We first described the daily cases reported 1 year following the transition from emergency response to long-term COVID-19 disease management. The dataset was used as the benchmark for correlation analysis with the results from 4 surveillance systems, including a hospital surveillance system, a wastewater surveillance system, a meteorological surveillance system, and an internet search engine system (Table S1 in Multimedia Appendix 1).

We integrated traditional public health surveillance data (eg, laboratory confirmations and clinical reports) with novel digital proxies and contextual environmental data. While clinical reports and wastewater monitoring were direct components of public health surveillance systems, internet search metrics, such as Baidu Search Index (BSI), were used as digital syndromic surveillance tools that proxy population-level health-seeking behavior. Meteorological factors were included as contextual confounding variables to control for their potential influence on both virus transmission and reporting behavior.

Data were initially sourced from the National Notifiable Disease Reporting Information System (NDRIS). Following the 2003 SARS outbreak, China established the web-based NDRIS for quick case identification and response [20]. Since 2008, all notifiable infectious disease data (including COVID-19 since January 2020) have been electronically reported to the China CDC within 24 hours. The system covered all 31 mainland provinces, with individual case data standardized by the National Health Commission’s guidelines. We analyzed COVID-19 confirmed cases reported by medical institutions via the NDRIS from April 1, 2023, to June 30, 2024. All individual case details were de-identified.

Our analysis focused on the comparison between April 1, 2023, and June 30, 2024. This timeframe centered on China’s policy shift in COVID-19 management, transitioning toward routine prevention and control measures. By the end of the study period, SARS-CoV-2 showed patterns consistent with endemic circulation in the study region. Nevertheless, episodic local epidemic waves continued to occur and formed the focus of our analysis. The surveillance systems we selected for this study have implemented standardized data collection and reporting protocols. The discrepancy in the time frames arose because the meteorological surveillance system covered a slightly extended period (March 1, 2023, to June 30, 2024) to ensure comprehensive monitoring and to capture potential early signals of disease activity. All time series were aligned by calendar date. We assessed data availability for each surveillance stream across the entire study period and report exact sample sizes and completeness in Table S2 in Multimedia Appendix 1 (data availability and completeness). Reported cases had 48 of 60 (10.50%) days of missing data, and missing values were imputed using linear interpolation prior to analyses.

From April 1, 2023, to June 30, 2024, a cumulative total of 62,236 cases were reported, predominantly characterized by sporadic occurrences. The first peak occurred in 2023 from late April to the end of June. Since the beginning of 2024, the number of reported cases has consistently remained low. February 18, 2024, the first working day after the Spring Festival holiday, marked the onset of an upward trend in reported cases. The second peak was reached between March and June, followed by a fluctuating downward trend, as illustrated in Multimedia Appendix 3.

Starting from January 8, 2023, the daily patient volume at fever clinics in medical institutions began to fluctuate and decrease, exhibiting a pattern of varying changes. It reached 3 peaks in March, May, and December of 2023. Since November 20, 2023, the patient volume had rapidly increased, reaching its peak on December 5, 2023, and remaining high for 52 days. The characteristics of low-level fluctuations were observed since February 2024. Peaks in the positivity rate of NAAT occurred from late April to late June and from July to mid-October in 2023. In January 2024, the positivity rate dropped to around 2% and showed a fluctuating upward trend, reaching its peak in late March (Multimedia Appendix 3). Wastewater SARS-CoV-2 detection rates declined early in 2023 before rising significantly post April 20, 2023, remaining elevated until June 2023 before trending downward with fluctuations. In 2024, from May 27 to June 3, the positive detection rate of sewage experienced a significant fluctuation and increase, with a weekly average increase of more than 30%, reaching the alert level. From June 6 to June 10, the overall positive detection rate fluctuated at a high level. From June 13 to June 24, it showed a trend of fluctuating decrease. The virus concentration followed a similar pattern (Multimedia Appendix 3). Daily weather variables and COVID-19–related BSI trends were detailed in Multimedia Appendix 3.

