Potential importance of residual confounding in systematic reviews of observational studies: Answer to Mathur and VanderTweele.