ASSESSING FIELDWORKER RELIABILITY IN A NATIONAL STUDY OF INJURY MORTALITY IN SOUTH AFRICA

Background Vital registration data indicate that injury deaths have decreased since the first National Burden of Disease (NBD) study in South Africa for 2000 but it is not clear which injury types may have been affected as a high proportion have undetermined intent. In order to quantity injury types, as part of the second NBD study, we conducted a national survey of non-natural deaths presenting to South African mortuaries in 2009. Aims/Objectives/Purpose To test inter-observer reliability between fieldworkers. Methods Lay interviewers were trained to identify the external cause of injury and mobile telephones were used for data capture. The main study sampled 22 000 records from 45 mortuaries. Reliability was tested by two field-workers independently collecting data from the same folder on the same day for 5% of the sample. The proportion of potential agreement beyond chance between two data collectors extracting the same data was determined using the Kappa score, with a value of 0.6 denoting substantial and a value of greater than 0.8 denoting near perfect agreement. Results/Outcome Based on data from 1025 cases, the agreement on the manner of death showed high agreement with a Kappa of 0.861 (95% CI 0.837 to 0.884). There was 88.8% agreement on the detailed information about the external cause of the injury death. Significance/Contribution to the Field Fieldworker coding errors are a potential source of bias in injury surveys, particularly when information is interpreted from secondary sources. The current study provides a simple and cost-effective method with which to measure the reliability of information.