Interestingly, so far as demos go, most things haven't changed all that much since DS was on because the most desired demographic is still 18-49 year old women - and women viewers are still far more desired than men because many advertisers still think that women make the most decisions when it comes to what products to buy (and that probably is still true to a major extent).
Regardless of sex, advertisers don't really desire 50+ viewers because the thinking is they're too set in their ways to try new things. Though while that may be true of some, it's certainly not true of all. Mostly everyone I know is 50+ and they're more than willing to try new things. It's certainly true of me. But then, no one can say advertisers' thinking always syncs with reality. However, it still remains true that any show for which 50+ viewers comprise the largest portion of its audience is very much in danger of being canceled. It happens time and time again no matter how many millions of 50+ viewers they are for the show.
On the other hand, the thinking when it comes to teens has changed drastically since DS was on. Advertisers now realize that if they can lure teens to buy their product, they may just have a customer for life - or at least the life of a product.
One thing, though, is that Nielsen ratings represent only a percentage of viewership. The numbers that are featured in the Syndication Promo Reel are the raw numbers that Nielsen compiled and they represent just a fraction of the audience watching. Nielsen would then take that raw data and, using a series of equations that they developed, would then extrapolate how many viewers the raw data indicated were actually watching. The 20 million figure for viewership of DS was an extrapolated number from the raw data. Nielsen worked very hard to make those extrapolated numbers as accurate as possible (though they've had tons of criticism through the years). But the point I'm making is that someone can't actually take the raw data numbers and use them with any extrapolated numbers.
One has to use raw data with raw data and extrapolated numbers with extrapolated numbers when making demographic comparisons.