Introduction
Smartphones are ubiquitous and convenient, but raising health concerns, including addiction. With most of the global population owning a smartphone, the potential negative impacts on health have also increased (Ratan, Parrish, Zaman, Alotaibi, & Hosseinzadeh, 2021). Problematic smartphone use may be inferred when individuals exhibit preoccupation with the device, progressive increases in use consistent with tolerance, a persistent inability to cut down despite intentions to do so, or a diminished interest in other activities (H.-K.
Lee, Kim, et al., 2017). Smartphone use has been linked to depression, anxiety, chronic stress, and low self-esteem (Elhai, Dvorak, Levine, & Hall, 2017). Prior research has shown that smartphone addiction correlates with higher GABA levels and a disrupted GABA–Glx balance.
Here, Glx refers to the combined signal of glutamate and glutamine. This neurochemical profile has been linked to poorer mental-health outcomes (Seo et al., 2020). This trend is especially alarming among youth estimated that approximately 25% of children and adolescents engage in problematic smartphone use, leading to deteriorating mental health, such as social isolation and a decline in face-to-face communication (Sohn, Rees, Wildridge, Kalk, & Carter, 2019).
Adults are also at higher risk, as emphasized in recent studies (Al-Kandari & Al-Sejari, 2021; Kusumota et al., 2022). Moreover, problematic use can exacerbate pre-existing mental health and disrupt sleep patterns (Dresp-Langley & Hutt, 2022; Gjoneska et al., 2022). Despite growing recognition of these issues, relatively few studies have examined the multifaceted effects of smartphone overuse among individuals with sleep disorders such as insomnia, including mechanisms implicating melatonin suppression via melanopsin-expressing intrinsically photosensitive retinal ganglion cells (ipRGCs) projecting to the suprachiasmatic nucleus (SCN), which delays circadian phase and sleep onset (Bozkurt, Demirdöğen, & Akıncı, 2024; Chang, Aeschbach, Duffy, & Czeisler, 2015).
Insomnia is a prevalent sleep disorder characterized by difficulty falling or staying asleep, waking up too early, and non-restorative sleep (Roth, 2007). It is often accompanied by a range of mental and physical health problems (S. Lee, Lee, & Cho, 2022; Morin & Buysse, 2024), with socioeconomic costs, such as reduced productivity, increased absenteeism, and healthcare.
Recent studies showed that approximately 10% of adults suffer from chronic insomnia, and 20% experience occasional insomnia, with prevalence rising globally (Garland et al., 2018; Morin & Jarrin, 2022). While acute insomnia can resolve spontaneously, persistent cases often evolve into chronic forms that significantly impair quality of life (Benca, 2005).
Cognitive behavioral therapy for insomnia (CBT-I) is the primary treatment for insomnia targeting cognitive distortions and dysfunctional behaviors (Morin & Buysse, 2024). Cognitive distortions include catastrophic thinking about poor sleep, unrealistic sleep expectations, and excessive worry about sleep loss. Dysfunctional behaviors include irregular sleep, spending time awake in bed, and stimulatory activities before sleep (Morin & Bélanger, 2011).
Smartphone overuse at bedtime is associated with insomnia symptoms likely due to circadian rhythm and delayed sleep onset (Schmid et al., 2021).
With recent advances in sensor technology and artificial intelligence improving measurement accuracy (Galatzer-Levy & Onnela, 2023), digital phenotyping has rapidly expanded in psychiatry (Chia & Zhang, 2022; Yeom et al., 2025). This approach leverages data from digital devices such as smartphones and wearables and provides objective measurements of mental health assisting traditionally subjective clinical assessments. The availability of data is estimated to lead to the expansion of digital phenotyping (De Boer et al., 2023).
It could be a novel opportunity to advance our understanding of the relationship between smartphone overuse and insomnia.
This study aimed to understand the effect of smartphone overuse on insomnia by integrating digital phenotyping with clinical assessments. We analyzed the demographic, psychological, and behavioral characteristics of insomnia group based on their level of smartphone use by Smartphone Overuse Screening Questionnaire (SOS-Q) (H.-K. Lee, Kim, et al., 2017).
