Abstract Body

Lack of nationally representative data hinders assessment of national HIV treatment programs in many low-resource settings. Where laboratory data are collected on a national scale, such data could be used to create a national monitoring cohort but only if information on treatment initiation can be determined. We developed a novel method to impute dates of antiretroviral treatment (ART) initiation from routine laboratory data in South Africa’s public sector HIV programme that could be applied to a national labs database such as South Africa’s National Health Laboratory Service (NHLS) database and assessed validity of this approach.

We analyzed data from two large clinical HIV cohorts: Hlabisa (rural primary care clinics and one sub-district hospital in KwaZulu-Natal) and Right to Care (network of urban clinics in Gauteng). Both cohorts contain known ART initiation dates and lab test results are imported directly from NHLS. While the ART initiation date for patients was known (gold standard), we imputed ART start dates using only lab data that would be available in a laboratory database. To do this, we identified the date of “ART workup”; the lab tests used to determine treatment readiness in national HIV treatment guidelines (first documented hemoglobin or alanine transaminase test among patients with a CD4 count in the 12 months prior to these tests). We then calculated the median time from the ART workup to ART initiation and imputed ART start date as the date of ART workup plus that median time. We calculated sensitivity (SE), specificity (SP), positive predictive value (PPV), and negative predictive value (NPV) of our imputed start date to be within 6 months of the actual ART start date.

We analyzed data from >80,000 HIV-positive adults of whom >90% had an ART workup. Among patients who had a workup and initiated ART, the median time to initiation was 16 days (IQR 7, 31) in Hlabisa and 21 (IQR 8, 43) in RTC cohort. Among patients with known ART start dates, SE of the imputed start date was 83% in Hlabisa and 88% in RTC, indicating this method will correctly estimate the ART start date for about 85% of those with a known ART workup. In Hlabisa, PPV was 95%. SP (100%) and NPV (92%) were also very high.

Routine lab data can be used to infer ART initiation dates in South Africa’s public sector with high rates of classification. Lab data can be used to monitor and evaluate health systems performance and improve the accuracy and completeness of clinical records.