Study population and design
In this nationwide anonymous cross-sectional survey, we used a stratified random sampling method via an online survey company established in 2006 (Wen Juan Xing; Changsha Ranxing Information Technology Co., Ltd., Hunan, China) from December 14, 2020, to January 31, 2021. As a specialized data science company, Wen Juan Xing [23] has a database comprising factual and well-characterized personal information (e.g., sex, region, and age) for over 2.6 million Chinese respondents. We used this platform to conduct stratified random sampling, recruit target participants, and distribute questionnaire surveys. Based on information recorded in the database, the platform has been used by many researchers in cross-sectional studies to collect data regarding health-related attitudes among the general population [24,25,26].
We recruited targeted participants in China for this study according to the following inclusion criteria: (1) women aged 18–49 years, and (2) women who agreed to participate in the study. We planned to recruit 3,000 participants. Sampling of participants was divided into three stages via the online survey platform (Wen Juan Xing). First, we divided targeted participants into three regions (eastern, central, and western regions) and randomly selected two provinces from each region. Second, the sample size for each province was allocated in proportion to the population of each province, according to the China Statistical Yearbook 2020 [27]. Third, we used the Wen Juan Xing online platform to randomly select and recruit target participants from the sample database according to the sample size requirements. This study was approved by the Ethical Committee of Peking University Third Hospital (IRB00006761-M2020528) and conducted according to the Declaration of Helsinki. Informed consent was obtained from all participants.
Assessment of risk perception
We estimated risk perception regarding influenza vaccination with a survey tool commonly used in previous studies for vaccination intention based on the Health Belief Model (HBM), which has good internal consistency and reliability [28, 29]. The HBM is an appropriate theoretical framework for understanding vaccination intent and illustrating the factors influencing people’s decision-making about vaccination [18, 28, 29].
The HBM covers five dimensions and includes nine questions—two on the dimension of “cues to action” and seven on risk perception, including perceived susceptibility, severity, barriers, and benefits. Thus, we used two questions to evaluate perceived susceptibility to infection for the participants themselves and for their children (when applicable), one question to evaluate the perceived severity of infection, three questions to evaluate perceived barriers to vaccination (vaccine safety, effectiveness, and the possibility of infection after vaccination), and one question to evaluate the perceived benefits of vaccination (protective effect). Each question was answered on a three-point Likert scale (“very concerned or agree,” “concerned or not sure,” and “not concerned or disagree,” assigned a score of 3, 2, and 1 points, respectively). The participants were categorized into three groups based on the summed scores for each HBM dimension by tertiles, with the top 33.3% of participants denoted the “high” group, the bottom 33.3% denoted the “low” group, and the middle denoted the “moderate” group. Generally, a Cronbach’s alpha coefficient between 0.6 and 0.8 indicates good internal consistency. In our study, the Cronbach’s alpha index for the different dimensions for influenza vaccination ranged from 0.77 to 0.82, which showed adequate internal consistency and reliability [30].
Measurement of vaccine hesitancy for children
The primary outcome was parental attitudes toward influenza vaccination for their children. If participants answered “no” to the question “If you have children under 18 years old and a vaccine becomes available, would you be willing to have your children receive the seasonal influenza vaccine?”, they were classified into the hesitancy group.
Covariates
Except for HBM and attitudes regarding influenza vaccination, the structured self-administered online questionnaire also included items on sociodemographic characteristics, health status, and knowledge about influenza.
Sociodemographic characteristics included age, residential region, education level, occupation, and monthly household income per capita (RMB). Health status covered gravidity, parity, history of chronic disease, and history of influenza vaccination. We evaluated knowledge about influenza in terms of six aspects: source of infection, route of transmission, susceptible populations, common symptoms, high-risk populations for severe illness and death, and individual preventive measures against infection. Each respondent received one point for each correct answer; no points were received for incorrect answers. We divided the total knowledge score into three groups (low, moderate, and high) by tertiles.
Data analysis
We presented continuous and categorical variables as mean (standard deviation, SD) or percentage (%). The characteristics of participants with vaccine hesitancy were compared using Pearson’s χ2 test. We used univariate and multivariable logistic regression models to estimate the crude odds ratios and adjusted odds ratios (aORs) of vaccine hesitancy in different risk perception groups. A sensitivity analysis was carried out by fitting different models to examine the robustness of the estimation. Model A was a univariate model. In model B, we adjusted sociodemographic characteristics. In model C, we adjusted all covariates, including sociodemographic characteristics, health status, and knowledge of influenza.
Subgroup analyses were performed on each covariate after adjusting the other covariates. The heterogeneity test was used to examine differences between groups, wherein a P-value of less than 0.05 indicated a statistically significant difference. All analyses were conducted using IBM SPSS version 25.0 (IBM Corp., Armonk, NY, USA), R version 3.4.0 (The R Project for Statistical Computing, Vienna Austria), and Stata version 16.0 (StataCorp LLC, College Station, TX, USA).