This study used a repeated cross-sectional design, conducting independent surveys annually from 2019 to 2024. Given the observational, repeated cross-sectional design without a control group, we interpret our findings as population-level associations over time, not causal effects.

The surveys were conducted in Zhejiang Province, a major coastal province in eastern China with a diverse population of over 64 million. The formal survey was conducted annually between August 1 and October 31 of each year from 2019 to 2024. Recruitment and data collection occurred simultaneously during this period. The surveys were carried out within 30 representative counties or districts surveillance sites across the province [19]. These sites were selected using stratified cluster random sampling based on per capita gross domestic product (tertiles) to ensure representativeness. The methods for sampling, sample size determination, and data collection were consistent with those described in our previously published two-round cross-sectional study using the 2019 and 2022 data from the same surveillance system [17]. This study extends this work by including four additional survey rounds (2020, 2021, 2023, and 2024) to cover the full pandemic cycle from prepandemic to postpandemic periods.

The primary outcome was adequate IDSHL, defined as a score of ≥12 on the IDSHL scale (see Instrumentation; Multimedia Appendix 1). The main exposure of interest was survey year (2019‐2024), which served as a proxy for the pandemic phase (prepandemic, acute pandemic, and postpandemic). Covariates included sex, age group (15-29, 30-39, 40-49, 50-59, and 60-69), ethnicity (Han vs minority), education level (primary school or lower, middle school, high school, technical school or college, undergraduate or higher), marital status (married vs unmarried, divorced, or widowed), occupation (farmer, worker, agency or institutional personnel, student, and other), and urban or rural residence. All covariates were self-reported and treated as potential confounders in multivariable analyses.

Data were collected via household visits using tablet computers. Prior to household visits, community workers or community health physicians (investigators) contacted the sampled household head by telephone, informed them of the selected family member, and scheduled a face-to-face visit. If the appointment failed, another was scheduled. If three consecutive attempts to schedule a successful visit failed, the household was replaced with the next household on the sampling list, and the recruitment process was restarted. Prior to the formal survey, both the surveyor and the quality control supervisor verified whether the participating family member matched the randomly selected subject. If not, the correct member was requested. If the selected member was unavailable, a new appointment was scheduled via telephone.

Questionnaire administration followed 2 modes: if the respondent was capable of self-completion, the surveyor remained present to provide assistance and correct obvious errors in basic information. If the respondent could not complete the questionnaire independently, the surveyor administered it through face-to-face interviewing.

To ensure data authenticity and quality, the following measures were implemented.

IDSHL was assessed using a validated 12-item subscale from the National Health Literacy Surveillance Survey in China [17]. The scale covers 3 dimensions: knowledge (6 items), behavior (4 items), and skills (2 items). It includes one true or false item, 8 single-choice items, and 3 multiple-choice items. Correct answers to multiple-choice questions were awarded 2 points, while single-choice and true or false items received 1 point each, for a total score of 15. Participants scoring ≥12 points (≥80%) were classified as having “adequate IDSHL” [19].

This cutoff has been widely used in China’s National Health Literacy Surveillance System and is recommended in the official surveillance guidelines [17].

Given the observational, cross-sectional survey design, no formal blinding or masking procedures were implemented. Surveyors were trained to administer the questionnaire in a standardized manner without revealing the study hypotheses or influencing responses. Participants were not informed of specific research hypotheses beyond the general purpose of the health literacy survey. No experimental manipulation or group assignment occurred, so masking of condition assignments was not applicable.

All 12 items demonstrated satisfactory content validity, with indices greater than 0.8. The overall internal consistency of the scale was acceptable (Cronbach α=0.67). For the 3 subscales, Cronbach α values were 0.71 for knowledge, 0.65 for behavior, and 0.63 for skills. Given the scale’s multidimensionality and diverse item content (covering multiple diseases and response formats), these coefficients are considered adequate for population-level surveillance [17]. Reliability estimates are based on the current sample (N=112,917). No additional validity coefficients from external samples are reported, as the scale is a national surveillance instrument with established content validity.

This was a nonexperimental (observational) design with no manipulated conditions. The study used a repeated cross-sectional design (six annual waves, 2019‐2024). No randomization or intervention assignment occurred. The primary comparison was across time (survey year) as a proxy for pandemic phases.

