OBJECTIVES: Effect modification in network meta-analysis occurs when differing levels of patient or study-level characteristics impact the relative treatment effect. A common practice is to adjust for effect modification using meta-regression. However, when there are few trials, researchers may be unable to meta-regress, or may need to employ a simplifying assumption of a common modifying effect for all comparators versus the reference treatment. In absence of data on effect modification between treatments (e.g. for indirect comparisons), the validity of effect modification-related assumptions can be assessed by evaluating subgroup effects on treatment response. Our objectives were to quantify the relationship between subgroup effects on treatment response versus treatment effect modification, and to demonstrate the benefit of a modeling approach that incorporates subgroup data from trials, rather than meta-regression, to address effect modification. METHODS: Using a hypothetical network of three treatments, A, B, and C, compared indirectly via B, we calculated the magnitude of effect modification by systematically varying the subgroup effect on treatment responses. Analysis was performed using the odds ratio scale. We used a hierarchical Bayesian model to demonstrate an approach for adjusting for effect modification by incorporating subgroup data, using both hypothetical and real-world examples.  RESULTS: Our analysis revealed that relatively small differences in subgroup effects on treatment response can result in potentially meaningful treatment effect modification in direct comparisons; the magnitude of the effect can become compounded in indirect comparisons. Incorporating subgroup data corrected the modeled values back to the true value in the hypothetical example, and produced updated and potentially less biased estimates in the real-world examples.  CONCLUSIONS: The findings will be valuable to researchers evaluating the need to adjust for effect modification and the validity of imposing simplifying assumptions. The model incorporating subgroup data offers an alternative modeling approach that obviates challenges of meta-regression in settings with few trials.