疫苗错误信息与流感疫苗接种率的关联:基于中国浙江省中断时间序列研究的论文
Associations between vaccine misinformation and influenza vaccine uptake: a population-based interrupted time-series study in China
作者团队基于浙江省2020年1月至2025年4月流感疫苗接种数据,识别出2022年9月至2022年12月期间与疫苗相关的错误信息事件(Sinovac Vaccine Rumours)和2024年8月期间的事件(Pasteur Influenza Vaccine Incident)。
原文明确指出关联性质为主题与时空的准实验分析,读者可据此区分关联与因果结论。
Introduction
Vaccination remains a cornerstone of public health for controlling infectious diseases. However, the proliferation of misinformation has become a critical barrier to sustaining vaccination coverage. Algorithm-driven social media platforms amplify vaccine-related misinformation, creating a “digital wildfire” effect that distorts scientific evidence on vaccine safety and efficacy1,2,3. According to the World Health Organization (WHO), misleading vaccine information has contributed to declining childhood immunisation coverage in more than 100 countries, accompanied by a global resurgence of measles and polio4. These developments illustrate how misinformation has evolved from a communication issue into a population-level behavioural threat, an “infodemic” with measurable consequences5.
A growing body of research has examined COVID-19 vaccine-related misinformation and its relationship with vaccine hesitancy or vaccination intention6,7,8,9,10,11,12. Beyond COVID-19 vaccines, misinformation and vaccine-related media coverage have also been associated with reduced acceptance of other vaccines, including HPV and influenza13,14,15. Some ecological studies have further suggested that online misinformation or social media activity may be associated with lower vaccination coverage or slower vaccine uptake at the population level8,13. However, most existing studies rely on small-scale observational and experimental designs, focusing on intentions rather than actual vaccination behaviour6,10,11,16. Evidence from non-COVID-19 settings remains particularly scarce, and how misinformation, especially its thematic content and temporal patterns, is temporally associated with real-world vaccination behaviour remains poorly understood.
Influenza vaccination is an important strategy for preventing severe influenza-associated disease resulting in hospitalisation or death, particularly among older adults17. In China, however, influenza vaccination uptake among older adults remains suboptimal; a recent systematic review and meta-analysis estimated a pooled seasonal influenza vaccination rate of 17% among adults aged ≥60 years in mainland China, with uptake as low as 7% in areas lacking free vaccination policies18. This low and policy-dependent uptake reflects the coexistence of self-paid vaccination and locally funded free vaccination programmes, creating heterogeneous vaccination contexts across regions. In such a large and diverse vaccination setting, confidence in influenza vaccination may be particularly vulnerable when vaccine-related misinformation spreads during vaccination campaigns, providing an important setting for studying how misinformation interacts with real-world vaccination behaviour.
In this study, we integrated province-wide daily influenza vaccination records from Zhejiang (January 2020-April 2025), a socioeconomically developed region in eastern China where government-funded influenza vaccination for older adults has been implemented since 2020 and expanded over time19, with misinformation events identified on major social media platforms. Using interrupted time-series (ITS) and counterfactual modelling, we evaluated how vaccination rates changed following identified vaccine-related misinformation events by vaccine type, age group, and region. By estimating expected vaccination trajectories in the absence of identified misinformation events, counterfactual ITS modelling provides a quasi-experimental framework for assessing whether observed vaccination outcomes deviated from patterns expected under pre-existing trends, seasonality, and measured contextual factors20,21. By linking population-level behavioural data with misinformation dynamics, this study provides empirical evidence on associations between vaccine-related misinformation events and collective vaccination behaviour, offering insights into the thematic and temporal patterns of misinformation and vaccination behaviour, and informing more responsive vaccine communication and infodemic management strategies worldwide.
Results
Total patterns of influenza vaccination
An overview of the study design is shown in Fig. 1. Between 2020 and 2025, influenza vaccination exhibited strong seasonality in Zhejiang Province, with vaccination rates rising in August or September, peaking in late autumn, and declining after November. The total number of influenza vaccinations increased steadily from 3.38 million in 2020 to 6.43 million in 2024, with an annual average of 4.7 million doses. Government-funded vaccinations rose steadily over these years, with a marked increase in 2024 (from 2.74 million in 2023 to 5.07 million). In contrast, self-paid vaccinations increased from 1.57 million in 2020 to 2.41 million in 2023, but declined to 1.35 million in 2024.
This framework outlines the analytical workflow linking three misinformation events: Sinovac Vaccine Rumours of Pulmonary Nodules (Sinovac Vaccine Rumours, 2022), Disciplinary Case of the Former Chief Engineer of Sinopharm (Disciplinary Case, 2024), and Sanofi Pasteur Influenza Vaccine Lot Recall Incident (Pasteur Influenza Vaccine Incident, 2024), with changes in influenza vaccination behaviour. Analyses were conducted on total vaccination rate and two vaccine types: government-funded influenza vaccines and self-paid influenza vaccines. Interrupted time-series and counterfactual models were applied to quantify changes in vaccination rates, with primary and stratified analyses conducted by vaccine type, age group, and region. The framework integrates thematic and temporal perspectives to characterise how misinformation corresponds to variations in vaccination rates.
Changes in total vaccination rates
ITS analysis revealed distinct temporal patterns across the three misinformation events (Table 1; Fig. 2; Supplementary Fig. 3).
This figure presents the two vaccine-related misinformation events that showed statistically significant changes in vaccination rates in the main analysis. Dots represent the logarithm of observed vaccination rates (per 100,000 residents) after removing the seasonality, autocorrelation, and pulse terms from the model. Panels (a) and (b) depict total vaccination trends during the Sinovac Vaccine Rumours of Pulmonary Nodules and Sanofi Pasteur Influenza Vaccine Lot Recall Incident, respectively. Panels (c) and (d) show trends stratified by vaccine type (government-funded and self-paid) for the corresponding events. Solid lines indicate fitted regression trends, while dashed lines represent projected baseline trends in the absence of misinformation.
