This longitudinal analysis used baseline data (2019) from the CNSSCH^[12]. A stratified random sampling approach selected eight provinces (Shanghai, Fujian, Shanxi, Henan, Hunan, Gansu, Chongqing, Guangxi) representing eastern, central, and western regions for follow-up assessment in November 2020. From the initial baseline cohort of 14,532 adolescents aged 9-18 years [Figure 1], 9,814 age-eligible participants were retained for descriptive statistics. Final analytical samples comprised 5,006 respondents meeting complete questionnaire criteria (aged 9-18 years), excluding younger children (6-8 years), lacking survey participation, and older participants with critical data deficiencies (e.g., physical activity metrics). The protocol received ethics committee approval from Peking University Health Science Center (IRB00001052-18002, IRB00001052-21001), with written informed consent obtained from all participants and legal guardians.

Certified staff conducted anthropometric assessments using standardized protocols, with participants barefoot in light clothing. Height and weight were measured to 0.1 cm and 0.1 kg precision, respectively. The body mass index (BMI) derived from weight(kg)/height(m)² informed obesity status classification according to China’s National Health Commission age- and gender-specific percentiles^[13]: overweight (OW) ≥ 85th percentile, obesity (OB) ≥ 95th percentile. Overweight transition was operationalized as progression from baseline normal weight, thinness status, to OW during follow-up. Obesity transition was operationalized as progression from baseline normal weight, thinness, or OW status to OB during follow-up.

The frequency of wearable device usage was assessed through a structured question: “How often do you use wearable devices (e.g., smartwatches, fitness trackers)?” As the questionnaire aggregated these two modalities, they were analyzed together as a single exposure variable in the present study. Participants selected from five predefined response options: “Frequently”, “Sometimes”, “Occasionally”, “Rarely”, and “Never”. Responses were subsequently coded as an ordinal variable (1-5) for analytical purposes. These terms reflected perceived frequency rather than quantifiable usage. To ensure consistency, pilot testing confirmed participants’ clear understanding of both the device scope (limited to movement-sensing wearables) and temporal definitions before nationwide implementation. While both tools target physical activity monitoring, their behavioral mechanisms differ. Aggregating these two modalities in the exposure variable may dilute modality-specific effects.

Certified field staff distributed structured paper questionnaires to participants aged 9-18 years during regular school hours. Utilizing a self-administered format with researcher oversight for clarification needs, the instrument systematically captured: Sociodemographic profiles: age, gender, residential location, grade level, singleton status, and parental educational attainment. Health-related behaviors: frequency of sugar-sweetened beverage consumption, daily sleep patterns, and physical activity engagement. According to the World Health Organization (WHO) Guidelines on Physical Activity for Children and Adolescents, which recommend that children and adolescents aged 5-17 years accumulate at least 60 min of moderate-to-vigorous physical activity (MVPA) daily. Our study used “≥ 1 h” to align with this guideline, as MVPA is the primary component of physical activity associated with obesity prevention. Based on the prior epidemiological studies using the 8-h threshold for sedentary time, which note that sedentary time exceeding 8 h per day is associated with an increased risk of obesity^[8]. Psychometric evaluation confirmed moderate reliability (Cronbach’s α = 0.76) for institutional-level measures within the survey framework^[14].

Missing values were imputed using hot-deck imputation^[15], where each missing case was matched to a donor with similar age, sex, and residential location. This method was selected due to the low proportion of missing data and the categorical nature of key covariates, which made multiple imputation less efficient and unnecessary for bias reduction in this context. Categorical variations by sex and residence were evaluated via χ² tests, whereas age-dependent trends were evaluated using Mantel-Haenszel linear trend analysis. Multilevel logistic regression models estimated adjusted odds ratios (ORs) with 95% confidence intervals (CIs) for behavioral predictors of overweight/obesity incidence and transitions, incorporating province-level random intercepts to accommodate hierarchical data structures. Fixed effects adjustments included demographic (age, sex, urban/rural residence, singleton status), dietary (breakfast frequency, sugar-sweetened beverage consumption), and socioeconomic covariates (parental education), with multicollinearity excluded through variance inflation factor verification [all variance inflation factor (VIF) < 5]. Analyses were executed in Statistical Package for the Social Sciences (SPSS) 26.0 and R 4.2.2 using two-tailed significance thresholds (α = 0.05). Sensitivity testing redefined OW/OB using International Obesity Task Force (IOTF) pediatric criteria^[16], a global reference standard, to confirm findings consistency, and the relevant results are provided in the supplementary materials [Supplementary Figure 1].

