Exploring sample bias in a longitudinal disability survey study with high retention rates

Longitudinal survey studies are important tools that can be used to evaluate causal factors that impact health outcomes. However, attrition is a significant concern for most longitudinal studies and can lead to sample bias that may impact results and conclusions. Our objective was to explore potential bias in a longitudinal study of individuals with multiple sclerosis (MS) and spinal cord injury (SCI) that had very high retention rates between measurement time points. Individuals with MS and SCI completed self-report surveys every four months for 20 months. The average response rate for each time point was 95.3% and the cumulative retention rate was 72.5%. Logistic regression was used to evaluate factors that predicted dropout between initial enrollment and the final data collection time point both across and within the two disability groups. In the combined sample of MS and SCI, of the ten variables included in the multivariate regression, younger age, male sex, SCI disability, and higher pain were significantly associated with likelihood of dropout. Analyses within disability groups indicate that only higher pain in persons with SCI was significantly associated with dropout. These results indicate that, even with high overall retention, significant sample bias can occur over time. In these disability populations, additional efforts may be needed to encourage younger persons, men, and those with higher pain to continue to participate in longitudinal studies. Future analyses of these longitudinal data also need to consider the impacts of non-random attrition in the sample.