Empowered by digital technologies, short videos have profoundly reshaped the overall landscape of information dissemination and consumption. According to the 56th Statistical Report released by the China Internet Network Information Center (CNNIC), as of June 2025, the user base of short videos in China has reached 1.068 billion, with an average daily viewing duration of approximately 151 min. This data indicates that short videos have evolved from a mere form of entertainment to a basic information service carrier covering the entire population (1). Its core competitiveness stems from “personalized recommendation” algorithms based on artificial intelligence (AI) technology: by deeply analyzing users’ browsing trajectories and interaction preferences, these algorithms construct tailored content streams with a “one-size-fits-none” approach. This not only significantly reduces users’ information search costs but also creates a highly immersive user experience (2, 3). As a globally leading social media platform, TikTok has built its content ecosystem by leveraging such recommendation algorithms, encouraging user participation in short video creation and sharing. Statistics from 2023 indicate that its global cumulative active users have reached 2 billion, with Monthly Active Users (MAU) exceeding 900 million (4). For patients with chronic diseases who require long-term disease self-management, short video platforms, by virtue of their convenience and personalized content advantages, have naturally become an important channel for them to access disease knowledge, rehabilitation guidance, and emotional support (5).

Lumbar disc herniation (LDH) is the most common disorder in spinal surgery, characterized by high prevalence, a prolonged disease course, and a high recurrence rate. The efficacy of its treatment and rehabilitation is highly dependent on patients’ long-term self-management behaviors and accurate disease perception (6, 7).Traditional health information access channels, such as doctor-patient communication, printed materials, or static webpages, are being supplemented or even partially replaced by dynamic, intuitive, and highly engaging short video content. Driven by the need to alleviate symptoms and seek rehabilitation methods, patients are actively or passively immersed in the information streams constructed by algorithms (8). This health information environment, driven by data and code, harbors non-negligible risks: the platform's recommendation logic is based on “user interaction data” and “traffic preferences” rather than “hierarchical medical evidence,” which may lead patients to be systematically exposed to an information environment with mixed quality—where scientific rehabilitation guidance coexists with unproven miracle cures and exaggerated commercial content. Furthermore, prolonged immersion in lumbar disease-related videos may exacerbate patients’ anxiety or disease burden. Bright et al. (9) found that over-reliance on short video recommendation algorithms may cause users to overlook their own intuitive feelings and judgments, or information overload may lead to decision fatigue, which to a certain extent increases users’ stress and anxiety. When this general mechanism acts on patients seeking clear rehabilitation plans, its negative effects may be amplified: patients may fall into cognitive confusion and psychological distress amid contradictory, intimidating, or false information.

Despite the increasingly prominent role of short videos in health communication, relevant academic research remains underexplored. First, most studies have remained confined to cross-sectional analyses of platform content quality or user attitude surveys (8, 10), with a critical lack of longitudinal causal evidence regarding the impact of immersion duration and content exposure on objective clinical outcomes. Second, existing research mostly treats social media use as a monolithic construct, failing to distinguish between the differential impacts that specific algorithm-recommended content types and immersive usage behaviors may independently exert. Finally, in chronic disease management contexts such as LDH, there is a paucity of in-depth empirical examination into whether and how health anxiety functions as a key mediating variable linking digital information behaviors to physical rehabilitation outcomes. Thus, this study designed and conducted a 6-month prospective cohort study. The primary objective is to systematically analyze how immersion duration in relevant short videos and algorithm-recommended content types independently and synergistically influence the symptoms of LDH patients by affecting health anxiety.

This study conducted a 6-month prospective cohort study to systematically investigate the differential impacts of viewing duration and content of lumbar disease-related videos—passively recommended by short video platform algorithms without active user search—on health anxiety and clinical symptoms in patients with lumbar disc herniation (LDH). It also verified the mediating role of anxiety in these associations. The results showed that longer daily average viewing duration of LDH-related content in short videos was associated with increased health anxiety and poor lumbar function recovery after 6 months. Among algorithm-recommended content types, exposure frequency to warning and mobilization-related content had the most significant negative impact on psychological and clinical outcomes, while science popularization and education content, as well as experience sharing content, showed no significant associations. Changes in health anxiety (ΔMCQ-HA) played an important mediating role between short video usage behaviors (including viewing duration and exposure to non-standardized content) and changes in lumbar function (ΔJOA). These findings provide crucial longitudinal empirical evidence for understanding how the digital information environment, particularly algorithm-driven passive information flow, subtly shapes the psychological state and physiological rehabilitation process of patients with chronic diseases.

