Assessment of facility-based tuberculosis data quality in an integrated HIV/TB database in three South African districts.

backgroundAssessment of data quality is essential to successful monitoring & evaluation of tuberculosis (TB) services. South Africa uses the Three Interlinked Electronic Register (TIER.Net) to monitor TB diagnoses and treatment outcomes. We assessed the quality of routine programmatic data as captured in TIER.Net.

methodsWe reviewed 277 records from routine data collected for adults who had started TB treatment for drug-sensitive (DS-) TB between 10/2018-12/2019 from 15 facilities across three South African districts using three sources and three approaches to link these (i.e., two approaches compared TIER.NET with the TB Treatment Record while the third approach compared all three sources of TB data: the TB treatment record or patient medical file; the TB Identification Register; and the TB module in TIER.Net). We report agreement and completeness of demographic information and key TB-related variables across all three data sources.

resultsIn our first approach we selected 150 patient records from TIER.Net and found all but one corresponding TB Treatment Record (99%). In our second approach we were also able to find a corresponding TIER.Net record from a starting point of the paper-based, TB Treatment Record for 73/75 (97%) records. We found fewer records 55/75 (73%) in TIER.Net when we used as a starting point records from the TB Identification Register. Demographic information (name, surname, date of birth, and gender) was accurately reported across all three data sources (matching 90% or more). The reporting of key TB-related variables was similar across both the TB Treatment Record and the TB module in TIER.Net (p>0.05). We observed differences in completeness and moderate agreement (Kappa 0.41-0.60) for site of disease, TB treatment outcome and smear microscopy or X-ray as a diagnostic test (p<0.05). We observed more missing items for the TB Treatment record compared to TIER.Net; TB treatment outcome date and site of disease specifically. In comparison, TB treatment start dates as well as HIV-status recording had higher concordance. HIV status and lab results appeared to be more complete in the TB module in TIER.Net than in the TB Treatment Records, and there was "good/substantial" agreement (Kappa 0.61-0.80) for HIV status. DISCUSSION AND

conclusionOur key finding was that the TB Module in TIER.Net was more complete in some key variables including TB treatment outcome. Most TB patient records we reviewed were found on TIER.Net but there was a noticeable gap of TB Identification patient records from the paper register as compared to TIER.Net, including those who tested TB-negative or HIV-negative. There is evidence of complete and "good/substantial" data quality for key TB-related variables, such as "First GeneXpert test result" and "HIV status." Improvements in data completeness of TIER.Net compared to the TB Treatment Record are the most urgent area for improvement, especially recording of TB treatment outcomes.