During the Sinovac Vaccine Rumours (September-December 2022), the pre-event vaccination rate increased by 4.8% per week (b₁ = 0.047, P = 0.009). The estimated trend shifted to a 3.5% weekly decrease during the transition period, but this slope change from the pre-event trend was not statistically significant (b1 + b₃ = −0.036; b₃ = −0.083, P = 0.203). After the transition period, the weekly trend attenuated from the pre-event 4.8% increase to a 1.8% increase, representing a statistically significant reduction in slope (b1 + b5 = 0.018; b5 = −0.029, P = 0.046). Fourier terms (Seasonality 2: b = 0.976, P = 0.007; Seasonality 3: b = 0.696, P < 0.001) and the lag term (b = 0.614, P < 0.001) confirmed pronounced seasonality and short-term dependence in weekly vaccination rates.
A similar reversal occurred during the Pasteur Influenza Vaccine Incident (August 2024). The vaccination rate had increased by 7.3% per week prior to the event (b₁ = 0.071, P = 0.003), but the post-event trend shifted to a 9.0% weekly decrease (b1 + b₃ = −0.094; b₃ = −0.165, P < 0.001). Seasonality remained strong (Seasonality 1: b = −1.533, P < 0.001; Seasonality 2: b = 2.144, P < 0.001; Seasonality 3: b = 0.457, P < 0.001; Seasonality 4: b = −1.146, P < 0.001). By contrast, the Disciplinary Case (April 2024) showed no significant change in total vaccination rate (b₃ = −0.255, P = 0.185).
Counterfactual analyses supported these findings (Fig. 3). For both the Sinovac Vaccine Rumours and Pasteur Influenza Vaccine Incident, observed vaccination rates diverged substantially from projected counterfactual trends, with the gap widening progressively following the dissemination of misinformation and persisting throughout the remaining duration of the vaccination season. For the Disciplinary Case, the difference between observed and counterfactual trajectories was minimal and statistically insignificant.
The lower panel displays the total log-transformed influenza vaccination rate (per 100,000 population) from January 2020 to April 2025, with the light green shading indicating the annual influenza vaccination peak season. The upper three panels compare observed (grey), fitted (blue), and counterfactual (orange, 95% CI shaded) vaccination rates during three identified vaccine-related misinformation events: a Sinovac Vaccine Rumours of Pulmonary Nodules (Sinovac Vaccine Rumours, September-December 2022), b Disciplinary Case of the Former Chief Engineer of Sinopharm (Disciplinary Case, April 2024), and c Sanofi Pasteur Influenza Vaccine Lot Recall Incident (Pasteur Influenza Vaccine Incident, August 2024).
Changes in government-funded and self-paid vaccination rates
Patterns differed between vaccine types (Table 1; Fig. 2; Supplementary Figs. 3–5). For government-funded vaccines, the Sinovac Vaccine Rumours was associated with a significant downward trend relative to the pre-event period (b = −0.044, P = 0.015). During the Pasteur Influenza Vaccine Incident, an immediate rise was detected (b = 12.186, P < 0.001), followed by a decline in the post-event trend (b = −0.172, P = 0.024). No significant changes were observed for the Disciplinary Case.
For self-paid vaccines, only the Pasteur Influenza Vaccine Incident showed significant reductions, with an immediate drop (b = −1.697, P = 0.001) and a sustained downward trend (b = −0.139, P < 0.001).
Furthermore, the post-event decline (b of Posttime) was more pronounced for government-funded vaccines than for self-paid vaccines, suggesting a stronger association among government-funded vaccination rates.
Stratified and regional patterns
Age-stratified analyses (Supplementary Table 1) showed that, during the Sinovac Vaccine Rumours, government-funded vaccination rates declined significantly among individuals under 60 years (b = −0.043, P = 0.022). During the Pasteur Influenza Vaccine Incident, total (b = −0.154, P < 0.001), government-funded (b = −0.183, P = 0.018), and self-paid vaccination rates (b = −0.144, P < 0.001) declined significantly among individuals under 60 years; total (b = −0.169, P = 0.002) and government-funded vaccination rates (b = −0.172, P = 0.013) also declined among individuals aged 60 years or older. No significant changes were observed for the Disciplinary Case. Spatially, Quzhou and Zhoushan exhibited notable declines in vaccination rates during the Sinovac Vaccine Rumours (Supplementary Table 2).
Sensitivity analysis
Sensitivity analyses revealed that shifting the event window by ±1 week did not result in any substantial differences from the primary analyses (Supplementary Table 3). In models excluding temperature and Stringency Index covariates, a few associations became no longer statistically significant (in the total vaccination rates during the Sinovac Vaccine Rumours), underscoring the importance of controlling for potential confounders in the primary analyses (Supplementary Table 4). In the calendar-matched placebo analyses, no significant downward post-event trends were observed during the corresponding non-event periods in the preceding year. For the placebo period corresponding to the Sinovac Vaccine Rumours, the only significant post-event trends were upward rather than downward, and were observed in total vaccination rates (b = 0.045, P = 0.019) and government-funded vaccination rates (b = 0.059, P = 0.006) (Supplementary Table 5). For the Rumours Linking COVID-19 Vaccines to Leukaemia, which was not included in the main analysis because of its prolonged and less clearly bounded public-opinion period, supplementary sensitivity analyses using both a single-interruption ITS model and a transition-period ITS model showed declining trends in total and self-paid vaccination rates (Supplementary Table 6).
