PLoS ONE (Jan 2021)

Using family planning service statistics to inform model-based estimates of modern contraceptive prevalence

  • Niamh Cahill,
  • Emily Sonneveldt,
  • Priya Emmart,
  • Jessica Williamson,
  • Robinson Mbu,
  • Airy Barrière Fodjo Yetgang,
  • Isaac Dambula,
  • Gizela Azambuja,
  • Alda Antonio Mahumane Govo,
  • Binod Joshi,
  • Sayinzoga Felix,
  • Clarisse Makashaka,
  • Victor Ndaruhutse,
  • Joel Serucaca,
  • Bernard Madzima,
  • Brighton Muzavazi,
  • Leontine Alkema

Journal volume & issue
Vol. 16, no. 10

Abstract

Read online

The annual assessment of Family Planning (FP) indicators, such as the modern contraceptive prevalence rate (mCPR), is a key component of monitoring and evaluating goals of global FP programs and initiatives. To that end, the Family Planning Estimation Model (FPEM) was developed with the aim of producing survey-informed estimates and projections of mCPR and other key FP indictors over time. With large-scale surveys being carried out on average every 3–5 years, data gaps since the most recent survey often exceed one year. As a result, survey-based estimates for the current year from FPEM are often based on projections that carry a larger uncertainty than data informed estimates. In order to bridge recent data gaps we consider the use of a measure, termed Estimated Modern Use (EMU), which has been derived from routinely collected family planning service statistics. However, EMU data come with known limitations, namely measurement errors which result in biases and additional variation with respect to survey-based estimates of mCPR. Here we present a data model for the incorporation of EMU data into FPEM, which accounts for these limitations. Based on known biases, we assume that only changes in EMU can inform FPEM estimates, while also taking inherent variation into account. The addition of this EMU data model to FPEM allows us to provide a secondary data source for informing and reducing uncertainty in current estimates of mCPR. We present model validations using a survey-only model as a baseline comparison and we illustrate the impact of including the EMU data model in FPEM. Results show that the inclusion of EMU data can change point-estimates of mCPR by up to 6.7 percentage points compared to using surveys only. Observed reductions in uncertainty were modest, with the width of uncertainty intervals being reduced by up to 2.7 percentage points.