Planned data diagnostics included the following prespecified criteria: (1) post data collection exclusion of participants was based solely on the 3 quality control criteria (leading questions, duration <10 min, and photo inconsistency); no additional exclusion criteria were applied. (2) Missing data: less than 2% of sampled households were replaced due to quality issues; no imputation was performed for missing item responses as the questionnaire was reviewed for completeness on-site. We conducted Little Missing Completely At Random (MCAR) test on the full dataset (N=112,917) using full information maximum likelihood. The result was nonsignificant (χ²₁₄₅=156.73; P=.24), indicating that missing data (at the household replacement level) were missing completely at random (degrees of freedom derived from the number of variables and observed missing patterns). (3) Outliers: no statistical outlier removal was performed; all valid responses were retained. (4) Data distributions: normality of IDSHL scores was assessed visually using histograms and Q-Q plots; for nonnormal distributions, nonparametric tests (Kruskal-Wallis) were used as specified. (5) Data transformations: no data transformations were applied.

Data were analyzed using SPSS (version 27.0; IBM Corp) and Joinpoint Regression Program (version 5.4.0; National Cancer Institute). Descriptive statistics were presented as frequencies (percentages) for categorical variables and means (SDs) for continuous variables. No qualitative analysis was performed.

A total of 112,917 valid questionnaires were analyzed across 6 years. Figure 1 presents the participant flow diagram, showing the number of households contacted, eligible, and finally included for each survey year. Across the 6 annual surveys (2019‐2024), the number of valid participants ranged from 17,131 (2021) to 19,257 (2019). Replacement of invalid questionnaires due to quality issues (leading questions, duration<10 min, and photo inconsistency) occurred in less than 2% of sampled households per year; no further exclusions were applied.

Recruitment and data collection occurred annually between August 1 and October 31 of each year from 2019 to 2024, as described in the Setting subsection.

Table 1 presents the detailed sociodemographic characteristics of participants across the six annual surveys (2019‐2024). Significant differences were observed across years in sex, age, education, and occupation distributions (all P<.001), highlighting the need for weighting. Weighting was applied to all subsequent analyses.

This 6-year study, leveraging repeated cross-sectional data from before, during, and after the COVID-19 pandemic, demonstrates that the pandemic period was temporally associated with a significant and sustained population-level elevation in IDSHL among residents of Zhejiang Province, China. We observed a sustained population-level elevation in IDSHL, with the lowest likelihood of adequate literacy occurring in the prepandemic year (2019) and the highest in the postpandemic year (2024). The most rapid gains were temporally associated with the acute pandemic phase (2019‐2021), after which the increase decelerated. The temporal pattern showed a sharper increase during the acute pandemic phase (2019‐2021) followed by a slower increase thereafter. Improvements were domain- and disease-specific: the largest gains occurred for pandemic-salient behaviors (eg, respiratory etiquette), while nonrespiratory diseases such as hepatitis B exhibited more complex, fluctuating patterns. Persistent disparities by age, education, and ethnicity indicated that the benefits associated with the pandemic were not uniformly distributed. Crucially, our observational design cannot establish causality; all interpretations refer to population-level temporal associations.

Our findings align with previous research showing reduced incidence of other infectious diseases during the pandemic [16] and improvements in respiratory-related health literacy [14,21]. The sharp increase in adequate IDSHL during 2019‐2021 coincides with the period of most intensive public health messaging, stringent nonpharmaceutical interventions, and peak societal attention to infectious disease prevention [12]. This pattern is consistent with evidence that large-scale health communication campaigns can be temporally associated with rapid improvements in targeted knowledge and behaviors [22]. The subsequent deceleration after 2021, as the pandemic transitioned to endemic management, suggests that crisis-driven momentum is not self-sustaining—a temporal pattern not previously documented in the health literacy literature. The U-shaped pattern for hepatitis B knowledge is suggestive of a potential “crowding-out” effect described in prior literature [17,23]. However, unlike earlier work, our six-wave design revealed that not all non–COVID-19 content stagnated; items with stable, long-standing messaging (eg, vaccination contraindications) improved steadily [24].

