A cross-sectional study was conducted in Chengdu using an online questionnaire survey. Participants were recruited using multistage stratified random sampling. Our objective was to recruit patients who had developed ARI within the past 14 days and had recovered by the time of the survey. The questionnaire included information on participants’ sociodemographic characteristics, ARI severity, and health-seeking behaviors following ARI onset. Inclusion criteria were (1) individuals who developed ARI within the past 14 days and had fully recovered prior to the survey, (2) ability to understand the study procedures and sign the informed consent form (for minors and older adults, consent could be signed by their parents, guardians, or children), (3) ability to use the internet and mobile phones to complete the study content and data collection (children and older adults may have their parents, guardians, or children complete the forms on their behalf), and (4) residence in Chengdu for more than 6 months. The exclusion criterion was an incomplete questionnaire response.

Within this cross-sectional framework, we designed a detailed retrospective behavioral recall module in the questionnaire, asking participants to recall and report their main health-seeking behaviors after the onset of ARI (eg, searching for health information via the internet or media, medication purchase, hospital visit, and other health-seeking behaviors) and to record the specific day on which each behavior occurred until recovery. Thus, each participant provided a complete, time-ordered behavioral sequence—a longitudinal trajectory from onset to outcome.

Data were collected from January 16 to 28, 2025, across 6 county-level administrative districts in Chengdu. A total of 19,712 responses were collected during the study period. After applying exclusion criteria, 2340 participants were included in the analysis.

Given the self-reported nature of this survey, we developed and adopted a standardized operational definition of ARI, drawing on the case definition frameworks from the World Health Organization [13,14], the US Centers for Disease Control and Prevention [15], and the Expert Consensus on the Diagnosis, Treatment, and Infection Control of Acute Respiratory Infections in Adult Outpatient and Emergency Departments [16]. Specifically, ARI was defined as the presence of 1 systemic symptom (eg, fever, chills, headache, or body aches) and 1 respiratory symptom (eg, sore throat, cough, nasal congestion, or runny nose) within 14 days, or the presence of 2 or more respiratory symptoms. For children younger than 2 years or those unable to communicate verbally, ARI was defined as the presence of any one of the following within 14 days: fever (axillary temperature ≥37.0 °C), cough, or respiratory distress.

Health-seeking behaviors refer to the online or offline medical actions individuals take after developing ARI. These behaviors fall into 2 main categories. The first category involves actions that can intervene in disease progression, including self-initiated hospital visits and medication purchases (defined as the self-purchasing of medications). The second category involves digital health monitoring and information retrieval: searching for health information via the internet, media, or AI; consulting online or by telephone; self-checking using wearable devices; or at-home antigen testing. To simplify the Markov model, we consolidated the second category into a single state labeled “information seeking and self-monitoring (ISM).” This consolidation was justified because (1) functionally, all behaviors in this category served the same purpose of information acquisition and self-assessment, rather than representing distinct decision-making stages with different downstream consequences; and (2) analytically, the central question of this study was whether individuals experienced an information-seeking and self-monitoring stage, rather than exhaustively mapping all possible sequences of digital health actions. Within this framework, the first occurrence of any behavior in the second category—along with its timing—was retained and used to define entry into this state. Any subsequent behaviors within the same category, if present, were considered repeated manifestations of the same functional stage and were not treated as separate state transitions to avoid overparameterization. Importantly, the model preserved all transitions between ISM and other behaviors (medication purchase or hospital visit), irrespective of which occurred first.

This study used latent class analysis (LCA) to identify the severity of ARI. The number of symptoms and the duration of symptoms were categorized as categorical variables. Combined with absence status, these 3 categorical variables served as manifest variables to construct an LCA model with 1 to 4 latent categories. The optimal number of categories was selected based on the Akaike information criterion (AIC), Bayesian information criterion (BIC), likelihood ratio test (LRT), and minimum category probability. AIC and BIC served as measures of model fit and complexity, where lower values indicated better performance [17]. The LRT was used to assess whether adding a latent category significantly improved model fit compared with a model with one fewer category [18]. The minimum category probability had to exceed 0.10 to prevent model overfitting that could lead to excessive identification of latent classifications within the population [19]. After determining the optimal number of latent categories, each individual was assigned the ARI severity level corresponding to their most probable category membership.

