Assessing data quality in the community health worker (CHW) program in Eastern province, Rwanda

Background In order to achieve global development goals, the international community has called for the use of community health workers (CHWs) to deliver important health services. CHWs are routinely collecting large amounts of information. As program managers use these data to monitor and evaluate community-based activities, achieving and maintaining high data quality is critical. Objectives Measuring: 1) data accuracy of household registers compared to household interviews and client records in one district; 2) data reliability of monthly village reports compared to program registers in three districts; and 3) CHW and program factors related to data accuracy and reliability. Methods We used lot quality assurance sampling (LQAS) to determine data quality May 2011- June 2012. We randomly sampled: 1) six CHWs per cell, six households per CHW and classified cells as having ‘poor’ or ‘good’ accuracy for household registers based on five health indicators and a composite one, calculating point estimates by health center; and 2) 19 villages per health center classifying health centers as having ‘poor’ or ‘good’ reliability for village reports for three program indicators and a composite one, calculating point estimates by district. We administered a structured interview to CHWs in three districts measuring CHW and program factors, using logistic regression to measure associations with binary dependent variables data accuracy and reliability. Results Accuracy of household data varied by health center: point estimates were 61-72% for the composite indicator. Data reliability was poor across all districts: point estimates for the composite indicator were 26-60%. CHW having logged a visit to the household in the last month in the household register significantly increased odds of accurate data (OR: 1.71; 95%CI: 1.22, 2.39). The more sick children seen by CHWs significantly worsened data reliability: for four or more sick children versus none, OR: 0.283; 95%CI: 0.180, 0.445. Conclusions Community level data quality is variable in Rwanda, with CHWs generally able to collect data accurately, but not aggregate it correctly due to some program factors. To ensure dependable utilization of information by program managers, we recommend improved supervision and training and LQAS-based routine data quality assessments