In the digital health era, attention resource allocation and threat information processing occur interactively. Influenced by short video algorithm recommendations, individuals may experience increased pressure and inhibition related to their health management behaviors (15). Algorithms continually learn user preferences from behavioral data (e.g., viewing duration, likes, comments) to generate personalized content feeds. This can lead to passive and sustained exposure to high-intensity information pertaining to one's own health status. Such an exposure paradigm may not only reshape health cognitions but, more critically, may alter the psychological underpinnings of health management behaviors via emotional arousal and increased cognitive load. This process can consequently heighten the perception of health-related stress and impair self-regulatory capacity (16).Previous research has demonstrated a significant positive association between the frequency of exposure to illness-related information on digital platforms and levels of pain catastrophizing and health anxiety among individuals with chronic pain (17). This finding offers a key theoretical foundation for understanding the link between short video information exposure and health outcomes in patients with lumbar spinal disorders. This study found a significant positive correlation between changes in health anxiety and daily viewing time (P < 0.001). From a cognitive psychology perspective, dedicating substantial daily time to viewing illness-related content—even passively—can result in the continuous allocation of limited working memory resources to illness-related themes. This cognitive preoccupation likely impairs the processing of recovery-relevant information (e.g., proper exercise techniques, positive prognosis cases) and concurrently amplifies attentional bias toward illness-threat cues (e.g., complications, recovery setbacks) (18). More importantly, the reinforcement learning nature of recommendation algorithms can exacerbate this effect. Even brief engagement with specific illness-related content may prompt the platform to persistently deliver similar material, potentially creating a ‘health-threat information silo’ (19). From a health anxiety standpoint, such sustained exposure to negative information may amplify anxiety primarily via two mechanisms. First, a high volume of fragmented threat-related content can disrupt the equilibrium between perceived threat and perceived coping ability, leading to an overestimation of health risks—for instance, misperceiving rare complications as high-probability events (20). Second, the passive consumption of content without professional context (e.g., anecdotal experiences shared by non-medical individuals or exaggerated symptom descriptions) can heighten cognitive uncertainty, which in turn may activate catastrophic thinking (18).

We also observed a significant negative correlation between daily viewing time and lumbar functional outcomes (P < 0.001).This association not only reflects a displacement effect, where viewing time supplants time for rehabilitative activities, but also suggests potential influences from dual mechanisms induced by algorithmic recommendations: information overload and cognitive narrowing (21). Specifically, prolonged exposure to algorithmically curated illness-related videos may not only displace time for active rehabilitation but also deplete finite cognitive and emotional resources. This can result in attentional fixation on health-threat information, thereby impairing both the processing efficiency of and the motivation to act upon positive recovery-oriented messages (e.g., rehabilitative guidance) (22). This mechanism aligns with insights from digital health communication research. For instance, a systematic review noted that among social media interventions designed to promote health behaviours, only a minority successfully improved adherence to behaviours such as vaccination (23). This suggests that within complex algorithmic information flows, merely increasing exposure to health information may not necessarily lead to positive behavioral change. If content fails to effectively penetrate users’ cognitive load and trigger deep processing, its impact will be extremely limited.Further mediation analysis indicated that health anxiety partially mediated this relationship. Specifically, daily viewing time was found to exert a negative effect on lumbar functional recovery partly by elevating levels of health anxiety.This finding delineates a clear pathway through which digital engagement can lead to psychological stress that, in turn, impedes physiological function (24). It suggests that algorithmically recommended content may act not merely as passively consumed information but as a persistent psychosocial stressor. This stressor, by amplifying anxiety-driven processes such as catastrophic interpretation of symptoms and excessive worry about recovery, may heighten pain perception sensitivity, foster avoidance of rehabilitative activities, and disrupt normal physiological repair mechanisms. These sequelae could ultimately manifest as impeded recovery of lumbar function. These results provide crucial empirical evidence for understanding the psychophysiological mechanisms by which digital technology engagement can influence physical health outcomes.

Notably, the absence of significant effects for science education and experience-sharing content in this study aligns with the core tenets of self-verification theory. Briñol et al. (25) posit that the influence of information on individual behaviour depends on the dual activation of cognitive and affective validation. While science-based content offers cognitive accuracy, it often lacks the emotional salience and self-relevance characteristic of awareness-motivation content, thereby failing to trigger deep cognitive processing or emotional resonance in patients. Conversely, the high heterogeneity of experience-sharing narratives, combined with the stochastic nature of algorithmic recommendations, may result in content that is poorly matched to an individual patient's specific situation, thus diminishing its potential for self-validation. This finding implies that the health impact of algorithmic information environments is not determined merely by the volume of information but by the content's capacity to activate dual pathways of cognitive and affective validation within the user. This insight offers a theoretical anchor point for future optimizations of algorithmically curated health content. Importantly, our analysis confirmed a significant association between exposure to awareness-motivation content and poorer lumbar functional outcomes (P < 0.001). This finding reveals a health risk amplification effect stemming from a ‘negativity bias’ inherent in algorithmic curation. This content category predominantly features negative narratives, such as severe complications, surgical risks, and cases of failed recovery from lumbar spinal disorders. Its dissemination logic is highly congruent with the platform economics of short-video platforms, which prioritize ‘attention capture maximization’ (26). From a neuropsychological standpoint, such content can rapidly induce coupling between the default mode network (DMN) and the ventral tegmental area (VTA). This coupling reinforces viewing behaviour via dopaminergic reward pathways while simultaneously impairing cognitive control functions, thereby exacerbating catastrophic illness-related cognitions (27).

