Citing prior literature is a complex process.

Sometimes, most often in new entrants to a field or in the core topic of the research, we either know the whole literature or research it exhaustively, going up and down the citation chains, covering all the reviews, etc.

However, most researchers only read _every_ (often annoyingly flawed) study in our core field. Otherwise, we just hit the high points, often only the abstracts and conclusions. That's how learning happens at an acceptable pace.

Then, we write something and remember a tangential fact - so we go find the reference we vaguely remember that said it. Grand statements like "vitamins may be used to prevent acute infections like rhinitis, and chronic conditions such as cancer and heart disease" might be followed by a string of such references. A similar result would occur by simply finding a paper with another grand statement, likely a review, and tracking down its canonical references. It would be too much work to look through every paper that cited these articles and find the one that contradicted, or cast in a different light, the whole (or part of the) conclusions. Then we'd also have to decide which was right (given that 1/20 papers can be completely wrong if they were significant at .95, double blind or not).

All of this presumes that the article and its contradiction are even mutually self aware - some problems span fields in such a way that the same people can't read both papers with any degree of comprehension, and until someone interdisciplinary comes along, the problem isn't even noticed.