Individuals who met the following criteria were included in this survey: (1) 18 years of age or older and (2) were able to communicate effectively with investigators. The exclusion criteria included (1) no cognitive disorders and (2) no mental disturbance. The data were collected using a structured questionnaire. Questionnaires with completion times below 5 minutes or above 50 minutes were excluded, as pilot testing indicated these thresholds captured inattentive or interrupted responses. Questionnaires with a response missing rate exceeding 50% for the purpose of quality control were also excluded.

Of the 7725 participants included in the final analysis, 3628 (46.96%) were male, and the average age was 65.49 (SD 9.39) years. A total of 4403 (57.0%) participants were users of short-video platforms. Further detailed characteristics of the participants, including demographic, socioeconomic, health-related, and social support factors, are presented in Table 1.

Multistage stratified sampling was applied to select participants in the survey from July to September 2024. First, initial stratification was conducted based on geographical region. Fujian province, Hubei province, and Chongqing municipality were chosen as the survey areas, which are respectively located in the eastern, central, and western regions of China. Second, within each province, 6‐12 urban subdistricts and rural townships were randomly selected as secondary sampling units, ultimately covering 24 subdistricts or townships. Finally, through the community registries, 300‐400 participants were randomly selected from each subdistrict or township. Sampling weights were not applied in the analyses, and all estimates are presented as unweighted statistics. The sampling design aimed to ensure structural representativeness across regional and urban-rural strata within the selected provinces. The data were collected through face-to-face interviews with the respondents by uniformly trained investigators, and the questionnaires were filled out simultaneously. This study was conducted and reported following the Journal Article Reporting Standards for quantitative research [26].

A total of 7725 valid samples were included from the survey, yielding an effective recovery rate of 85.48%. All the analyses were conducted on this sample of valid questionnaires, and there were no missing data for the variables required by this study. Sample size estimation followed the standard formula for proportion-based calculations: [n=uα2 pq∕d2], where uα represents the critical value for a 95% CI (uα=1.96), p is the estimated proportion of short-video platform users based on the 52nd Statistical Report on Internet Development in China—Research on Internet Development [12], q is the complementary proportion (q=1-p), and d denotes the allowable error (1.2%). Using these parameters, the minimum required sample size was calculated as 5283.

First, 2-tailed t test and variance analysis were used to analyze the differences in self-rated health and relative health deprivation among different groups. In the second step, propensity score matching analysis with no replacement was performed to control for selection bias based on the probability of short-video platform use (users and nonusers). Afterward, we fitted quantile regression models on the 10th, 20th, 30th, 40th, 50th, 60th, 70th, 80th, and 90th quantiles of the self-rated health and relative health deprivation distribution to investigate the differences in health between short-video platform users and nonusers. Quantile indicates the location of individuals within the outcome distribution [32]. For example, the individual in the 80th quantile of the conditional distribution has a self-rated health larger than 79% of those with similar short-video platform use behaviors. Quantile regression is a flexible and powerful tool that can assess how health relates to short-video platform use at any quantile distribution. Then, a linear regression model was applied to explore the association of different dimensions of short-video platform use with health, using the full sample. Independent variables consisted of the years, frequency, daily duration, and purpose diversity of use. Finally, we used moderating effect analysis to explore the moderating role of SES between the daily duration of use and health. A series of sensitivity analyses was conducted to test the robustness of the results. Quantile regression was conducted again based on propensity score matching with replacement. In the linear regression model, years, frequency, daily duration, and purpose diversity of use were regarded as continuous variables to be analyzed. Further, we limited the analysis of these dimensions to users of short-video platforms, so as to make the coverage more targeted. The urban-rural residence was used to replace the self-assessed SES to evaluate the moderating role of socioeconomic factors. In all models, coefficients were tested using 2-tailed t tests.

Stata 18.0 was used for all analyses. Statistical significance was defined as a 2-sided P value less than .05.

The final sample consisted of 7725 participants in this study (Figure 1). Characteristics of eligible participants are presented in Table 1. Of the 7725 participants, 4403 (57%) used short-video platforms, 3628 (46.96%) were male, 6492 (84.04%) were married or cohabiting, and 2342 (30.32%) lived in urban areas. Participants who were spousal (n=3966, 90.07% vs n=2526, 76.04%) and had received education (n=6621, 74.93% vs n=5869, 44.13%) were more likely to use short-video platforms. Compared with nonusers of short-video platforms, users had lower scores of multimorbidity-weighted index (2.40, SD 1.96 vs 2.66, SD 2.09), depression (5.28, SD 4.50 vs 6.65, SD 5.02), anxiety (1.82, SD 3.22 vs 2.39, SD 3.71), and loneliness (3.54, SD 1.19 vs 3.90, SD 1.39). The differences in scores of self-rated health and relative health deprivation among different groups are shown in Table 1.

