Moving towards a person-centred HIV care cascade: An exploration of potential biases and errors in routine data in South Africa.

In 2022, in recognition of lags in data infrastructure, the World Health Organisation (WHO) recommended the use of routinely linked individual patient data to monitor HIV programmes. The WHO also recommended a move to person-centred care to better reflect the experiences of people living with HIV. The switch from aggregated service level data to person-centred data will likely introduce some biases and errors. However, little is understood about the direction and magnitude of these biases. We investigated HIV-testing and HIV-care engagement from 2014 to 2018 in the Agincourt Health and Demographic Surveillance System (HDSS). We digitised and linked HIV patient clinic records to HDSS population data in order to estimate biases in routine clinical data. Using this linked data, we followed all individuals linked to HIV-related clinic data throughout their care pathway. We built sequences to represent these pathways. We performed sequence and cluster analyses for all individuals to categorise patterns of care engagement and identified factors associated with different engagement patterns using multinomial logistic regression. Our analyses included 4947 individuals who were linked to 5084 different patient records. We found that routine data would have inflated patient numbers by 2% due to double counting. We also found that 2% of individuals included in our analyses had received multiple HIV tests. These phenomena were driven by undocumented transfers. Further analysis of engagement patterns found a low level of stable engagement in care (<33%). Engagement fell into three distinct clusters: (i) characterised by high rates of late ART initiation, unstable engagement in care, and high mortality (53.9%), (ii) characterised by early ART initiation followed by prolonged periods of LTFU (13.7%), and (iii) characterised by early ART initiation followed by stable engagement in care (32.4%). Compared to cluster (i) older individuals were less likely to be in cluster (ii) and more likely to be in cluster (iii). Those who initiated ART prior to 2016 were more likely to be in cluster (ii) and (iii) compared to cluster (i). Those who initiated ART for PMTCT (RRR: 1.88 (95% CI: 1.45, 2.44)) or TB coinfection (RRR: 2.11 (95% CI: 1.27, 3.50)) were more likely to be in cluster (ii) when compared to those who initiated ART due to CD4 eligibility criteria. Males (RRR: 0.63 (95% CI: 0.51, 0.77)) were less likely to be in cluster (iii) compared to cluster (i) as were those who initiated ART for PMTCT (RRR: 0.77 (95% CI: 0.62, 0.97)) or under test and treat guidelines when compared to those who initiated ART due to CD4 eligibility. Only a minority of patients are consistently engaged in care while the majority cycle between engagement and disengagement. Individual level data could be useful in monitoring programmes and accurately reporting patient figures if it is of high quality, has minimal missingness and is properly linked in order to account for biases that accrue from using this kind of data.