Discussion
This study combined large-scale vaccination records with online misinformation timelines to examine whether identified vaccine-related misinformation events were temporally associated with changes in influenza vaccination behaviour in Zhejiang, China. We identified significant attenuations or declines in vaccination trends after two vaccine safety- or effectiveness-related misinformation events that occurred during the vaccination season. Specifically, for the Sinovac Vaccine Rumours, the vaccination rate attenuated significantly from a 4.8% weekly increase before the event to a 1.8% weekly increase after the event. For the Pasteur Influenza Vaccine Incident, the rate changed from a 7.3% weekly increase at the baseline to a 9.0% decrease afterward. In contrast, the administrative misinformation (Disciplinary Case) did not show a significant change. Government-funded vaccination rates appeared to be more consistently affected, with declines observed during both the Sinovac Vaccine Rumours and the Pasteur Influenza Vaccine Incident, whereas significant reductions in self-paid vaccination were mainly observed during the Pasteur Influenza Vaccine Incident. These findings provide population-level, real-world evidence that misinformation may coincide with sustained changes in vaccination behaviour, and suggest that its thematic framing, temporal context, and vaccine type jointly shape public responses.
The differential patterns of the three events may support the hypothesis that the thematic content and intensity of misinformation discourse are associated with different vaccination responses. The Sinovac Vaccine Rumours and Pasteur Influenza Vaccine Incident both centred on vaccine safety or effectiveness, core concerns in vaccination decision-making, although they differed in information type: the former was a fabricated or unverified rumour, whereas the latter was an actual, documented precautionary recall that was amplified or misinterpreted online as a broader concern about influenza vaccine effectiveness or safety. Despite this distinction, both events were associated with attenuated or declining vaccination trends, consistent with prior research showing that safety-focused misinformation is strongly associated with COVID-19 and influenza vaccine hesitancy1,6,22,23,24. Although the Sinovac Vaccine Rumours concerned COVID-19 rather than influenza vaccines, the observed association with influenza vaccination may reflect broader vaccine-confidence spillover rather than an influenza-specific misinformation effect. Prior studies have reported links between COVID-19 and influenza vaccination behaviours, suggesting that concerns about one vaccine may generalise to broader perceptions of vaccine safety, production quality, and institutional trust25,26,27. In contrast, the administrative trust-related Disciplinary Case showed no significant association with vaccination rates, suggesting that not all vaccine-related misinformation events are followed by measurable behavioural changes. Although political and cultural narratives can influence attitudes9,22,28, the real-world behavioural consequences of administrative trust-related events appear limited. Together, these findings suggest that not all misinformation events are equally consequential, and that public health responses should prioritise rapid detection and targeted, trust-oriented communication strategies for safety- or effectiveness-related misinformation events, which appear most likely to coincide with real-world behavioural change.
Temporal context also appeared to modulate the associations between misinformation events and vaccination rates. The Sinovac Vaccine Rumours (September-December 2022) and Pasteur Influenza Vaccine Incident (August 2024) occurred around China’s annual influenza vaccination campaigns (typically September-November), when public attention and vaccine decision-making peak. Misinformation emerging during these windows may be more closely associated with vaccination decisions, as individuals actively seeking vaccination information were more susceptible to misleading claims. This pattern echoes the “temporal salience” principle in behavioural science, where misinformation encountered near decision-making moments is more likely to be associated with persistent changes29,30. Consistent with this pattern, our study found that vaccination rates declined persistently following the misinformation events. Previous studies have similarly reported that exposure to misinformation shortly before vaccination seasons, even if later corrected, can lead to persistent distrust in vaccines31,32,33. From a policy perspective, these findings indicate that the timing of misinformation may be as important as its content. Public health systems should prioritise early warning mechanisms and rapid-response communication during peak decision-making windows, when populations appear most vulnerable to misleading claims.
Government-funded vaccination rates appeared to be more consistently affected. During the Sinovac Vaccine Rumours, declines were observed in government-funded vaccination rates, but not in self-paid vaccination rates. During the Pasteur Influenza Vaccine Incident in 2024, although the government-funded rate showed an immediate post-event surge, likely reflecting the Zhejiang provincial policy that expanded eligibility from ≥70 to ≥60 years, it was followed by a subsequent decline in the post-event trend. Prior research on the 2018 Changchun Changsheng vaccine incident in China similarly found more substantial behavioural impacts on National Immunisation Programme (NIP) vaccines than on non-NIP ones34. This pattern may partly reflect differences in public perceptions of government-funded and self-paid vaccines, including concerns that publicly provided vaccines may be perceived as lower in quality or less trustworthy by some individuals35,36. However, because vaccine-type differences may also be influenced by eligibility criteria, programme timing, public campaigns, denominator changes, vaccine access, and the proportions of different manufacturers or brands, this interpretation requires further investigation. Nevertheless, given that government-funded influenza vaccines primarily target older adults and account for a large share of total uptake, these findings highlight the need for real-time monitoring of misinformation and rapid communication responses to sustain confidence in public immunisation programmes37,38.
The strengths of this study lie in its use of large-scale, real-world data and the integration of behavioural, thematic, and temporal perspectives. The combination of ITS and counterfactual modelling approaches visualised and quantified deviations in observed vaccination rates from expected trajectories after identified misinformation events. Stratified analyses by vaccine type revealed heterogeneous patterns, offering actionable insights for risk communication.
However, several limitations should be noted. First, as an ecological study, this analysis lacked individual-level data on misinformation exposure, beliefs, and vaccination decision-making. Therefore, the observed associations should not be interpreted as individual-level effects of misinformation exposure. Second, the definition and operationalisation of misinformation had limitations. We did not infer intent to mislead, and the included events should be interpreted according to our operational definition rather than as uniformly fabricated rumours or deliberate disinformation. In particular, the Disciplinary Case and Pasteur Influenza Vaccine Incident were factual events, but their online circulation could raise concerns about vaccine production, regulatory trust, effectiveness, or safety. This heterogeneity in information type should be considered when comparing findings across events. Third, event identification and public attention measurement were incomplete. The study focused on three major events identified mainly through REDnote, Weibo, and Baidu Index, and may have underrepresented discussions on other platforms, lower-attention events, prolonged events, or less keyword-specific misinformation. Baidu Index is only a proxy for public attention and cannot directly measure misinformation exposure, sentiment, belief change, or media coverage unrelated to misinformation. Generalisability beyond Zhejiang Province may also be limited because vaccination policies, media environments, and health service accessibility differ across regions. Fourth, unmeasured confounders, including policy changes, vaccine supply or access, influenza activity, COVID-19-related events, broader vaccine-confidence spillover, and changes in health-seeking behaviour, may have influenced vaccination trends. Finally, different ITS model specifications can affect estimated level and slope changes and statistical significance39, and counterfactual trajectories depend on model-based assumptions about underlying trends and event timing. Therefore, model dependence and specification uncertainty cannot be fully excluded.
