Limited evidence of non-response bias despite modest response rate in a nationwide survey of long-term cancer survivorsresults from the NOR-CAYACS study

PurposeDeclining response rates threaten the generalizability of health surveys. We investigate (1) the effect of item order on response rate; (2) characteristics of early , late and non-responders; and (3) potential non-response bias in a population-based health survey of childhood, adolescent and young adult cancer survivors (CAYACS).MethodsWe mailed a questionnaire survey to 5361 eligible CAYACS identified by the Cancer Registry of Norway (CRN), representing a range of cancer diagnoses. The 302-item questionnaire included a range of survivorship-related questions and validated patient-reported outcome measures. To investigate item-order effects on response rates, we constructed two versions of the questionnaire presenting cancer-related or socio-demographic items first. The CRN provided demographic and clinical information for the total population. Risk of non-response bias was estimated by (1) comparing outcomes between early and late responders (answered after a reminder), and (2) by applying inverse probability of participation weights to construct a total population (with 100% response) and then compare 21 a priori selected outcomes between early responders, all responders (early + late) and the total population (all eligible).ResultsSurvey item order did not affect response rates (cancer first 49.8% vs socio-demographic first 50.2%). Shorter time since diagnosis, male gender and a malignant melanoma diagnosis remained significant predictors of non-response in a multivariable multinomial regression model. There were no significant differences on 16/21 survey outcomes between early and late responders, and 18/21 survey outcomes between early responders, all responders and the total population.ConclusionDespite a modest response rate, we found little evidence for a response bias in our study.Implications for Cancer SurvivorsSurveys of survivor-reported outcomes with low response rates may still be valuable and generalizable to the total survivor population.