Found another crappy test thanks to random order. In some cases, the
patient needed for the edit test was being deleted before the test could
execute. So selecting a specific patient for the delete test should make
things more robust.