In conclusion, by analysing large-scale real-world data, this study quantified the association between vaccine misinformation and reductions in influenza vaccination rates. The magnitude of this association varied by the thematic and temporal patterns of misinformation, with safety-related misinformation emerging near vaccination peaks showing the most pronounced temporal associations. Government-funded vaccines appeared more consistently affected by these events. Together, these findings advance understanding of how vaccine-related misinformation events are associated with measurable changes in vaccine uptake in real-world settings, highlighting the importance of considering both thematic content and timing when assessing the population-level impacts of misinformation and informing vaccine communication strategies.
Methods
Data sources
We used province-wide daily influenza vaccination records from Zhejiang, a socioeconomically developed region in eastern China, from January 2020 through April 2025. These data were retrospectively extracted from routinely collected administrative records in the Intelligent Service Information System of Zhejiang Provincial Immunisation Programme. The database provides information on vaccine type (government-funded/self-paid), age, sex, and region (11 prefecture-level cities). Influenza vaccines in China are classified as non–National Immunisation Programme vaccines. In Zhejiang Province, government-funded influenza vaccination mainly targets children and older adults. Residents aged ≥70 years with Zhejiang household registration were eligible to receive influenza vaccines free of charge between 2020 and 2023; since 2024, the programme has been expanded to include all residents aged ≥60 years.
To minimise bias from reporting delays, data were aggregated into weekly counts (52 weeks per year). Population denominators (total population and age-specific population) were extracted from the Zhejiang Statistical Yearbook to compute weekly influenza vaccination rates per 100,000 residents.
Identification of misinformation events
Following recent public health definitions, we operationalised vaccine-related misinformation events as online events involving vaccine-related information that was false, inaccurate, or misleading according to the best available evidence at the time40. We did not attempt to determine whether individual posts were shared with intent to mislead, because intentionality is difficult to infer from public social media posts and search data41. Informed by the information disorder framework, we used “vaccine-related misinformation events” pragmatically in this study to include fabricated or unverified vaccine-related rumours, as well as misleading interpretations, amplification, or unverified extrapolations of factual vaccine-related events42.
Candidate misinformation events related to vaccination were identified through keyword tracking and verification on major Chinese social media platforms (Weibo and REDnote/Xiaohongshu) and Baidu Index. Weibo is a large microblogging service often described as China’s equivalent to Twitter, and REDnote/Xiaohongshu is a widely used user-generated content platform similar to Instagram. We systematically collected and analysed public discourse on vaccine-related misinformation across these platforms from January 2020 through April 2025. We employed a comprehensive keyword search strategy that included terms such as “vaccine side effects”, “vaccine accident”, “vaccine safety”, and “vaccine efficacy”.
Baidu Index (https://index.baidu.com/v2/index.html#/) provides aggregated statistics on keyword search frequency and quantifies public attention by calculating weighted search frequencies of selected keywords across Baidu web pages. This index reflects the level and persistence of public interest in specific topics over time, allowing dynamic monitoring of information exposure and public concern during vaccine-related events. Fluctuations in this index were used to track the intensity and duration of public concern surrounding vaccine-related issues. For each candidate event, event-specific Chinese search terms were selected according to the dominant wording observed in online discussions and were used to assess whether the event generated an interpretable Baidu Index trend. The onset week of an event was primarily defined as the week in which the event-specific Baidu Index showed the first abrupt rise, and this timing was cross-checked against corresponding social media discussions and event background. For events with a prolonged public-opinion process, the event window was further assessed according to the sustained elevation and subsequent attenuation of the Baidu Index trend.
Based on social media search results and Baidu Index trends, we initially identified six candidate vaccine-related misinformation or misleading information events: (1) Fake Mineral Water COVID-19 Vaccine Case (January-February 2021): Posts falsely claimed that counterfeit COVID-19 vaccines were produced using mineral water, prompting widespread online concern despite immediate clarification from authorities. (2) Rumours Linking COVID-19 Vaccines to Leukaemia (May-June 2022): Misinformation suggested that COVID-19 vaccine–induced antibodies could cause leukaemia, circulating widely before being refuted by medical experts. (3) Sinovac Vaccine Rumours of Pulmonary Nodules (September-December 2022): Unsubstantiated claims alleged that inactivated COVID-19 vaccines caused pulmonary nodules, contradicting available clinical evidence. (4) Sinovac Pharmaceutical Filing Misinterpreted Event (January 2023): Routine regulatory filings by Sinovac were misinterpreted online as evidence of vaccine quality issues, leading to transient speculation. (5) Disciplinary Case of Former Chief Engineer of Sinopharm (April 2024): The disciplinary investigation of former chief engineer Yang Xiaoming led to speculation about the integrity of vaccine research and development processes. (6) Sanofi Pasteur Influenza Vaccine Lot Recall Incident (August 2024): A precautionary recall based on potency concerns was misinterpreted in online discussions as evidence of broader vaccine safety or effectiveness problems.
These events were selected according to the following criteria:
-
1.
vaccine-related misinformation events related to vaccine production, safety, effectiveness, administration, or regulatory trust;
-
2.
a clearly defined onset week and an interpretable event window suitable for ITS analysis, with the onset week identified by the first abrupt rise in the event-specific Baidu Index;
-
3.
notable online dissemination, reflected by visible spikes in social media engagement or search activity, with the highest Baidu Index value during the public-attention peak reaching at least 800.
