Attrition and bias in a longitudinal population-based study on back pain

Background and Aims: Potential bias because of attrition has received little attention in primary studies on back pain. This study aims to identify those back pain related indicators most susceptible to bias and to discuss practical consequences for back pain research. Methods: Analyses were based on a population-based longitudinal multi-centre postal back pain survey with two postal follow-up measurements within two years. The baseline sample comprised 9,263 subjects. Different sets of measures at entry were used to predict subsequent attrition: Socio-demographic variables, indicators of back pain, health related measures, and response behaviour. Back pain related indicators comprised prevalence estimates, pain intensity, disability, and radiating pain. Weighted and unweighted back pain outcomes were compared at the first and second follow-up to assess bias. Results: Little more than half of the eligible participants at baseline continued participation till the second follow-up. Age and prior response behaviour were the best predictors of attrition while health and back pain related variables were less important. Differences between weighted and unweighted estimates of back pain related indicators were small to negligible. Against our expectations, the reported back pain burden slightly declined over time. Conclusion: Differenzial attrition over the different measurement points consecutively reduces the representativeness of the sample. Despite this, bias due to attrition has a small effect on the point estimates of most back pain related outcomes.