The baseline characteristics of the participants were summarized using standard descriptive statistics. Categorical variables were presented as frequencies and percentages, while continuous variables were presented as the median and IQR. Where appropriate, comparisons were performed using the Kruskal-Wallis test, Mann-Whitney U test, Pearson chi-square test, or Fisher exact test.

An MSM model was constructed to simulate the changes in the sequence of health-seeking behaviors from the onset of ARI to recovery. A first-order time-homogeneous MSM model was used in this study, which assumed that transition intensities between states did not vary over time and depended only on the individual’s current state, not on the historical path by which they arrived at that state. For the MSM models, we defined 4 possible states through which an individual could move (illness, ISM, medicine, and hospital) and one absorbing state (recovery). We defined ARI as entering the “illness” state, with day 0 corresponding to illness onset for each individual. Subsequently, any of the other 3 behaviors might or might not occur. To further simplify the model, we treated “hospital” as the terminal state of the health-seeking pathway and did not track subsequent behaviors after “hospital.” However, transitions from “hospital” to “recovery” (as the absorbing state) were allowed, focusing instead on behavioral changes prior to seeking medical care. MSM models incorporating single covariates (age, sex, urban-rural status, mobile phone use, and severity of ARI) were constructed separately, with significant factors subsequently integrated into a multivariate MSM model. Subgroup analyses of the MSM models containing multiple covariates were performed by age group and urban-rural area. The 14-day transition probabilities and transition intensity matrices between states in the overall and subgroup multivariate MSM models, along with the hazard ratios (HRs) and 95% CIs of the overall multivariate model, were extracted.

The transition probabilities for days 1 through 14 were then entered as the dependent variable into beta regression models (which are appropriate for outcomes bounded within the [0, 1] interval [20]) to examine temporal trends within urban-rural and age subgroups. Three models were fitted: a main effects-only model, a main effects plus time variable model, and a model incorporating the interaction between main effects and time variables. Optimal model selection was based on a comprehensive assessment of AIC values, and coefficients were log-transformed to derive odds ratios (ORs) with 95% CIs.

Two-tailed tests were performed, and P value <.05 was considered statistically significant. All statistical analyses were performed using the “msm,” “poLCA,” and “betareg” packages in R software (version 4.5.1; R Foundation for Statistical Computing).

The study was approved by the Ethics Committee of the Chinese Academy of Medical Sciences and Peking Union Medical College (approval CAMS&PUMC-IEC-2025‐010) and the Chengdu Center for Disease Control and Prevention (Chengdu Institute of Health Supervision; approval 2025003) and was conducted in accordance with the principles of the Declaration of Helsinki. Participants were required to read and agree to the informed consent form online before beginning the questionnaire. No identification of individual participants in any images of the manuscript or appendices is possible.

The participant selection process is depicted in Figure 1. Of the 2340 participants with ARI included in the analysis, 487 (20.8%), 1592 (68%), and 261 (11.2%) were aged 0 to 17 years, 18 to 59 years, and ≥60 years, respectively, and 1828 (78.1%) and 512 (21.9%) lived in urban and rural areas, respectively (Table 1). On the basis of LCA model fit indices and interpretability (Table S1 in Multimedia Appendix 1), the 2-class model was optimal for ARI severity. This model identified 976 (41.7%) participants in the moderate to severe group and 1364 (58.3%) participants in the mild group. The moderate to severe group was characterized by a high probability of 3 to 5 symptoms (0.60) or ≥6 symptoms (0.36), prolonged symptom duration ≥7 days (0.67), and some absenteeism (0.24). In contrast, the mild group was characterized by a high probability of having only 2 symptoms (0.60), shorter symptom duration of 2 to 6 days (0.62), and minimal absenteeism (0.07). The full conditional probabilities defining these 2 classes were presented in Table S2 in Multimedia Appendix 1. Regarding health-seeking behaviors, significant differences among the 3 age groups were observed in ISM-based behaviors, purchasing medicine, and hospital visits, while differences between urban and rural groups were primarily evident in the latter two behaviors (P<.05; Table 1)

