Background: Meta-analyses using individual patient data (IPD) allow for a detailed evaluation of differences in treatment effects among subgroups. If, however, patients from different trials have different baseline risks, an observed difference in treatment effect may be caused by another variable than the one indicating the subgroup. Therefore, in IPD meta-analysis observed subgroup effects may be confounded. The so-called two-stage approach (i.e. subgroups are analyzed within trials before pooling the results over trials) has been recommended to control for such confounding. Subgroup effects in a single trial, however, can also be confounded. Methods: To illustrate confounding of subgroup effects in an individual trial a numerical example on the effects of aspirin use on the risk for stroke is presented. The effects of aspirin are modified by age, but not by sex. Age and sex are associated (OR=2.25). Effects of aspirin use are first assessed within sex subgroups and subsequently within sex subgroups that are stratified by age. Results: The effects of aspirin on the risk for stroke differed among sex subgroups: risk ratios (RR) 1.00 and 0.83, respectively. After additional stratification by age, age appeared the effect modifier rather than sex: risk ratios were equal for young males and young females (RR 1.33) as well as for older males and older females (RR 0.5). Conclusion: Since subgroup effects can be confounded within an individual trial a two-stage approach will not solve this problem of confounding within trials that are included in meta-analyses. Consequently, a one-stage approach (i.e. pooling over trials before subgroups are analyzed) to control for confounding is considered at least as appropriate as a two-stage approach.