Article type
Year
Abstract
Background: The synthesis of individual patient data (IPD) from randomized clinical trials (RCTs) can strengthen evidence used for decision-making. Network meta-analyses (NMA) modelling IPD usually include non-sponsored or publicly sponsored RCTs. Evidence suggests that IPD sharing may depend on study characteristics, such as funding type, RCT size, RCT risk of bias, and treatment effect. However, retrieval bias in IPD-NMA of sponsored RCTs has not been assessed before.
Objectives: To explore retrieval bias in IPD-NMAs of sponsored RCTs and address challenges and barriers.
Methods: We contacted authors and sponsors of RCTs eligible for 2 IPD-NMAs to obtain IPD. If a study had multiple sponsors, we contacted all of them. To facilitate IPD retrieval, we contacted data sharing platforms. All IPD were checked for consistency with results from published RCTs. We explored whether IPD studies suggested different findings with those of studies not sharing IPD and outlined the IPD availability from sponsors. We noted all barriers and resource requirements associated with the IPD acquisition during the author and sponsor contact processes.
Results: We included 137 RCTs and received IPD for 29 (21%) RCTs (1058 total waiting days). None of the 137 authors shared their IPD. Instead, 17 sponsors for 107 studies were contacted and 7 (41%) sponsors shared their data through proprietary sponsor-specific platforms. The 7 sponsors held data for 94 RCTs and we obtained data from 31% (29/94) of these RCTs. Of the 29 RCTs, we were able to include 23 RCTs in our NMA due to incompleteness of provided data. For example, a study included only IPD for the placebo arm and thus was excluded from the NMA. A big challenge in the IPD was the high dropout rate (up to 72%) from the RCTs, for which many original authors applied inappropriate imputation methods. Hence, our findings differed from published RCT results. We also encountered outcome reporting bias; specifically, some outcomes were missing from the publications but were available as IPD. The use of different platforms restricted us from combining IPD in a single NMA model and a one-stage analysis was impossible. Time restriction to remote-access platforms and frequent changes of these platforms added another challenge to the analysis, given that IPD from different RCTs were available at different time points.
Conclusions: Retrieval bias can severely impact NMA findings and decision-making. Our study highlighted challenges encountered during an IPD-NMA of sponsored RCTs.
Patient or healthcare consumer involvement: Personalized medicine is required to optimize health care. Well-conducted meta-analyses of IPD are considered the ‘gold-standard’ and influence patient care since patient-level data can be provided to facilitate tailored decision making. However, results from meta-analyses of IPD are likely subject to retrieval bias and awareness of these limitations and their potential impact on findings is required.
Objectives: To explore retrieval bias in IPD-NMAs of sponsored RCTs and address challenges and barriers.
Methods: We contacted authors and sponsors of RCTs eligible for 2 IPD-NMAs to obtain IPD. If a study had multiple sponsors, we contacted all of them. To facilitate IPD retrieval, we contacted data sharing platforms. All IPD were checked for consistency with results from published RCTs. We explored whether IPD studies suggested different findings with those of studies not sharing IPD and outlined the IPD availability from sponsors. We noted all barriers and resource requirements associated with the IPD acquisition during the author and sponsor contact processes.
Results: We included 137 RCTs and received IPD for 29 (21%) RCTs (1058 total waiting days). None of the 137 authors shared their IPD. Instead, 17 sponsors for 107 studies were contacted and 7 (41%) sponsors shared their data through proprietary sponsor-specific platforms. The 7 sponsors held data for 94 RCTs and we obtained data from 31% (29/94) of these RCTs. Of the 29 RCTs, we were able to include 23 RCTs in our NMA due to incompleteness of provided data. For example, a study included only IPD for the placebo arm and thus was excluded from the NMA. A big challenge in the IPD was the high dropout rate (up to 72%) from the RCTs, for which many original authors applied inappropriate imputation methods. Hence, our findings differed from published RCT results. We also encountered outcome reporting bias; specifically, some outcomes were missing from the publications but were available as IPD. The use of different platforms restricted us from combining IPD in a single NMA model and a one-stage analysis was impossible. Time restriction to remote-access platforms and frequent changes of these platforms added another challenge to the analysis, given that IPD from different RCTs were available at different time points.
Conclusions: Retrieval bias can severely impact NMA findings and decision-making. Our study highlighted challenges encountered during an IPD-NMA of sponsored RCTs.
Patient or healthcare consumer involvement: Personalized medicine is required to optimize health care. Well-conducted meta-analyses of IPD are considered the ‘gold-standard’ and influence patient care since patient-level data can be provided to facilitate tailored decision making. However, results from meta-analyses of IPD are likely subject to retrieval bias and awareness of these limitations and their potential impact on findings is required.