It depends a lot on what remote resources you expect or can
insist upon.
If you want minimal wiring and no opportunity for supply
integrity problems, you might make the signals more like
open/short to the return/ground; this would be active low
open drain / open collector.
If you want to push a positive voltage or current then
you start to care about things like whether this will
power the remote unit via input protection networks,
perhaps into a non-valid or marginal state. But if the
load is dumb and has no power, then you would be
looking to supply energy, not just a little-as-possible
bit of state information.
And speed matters; schemes like open collector do
depend on the pullup resistance for one of the transitions,
long wires can be capacitive and slow way down when
you get past inches of run and use pullup impedances
consistent with logic drive strengths. If you're sub-MHz,
OK; above this, at that length, you want to consider
transmission line effects and styles (the phone company
does, for audio at neighborhood scale; power company
does, at 60Hz at county/state/continent scale).
In my experience this choice often comes down to the
available remote and head-end resources, fault modes
and such. You mention signal reliability but nothing about
the "threat set" (against which it would be deemed reliable).
Give that some thought. You might like to look at app
notes for industrial long wire interfaces, even if you don't
intend to adopt any of them.