Figure 2 illustrates the intensities of state transitions for the total population, urban and rural populations, and age-specific cohorts calculated using the multivariate MSM model. The effects of covariates on the HR of transitions between different states in the total population are presented in Table S3 in Multimedia Appendix 1. Among the total population (Figure 2A), the intensity from illness to medicine use was highest (0.35, 95% CI 0.34‐0.37), while the intensity from illness to recovery was only 0.19 (95% CI 0.18‐0.20). The intensity of transition from illness to ISM among urban populations (0.21, 95% CI 0.20‐0.23) was close to that of rural populations (0.19, 95% CI 0.16‐0.22). However, the transition intensity from illness to hospital visit was 1.35 times higher among rural populations than among urban populations (0.31 vs 0.23; Figures 2B and 2C). The intensity of transition from ISM to medicine use among urban populations (0.12, 95% CI 0.08‐0.15) was 2.4 times that among rural populations (0.05, 95% CI 0.01‐0.07; Figures 2B and 2C). Similarly, the intensity of transition from ISM to medicine use among individuals aged 18 to 59 years (0.12, 95% CI 0.10‐0.14) was 2.4 times higher than that among those aged 0 to 17 years (0.05, 95% CI 0.01‐0.10) and 12 times higher than that among individuals aged 60 years and older (0.01, 95% CI 0‐0.03; Figures 2D, 2E, and 2F). The intensity of transition from illness to ISM among individuals aged 18 to 59 years (0.26, 95% CI 0.24‐0.27) was 2 times higher than that among those aged 0 to 17 years (0.13, 95% CI 0.10‐0.16) and 2.6 times higher than that among individuals aged 60 years and older (0.10, 95% CI 0.06‐0.13; Figures 2D, 2E, and 2F). The conversion intensity from illness to hospital visit for individuals aged 18 to 59 years (0.22, 95% CI 0.20‐0.23) was 0.63 times that of individuals aged 0 to 17 years (0.35, 95% CI 0.32‐0.39) and 0.76 times that of individuals aged 60 years and older (0.29, 95% CI 0.24‐0.35; Figures 2D, 2E, and 2F).

During the observation period from day 1 to day 14, the transition probabilities from the illness state were estimated and plotted in Figure 3. We used a beta regression model to examine transition probabilities over time within urban and rural areas and age subgroups. On the basis of AIC, we selected the optimal model incorporating main effects and time variables, with specific ORs (95% CIs) presented in Table S4 in Multimedia Appendix 1. The beta regression indicated that transition probabilities changed over time (adjusted P<.05). The transition probability showed an initial upward trend followed by a decline, with behavioral transitions primarily occurring during the early stages of illness (Figure 3). The transition probability from illness to ISM was significantly lower in both the 0- to 17-year and the ≥60-year age groups compared with the 18- to 59-year age group (both adjusted P<.001; Figure 3A). The transition probability from illness to medicine use was significantly lower in the rural group than in the urban group (P=.03; Figure 3B). The transition probability from illness to hospital visit was significantly higher in the 0- to 17-year age group than in the 18- to 59-year age group (adjusted P<.001), and it was also higher in the rural group than in the urban group (P=.048; Figure 3C).

We estimated the transition probability from ISM over days 1 to 14 and plotted it in Figure 4, also using a beta regression model for analysis. Results indicated that the transition probability changed over time (adjusted P<.05). Similar to Figure 3, behavioral transitions primarily occurred during the early stages of the disease (Figure 4). The transition probability from ISM to medicine use was significantly lower in both the 0- to 17-year and the ≥60-year age groups compared with the 18- to 59-year age group (both adjusted P<.001), with the lowest probability observed in the ≥60-year group (Figure 4A). In addition, the transition probability was lower in the rural group than in the urban group (adjusted P<.001; Figure 4A). The transition probability from ISM to hospital visit was significantly higher in the ≥60-year age group compared with the 18- to 59-year age group (adjusted P<.001), but it was also significantly lower in the rural group than in the urban group (adjusted P=.02; Figure 4B).