The dramatic increase in correct responses for respiratory etiquette, a direct target of COVID-19 messaging, illustrates the focused impact of crisis-driven education [25]. In contrast, the U-shaped pattern for hepatitis B knowledge implies that intense focus on a novel pathogen may temporarily divert public attention from endemic threats, though knowledge may recover as the crisis abates [26]. Notably, items unrelated to respiratory illnesses but with consistent messaging (eg, vaccination contraindications) improved steadily, indicating that generalizable knowledge can still accrue during a pandemic. These nuanced patterns highlight a key limitation of crisis-driven education: it powerfully enhances literacy in highlighted areas but does not automatically generalize across all infectious diseases.

Despite overall improvements, preexisting disparities by age, education, and ethnicity persisted. This finding echoes evidence that health literacy initiatives often benefit more advantaged groups to a greater extent [27,28]. The pandemic’s heavy reliance on digital platforms (eg, WeChat, news apps) for information dissemination may have exacerbated these gaps, as older adults and less-educated groups face documented barriers to digital access and literacy [29]. Although this study did not directly measure digital health literacy, prior literature has established its mediating role [30]. Therefore, postpandemic strategies should consider addressing digital divides through age-friendly interfaces, community-based outreach, and tailored messaging. Our data do not directly examine causal mechanisms, but personal experience of COVID-19 infection, including long COVID, could have contributed to knowledge and behavior change [31], a hypothesis warranting future investigation

The study was conducted in a single developed Chinese province (Zhejiang) with a population of over 64 million, which limits direct generalizability to other regions or countries with different pandemic responses, health systems, or cultural contexts. However, the sampling strategy (multistage stratified random sampling, weighting to census) supports generalizability within Zhejiang Province. The findings may be most applicable to other economically developed regions in China or similar settings with intensive digital information environments. Ecological validity is supported by the real-world pandemic context and the use of routine surveillance methods.

Several limitations should be considered. First, the repeated cross-sectional design captures population trends but cannot follow individuals over time or establish causality. Unmeasured confounding factors (eg, other policy changes and media coverage) may have influenced IDSHL. Second, the study was conducted in a single developed Chinese province, limiting generalizability to other regions or countries with different pandemic responses and health systems. Third, the IDSHL scale primarily measures functional literacy and does not assess critical or interactive dimensions, nor digital health literacy directly. Fourth, self-reported data may be subject to social desirability bias, particularly during the pandemic. Fifth, while the Cronbach α values for the behavior (0.65) and skills (0.63) subscales are within the acceptable range for population surveillance given the scale’s multidimensionality [32], they are modest; observed changes in these domains should therefore be interpreted with some caution. Finally, the weighting procedure reduces but does not eliminate potential bias from sample composition changes.

The findings carry several practical implications. First, public health authorities should leverage the postpandemic elevated baseline to broaden comprehensive infectious disease education, ensuring that respiratory-focused gains do not come at the expense of other threats (eg, hepatitis B and rabies). Second, targeted interventions are needed to reach older adults, less-educated populations, and ethnic minorities using hybrid (digital + traditional) channels. For example, in Zhejiang, community health workers could deliver face-to-face sessions on nonrespiratory diseases in rural senior centers, complemented by WeChat mini-programs with large-font, audio-assisted content, and monthly live question-and-answer sessions via local health official accounts. Third, routine surveillance of IDSHL should be sustained to detect declines in nonpandemic disease knowledge. By learning from the patterns observed during this global health crisis, short-term crisis responses can be translated into long-term, equitable public health preparedness [33].

This 6-year study offers the first population-level evidence spanning the full COVID-19 pandemic cycle that a major public health crisis can be temporally associated with substantial and sustained gains in IDSHL. However, the observed improvements were not self-sustaining: they accelerated during the acute crisis phase and decelerated thereafter, and they did not automatically generalize to nonrespiratory diseases. Moreover, preexisting disparities by age, education, and ethnicity persisted. These findings underscore a core tension: crises can rapidly elevate targeted literacy, but maintaining and broadening those gains requires deliberate, ongoing investment. Policymakers should therefore view crisis-driven literacy improvements as a foundation for, not a replacement of, routine health education and equity-focused communication strategies.