19 patients completed, according to the article.

And List Sorting, Oral reading, and Flanker only? The first and last are part of global and fluid composites, so those have to be excluded from comparison. That leaves us with 3 improved scores out of 12 tests. So 9 did not improve, or got worse. Figure 3 (of the original article) shows that the changes aren't big. Just "significant". Since the participants were in the early stages of dementia, this seems well within expectations.

So I can't see those numbers as impressive.