What Drives Respondents to Seroepidemiological Surveys? Insights From COVID-19 and Implications for Future Pandemics

Non-random participation can undermine the representativeness of seroepidemiological surveys. Despite their critical role in estimating disease spread during pandemics, non-response bias and methods to correct it require further investigation. This study aimed to examine sociodemographic characteristics and coronavirus disease 2019 (COVID-19)-related factors influencing participation in a seroepidemiological survey.

We analyzed data from a national COVID-19 seroepidemiological survey in Japan between December 2022 and March 2023. We performed multivariable logistic regression analyses to estimate adjusted odds ratios (aOR) and their confidence intervals (CIs) after variable selection with the Group Least Absolute Shrinkage and Selection Operator.

Among 6,091 participants, factors associated with higher odds of seroepidemiological surveys participation included being female (aOR 2.08; 95% CI, 1.25-3.47), living in larger households versus living alone (two: aOR 2.34; 95% CI, 1.20-4.55; four or above: aOR 2.05; 95% CI, 1.03-4.06), higher education levels versus junior high school education (high school: aOR 2.66; 95% CI, 1.06-6.15; junior colleges, technical colleges, vocational schools: aOR 5.51; 95% CI, 1.94-15.07; university and above: aOR 3.30; 95% CI, 1.26-7.98), and having a higher household income versus earning <2 million yen (2-4 million yen: aOR 3.32; 95% CI, 1.52-7.33; 4-6 million yen: aOR 2.73; 95% CI, 1.20-6.23, ≥6 million yen: aOR 4.51; 95% CI, 1.91-10.59). Lower seroepidemiological survey participation odds were observed in those hesitant or unwilling to vaccinate (aOR 0.16; 95% CI, 0.09-0.29) and those perceiving a higher COVID-19 positivity rate among close contacts (aOR 0.98; 95% CI, 0.98-0.99).

Education, income, household size, sex, vaccination status, and perceived infection risk influenced seroepidemiological survey participation. The findings highlight the need to account for non-response bias using weighted methods like inverse probability weighting.