paper/paper.md

In brittanyblouin/ANCRTAdjust: Adjusting HIV prevalence from ANC-RT data for common biases

title: 'ANCRTAdjust: An R package to adjust routine HIV testing data from antenatal care to reduce bias in estimating HIV prevalence trends' tags: - R - antenatal care routine HIV testing data - HIV prevalence trends - data cleaning - bias adjustment authors: - name: Brittany Blouin orcid: 0000-0001-7447-7467 affiliation: 1 - name: Jeffrey W Eaton orcid: 0000-0001-7728-728X affiliation: 2 - name: Mary Mahy affiliation: 3 - name: Mathieu Maheu-Giroux orcid: 0000-0002-8363-4388 affiliation: 1 affiliations: - name: McGill University, School of Global and Population Health, Department of Epidemiology, Biostatistics, and Occupational Health index: 1 - name: Imperial College London, Department of Infectious Disease Epidemiology index: 2 - name: The Joint United Nations Programme on HIV/AIDS (UNAIDS), Department of Strategic Information index: 3 date: September 16, 2019 bibliography: paper.bib

Summary

HIV surveillance in high-burden countries has traditionally been conducted using sentinel surveillance of pregnant women attending antenatal care (ANC-SS). Presently, administrative reporting data about routine HIV testing of all pregnant women attending antenatal care (ANC-RT) are increasingly being used for monitoring HIV epidemic trends [@WHO:2015]. Importantly, using ANC-RT data could increase representativeness, geographical coverage, and granularity of information about HIV. If ANC-RT data on HIV serostatus are to be used for HIV surveillance, several challenges need to be addressed [@Diaz:2005]. First, inferences from routinely collected program data could be biased due to imperfect data completeness - for example, some health facilities might not report HIV testing data [@WHO:2013]. Second, ANC-RT data could be inconsistent with missing and/or invalid data. Missing data problems include: missing entire reporting periods or missing one or more variables within a reporting period. Invalid data problems include: negative count values, the ‘number of HIV-positive women’ exceeding the ‘number of women tested for HIV’, and, the ‘number of women tested for HIV’ exceeding the ‘number of women who attended the health facility’. Finally, imperfect testing coverage during antenatal care could result in selection bias, leading to overestimation of HIV prevalence [@Maheu-Giroux:2019].

ANCRTAdjust is an R package that facilitates assessment of ANC-RT data quality, helps clean ANC-RT data, and, if warranted, can perform specific adjustments for common data inconsistencies as to minimize both information and selection bias in HIV prevalence estimates. Functionalities to longitudinally visualize and tabulate data quality indicators, as well as to identify outliers, are provided. A standard data cleaning function can be used to impute missing data (when possible), and to eliminate impossible negative count values and the possibility that the ‘number of HIV-positive women’ exceeds the ‘number of women tested’. Specifically, different functions can be used to:

1. Evaluate data quality and completeness;
2. Perform data cleaning;
3. Identify outlying observations; and
4. Perform adjustments for common data inconsistencies due to incomplete reporting periods (using inverse probability of censoring weighting), multiple testing (occurring when the same women are tested multiple times during the reporting period leading to the ‘number of women tested for HIV’ exceeding the ‘number of women who attended the health facility’), and/or imperfect testing coverage (using empirically derived adjustment factors).

To the best of our knowledge, there are currently no alternative package to performe these specific tasks and existing data cleaing packages generally provide high-level functions that do not necessarily respond to the idiosyncracies of national ANC-RT data. As several countries transition from ANC-SS to ANC-RT data, it is imperative that clear and simple guidelines be established for the use of such data so that valid conclusions can be drawn for HIV surveillance purposes. The ANCRTAdjust R package was designed to be used by program managers, and public health practitioners and researchers alike, to help clean and adjust (if warranted) ANC-RT data for monitoring HIV trends in endemic countries. A standard protocol for data validation, implemented across all countries providing ANC-RT data for monitoring HIV trends, will ensure that HIV prevalence estimates are consistent, reproducible, and valid.

Acknowledgements

This project was funded by the Joint United Nations Programme on HIV/AIDS (UNAIDS). MMG’s research program is funded by a career award from the Fonds de recherche du Québec – Santé.

References

brittanyblouin/ANCRTAdjust documentation built on Oct. 28, 2019, 4:53 a.m.