Using a multidimensional approach, including clinical scales and data from wearable devices, this study sought to identify the key differences between high- and low-risk smartphone users and observe insights for clinical interventions. The study plan was informed by prior research on insomnia and circadian rhythm disorders, including guidance to capture habitual sleep–wake variability through multi-week monitoring (Smith et al., 2018). By analyzing the interaction between digital behavior and sleep health, we can contribute to a new approach to modern insomnia and the possibilities of sleep therapy.
Methods
Study population
From January 2023 to July 2024, 398 participants, including patients with insomnia and non-insomnia patients, were recruited from the Korea University Anam Hospital and Datamaker Inc. (Daejeon, South Korea), a specialized data collection service company that conducted participant recruitment and data collection under the same research design and ethical protocols. Participants aged 19–65 were recruited from the local community through online community platforms and campus bulletin boards.
This study was conducted on 249 participants with insomnia, defined as individuals who experienced subjective insomnia symptoms at least three times per week over the past 3 months, with an Insomnia Severity Index (ISI) score of 8 or higher (Bastien, Vallières, & Morin, 2001). Exclusion criteria for the insomnia group included intellectual disability, organic brain damage, a diagnosis of a schizophrenia spectrum disorder, current treatment for sleep disorders, and not owning a smartphone.
The participants provided informed consent at the onset of the study and completed clinical report forms covering demographic information, family medical history, and current health conditions using various scales. The participants were equipped with wearable devices (Fitbit Inspire 3, Fitbit Inc., USA) to gather data on step count, heart rate, and sleep patterns. They were also asked to download the “SOMDAY” smartphone application (Lumanlab Inc., Seoul, South Korea) to log their daily habits.
Three participants were excluded due to missing data, leaving 246 individuals for the final analysis (Fig. 1).
Measures
Clinical assessment
Demographic information, including age, sex, marital status, education, job, life pattern (e.g., morningness/eveningness and related chronotype categories), sleep environment (e.g., alone, with a partner, with children, or other), weight, and height, was collected. In addition, the participants completed a series of standardized self-reported questionnaires to assess key study variables.
To evaluate insomnia severity, the Insomnia Severity Index (ISI) (Morin, Belleville, Bélanger, & Ivers, 2011) was used as a validated measure to assess both the presence and severity of insomnia symptoms. The Smartphone Overuse Screening Questionnaire (SOS-Q) was used to classify participants into risk groups based on their smartphone usage patterns (H.-K. Lee, Kim, et al., 2017).
Participants with an SOS-Q score exceeding 48 were categorized into the smartphone overuse high-risk group (SOHG) (n = 141), whereas those with a score of 48 or lower were assigned to the smartphone overuse low-risk group (SOLG) (n = 105).
Additionally, as sleep-related scales, results were collected from Dysfunctional Beliefs and Attitudes about Sleep (DBAS-16) (Morin, Vallières, & Ivers, 2007), the Korean version of Biological Rhythms Interview of Assessment in Neuropsychiatry (K-BRIAN) (Cho, Jung, Kapczinski, Rosa, & Lee, 2018), Pittsburgh Sleep Quality Index (PSQIK) (Sohn, Kim, Lee, & Cho, 2012), Berlin Questionnaire (BERLIN) (Senaratna et al., 2017), Morningness-Eveningness Questionnaire (MEQ) (Hwang et al., 2024), Bedtime Procrastination Scale (BPS) (An, ju Chung, & Suh, 2019), Ford Insomnia Response to Stress Test (FIRST) (Chang & Suh, 2018). Further, as non-sleep-related scales, International Restless Leg Scale (IRLS) (Group, 2003), The Korean version of Alcohol Use Disorders Identification Test (AUDIT-K) (Chang et al., 2016), Patient Health Questionnaire-9 (PHQ-9) (Han et al., 2008), Generalized Anxiety Disorder-7 (GAD-7) (S. H.
