OK, I've found the answer myself.
When the Node.js agent captures a complete call graph, the call graph tab of the transaction snapshot displays the total execution total time of the transaction, including wait times, at the top the call graph.
The execution time shown in the call graph pane is the sum of the times of the individual method calls inside the call graph. This value may be less than the total execution time because it does not include wait times between calls.
As a backend newbie, I have not idea what these wait times between calls might be? Database? Where this call graph has made an async call and node.js runs another thread?
And what do I do to reduce the total transaction time where most of the time is spent "waiting"?