Globally, work-related deaths (injuries and diseases) are a major social and public health problem. Register data on fatal occupational injuries in high-income countries may be considered to have high quality, especially when reporting is mandatory and regulated by law. We aimed to assess the accuracy of work-related injury death statistics in Norway, with reference to the Labour Inspection Authority and three other on-going registration systems (the cause-specific mortality register, the register for governmental compensations, and the register for insurance companies). In this register-based study, we used the capture-recapture technique to adjust for undercounting. We investigated whether the capture-recapture method using two or three sources gave a valid estimate of fatal occupational injuries as compared with the number of cases identified in four registers administrated by the Norwegian Labour Inspection Authority, Statistics Norway, the Labour and Welfare Administration, and Finance Norway. The inclusion criteria were fatal unintentional injuries among residents of Norway between 2000 and 2003 that occurred while working for income in private and public land-based industries. We obtained ethical and legal approvals. In a period of four years (2000-2003), the Labour Inspection Authority registered 171 occupational injury deaths among residents employed in land-based industries. Two combinations of data sources gave capture-recapture estimates of 246 [95% CI 216; 279] and 265 [95% CI 234; 299] deaths. In total, 246 cases were identified in the four registration systems, which was 44% higher than the number of deaths registered by the Labour Inspection Authority. The Labour Inspection Authority had the most complete register out of the four registration systems. The capture-recapture method used on two overlapping data sources gave highly valid estimates of the total deaths. We demonstrated the existence of significant weaknesses in the registration systems in a country considered to have high-quality register data.
Read full abstract