This is one of those fake stats, where

if among people not taking it you would see 4 out of 100 get dementia within 5 years of when you started measuring,

if they were taking it you'd instead see 3 out of 100 get dementia within 5 years,

and the number is either a projection ("predicted risk") because they didn't actually study it for 5 years,

or the number is a p-hack because they checked every possible thing they could with the 5-year data they had and this was the only interesting thing that they found ("post hoc analysis"),

or both (the projection of a p-hack), which I suspect because they also tell you about the 20-year risk,

and it is a study sponsored by the manufacturers of the drug,

being suggested to cure a disease diagnosed by checklist (rather than any physical test.)

Will that keep it out of the headlines? No, it was written to become a "free" ad that will be published by the people they send enormous amounts of money to, but by their "news" departments.

The best thing we can do for the progress of medicine in the US is to make direct-to-consumer advertising of prescription drugs illegal again. There would be no use for this if the media hadn't been bribed to become cheerleaders for any old shit. It's illegal everywhere in the world other than the US and New Zealand.