Table 1 presents the distribution of wearable device usage and characteristics of the study population stratified by gender groups in the 2019-2020 follow-up study. A total of 5,006 participants were included, with 2,501 boys and 2,505 girls. The incidence of overweight and obesity was higher in boys than in girls (overweight: 17.3% vs. 11.3%; obesity: 12.2% vs. 10.5%), with significant differences (P < 0.001). The transition to overweight and obesity also had gender-related variations, with higher proportions in boys. Supplementary Table 1 summarizes baseline and transition characteristics stratified by wearable device usage frequency. For wearable device usage, boys were more likely to use them frequently (13.6%) compared to girls (11.0%), while girls had a higher proportion of never using wearable devices (34.8% vs. 38.8% in boys), and these differences were significant (P < 0.001).

In crude models, compared to frequent users, occasional and sometimes users exhibited the lowest risks of transitioning to overweight and obesity, while both rare and never users showed elevated risks [Figure 2]. After adjusting for age, sex, urban/rural residence, and baseline BMI, the dose-response relationship became more pronounced: the adjusted OR for transitioning to overweight increased progressively with reduced usage frequency. Specifically, occasionally users (OR = 1.61, 95%CI: 1.00-2.58, P = 0.050), rare users (OR = 1.55, 95%CI: 0.95-2.51, P = 0.077), and never users (OR = 1.56, 95%CI: 0.99-2.46, P = 0.058) showed numerically elevated risks compared to frequent users, but these associations did not reach conventional statistical significance. Similarly, for transitioning to obesity, rarely users exhibited a significantly higher risk (OR = 1.50, 95%CI: 1.04-2.18, P = 0.030), while never users demonstrated a borderline trend (OR = 1.39, 95%CI: 0.98-1.98, P = 0.066) that was not statistically significant. Collectively, this trend indicates a nonlinear, U-shaped dose-response relationship, suggesting that moderate usage (sometimes/occasional) provides the most protective effect.

We further explored the heterogeneity in the association between wearable device/mobile fitness app usage frequency and obesity transition risks through subgroup analyses [Figure 3]. Stratified by age, the ORs for transitioning to overweight or obesity across usage frequency categories showed no statistical significance in either the 9-12-year or 13-18-year age groups, though the OR trends aligned with the primary analysis, suggesting a potential link between low-frequency usage and elevated risk.

Sex-stratified analyses revealed that boys in the rare-use group had significantly higher risks of transitioning to overweight (OR = 1.96, 95%CI: 1.05-3.68, P = 0.035) and obesity (OR = 1.62, 95%CI: 1.01-2.60, P = 0.045), whereas girls exhibited a significant association only in the occasional-use group for obesity transition (OR = 1.84, 95%CI: 1.00-3.38, P = 0.049).

Residence-stratified results demonstrated that rural participants in the occasional-use (OR = 2.99, 95%CI: 1.23-7.25, P = 0.016) and rare-use (OR = 2.55, 95%CI: 1.07-6.08, P = 0.034) groups had significantly elevated risks of transitioning to overweight compared to frequent users, while no significant associations were observed in urban areas. These findings indicate gender and residence-specific disparities in the relationship between wearable device usage frequency and obesity transition. While subgroup analyses suggest higher risks in rural children and boys when device usage is low, this does not imply that frequent use always confers greater benefit. Instead, it highlights stronger sensitivity to device engagement in these subgroups, potentially due to contextual factors such as limited access to alternative health resources or gender-related behavioral responses.

Figure 4 shows the association between wearable device usage frequency and the risk of insufficient physical activity. Using never-users as the reference group, the analysis revealed that frequent users had a non-significant reduction in physical inactivity risk (OR = 0.86, 95%CI: 0.71-1.04, P = 0.110). In contrast, occasional users (OR = 0.61, 95%CI: 0.51-0.73, P < 0.001) and intermittent users (OR = 0.68, 95%CI: 0.57-0.81, P < 0.001) exhibited significantly lower risks of inadequate physical activity (PA), suggesting a higher likelihood of meeting the recommended daily PA guideline of ≥ 1 h. However, rare users showed no significant difference compared to never-users (OR = 0.97, 95%CI: 0.82-1.14, P = 0.695). These results indicate a descriptive U-shaped trend in the associations: moderate device usage (i.e., occasional or intermittent use) was associated with the most substantial PA improvement, whereas both excessive (frequent use) and minimal (rare use) engagement failed to yield significant benefits. This pattern implies that periodic, rather than constant, device interaction may optimally motivate PA adherence in children and adolescents, potentially by balancing behavioral reinforcement with avoiding technology fatigue.

