Abstract
INTRODUCTION: Definitions of retention-in-care in Prevention of Mother-to-Child Transmission of HIV (PMTCT) vary substantially between studies and programmes. Some definitions are based on visits missed/made, others on a minimum total number of visits, or attendance at a final clinic visit at a specific time. An agreed definition could contribute to developing evidence-based interventions for improving retention-in-care. In this paper, we estimated retention-in-care rates according to different definitions, and we quantified and visualized the degree of agreement between definitions. METHODS: We calculated retention in care rates using nine definitions in the six INSPIRE PMTCT intervention studies, conducted in three sub-Saharan African countries between 2013 and 2017. With data from one of the studies (E4E), we estimated the agreement between definitions using Gwet's agreement coefficient (AC1) and concordance. We calculated positive predictive values (PPV) and negative predictive values (NPV) for all definitions considering successively each definition as the reference standard. Finally, we used a Multiple Correspondence Analysis (MCA) to examine clustering of the way different definitions handle retention-in-care. RESULTS: Retention-in-care rates among 5107 women ranged from 30% to 76% in the complete dataset with Gwet's AC1 being 0.56 [0.53; 0.59] indicating a moderate agreement between all definitions together. Two pairs of definitions with high inner concordance and agreement had either very high PPV or very high NPV, and appeared distinct from the other five definitions on the MCA figures. These pairs of definitions were also the ones resulting in the lowest and highest estimates of retention-in-care. The simplest definition, that only required a final clinic visit to classify women as retained in care, and classified 55% of women as retained in care, had a PPV ranging from 0.7 to 1 and a NPV ranging from 0.69 to 0.98 when excluding the two pairs afore-mentioned; it resulted in a moderate to substantial agreement and a 70% to 90% concordance with all other definitions. CONCLUSIONS: Our study highlights the variability of definitions in estimating retention-in-care. Some definitions are very stringent which may be required in some instances. A simple indicator such as attendance at a single time point may be sufficient for programme planning and evaluation.