The Baidu Index threshold of ≥800 was used as an operational public-attention criterion rather than a validated cutoff in the literature. It was intended to identify major events with a sufficiently visible online signal and a temporally interpretable pattern for quantitative interruption analysis. Candidate events were screened before outcome modelling and were not included or excluded based on their association with influenza vaccination rates. Candidate events were reviewed jointly by the study team using the predefined criteria, and uncertain cases were resolved through discussion until consensus was reached. The overall process of candidate event identification, selection, and handling is illustrated in Supplementary Fig. 1.
According to the above criteria, three candidate events were not included in the main ITS analysis. For the Rumours Linking COVID-19 Vaccines to Leukaemia, the event-specific Baidu Index trend showed an abrupt initial peak exceeding 800, indicating substantial public attention. However, the subsequent attention pattern resembled the Sinovac Vaccine Rumours in showing more than one peak, while the second peak did not reach the predefined threshold of 800. In addition, the related public-opinion period was prolonged and diffuse, making it less suitable for the main ITS design requiring a clearly bounded interruption window. Therefore, this event was not included in the main three-event analysis. For the Fake Mineral Water COVID-19 Vaccine Case and the Sinovac Pharmaceutical Filing Misinterpreted Event, discussions on Weibo and REDnote/Xiaohongshu suggested that factual events had triggered concerns about vaccine quality, safety, or regulatory processes. However, these events were primarily related to a criminal/counterfeit vaccine case or routine filing and regulatory information, and the misleading interpretation component was less clearly separable from factual reporting than for the events included in the main analysis. In addition, we could not identify suitable event-specific Baidu Index keywords that reliably captured these two events. Searches using representative event-specific Chinese terms did not produce interpretable term-specific Baidu Index trends, suggesting that online discussion of these events was relatively limited or less concentrated around distinct searchable keywords.
The three events identified for the main analysis were: (1) Sinovac Vaccine Rumours of Pulmonary Nodules (Sinovac Vaccine Rumours, September-December 2022), a fabricated or unverified safety rumour linking COVID-19 vaccines to pulmonary nodules; (2) Disciplinary Case of the Former Chief Engineer of Sinopharm (Disciplinary Case, April 2024), a documented administrative and anti-corruption-related event involving a senior vaccine scientist that was interpreted online as a vaccine production or regulatory trust concern; and (3) Sanofi Pasteur Influenza Vaccine Lot Recall Incident (Pasteur Influenza Vaccine Incident, August 2024), an actual, documented precautionary recall over potency concerns that was amplified or misinterpreted online as a broader concern about influenza vaccine effectiveness or safety. The event-specific Baidu Index trends, based on Chinese search terms, were used to characterise the timing, intensity, and duration of public attention and to guide the event-specific ITS model specification. The Baidu Index trends for the selected and sensitivity-analysis events are shown in Supplementary Fig. 2.
To further assess the dependence of event selection on the Baidu Index threshold, we found that lowering the threshold to ≥500 would have captured both attention peaks of the Rumours Linking COVID-19 Vaccines to Leukaemia (Supplementary Fig. 2d). Given its relevance to vaccine safety concerns, this event was therefore analysed separately in additional sensitivity analyses. Specifically, the single-interruption ITS analysis used the week of 30 May to 5 June 2022 as the interruption point, and the transition-period specification was analogous to that used for the Sinovac Vaccine Rumours.
Interrupted time series analysis
We applied segmented regression models to estimate the changes in level and slope of weekly vaccination rates following each misinformation event. The general model was specified as:
$$\begin{array}{c}\log \left({Y}_{t}\right)={\beta }_{0}+{\beta }_{1}{{\cdot }}{Time}+{\beta }_{2}{{\cdot }}{Startpoint}+{\beta }_{3}{{\cdot }}{Posttime}+\sum {\beta }_{p}{X}_{p}+{Seasonality}\\+{\beta }_{m}{{\cdot }}{Pulse}+{\beta }_{n}{{\cdot }}\log \left({Y}_{t-1}\right)+{\epsilon }_{t}\end{array}$$
(1)
where \({Y}_{t}\) represents the weekly vaccination rate (/100,000) at time t; \({Time}\) is the duration in weeks since the baseline period, with \({\beta }_{1}\) indicating the natural trend of vaccination rates during the pre-misinformation period; \({Startpoint}\) is a dummy variable for the occurrence of misinformation (0 before the event onset, 1 after), with \({\beta }_{2}\) representing the immediate level change upon the emergence of misinformation; \({Posttime}\) measures the time in weeks elapsed since the onset of misinformation, with \({\beta }_{3}\) reflecting the change in slope during the post-misinformation period compared with the pre-misinformation period.
Using public search data from the Baidu Index platform43, we selected appropriate ITS models to analyse the three misinformation events. For both the Disciplinary Case (event week identified around April 28, 2024; analysis window: November 2023 to November 2024; Supplementary Fig. 2b) and the Pasteur Influenza Vaccine Incident (event week identified around August 27, 2024; analysis window: January 2024 to April 2025; Supplementary Fig. 2c), which were characterised by shorter public-opinion durations, no transition term was introduced in the analysis. Both events were therefore modelled using the base segmented regression described in Eq. (1). For the Sinovac Vaccine Rumours (public-attention period: September to December 2022; analysis window: January 2022 to December 2023), which had a relatively long duration with noticeable public-opinion development (Supplementary Fig. 2a), we set September 20 to December 20, 2022, as the transition period for public-opinion. The specific ITS model is as follows:
$$\begin{array}{c}\log \left({Y}_{t}\right)={\beta }_{0}+{\beta }_{1}{{\cdot }}{Time}+{\beta }_{2}{{\cdot }}{Startpoint}+{\beta }_{3}{{\cdot }}{Transition}+{\beta }_{4}{{\cdot }}{Endpoint}+\\ {\beta }_{5}{{\cdot }}{Posttime}+\sum {\beta }_{p}{X}_{p}+{Seasonality}+{\beta }_{m}{{\cdot }}{Pulse}+{\beta }_{n}{{\cdot }}\log \left({Y}_{t-1}\right)+{\epsilon }_{t}\end{array}$$
(2)
where \({Y}_{t}\) represents the weekly vaccination rate (/100,000) at time t; \({Time}\) is the duration in weeks since the baseline period, with \({\beta }_{1}\) indicating the natural trend of vaccination rates during the pre-misinformation period; \({Startpoint}\) and \({Endpoint}\) are two dummy variables for the onset and conclusion of the misinformation transition period, respectively (0 before, 1 after), with \({\beta }_{2}\) and \({\beta }_{4}\) representing the immediate level changes upon the onset and conclusion of the transition period; Transition measures the time in weeks since the transition period began during the transition period and was held constant thereafter, with \({\beta }_{3}\) reflecting the change in slope during the transition period compared with the pre-misinformation period; \({Posttime}\) measures the time in weeks elapsed since the end of the transition period, with \({\beta }_{5}\) reflecting the change in slope during the post-misinformation period compared with the pre-misinformation period. The difference between the post-transition and transition-period slopes is \({\beta }_{5}-{\beta }_{3}\).
