How to avoid bias in systematic reviews of observational studies