Randomized Comparison of Health Examination Nonparticipation and Questionnaire Nonresponse in the Healthy Finland Survey

Background Participation rates in health examination surveys (HES), and questionnaire or interview surveys (QS) have declined over time. Although low participation does not bias results per se , non-participation is often selective, reducing representativeness even after adjustments on sociodemographic factors. Range of incentives have been tested to improve participation, and health examinations themselves may act as non-monetary, conditional incentives by offering personalized feedback, but direct comparisons between HES and QS remain limited. We aimed 1) to compare participation between individuals invited to the HES versus the questionnaire only in a randomized setting and using register-based characteristics, and 2) to examine how this information affects estimates. Methods The Healthy Finland Survey (2022–2023) was based on a stratified, nationally representative QS sample (n=61,600). For the HES, a random subsample was selected from the QS sample, providing a randomized setting to estimate the incentive effect of the health examination. Comprehensive register data were individually linked with the full sample for the missing data analyses. Inverse probability weights (IPW) were based on pre-2022 register data and supplemented with diagnoses recorded during and after the survey, as symptoms preceding diagnosis may influence participation. Logistic regression models and non-parametric, machine learning methods (e.g. random forests) were used to identify predictors and estimate IPWs. Clustering analysis was applied to assess differences of the estimated participation probabilities based on the parametric main-effects and random forest models. Results Participation rates were substantially higher in the HES sample compared to the QS-only sample among individuals aged 20–75 years, with a difference of 15–20 percentage points under age 50. Based on the Bayesian information criteria, Age*HES, Age*Sex, and Sex*Married interactions were included in the QS weighting model. Random forest variable selection identified a relatively small set of influential predictors. Age-standardized prevalence estimates of selected questionnaire variables did not differ between HES and QS participants. Conclusion Health examinations markedly increased participation among working-age individuals compared to questionnaire alone. However, prevalence estimates based on questionnaire variables were similar in the HES and QS only groups, suggesting that lower QS participation did not compromise validity. Trial registration Clinical trial number: not applicable.