Covariates (\(\sum {\beta }_{p}{X}_{p}\)) included the Stringency Index from Oxford COVID-19 Government Response Tracker (OxCGRT)44 and ambient temperature from the U.S. National Centres for Environmental Information45. Seasonality was adjusted using four Fourier terms (sine and cosine functions with a 52-week period). Pulse terms were introduced to capture short-term surges associated with Zhejiang’s free influenza vaccination programme each August or September. Because Zhejiang’s government-funded influenza vaccination programme for older adults has usually started in August or September each year since 2020, government-funded and total vaccination rates showed immediate pulse-like surges at programme initiation, unlike self-paid vaccination rates. Therefore, based on changes in government-funded vaccination uptake, Pulse terms were set for one or two programme-start weeks to capture these externally driven increases (1 for the specific week, 0 otherwise). In 2024, two pulse weeks were specified, which may reflect the expansion of the government-funded vaccination programme from residents aged ≥70 years to those aged ≥60 years. Pulse terms were included in the models for total and government-funded vaccination rates, but not in the self-paid vaccination models. Considering the persistence and short-term dependence of weekly vaccination behaviour, \(\log \left({Y}_{t-1}\right)\) was included as a lagged outcome covariate. Serial autocorrelation was tested using residual autocorrelation plots and the Durbin-Watson (DW) test. If residual autocorrelation remained after inclusion of the lagged outcome term, an AR(1) residual correlation structure would be included [\(\text{Corr}\left({{\varepsilon }_{t},\varepsilon }_{t-1}\right)=\rho\)]. Models without an AR(1) residual structure were estimated using ordinary least squares (OLS), whereas models with AR(1) residual correlation were estimated using generalised least squares (GLS). In the final models, inclusion of the lagged outcome term adequately reduced residual autocorrelation, and no AR(1) residual correlation structure was retained.
The fitted coefficients \({b}_{1},{b}_{2},{b}_{3},\ldots\) represent the estimates of the corresponding model parameters \({\beta }_{1},{\beta }_{2},{\beta }_{3},\ldots\). Because the outcome was log-transformed, coefficients were converted to percentage changes as (\({e}^{b}-\)1) × 100%, with combined slopes calculated by exponentiating the sum of the relevant coefficients.
Counterfactual estimation
Counterfactual predictions were generated to estimate potential influenza vaccination rates under the hypothetical scenario in which no vaccine misinformation occurred. Specifically, the misinformation-specific terms were set to zero, including Startpoint and Posttime in the base ITS model, and Startpoint, Transition, Endpoint, and Posttime in the transition-period model. In contrast, the fitted intercept and coefficients for the underlying time trend, covariates, seasonality, pulse terms, and lagged outcome term were retained. The counterfactual prediction can be expressed as:
$$\begin{array}{c}\log \left({\hat{Y}}_{t}\right)={\beta }_{0}+{\beta }_{1}{{\cdot }}{Time}+\sum {\beta }_{p}{X}_{p}+{Seasonality}+{\beta }_{m}{{\cdot }}{Pulse}\\ +{\beta }_{n}{{\cdot }}\log \left({Y}_{t-1}\right)\end{array}$$
(3)
where \({\beta }_{0}\), \({\beta }_{1}\), \({\beta }_{p}\), \({\beta }_{m}\), \({\beta }_{n}\) correspond to parameters defined in Eqs. (1) or (2); \({\hat{Y}}_{t}\) denotes the counterfactual predicted vaccination rate at week t. Confidence intervals for the counterfactual trajectories were calculated using the variance-covariance matrix of the fitted ITS model coefficients, reflecting the uncertainty in the estimated counterfactual mean vaccination rates, and were then transformed from the log scale back to the original vaccination-rate scale.
Sensitivity analysis
We estimated changes in total, government-funded, and self-paid vaccination rates and conducted stratified analyses by age group ( < 60 and ≥60 years) and by region (11 prefecture-level cities). If zero values were present in the weekly vaccination rate, they were replaced by half the smallest non-zero value before the log transformation46,47. Meanwhile, sensitivity analyses were conducted by varying event window definitions by ±1 week, and by testing models with and without adjustment for the Stringency Index and temperature variables. To examine whether similar deviations would be observed during comparable seasonal periods without corresponding misinformation events, we additionally conducted calendar-matched placebo analyses for events showing statistically significant associations in the main analysis by applying the same ITS specifications to the corresponding calendar periods in the preceding year. In addition, the Rumours Linking COVID-19 Vaccines to Leukaemia, which was not included in the main analysis because of its prolonged and less clearly bounded public-opinion period, was also analysed separately in supplementary sensitivity analyses using both a single-interruption ITS model and a transition-period model.
Statistical analysis
All analyses were conducted in R (Version 4.4.2) using the packages forecast, lmtest, and tidyverse, among others. Statistical significance was defined as P < 0.05 (two-sided).
Ethical approval
This study was approved by the Institutional Review Board of Zhejiang Provincial Centre for Disease Control and Prevention (No. 2025-038-01). All vaccination data were extracted and anonymised with no personal identifiers.
