gms | German Medical Science

Background/research question: Network meta-analysis (NMA) enables the simultaneous comparison of multiple interventions, while individual participant data meta-analysis (IPD) allows to unify definitions and analyses across studies as well as to account for participant-level covariates that might modify the effect of an intervention. In a BMBF-funded Cochrane review, we combined both approaches to investigate the effects of oral nutritional interventions on patient-relevant outcomes in hospitalized older people at nutritional risk [1]. The aim of this work is to outline our methodological approach and to share gained experiences.

Methods: Before initiating the review, we established the necessary infrastructure for IPD-analysis (e.g. data protection). At the full text screening stage, we requested IPD from authors either for the whole sample or a subsample, in case the study met our inclusion criteria only partially. If no response was received or IPD was unavailable, we used published aggregated data and excluded studies that only partially fulfilled our inclusion criteria. In cases where authors were willing to share IPD, a data-use agreement was signed and authors received guidance on preparing and transferring data. Following an initial quality check, data were harmonized for analysis using a data dictionary developed by the review team. Evidence synthesis was conducted in two stages: (1) analyzing IPD within each study and (2) pooling these results with published aggregated data as NMA.

Results: In total, 27 of our 36 IPD-requests were answered. However, IPD was often no longer available (n=8) or data sharing was not possible for other reasons (n=5). In two cases, contact was discontinued after initial positive response regarding data sharing. Most of these studies were subsequently excluded because the samples only partially met our eligibility criteria. Time from initial contact to receipt of IPD ranged from a few weeks to 7 months. Finally, we included 21 trials, 12 with IPD and 9 using published aggregated data. Due to small sample sizes – particularly when only subsamples were included – we frequently downgraded certainty of results for imprecision.

Conclusion: Taking into account sufficient time resources, we were able to successfully implement the IPD-NMA approach in our review. The approach allowed the inclusion of additional studies by incorporating subsamples. However, using subsamples from smaller trials may have led to more heterogeneous comparison groups and may have affected the precision of results.

Competing interests: This work is funded by the German Federal Ministry of Education and Research (BMBF) (grant number: 01KG2102). The authors have no relevant financial or non-financial interest to disclose.