No evidence of response bias in a population-based childhood cancer survivor questionnaire survey - Results from the Swiss Childhood Cancer Survivor Study.

Claudia E Kuehni,Corina S Rueegg,Nicolas X Von Der Weid,Marcel Zwahlen,Gisela Michel,Hajo Zeeb,Micòl E Gianinazzi

doi:10.1371/journal.pone.0176442

Abstract

PurposeThis is the first study to quantify potential nonresponse bias in a childhood cancer survivor questionnaire survey. We describe early and late responders and nonresponders, and estimate nonresponse bias in a nationwide questionnaire survey of survivors.MethodsIn the Swiss Childhood Cancer Survivor Study, we compared characteristics of early responders (who answered an initial questionnaire), late responders (who answered after ≥1 reminder) and nonresponders. Sociodemographic and cancer-related information was available for the whole population from the Swiss Childhood Cancer Registry. We compared observed prevalence of typical outcomes in responders to the expected prevalence in a complete (100% response) representative population we constructed in order to estimate the effect of nonresponse bias. We constructed the complete population using inverse probability of participation weights.ResultsOf 2328 survivors, 930 returned the initial questionnaire (40%); 671 returned the questionnaire after ≥1reminder (29%). Compared to early and late responders, we found that the 727 nonresponders (31%) were more likely male, aged <20 years, French or Italian speaking, of foreign nationality, diagnosed with lymphoma or a CNS or germ cell tumor, and treated only with surgery. But observed prevalence of typical estimates (somatic health, medical care, mental health, health behaviors) was similar among the sample of early responders (40%), all responders (69%), and the complete representative population (100%). In this survey, nonresponse bias did not seem to influence observed prevalence estimates.ConclusionNonresponse bias may play only a minor role in childhood cancer survivor studies, suggesting that results can be generalized to the whole population of such cancer survivors and applied in clinical practice.

Highlights

Nonresponse bias can affect inferences drawn from questionnaire surveys across different populations, countries, and topics [1,2,3,4,5,6]
Compared to early and late responders, we found that the 727 nonresponders (31%) were more likely male, aged
The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript

Summary

Introduction

Nonresponse bias can affect inferences drawn from questionnaire surveys across different populations, countries, and topics [1,2,3,4,5,6]. In studies of general population samples in North America and Europe, nonresponders were more often male, less educated, single or divorced, childless, or immigrants [5,6,7, 10, 11]. This underrepresentation of certain subpopulations in surveys could lead to an under- or overestimation of effects and estimates in results [1,2,3,4,5,6]. The relevance of psychological late effects in childhood cancer survivors might be under-recognized, resulting in insufficient mental health follow-up care

Methods

Results

Conclusion