Taking into account house 2016 vote is problematic, as factors such as variations in the quality of the opposing candidate can wildly swing the seat(EG AZ-02 swinging 10 points R in 2016 because Heinz was a terrible candidate with effectively no campaign.) Seats won in waves often swing quite wildly from the house vote in the previous year.
Of course it is problematic due to the factors you mention. But I can't see a better solution than that, since Pennsylvania, Minnesota, and New York all have at least one Trump district D-held seat and one HRC district R-held seat. I, for one, don't think Dave Reichert's seat in WA-08 will be more likely to go Democrat in 2018 than AZ-01.
Also, added two-party House Dem vote performance over two-party HRC vote in 2016, which I missed last time.
You have to accept that a map based on data isn't going to be perfect, and a map that underestimates a few strong incumbents is better then a map that overestimates a lot of mediocre incumbents who had bad challengers.