Non-response and attrition in a longitudinal study: results from the Swiss Spinal Cord Injury (SwiSCI) Community Surveys

Abstract Introduction Epidemiological surveys targeting individuals with disabilities are challenging, as several factors related to physical impairment and disability may affect study participation. Longitudinal community surveys are particularly prone to bias related to non-response and attrition. This study describes characteristics of responders and non-responders in the Swiss Spinal Cord Injury Cohort Study (SwiSCI) community survey. Methods The study included individuals with chronic SCI eligible for SwiSCI community surveys in 2012, 2017, and 2022. Study population characteristics and response patterns were described using descriptive statistics. Determinants of response and attrition were assessed using logistic regression with clustered standard errors. Predictors included age, sex, SCI etiology, lesion level and severity, time since injury, and preferred language (German, French, Italian). Survey history was captured using two lagged variables and their interaction, including categories for undefined lags. Non-linear effects of age and time since injury were modeled using fractional polynomials. Model performance was evaluated by comparing the observed and predicted transition probabilities at the group level. Results Overall, 5,854 individuals were included in the study, of whom 3,621 were eligible for more than one survey wave. Participation declined over time, from 49% in 2012 to 32% in 2022. Observed transition probabilities for response varied between 65.3% and 8.8% across the three waves. Cumulative attrition rates for 2012 respondents were 50.2% in 2017 and 70.6% in 2022. Response peaked between approximately 50–70 years of age and declined sharply thereafter. Higher response rates were also observed among males, those with paraplegia, and those invited in the early years following chronic SCI, with the lowest response rates around 15–30 years post-injury. Survey history was a strong predictor of subsequent response. Conclusion The SwiSCI community survey is affected by non-response and attrition. Survey history was strongly associated with subsequent participation and may reflect both state dependence and individual or contextual factors. These findings may support targeted strategies to maintain participation and careful analytical approaches. However, the underlying mechanisms related to non-response and attrition require further investigation to be adequately addressed in recruitment and retention strategies. The analytical approach used in this study may inform other longitudinal surveys.