A cross-sectional survey was conducted during July and October 2022 in Zhejiang and Henan Provinces respectively, either of which is the major labor-importing or labor-exporting province in China and exists a large number of migrant families or left-behind families. The target population of this survey was made up of caregivers of children aged 1–6 years, who resided in urban areas of Xiaoshan District, Hangzhou City, Zhejiang Province and rural areas of Song County, Luoyang City, Henan Province. Families of caregivers included in Xiaoshan district were migrant families and local urban families, and those of caregivers included in Song County were left-behind families and non-left-behind families. Caregivers were excluded if (1) they have been residing locally for < 6 months; or (2) their children could not be vaccinated because of severe illness or disabilities.

Five towns/communities were selected in Song County and Xiaoshan District using the simple random sampling method, and caregivers were recruited in local township health centers and community health service centers. By reviewing the immunization coverage of non-NIP vaccines among urban and rural children (75.8%), using a desired precision of ± 2% with 95% confidence intervals, and assuming a non-response rate of 10.0%, the required number of surveyed caregivers of children was 1937.

The questionnaire included five sections (characteristics of children, characteristics of caregivers, immunization coverage of children, immunization knowledge of caregivers, and immunization satisfaction of caregivers) and 51 questions. The questionnaire was administered on the online platform Wenjuanxing (https://www.wjx.cn/), the most popular web-based questionnaire platform in China. There were standard instructions for the questionnaire survey, and a total of 35 primary care physicians were invited to administrate the questionnaires. It would take about 10 min to complete the questionnaire, and the caregivers were incentivized to participate in the survey by providing reasonable money. In our survey, only one caregiver per child was face-to-face interviewed, whose answers were entered real-time to the Wenjuanxing platform by primary care physicians. Before the survey, each caregiver of the child enrolled was asked to provide electronic informed consent on the Wenjuanxing online platform. Zhejiang University School of Public Health Medicine Ethics Committees approved the study protocol (ZGL202206-6).

The five types of non-NIP vaccines included in our survey were Haemophiles influenza b (Hib) vaccine, varicella vaccine, rotavirus vaccine, enterovirus 71 vaccine (EV71) and 13-valent pneumonia vaccine (PCV13). These five vaccines were commonly used in China, but have not been included in the NIP [22]. Caregivers were asked whether their children had been vaccinated the aforementioned non-NIP vaccines, and the coverage of non-NIP vaccines was calculated among local urban, migrant, non-left-behind and left-behind children respectively.

The immunization knowledge of caregivers was measured using seven items, including convenience, category, efficiency, continuity, time, schedule, and adverse events of non-NIP vaccines, which were calculated among local urban, migrant, non-left-behind and left-behind families respectively. The description of each item was provided to caregivers, and they were asked whether they aware of these seven items. For example, the knowledge item of “convenience” was measured by asking caregivers whether they were aware of that children could accept non-NIP vaccination nationwide if they had a vaccination certificate.

The immunization satisfaction of caregivers was investigated using nine items, including convenience, vaccination reminder, vaccination environment, consultation, vaccination skills, service quality, vaccination process, vaccination education and time of vaccination, which were calculated among local urban, migrant, non-left-behind and left-behind families respectively. Similarly, a description of each item was provided to caregivers, and caregivers were asked whether they were satisfied with the aforementioned nine items. For example, the experience item of “schedule” was measured by asking caregivers whether they were satisfied with the non-NIP vaccination schedule of local clinics.

Questions for immunization knowledge and satisfaction were general to all non-NIP vaccines, and the whole questionnaire was presented in the Additional file 1: Questionnaire.

The characteristics of children included age, sex (male; female) and birth order (first-born; later-born). The characteristics of caregivers included family role (parents; others), age (≤ 35 years; 35–40 years; ≥ 40 years), sex (male; female), education level (elementary school or lower; middle school; junior college or higher), total household income (less than average; more than average), physical health [assessed by 12-Item Short Form Survey (SF-12)] and mental health (assessed by SF-12). Total household income was measured by the sum of earning income, capital income, pension income, income from government transfers, other income and the total income from other household members during last year.

Characteristics of families were described as mean [standard deviation (SD)] for continuous variables and as number (percentage) for categorical variables. The differences of variables across groups by family types were compared using one-way analysis of variance (ANOVA) and Chi-square test.

Log-binomial regression models were used to calculate prevalence ratios (PRs) and 95% confidence intervals (CIs) for the difference on immunization coverage of children, and knowledge and satisfaction of caregivers of different family types, regarding local urban families as the reference. Age and sex of children were adjusted in the log-binomial regression models. The coverage-knowledge-satisfaction network models of non-NIP vaccination were conducted through Mixed Graphical Model (MGM) using the R package “qgraph” [23]. Least Absolute Shrinkage and Selection Operator (LASSO) and Extended Bayesian Information Criterion (EBIC) were utilized to reduce the pseudo-correlation of connections in the network and to find the optimal fitting model [24]. The nodes of the network represent different items in coverage, knowledge, and satisfaction of non-NIP vaccines, while the bridges represent the correlation between nodes, which are represented by calculating the coefficient of partial correlation among the items of vaccines coverage, vaccination knowledge and satisfaction. The closer the nodes are and the thicker the connected line segments are, the stronger the correlations between nodes are. Strength, defined as the absolute value of the shortest distance between all connections of a node, was chosen to represent the centrality of each node, and its magnitude could represent the extent of connection between nodes in the network. Bridge expected influence refers to the sum of the absolute value of the shortest distance between a specific node and all other connected nodes, which was proportionable to the degree of network conduction through the node [25]. In addition, we used stability coefficient (CS) and 95% CI of edge weights to evaluate the stability of the network. CS coefficient represents the maximum sample attenuation ratio when the correlation value between the original network and the regenerative network parameters is 0.70 [26].

The multivariable logistic regression models to conducted to explore the associated factors of non-NIP vaccination among four types of families, with the outcomes of non-NIP vaccination divided into two groups: “having received all five non-NIP vaccines” and “having not received at least one non-NIP vaccines”. Odds ratios (ORs) and 95% CIs were calculated to compare the association of the characteristics of children and caregivers with non-NIP vaccination for each type of families.

Analyses were performed using SAS (Version 9.4, SAS Institute Inc., Cary, NC) and R (Version 3.6.1; R studio, Boston, Massachusetts). The study was conducted and reported in line with the Consensus-Based Checklist for Reporting of Survey Studies (CROSS, Additional file 2: Table).