Figures S1-S2 in Multimedia Appendix 1 show the distribution of self-rated health and relative health deprivation in quantiles. The 2 purple curves fluctuate to a large extent near the gray uniform line, indicating that the scores of self-rated health and relative health deprivation are not normally distributed. So, it is necessary to apply quantile regression. In total, 3322 pairs of users of short-video platforms versus nonusers of short-video platforms were formed based on the propensity matching scores. The characteristics of the participants in the propensity-matched dataset are provided in Table S1 in Multimedia Appendix 1.

The quantile regression results in the propensity-matched dataset are reported in Table 2. Compared with the nonusers of short-video platforms, the self-rated health of users was higher at the 70th to 90th quantile (P<.05), and the score of the relative health deprivation of users was lower at the 10th to 30th quantile (P<.05). For self-rated health, the correlation was the strongest at the 90th quantile (β=1.931, 95% CI 1.299-2.563; P<.001). Compared to the users at the 20th and 30th quantiles, the score of relative health deprivation decreased to a lesser extent among those at the 10th percentile (β=−0.003, 95% CI −0.005 to −0.001; P<.001). Moreover, the self-rated health of users is lower than nonusers below the 10th quantile (β=−2.224, 95% CI −3.835 to −0.613; P=.007), while no significant association was observed at the 90th quantile of relative health deprivation. There was no significant association at the 20th to 60th quantile of self-rated health and the 40th to 90th quantile of relative health deprivation.

Correlation analysis indicated that there was a correlation among years, frequency, daily duration, and the number of use purposes of short-video platforms usage. The results of the multicollinearity test showed that there was no collinearity among these variables (variance inflation factor<5). The specific results are shown in Tables S2-S3 in Multimedia Appendix 1.

As shown in Table 3, compared with participants who have used short-video platforms for ≤1 year, those who have used them for ≥3 years showed significantly better self-rated health (β=1.970, 95% CI 0.308-3.632; P<.05). Yet, after adjusting for the control variables, there was no significant correlation between the years of use and relative health deprivation. There was no significant correlation between frequency of use and self-rated health and relative health deprivation. Participants who spent 61‐120 minutes and 120‐240 minutes per day had significantly worse self-rated health (β=−3.385, 95% CI −4.872 to −1.898; P<.001; β=−3.038, 95% CI −5.054 to −1.022; P<.01) and higher score of relative health deprivation (β=0.035, 95% CI 0.021-0.050; P<.001; β=0.034, 95% CI 0.014-0.054; P<.01). However, there was no significant relationship between the participants’ daily duration of use exceeding 240 minutes and their self-rated health or relative health deprivation. Compared to no purposeful use, using short-video platforms for 2 purposes was associated with better self-rated health (β=6.082, 95% CI 0.250-11.914; P<.05) and lower relative health deprivation (β=−0.063, 95% CI −0.120 to −0.005; P<.05). The association was stronger for use with ≥3 purposes (self-rated health: β=6.213, 95% CI 0.276-12.151; P<.05; relative health deprivation: β=−.068, 95% CI −0.127 to −0.009; P<.05).

As shown in Figures 2 and 3, SES was a significant moderator of the association between daily duration of use and health (all P for interaction <.05). The corresponding coefficients and P for interaction can be found in Table S4 in Multimedia Appendix 1. Specifically, compared to the high SES group, participants with middle SES showed significantly lower self-rated health (β=−2.257, 95% CI −3.540 to −0.975; P=.001) and higher relative health deprivation (β=0.018, 95% CI 0.005-0.031; P=.005). Similarly, the low SES group showed significantly poorer self-rated health (β=−2.022, 95% CI −3.349 to −0.695; P=.003) and a higher score of health deprivation (β=0.014, 95% CI 0.001-0.027; P=.04).

