Risk of positive selection bias in longitudinal surveys among cancer survivors: Lessons learnt from the national Norwegian Testicular Cancer Survivor Study.

Background: Selection bias due to non- or incomplete compliance is challenging in surveys. Using data from a longitudinal survey in testicular cancer survivors (TCSs), we identify factors predicting incomplete compliance.

Method: In a questionnaire-based national survey (1998-2016; three waves) 1,813 > 5 year TCSs were invited to report post-treatment adverse health outcomes (AHOs). We separated complete from partial participants (participation in all three waves versus participation only once or twice). At each wave we additionally identified responders and non-responders based on their questionnaire return at the respective wave. Multivariable logistic regression analysis identified associations between AHOs reported at the first wave and partial participation. Survival differences between Responders and Non-Responders were assessed by the Kaplan-Meier estimate and the logrank test. Level of significance: p < 0.05.

Results: Of 1813 TCSs 1,346 TCSs (79 %) completed the first wave's questionnaire, and 783 (58 %) became complete and 653 (42 %) partial participants. Poor socio-economics, unhealthy life style, major co-morbidity and chemotherapy-related AHOs reported at the first survey wave were associated with a significant 1.5-1.9 times increased risk for partial participation. At the two last waves non-responders had significantly decreased overall survival compared with responders.

Conclusion: Our longitudinal study indicates positive selection bias during the 17 years of a longitudinal survey among TCSs, with fewer AHOs among Complete than among Partial Participants. If not sufficiently compensated for by data from external sources and/or statistical methods, attrition bias in longitudinal surveys may limit the external validity of findings related to cancer survivors' self-reported AHOs.