We also identified a significant inverse association between wearable device usage frequency and prolonged sedentary behavior risk [Figure 4]. Compared to frequent users, all lower-frequency user groups exhibited elevated sedentary risks: never-users had an OR of 1.36 (95%CI: 1.11-1.67, P = 0.003), while sometimes-users showed the highest risk (OR = 1.79, 95%CI: 1.42-2.26, P < 0.001). Similarly, occasional users (OR = 1.46, 95%CI: 1.16-1.85, P = 0.001) and rare users (OR = 1.36, 95%CI: 1.09-1.70, P = 0.007) demonstrated significantly increased risks.

These results indicate a clear dose-response gradient: as device usage frequency decreased (from sometimes-use to never-use), sedentary risk escalated progressively, with the steepest rise observed in sometimes-users (OR = 1.79). This pattern suggests that intermittent use may fail to counteract sedentary habits due to insufficient sustained behavioral reinforcement. Notably, the findings align with the PA time analysis, collectively supporting a mechanistic pathway whereby wearable devices improve health behaviors through activity reminders and sedentary interruption.

The consistency across outcomes underscores that frequent device engagement may optimize behavioral regulation by balancing real-time feedback with habituation avoidance, whereas sporadic use limits sustained benefits. These insights advocate for integrating wearable devices into pediatric obesity interventions, emphasizing consistent but non-intrusive usage to maximize long-term adherence.

While our findings advance understanding of pediatric wearable interventions, several limitations warrant cautious interpretation. First, residual confounding from unmeasured socio-technological factors, such as familial eHealth literacy gradients and device functionality disparities, may influence observed associations. Second, reliance on self-reported data for wearable device usage frequency and physical activity duration introduces the risk of recall bias. Participants may have overestimated or underestimated their actual device use or physical activity levels (e.g., forgetting occasional device use or overreporting compliant physical activity), which could potentially distort the observed associations between wearable device usage and obesity transition. While the CNSSCH survey was pilot tested for comprehension, no formal validation study of this specific question has been conducted. Third, the survey item aggregated wearable devices and mobile fitness apps into a single exposure measure. While both aim to monitor and promote physical activity, their behavioral mechanisms may differ: fitness trackers provide continuous feedback through sensors, whereas mobile apps may offer more intermittent or self-initiated feedback. Combining these modalities could therefore dilute modality-specific effects and obscure nuanced differences in behavioral pathways. Future research should differentiate between device-based and app-based usage, ideally using objective log data, to better capture their distinct contributions to obesity prevention. Fourth, we were unable to apply formal nonlinear modeling techniques because wearable device usage was assessed as an ordinal categorical variable rather than a continuous measure. This limited our ability to fully characterize the shape of the dose-response curve beyond categorical comparisons. Future studies that collect continuous or log-based usage data from devices or apps would enable more robust nonlinear modeling and provide deeper insight into the precise turning points of the U-shaped relationship observed in this study. Finally, several important potential confounders, such as parental BMI, household income, overall diet quality, and daily screen time, were not available in the CNSSCH dataset. The omission of these variables may have introduced residual confounding, as family background, socioeconomic status, and lifestyle factors are closely associated with both wearable device usage and obesity risk. Although we adjusted for a wide range of demographic and behavioral covariates, the possibility of unmeasured confounding cannot be excluded. Moreover, given the low level of missingness, hot-deck imputation was used to preserve sample representativeness; However, future work could apply multiple imputation for sensitivity analyses. Therefore, our effect estimates should be interpreted with caution, and future studies incorporating more comprehensive measures are warranted.

Moderate use of wearable devices appears to offer an optimal reduction in obesity risk among children, while potentially mitigating technology-related fatigue associated with excessive usage. Disparities related to geographic setting (rural vs. urban) and gender underscore the importance of developing context-specific intervention strategies. The beneficial effect of wearable devices on pediatric obesity risk is primarily mediated through the reduction of sedentary behavior and the increase in time spent in physical activity, with the greatest benefits observed at moderate frequencies of use. These findings highlight the need for public health initiatives to prioritize the quality and pattern of device usage over mere adoption rates. Such an approach may enhance the effectiveness of technology-based interventions aimed at combating childhood obesity.