After changing the propensity score matching method, the conclusions of the quantile regression remained unchanged. Regarding years, frequency, daily duration, and purpose diversity of usage as continuous variables in linear regression, the relevant conclusions are broadly consistent. The analysis results for the users of short-video platforms also found similar results. Moreover, compared to the urban residence group, the rural residence group had significantly lower self-rated health (β=−0.890, 95% CI −1.667 to −.113; P=0.03) and higher relative health deprivation (β=0.008, 95% CI 0.000-0.016; P=.038). This is also consistent with the previous analysis of the moderating role. All sensitivity analysis results are provided in Tables S5-S8 in Multimedia Appendix 1.

This study explored the heterogeneous association between multidimensional short-video platform use and individual health among Chinese populations. The results reveal a series of nuanced findings. Only among the health-advantaged group with relatively high self-rated health (70th-90th quantile) and low relative deprivation (10th-30th quantile), using short-video platforms was associated with better self-rated health and lower relative health deprivation compared to nonuse. Moreover, short-video platform use was linked to worse self-rated health in the least healthy group (10th percentile) yet showed no significant association with health deprivation in the most relatively deprived group (90th percentile). Another finding showed that greater purpose diversity (≥2 purposes) and longer engagement (≥3 y) with short-video platforms were positively associated with better self-rated health. However, frequency of use showed no significant link to health. A daily use duration of 1 to 4 hours was associated with poorer self-rated health and higher relative deprivation, yet no such negative association was found for duration exceeding 4 hours. SES significantly moderated the association between daily use duration and relative health deprivation, with the strength of this association being more pronounced among middle- and low-SES groups than in the high-SES group.

Our findings demonstrate theoretical congruence with the diffusion of innovations framework. This theory posits that technology adoption follows a social gradient, often privileging early adopters who possess greater resources [33]. The health-advantaged group has preexisting health and social resources that are better equipped to navigate digital environments selectively. They can make use of these advantages for information, social connection, and stress relief, thereby further enhancing their well-being [34]. Their lower baseline deprivation may also buffer them against negative social comparison, a known risk of short-video platforms use. Conversely, the significant negative association observed in the least healthy group (10th percentile of self-rated health) suggests potential pathways of harm. Individuals in poor health may be more vulnerable to algorithmic traps and engage in passive consumption of content. Their use may exacerbate sedentary behavior or increase exposure to distressing or misleading health information [35-37]. Another noteworthy finding is that there is no significant correlation between short-video platform use and relative health deprivation among the group with the highest level of health deprivation (10th percentile). It indicates that for individuals who perceive themselves as severely disadvantaged compared to others, engagement with short-video platforms does little to relieve this sense of inequity [38]. Their deprivation likely depends on structural factors, such as the severity of the disease and socioeconomic disadvantage. These cannot be easily relieved by the fragmented social support and information on short-video platforms. For individuals at the 20th to 60th percentiles of self-rated health and at the 40th to 90th percentiles of relative health deprivation, no significant association was observed. This group typically fits the description of the early majority as defined in diffusion of innovations theory [33]. Their habits of use are more likely to be influenced by the inherent characteristics of short videos, which are typically characterized by frequent but brief, passive, and entertainment-oriented consumption [39]. Together, this pattern of use tends to make short videos a relatively insignificant supplementary tool. Its association with cognitive, emotional, and behavioral remains diffuse and thus below the threshold of detectable statistical association in a cross-sectional design. Notably, our finding of a modest reduction in relative health deprivation (β=−0.003) at the 10th percentile requires careful interpretation. While this change is small at the individual level, its consistent detection at the lower tail of the deprivation distribution holds broader significance. It suggests that even among those who already feel relatively less deprived, short-video platform use may contribute a marginal, yet statistically robust, psychological sense of gain. From a population health perspective, a small shift that is reliably present across a large subgroup can translate into a meaningful aggregate benefit [40].

