of an increased likelihood of specific types of error in DNA replication and repair.