Table 1 shows the correlations between all surveillance datasets and the reported cases throughout the entire monitoring period (from April 1, 2023, to June 30, 2024). Most of the correlations were significant (P<.05), indicating evidence of relationships. Among them, the positive rate of NAAT showed a strong correlation (ρ=0.834, 95% CI 0.803‐0.860). The 4 variables of the wastewater surveillance system also exhibited a moderate correlation (all ρ>0.5; all P<.001). The correlation results of meteorological surveillance data revealed a moderate positive correlation between ambient temperature (ρ=0.346, 95% CI 0.263‐0.425), absolute humidity (ρ=0.330, 95 %CI 0.245‐0.409), and COVID-19–reported cases; a low positive correlation was observed between visibility and COVID-19–reported cases (ρ=0.179, 95% CI 0.089‐0.226). Atmospheric pressure exhibited a moderate negative correlation (ρ=−0.322, 95% CI –0.402 to –0.237).

To evaluate time-series correlation with 60-day moving windows, median Spearman values and the IQR of Spearman estimates were used for comparison with the reported cases (Figure 1). The wastewater data showed a positive correlation with reported cases, but with significant variability. Among the sewage monitoring indicators, N gene concentration was the most consistently and strongly correlated indicator, with 60.8% of rolling windows showing ρ>0.5 (median 0.585, IQR 0.214‐0.766) and the N gene positivity rate in 53.3% of rolling windows (median 0.530, IQR 0.222‐0.742). ORF1ab gene measures were slightly weaker but still relatively stable (ORF1ab gene concentration: 52%; ORF1ab gene positive rate: 47.2%). Among the hospital surveillance variables, the daily NAAT positive rate (positive_rate) was consistently strongly correlated indicator, with 58% of rolling windows showing ρ>0.5 (median 0.534, IQR 0.394‐0.724). The correlation between the number of fever clinic consultations and reported cases showed substantial temporal variability (27.9% of ρ>0.5 of rolling windows). The correlation approached zero around September 2023 and moderate negativity from October 2023 to March 2024. The median correlation coefficients between meteorological surveillance data and reported cases were mostly close to 0. Several BSIs showed meaningful associations: BSI6 (44.5%) and BSI7 (30.2%) had the highest shares among the BSI series, while other BSI terms declined in stability (BSI2 19.6%, BSI8 18.3%, BSI3 7.3%, BSI10 3.5%, BSI9 2.3%, BSI1 0.8%, BSI4 and BSI5 0%). The keyword-based BSI for name-related queries (eg, “COVID-19”) demonstrated stronger and more consistent correlations than symptom-related queries (median 0.475, IQR 0.192‐0.685). Median correlation coefficients for meteorological variables were generally close to zero, and their stability was low: temperature (Tmean) 18.8% of ρ>0.5 of rolling windows, absolute humidity (AHmean) 17.8%, visibility (VISmean) 1.8%, wind speed (WSmean), and precipitation (PRCPmean) 0.0%. These findings indicated the limited and inconsistent predictive value of meteorological variables when considered alone (Multimedia Appendix 4).

Correlation analyses showed the same association results with the lag (–7 to +7 d). Notably, the weather variables from the meteorological surveillance system demonstrated a stronger correlation coefficient with a longer lag period, indicating a stronger correlation with the reported cases, as detailed in Figure 2 and Multimedia Appendix 4.

To evaluate the predictive use of the selected surveillance indicators, we analyzed their association with reported case counts across a range of lead times from 0 to 7 days. The results of time-series regression analyses are summarized below.

DLNM analysis with a 7-day lag was to explore the association between each meteorological factor and the reported cases, and the results were consistent with the correlation analyses, as detailed in Figure 3.

The timeline plots revealed key insights into the estimated effect sizes for each standardized variable derived from quasi-Poisson regression models (Figure 4). Most surveillance variables demonstrated their strongest associations at a lead time of 0 days, indicating a high degree of contemporaneous alignment with case reporting. The analysis identified N gene positive rate of wastewater surveillance as the most consistent leading indicator (maximum effect size: β = 1.023, 95% CI 0.931‐1.115 at lag 0 d), followed by the positivity rate of NAAT (maximum effect size: β = .819, 95% CI 0.768‐0.870 at lag 0 d), supporting its potential use in contemporaneous alignment with short lead potential. Importantly, comparisons of effect estimates before and after adjusting for meteorological confounders (eg, Tmean and AHmean) showed that relationships for certain variables, particularly internet search indicators, were significant. This highlighted the critical role of controlling for meteorological confounding in deriving unbiased effect estimates, as detailed in Multimedia Appendix 5.