The differentiated associations observed between behaviors of short-video platform use and health suggest that how and why individuals engage with these platforms may matter more than how often they do so. The positive association of purpose diversity (≥2 purposes) with health underscores the importance of an active, agentic attitude for short-video platforms use. Short-video platforms are engineered for high engagement through infinite scroll and personalized feeds, which typically promote passive viewing [41]. Users who could actively use the platform for information seeking, social interaction, and content creation likely mitigate risks of mindless scrolling [42-44]. The association linked to longer engagement (≥3 y) may reflect digital literacy increases with the accumulation of use time. New users need to spend time exploring and mastering the functions of the short-video platforms, while long-term users may develop more regular and proactive habits. They might be able to use the platform’s functions more flexibly and benefit from it [45,46]. No significant correlation between frequency of use and health is observed, which is inconsistent with the previous studies regarding internet usage [10,47,48]. It might be due to the inherent characteristics of the short-video platforms. Unlike traditional media consumption, where each session constitutes a discrete, sustained period of attention, short-video use is characterized by brief and repetitive engagements [49]. Therefore, the frequency of use may not accurately reflect its association with health, as it fails to distinguish between these short usage periods and more prolonged, immersive usage periods. For the daily duration of use, using short-video platforms for 1 to 4 hours per day is associated with poorer health. This might indicate the psychological fatigue caused by excessive use and the adverse physical activities, such as sedentary behavior [5,50,51]. The daily duration of use exceeding 4 hours was not significantly associated with self-rated health and relative health deprivation. This might be related to the user’s occupation, pattern of use, and so on. This requires further analysis in subsequent studies.

The moderating role of SES observed in this study also aligns with the innovation diffusion theory and the digital divide theory [33]. From the perspective of innovation diffusion, high-SES individuals often act as early adopters. Their greater material and cognitive resources facilitate not only access but also more adaptive and beneficial engagement with new technologies, such as short-video platforms. This translates into higher digital health literacy, better self-regulatory capacity, and the ability to use these platforms more selectively [52-54]. Conversely, the theory of the second-level digital divide explains the heightened vulnerability of middle- and low-SES groups [55]. Disparities in digital skills and resources often confine their engagement to more passive, entertainment-focused consumption patterns [56]. This pattern of use, characterized by prolonged but less intentional exposure, makes them more susceptible to the platforms’ risks, such as unfavorable social comparison, algorithmic traps, and sedentary behavior [57]. Thus, preexisting socioeconomic inequalities are reproduced and potentially amplified in the digital realm, exacerbating feelings of relative health deprivation among the already disadvantaged.

Several limitations of this study should be acknowledged. First, despite the application of propensity score matching to mitigate selection bias, the inherent constraints of cross-sectional observational design preclude definitive causal inferences regarding short-video platform use and health outcomes [24]. Second, while multistage stratified sampling was implemented to ensure regional diversity, potential sampling biases may exist due to nonprobability sampling methods and self-reported participation rates, limiting the generalizability of the findings to the broader Chinese adult population. Third, self-rated health as a subjective measure is susceptible to social desirability bias and cultural interpretation variance, potentially underestimating or overestimating actual health disparities. Fourth, although multiple covariates were statistically controlled, residual confounding effects from unmeasured variables, such as cognitive function and digital literacy, cannot be entirely ruled out, particularly given their demonstrated associations with both health behaviors and technology engagement patterns. Fifth, SES relying on self-report may have measurement error. Future studies would benefit from incorporating objective indicators, such as household income brackets, educational attainment levels, and occupational prestige scores to enhance construct validity. Sixth, the categorization of screen time into discrete intervals limits our ability to model dose-response relationships or detect nonlinear threshold effects within usage categories. This limitation could be addressed through device-based time-use monitoring in subsequent research. Seven, the lack of content type differentiation obscures the understanding of how specific platform features link to health, an issue that could be resolved through qualitative content analysis. Finally, the study focused on short-video platforms and did not capture the potential association between other digital media forms and health. Future longitudinal studies should incorporate specific contents of usage to better identify for whom and under what conditions digital engagement supports health.

This study advances the field by moving beyond the examination of overall associations in the novel context of short-video platforms. It investigates for whom and under what conditions the use of these platforms is linked to health. By exploring multidimensional usage behaviors, it identifies distinct patterns of benefit and risk across population subgroups, providing a nuanced understanding of the relationship between digital engagement and health. Self-rated health and relative health deprivation help to clarify an individual’s subjective perception of health, as well as the sense of health inequality derived from social comparisons. This study carries significant associations between short-video platform use and individual health, which are heterogeneous across population subgroups. Policy must therefore shift from ensuring universal access to actively governing for equitable digital well-being. This requires targeted protections for vulnerable groups, especially those in poor health or with lower SES. They are most susceptible to harm from passive or prolonged use. Meanwhile, the observed disparities in relative health deprivation suggest public policy should aim not only to improve material health access but also to intentionally shape digital environments to address both material and perceived health disparities.