International collaborations using linked administrative data: Lessons from the MARIC study

Abstract

Introduction: The Mortality After Release from Incarceration Consortium (MARIC) is a multi-disciplinary collaboration of researchers, clinicians and decision-makers representing 29 cohorts of adults from 11 countries with a history of incarceration. The MARIC study is the world’s largest coordinated effort to examine mortality in this population, with a combined sample of >1.2 million individuals and >8.2 million person-years of follow-up. Due to both the sensitive nature of incarceration data and the volume of data which had not yet been published, sharing raw data internationally with external collaborators was not possible. Limiting the analyses to published data restricted the potential meta-analyses substantially and decreased the overall population sample size.

Methods: Our analyses of the epidemiology of mortality in formerly incarcerated adults will be conducted using a two-step, individual participant data meta-analysis (IPDMA). In this process, detailed statistical analysis plans (SAPs) are constructed by the central team – including data harmonization standards, sample outputs, and generalised processing scripts in multiple computing languages. Next, individual-level data from each cohort are analysed locally by each cohort team using the instructions detailed in the SAP, resulting in aggregate outputs for each cohort. These outputs are then synthesised and analysed using standard mixed-effects meta-analysis techniques, which account for variation between cohorts inherent to their specific population samples.

Results: A simple analysis was trialled in order to refine the SAP specification guidelines. A data dictionary was provided by each cohort, such that the variations in data structure could be understood and addressed. Deviations from the expected format of the outputs were minimised when a detailed explanation of how the output would be analysed was combined with an example output.

The analyses were well powered to investigate mortality in the cohort at great granularity. However, we noted several limitations. First, the heterogeneity of the cohorts is likely to impact the aggregates, and some mixed-effects analysis inferences are limited by the IPDMA methodology. Second, the data collected across sites were not consistent. As such, some analyses into less commonly collected risk factors and causes of death will only be possible to conduct in a subset of the included cohorts. This reduces the power of the analysis, as well as being a source of possible bias.

Conclusions: The MARIC study greatly increases the scope and precision of research on mortality after release from incarceration and brings international attention to this problem. The IPDMA methodology allows a rigorous and comprehensive analysis of this data to be conducted internationally and collaboratively.