Figure 5 summarizes the interaction coefficients between meteorological factors and multiple surveillance indicators over lead times of 0 to −7 days. Overall, Tmean and AHmean exhibited the most frequent and pronounced modifications of surveillance signals, whereas Pmean and VISmean showed weaker and less consistent interactions. Interaction effects were temporally heterogeneous: many significant interactions clustered at short negative lags (typically lag −1 to lag −3), indicating that recent meteorological conditions can modify surveillance-case associations within a few days. Laboratory- and wastewater-based indicators (NAAT positive rate, N gene and ORF1ab gene positivity, and concentration) displayed relatively few significant interactions overall, suggesting that these signals were robust to short-term meteorological variation. By contrast, internet-search indicators (BSIs) showed more numerous and larger interaction coefficients, with several positive interactions with Tmean and negative interactions with AHmean reaching statistical significance (P<.05), as detailed in Multimedia Appendix 5.

In this city-level study spanning April 1, 2023, to June 30, 2024, we compared multiple surveillance streams and found that no single indicator was sufficient to capture the full dynamics of COVID-19. Rather, a complementary, multichannel approach—combining hospital, wastewater, digital data with meteorological variables—provided the most robust situational awareness and generally showed contemporaneous alignment with short lead potential up to 7 days. Integrating these streams allowed us to exploit their distinct temporal properties and public health uses, supporting timely detection, interpretation, and response to evolving transmission patterns.

Wastewater surveillance emerged as a particularly valuable early signal. Our time-series models, which controlled for key meteorological confounders, demonstrated that N gene concentration and positivity rate were significant predictors of reported cases, with a lead time of 0‐7 days. These associations were relatively robust to short-term weather variation, consistent with wastewater’s capacity to capture community-level viral circulation, including asymptomatic and presymptomatic infections, before clinical case ascertainment [25-27]. While its rolling correlation exhibited variation (N gene concentration, median 0.585, IQR: 0.214‐0.766), this likely reflected its sensitivity to rapid shifts in transmission dynamics that case-based reporting only later captured. Our study also indicated that the wastewater monitoring of SARS-CoV-2 positivity rate and concentration exhibited moderate correlations with reported cases (all ρ>0.5).

Hospital-based NAAT positive rate from sentinel hospitals proved to be the most stable and reliable indicator for real-time tracking (median 0.534, IQR 0.394‐0.724; 58% of rolling windows ρ>0.5), showing peak association at lag 0 days. The substantial IQR of correlation coefficients suggested a lack of robustness in the associations between most variables and reported cases, indicating significant fluctuation in the correlation intensity over time. The high sensitivity of clinic-based NAAT testing highlighted its critical role in identifying infected individuals and accurately tracing the trajectory of the pandemic [28,29]. Our results indicated that the correlation between fever clinic visits and reported cases exhibited significant fluctuation over time (IQR –0.285 to 0.531). As the pandemic progresses and vaccination campaigns are rolled out, the symptoms of COVID-19 infections have diverged from the peak pandemic period [30,31]. In addition, in the postpandemic era, the prevalence of influenza may also impact the number of consultations for patients with COVID-19 [32].

Notably, our study added to the evidence on digital surveillance systems by specifically evaluating China’s Baidu Index data. Unlike some prior studies based on other platforms, which showed limited correlation with case counts (ρ=0.04‐0.08) [33], our analysis revealed that Baidu search metrics for COVID-19–related terms exhibited meaningful correlations with reported cases. Using a validated 7-day moving window [34], digital surveillance data provided a proxy for public awareness and health-seeking behavior, offering real-time insight into public perception and sentiment during outbreaks [35,36]. Several terms demonstrated intermediate temporal stability over a 60-day window (eg, BSI6 44.5% and BSI7 30.2%), supporting their use as indicators of behavioral intent. However, their temporal variability meant that they might be interpreted alongside clinical and wastewater data to mitigate false positives from media-driven or non–outbreak-related trends. Digital surveillance from Baidu showed heterogeneous performance and should primarily be treated as a behavioral and awareness signal rather than a direct proxy for case burden. Name-based queries (terms referring to COVID-19 or SARS-CoV-2) exhibited modest positive associations with reported cases, while symptom-based queries (terms for fever, cough, etc) were generally weaker and sometimes showed neutral or negative associations. Meteorological conditions and seasonal behaviors could modify these search patterns (eg, warmer weather or influenza seasonality could alter health-seeking and search behavior). Thus, digital indices were best seen as reflecting population concern and potential care-seeking. Consequently, we recommend using Baidu metrics as an early behavioral indicator. When corroborated by wastewater and NAAT data, they could help prioritize investigations, rather than being used to infer changes in actual community incidence.

