I wonder if it would make sense to base the comparison on road network density in areas that are above some threshold for population density?

i.e. Try and measure how many roads there are in areas where most people actually live?

Only if you're trying to intentionally cherry pick the data. Population density inherently affects road networks, and that will be reflected in the data.