Conceptual and statistical considerations for producing global estimates of health indicators: learnings from perinatal health estimates

Abstract Background Global estimates are esssential for tracking international targets such as the Sustainable Development Goals (SDGs), therefore vital for planning and priority setting. Modelling and estimation of health indicators vary greatly in aspects of methodology, assumptions, data sources, data quality, and data availability. To aid with consistent reporting of global estimates, a reporting guideline; The Guidance for Accurate and Transparent Health Estimates Reporting (GATHER) was developed in 2016. However, the GATHER statement does not detail the process or statistical considerations for developing global estimates of health indicators. There is no standard statistical framework and methods for producing global estimates of health indicators, even for the same indicator such as low birthweight or preterm birth. This poses challenges for making direct comparisons and trends over time for assessing progress achieved by countries. Methods We developed a framework for conceptual and statistical considerations in an iterative process. An informal review of the methodology used the produce global estimates of perinatal health was conducted and summarised, and expert consultative groups were convened regularly to develop methodologies for the low birthweight and preterm birth estimates (2020). The meetings occurred regularly throughout the estimation process, with methods reviewed and updated accordingly. Following the publication of the estimates, the authors critically reviewed the process and summarised the analytical framework described in the results. Results In this paper, we discuss key considerations for producing global estimates of health indicators to ensure a standardised approach. We focus on: conceptual framework, input data, assessment of data quality, statistical model considerations, and publication of global estimates. We define each concept and give practical examples of how each consideration could be identified and included for global estimate modelling. In addition, we provide a checklist with examples from our recent modelling of global estimates for low birthweight.