Meteorological signals provide contextual information but do not demonstrate robust independent predictive power in multivariable models. Most studies have demonstrated negative correlations between temperature, humidity, and COVID-19 cases, but some have contested the extent of the influence of meteorological factors on the epidemic [37-41]. In this study, bivariate analyses showed mixed associations between meteorological variables and reported cases (positive associations with temperature, absolute humidity, visibility; negative association with air pressure), but these associations were substantially attenuated in multivariable time-series models that included surveillance streams. This aligned with the consensus that meteorological variables alone are insufficient to curb COVID-19 resurgence without concomitant public health interventions [42,43]. Meteorological indicators exhibited low rolling-window stability, with many showing a low proportion of windows where ρ exceeded 0.5. This implied that their primary use was more contextual, helping to interpret seasonal variations or behaviorally mediated changes, rather than functioning as reliable standalone predictors.

Data from diverse surveillance channels allows for scientific and rapid characterization of epidemiology, trend tracking, and outbreak prediction [44]. Drawing lessons from the COVID-19 pandemic, it is imperative to establish routine surveillance for infectious diseases. No single surveillance system can perfectly fulfill all information needs. Our findings supported a sequential, multilayered surveillance strategy: wastewater monitoring can serve as an early alert, digital indices can signal shifts in public concern and health-seeking that may precede care-seeking, and NAAT-based hospital surveillance provides confirmation and real-time tracking. Meteorological monitoring adds contextual information that can refine interpretation, particularly of behavioral signals, but is insufficient as a standalone predictor. During the year after policy relaxation, the city experienced 3 peaks in reported cases, with the largest wave from mid-February to late May 2024. By the end of the study period, SARS-CoV-2 showed patterns consistent with endemic circulation in the region, and integrated surveillance streams captured the rise and fall of transmission and informed timely public health action.

Our study had several limitations. First, taking COVID-19–reported cases as the benchmark might lead to an underestimation of the actual case count. In this study, we used the local CDC’s Information System for Disease Control and Prevention (CISDCP) to obtain the daily reported cases of COVID-19, which enabled us to gain the endemic situation to the greatest extent. However, since some COVID-19–infected individuals were asymptomatic and would not seek medical treatment without mandatory testing, our research might underestimate the actual number of infected people. Second, reported cases served as an operational benchmark but likely underestimated the actual incidence. Verification bias from variable care-seeking behavior and differential testing access (by time, locality, and population subgroup) might alter the relationship between true infections and reported cases. Such bias could affect correlations and lag estimates. To better address this, we recommend using a multichannel surveillance approach that combines wastewater and digital indicators, as well as representative serosurveys or adjustments for testing effort, to identify the trends. Nevertheless, it is imperative for future studies to incorporate hospitalization statistics to effectively and promptly monitor the severity of emerging infectious diseases. In addition, we conducted a short-term evaluation of the surveillance systems in only 1 city, and we did not consider the population heterogeneity and geographical heterogeneity. Future research should undertake more in-depth regional studies, adjust for covariates, and use multichannel surveillance systems to develop more accurate prediction models for infectious disease responses.

In the postpandemic era, effective endemic preparedness requires a hybrid surveillance architecture that combines traditional clinical testing with environmental and digital streams. Wastewater and NAAT surveillance together provide timely, complementary signals of community transmission, while syndromic and internet-derived indicators offer rapid insight into health care burden and public behavior. Meteorological data contribute important context, but are insufficient alone. The optimal mix of systems might vary depending on local resources, technological infrastructure, and cultural factors, but should invariably include multiple complementary data streams.

Future research should focus on integrating multichannel data, expanding geographic scope, and incorporating hospitalization and pathogen-specific signals to strengthen early warning systems and public health decision-making. The integrated strategy will be key to enhancing the effectiveness of public health responses to respiratory pathogens.