More critically, excessive exposure to awareness-motivation content can reconfigure a patient's illness identity. Prolonged immersion in such narratives may facilitate a shift from viewing oneself as ‘a person in recovery’ to identifying as ‘a potential severe case’. This identity reconfiguration can influence rehabilitation behaviours via a self-fulfilling prophecy mechanism. For instance, excessive worry may lead to avoidance of necessary functional exercises, while unrealistic recovery expectations may breed frustration. Collectively, these responses can initiate a vicious cycle: elevated anxiety fosters behavioural withdrawal from rehabilitation, which impedes functional recovery, in turn further heightening anxiety (28). This mechanism elucidates the mediating role of health anxiety in the link between short-video engagement and lumbar functional changes. Health anxiety thus acts not merely as a psychological affective state but as a critical ‘pathopsychological bridge’ linking digital information exposure to physiological recovery. At its core, health anxiety in this context represents a psychological stress response triggered by algorithmically amplified negative information exposure. This stress response, mediated through activation of the hypothalamic–pituitary–adrenal (HPA) axis, can subsequently modulate pain perception and interfere with tissue repair processes (13). In contrast, the observed ‘ineffectiveness’ of science education and experience-sharing content is not coincidental. Science-based content often suffers from a lack of personalization, failing to address the specific rehabilitation dilemmas of individual patients, which leads to inadequate cognitive validation. While experience-sharing content holds potential for emotional resonance, algorithmic recommendations’ “homogeneous lock-in” may expose patients to narrow rehabilitation experiences, preventing the development of comprehensive health awareness. This phenomenon highlights the “pseudo-personalization” dilemma inherent in algorithmic recommendations (29). Algorithms push homogeneous warning-based content based on traffic preferences, seemingly catering to user needs. In reality, this limits patients’ access to diverse health information, trapping those with lower education levels in negative information echo chambers and exacerbating withdrawal from recovery behaviors (30).

This study has several limitations. First, the relatively single sample source may limit the generalizability of the results. Second, daily viewing duration of LDH and spine-related disease videos relied on self-reported retrospective assessment by participants, lacking objective recording data. Memory bias and time perception deviation may lead to overestimation or underestimation of duration data, making it difficult to reflect true exposure levels. Third, the differential impacts of different recommendation algorithms were not distinguished, hindering the accurate identification of associations between algorithmic technical characteristics and health risks. Fourth, potential moderating variables such as patients’ social support and baseline psychological traits were not included to analyze potential interaction effects. Future research can be expanded in three directions: 1. adopting a multi-center, large-sample cohort design that includes LDH patients of different age groups and disease durations to improve the generalizability of results; 2. integrating algorithm log data to analyze the impacts of different recommendation mechanisms on information exposure quality and health outcomes, providing more precise technical basis for algorithm optimization; 3. conducting interventional studies to verify the effectiveness of an intervention model combining algorithmic content optimization and digital health literacy education, exploring practical pathways to mitigate algorithm-induced health risks.

This 6-month prospective cohort study confirmed that longer daily viewing duration of disease-related short videos and more frequent exposure to algorithm-recommended warning and mobilization content among LDH patients were associated with poorer lumbar function recovery and higher levels of health anxiety after 6 months. Health anxiety played a partial mediating role in these associations, with a mediated proportion of 16.13% for viewing duration and lumbar function recovery, and 20.80% for exposure to warning and mobilization content and lumbar function recovery. These findings indicate that passive exposure to disease-related content on short video platforms—especially prolonged viewing and high-frequency exposure to warning and mobilization content—significantly hinders the clinical rehabilitation of lumbar function through the key psychological pathway of exacerbating health anxiety. This study not only reveals the specific mechanism by which algorithmic recommendations, as a novel environmental stressor, affect chronic disease rehabilitation but also provides important empirical support for LDH patients’ information self-protection, the optimization of the health content ecosystem on short video platforms, and the consideration of patients’ digital media exposure in clinical practice.