Data availability
Aggregated vaccination data were obtained from the Intelligent Service Information System of the Zhejiang Provincial Immunization Programme, through the Zhejiang Provincial Center for Disease Control and Prevention under data-sharing agreements, and are not publicly available.
References
Ngai, C. S. B., Singh, R. G. & Yao, L. Impact of COVID-19 vaccine misinformation on social media virality: Content analysis of message themes and writing strategies. J. Med Internet Res. 24, e37806 (2022).
Lenti, J. et al. Global Misinformation Spillovers in the Vaccination Debate Before and During the COVID-19 Pandemic: Multilingual Twitter Study. JMIR Infodemiol. 3, e44714 (2023).
Larson, H. J. Stuck: How Vaccine Rumors Start and Why They Don’t Go Away. (Oxford University Press, United Kingdom, 2020).
Millions of children deprived of life-saving vaccinations amid COVID pandemic, misinformation surge | UN News. https://news.un.org/en/story/2023/04/1135797.
Infodemic. https://www.who.int/health-topics/infodemic.
Loomba, S., de Figueiredo, A., Piatek, S. J., de Graaf, K. & Larson, H. J. Measuring the impact of COVID-19 vaccine misinformation on vaccination intent in the UK and USA. Nat. Hum. Behav. 5, 337–348 (2021).
Wolfe, C. R. et al. Understanding the landscape of web-based medical misinformation about vaccination. Behav. Res Methods 55, 348–363 (2023).
Pierri, F. et al. Online misinformation is linked to early COVID-19 vaccination hesitancy and refusal. Sci. Rep. 12, 5966 (2022).
Neely, S. R., Eldredge, C., Ersing, R. & Remington, C. Vaccine Hesitancy and Exposure to Misinformation: A Survey Analysis. J. Gen. Intern Med 37, 179–187 (2022).
Lee, S. K., Sun, J., Jang, S. & Connelly, S. Misinformation of COVID-19 vaccines and vaccine hesitancy. Sci. Rep. 12, 13681 (2022).
de Saint Laurent, C., Murphy, G., Hegarty, K. & Greene, C. M. Measuring the effects of misinformation exposure and beliefs on behavioural intentions: A COVID-19 vaccination study. Cogn. Res Princ. Implic. 7, 87 (2022).
Allen, J., Watts, D. J. & Rand, D. G. Quantifying the impact of misinformation and vaccine-skeptical content on Facebook. Science 384, eadk3451 (2024).
Wilson, S. L. & Wiysonge, C. Social media and vaccine hesitancy. BMJ Glob. Health 5, e004206 (2020).
Suppli, C. H. et al. Decline in HPV-vaccination uptake in Denmark–the association between HPV-related media coverage and HPV-vaccination. BMC Public Health 18, 1360 (2018).
Brilli, Y., Lucifora, C., Russo, A. & Tonello, M. Influenza vaccination behavior and media reporting of adverse events. Health Policy 124, 1403–1411 (2020).
Rodrigues, F. et al. The Impact of Social Media on Vaccination: A Narrative Review. J. Korean Med Sci. 38, e326 (2023).
Health Organization, W. orld Vaccines against influenza: WHO position paper – May 2022. Wkly. Epidemiol. Rec. 97, 185–208 (2022).
Ma, R. et al. Seasonal influenza vaccination rate and vaccine effectiveness among older adults in mainland China: A systematic review and meta-analysis. Age Ageing 54, afaf252 (2025).
Wang, X. et al. Government livelihood project - free influenza vaccination for the older population in China: The example of Zhejiang Province. Front Public Health 13, 1571499 (2025).
Bernal, J. L., Cummins, S. & Gasparrini, A. Interrupted time series regression for the evaluation of public health interventions: a tutorial. Int J. Epidemiol. 46, 348–355 (2017).
Maquiling, A., Jeevakanthan, A. & Ho Mi Fane, B. The effect of vaccine mandate announcements on vaccine uptake in Canada: An interrupted time series analysis. Vaccine 41, 2932–2940 (2023).
Kumar, S., Shah, Z. & Garfield, S. Causes of vaccine hesitancy in adults for the influenza and COVID-19 vaccines: A systematic literature review. Vaccines (Basel) 10, 1518 (2022).
Schmid, P., Rauber, D., Betsch, C., Lidolt, G. & Denker, M.-L. Barriers of influenza vaccination intention and behavior – A systematic review of influenza vaccine hesitancy, 2005–2016. PLoS One 12, e0170550 (2017).
Mäki, K. O. et al. COVID-19 and influenza vaccine-hesitancy subgroups. PLoS One 19, e0308159 (2024).
Leuchter, R. K., Jackson, N. J., Mafi, J. N. & Sarkisian, C. A. Association between Covid-19 vaccination and influenza vaccination rates. N. Engl. J. Med. 386, 2531–2532 (2022).
Lunz Trujillo, K. et al. COVID-19 spillover effects onto general vaccine attitudes. Public Opin. Q 88, 97–122 (2024).
Luo, S. et al. Spillover between influenza and COVID-19 vaccination behaviors across pandemic phases and implications for general vaccine hesitancy. npj Vaccines 11, 70 (2026).
Rasul, M. E. & Ahmed, S. Not all conservatives are vaccine hesitant: examining the influence of misinformation exposure, political ideology, and flu vaccine acceptance on COVID-19 vaccine hesitancy. Vaccines (Basel) 11, 586 (2023).
Pineño, O. & Miller, R. R. Primacy and recency effects in extinction and latent inhibition: A selective review with implications for models of learning. Behav. Process. 69, 223–235 (2005).
Henne, P., Kulesza, A., Perez, K. & Houcek, A. Counterfactual thinking and recency effects in causal judgment. Cognition 212, 104708 (2021).
Pluviano, S., Watt, C. & Della Sala, S. Misinformation lingers in memory: Failure of three pro-vaccination strategies. PLoS One 12, e0181640 (2017).
