If I haven't linked to this author, I should have. Shannon Brownlee wrote a good book, and she reiterates some of her arguments in this Washington Post article.
She gets a lot of milage from debunking osteopenia in this article, but as I read her criticism, something struck me as being inconsistant. True, most women with osteopenia will not break a hip, and true, most people with actinic keratoses will not have skin cancer. But my perspective has always been from the side effect world. The statistics that we use to determine a "signal" is similar to the statistics used to predict whether someone will come down with a serious ailment if they have a "pre-disease." So why is one predictive model more valid than another? Well, if you're taking a medication, there's an assumption that its actually doing something positive for you, to make the benefit/risk ration worthwhile. She's right about one thing - there is definitely a growing movement to treat conditions before they even are manifest by using surrogate markers such as low cholesterol or bone density. But don't throw out that surrogate marker philosophy completely. Isn't blood pressure a surrogate marker? Does anybody feel bad just because they have it, or don't they have to wait for the kidney damage first?