Currently the group distinct scan optimization works if all the accumulators are $first or $last. However, if the index spec directions correspond to the $first/$last accumulators, then we should be able to use a distinct scan as well.

For example:

db.c.createIndex({b: 1, c: -1, a: 1}) 

[
{ '$sort': { b: 1, c: -1, a: 1 } },
{ '$group': {
     _id: '$b',
     c_accum: { '$last': '$c' },
     a_accum: { '$first': '$a' }
   }
}
]