Ecker, U. K. H., Sharkey, C. X. M. & Swire-Thompson, B. Correcting vaccine misinformation: A failure to replicate familiarity or fear-driven backfire effects. PLoS One 18, e0281140 (2023).
Lewandowsky, S., Ecker, U. K. H., Seifert, C. M., Schwarz, N. & Cook, J. Misinformation and its correction: Continued influence and successful debiasing. Psychol. Sci. Public Interest 13, 106–131 (2012).
Hou, Z. et al. Assessing the impact of the 2018 Changchun Changsheng vaccine incident on childhood vaccination in China. Commun. Med (Lond.) 3, 114 (2023).
Baik, Y., Bien-Gund, C. H., Bisson, G. P., Gross, R. & Fishman, J. Do some prefer to pay? Identifying bias against free COVID-19 tests. Public Health Pract. 7, 100483 (2024).
Wolinsky, A. Prices as signals of product quality. Rev. Econ. Stud. 50, 647–658 (1983).
Larson, H. J. et al. Measuring vaccine confidence: Analysis of data obtained by a media surveillance system used to analyse public concerns about vaccines. Lancet Infect. Dis. 13, 606–613 (2013).
Wood, R. M., Juanchich, M., Ramirez, M. & Zhang, S. Promoting COVID-19 vaccine confidence through public responses to misinformation: The joint influence of message source and message content. Soc. Sci. Med 324, 115863 (2023).
Turner, S. L. et al. Comparison of six statistical methods for interrupted time series studies: Empirical evaluation of 190 published series. BMC Med. Res. Methodol. 21, 134 (2021).
U.S. Department of Health and Human Services. Confronting Health Misinformation https://www.hhs.gov/surgeongeneral/reports-and-publications/health-misinformation/index.html (2021).
Mikati, I. K. E. et al. Defining misinformation and related terms in health-related literature: Scoping review. J. Med. Internet Res. 25, e45731 (2023).
Wardle, C. & Derakhshan, H. Information Disorder: Toward an Interdisciplinary Framework for Research and Policy Making https://edoc.coe.int/en/media/7495-information-disorder-toward-an-interdisciplinary-framework-for-research-and-policy-making.html (2017).
Baidu Index. https://index.baidu.com/v2/index.html#/.
COVID-19 Government Response Tracker | Blavatnik School of Government. https://www.bsg.ox.ac.uk/research/covid-19-government-response-tracker (2020).
National Centers for Environmental Information (NCEI). https://www.ncei.noaa.gov/.
Chen, J. & Roth, J. Logs with Zeros? Some Problems and Solutions*. https://doi.org/10.1093/qje/qjad054.
Martín-Fernández, J. A., Barceló-Vidal, C. & Pawlowsky-Glahn, V. Dealing with zeros and missing values in compositional data sets using nonparametric imputation. Math. Geol. 35, 253–278 (2003).
Acknowledgements
We appreciate all healthcare workers for their contributions to the study's data collection. This work was supported by the Shanghai Municipal Science and Technology Major Project (ZD2021CY001), the Disease Prevention and Control Innovation Team of Zhejiang Province (2026JKC-01), and the Shanghai Institute for Mathematics and Interdisciplinary Sciences (SIMIS, SIMIS-ID-2025-NW).
Author information
Author notes
These authors contributed equally: Xiangting Zhang, Xuan Deng.
Authors and Affiliations
Shanghai Institute of Infectious Disease and Biosecurity, Fudan University, Shanghai, China
Xiangting Zhang, Yihan Lu & Ye Yao
Department of Epidemiology, Ministry of Education Key Laboratory of Public Health Safety, School of Public Health, Fudan University, Shanghai, China
Xiangting Zhang & Yihan Lu
Department of Immunisation Programme, Zhejiang Provincial Center for Disease Control and Prevention, Hangzhou, China
Xuan Deng, Yuxia Liang, Hanqing He, Ying Wang, Yingping Chen & Shenyu Wang
Zhejiang Key Lab of Vaccine, Infectious Disease Prevention and Control, Zhejiang Provincial Center for Disease Control and Prevention, Hangzhou, China
Shenyu Wang
Department of Biostatistics, Ministry of Education Key Laboratory of Public Health Safety, School of Public Health, Fudan University, Shanghai, China
Ye Yao
Authors
- Xiangting Zhang
- Xuan Deng
- Yuxia Liang
- Hanqing He
- Ying Wang
- Yingping Chen
- Shenyu Wang
- Yihan Lu
- Ye Yao
Contributions
X.D. and Y.Y. conceived of this study. Y.L., H.H., Y.W. and X.Z. designed the study and collected the data. X.Z. analyzed data. X.Z., X.D. and Y.C. drafted the initial manuscript. Y.Y., Y.L. and S.W. critically revised the manuscript. All authors reviewed the final manuscript.
Corresponding authors
Correspondence to Shenyu Wang, Yihan Lu or Ye Yao.
Ethics declarations
Competing interests
Y.L. received the grant from MSD China (4-valent and 9-valent HPV vaccines licensed in China’s mainland) for research in modelling and health economics of HPV vaccination in adolescent girls; and consultation fee from Walwax Biotechnology Co., Ltd. (2-valent HPV vaccine licensed in China’s mainland, and a 9-valent HPV vaccine candidate in the clinical trial) and Xiamen Innovax Biotech Co., Ltd (2-valent and 9-valent HPV vaccines licensed in China’s mainland).
Additional information
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Zhang, X., Deng, X., Liang, Y. et al. Associations between vaccine misinformation and influenza vaccine uptake: a population-based interrupted time-series study in China. npj Digit. Public Health 1, 40 (2026). https://doi.org/10.1038/s44482-026-00047-1
Received: 28 April 2026
Accepted: 24 August 2026
Published: 05 October 2026
Version of record: 05 October 2026
DOI: https://doi.org/10.1038/s44482-026-00047-1
来源:Nature 健康护理主题 · nature.com