Thank you @andresrcs for your feedback, and helping me find my mistakes using RPostgreSQL and odbc. The sql chunk headers were missing intentionally, since they were giving me errors -- and I wanted knitr to create an output file.
My problem was I needed to fully qualify the tables as mimiciii.patients:
I fooled myself since I didn't need to fully qualify the tables in SQL chunks with RPostgres since I had included this option with dbConnect:
So, my conclusion now is odbc and RPostgreSQL work fine with dplyr or SQL chunks. RPostgres still has the 1970 date problem, but has Hadley Wickham's recommendation from a tweet in January.
Thanks again for your help.