Lee, Lee, & Cho, 2022), Korean version of Mood Disorder Questionnaire (KMDQ) (Jon et al., 2009), Body Sensation Questionnaire (BSQ) (Chambless, Caputo, Bright, & Gallagher, 1984), Korean Resilience Questionnaire-53 (KRQ-53) (Ko, Kim, Bartone, & Kang, 2018), Multidimensional Fatigue Scale (MFS) (Song et al., 2018), Spiritual Well-Being Scale (SWBS) (You & Yoo, 2016), and World Health Organization Quality of Life Brief Version (WHOQOL-BREF) (Min et al., 2002) results were collected (Supplementary Table 1).
Digital phenotypes (ecological momentary assessment and wearable data)
SOMDAY, a smartphone application, was developed as a comprehensive tool to gather data on the participants' daily habits and lifestyle choices. It is compatible with both the Android (Google) and iOS (Apple) platforms, ensuring accessibility to a broad user base. Specifically designed to collect ecological momentary assessment data, the app prompts participants to log their daily habits each night at 9 PM to ensure consistent data entry, including details on alcohol intake, caffeine consumption, napping behavior, stress levels, self-reported sleep duration, and frequency of night-time awakenings.
In parallel, the Fitbit device automatically recorded various metrics such as sleep duration, number of awakenings, sleep quality, exercise time, heart rate, step count, and walking distance for four weeks (H.-A. Lee, Kim, et al., 2017). By integrating SOMDAY with Fitbit, the data from Fitbit was synchronized and made accessible through the SOMDAY app, which was then seamlessly connected to an online database for easy access by the research team for further analysis.
Datasets
Heart rate data were collected at 5-min intervals, and sleep duration was recorded daily. Because the step and walking distance data were cumulative, we applied differencing and calculated daily averages to derive features and examine circadian rhythms.
The sleep duration parameters included total sleep time, wake time, number of nighttime awakenings, and time spent in REM, light, and deep sleep. To assess sleep quality, we used total sleep and wake times. Daily maximum, minimum, and average values were calculated for the walking distance.
To analyze circadian rhythms in heart rate, we performed Cosinor analysis using a two-day time window, producing the following key parameters: mesor (midline estimating statistic of rhythm), amplitude, acrophase, and goodness of fit (Cornelissen, 2014). Mesor represents the rhythm-adjusted mean, which indicates the average value of a parameter over the analyzed period. The amplitude, defined as the difference between the mesor and peak value, quantifies the range of variation within the circadian cycle.
Acrophase denotes the timing of the peak within each cycle and provides insights into circadian alignment. The goodness-of-fit measures how well the data aligns with a 24-h cosine model, reflecting the strength and clarity of the observed circadian rhythm. Additionally, we calculated the daily maximum, minimum, and average heart rates.
Steps were analyzed similarly to heart rate, with maximum, minimum, and average values computed for weekdays versus holidays (including Korean public holidays), as well as for day (8 AM–6 PM) and night (6 PM–8 AM). Further insights into activity cycles were derived using L5 (least active 5-h period) and M10 (most active 10-h period), as well as intradaily variability (IV) and interdaily stability (IS) (Krafty et al., 2019). L5 and M10 were calculated through a moving average approach, with L5 representing the lowest activity over any 5 h, and M10, the highest activity over any 10 h.
RA (relative amplitude) was caculated using L5 and M10.
Cosinor analysis features and sleep duration data were averaged over a four-week period, while maximum and minimum daily values were averaged for weekdays versus holidays and for day versus night. Similarly, L5, M10, RA, IS, and IV were separately averaged on weekdays and holidays.
Sample size, analytic hierarchy, and power
This study used rolling enrollment rather than a fixed sample size. Prior to closing recruitment, we defined a minimum sample size threshold to ensure adequate power for the primary analysis focusing on the insomnia-symptom cohort, which compared participants at high risk and low risk of problematic smartphone use.
Assuming a two-sided α = 0.05, power = 0.80, and a mid–small standardized mean difference (d = 0.36) with the observed group-size ratio (low-risk:high-risk ≈ 0.75), the required total sample size was approximately N = 247. Recruitment continued until this threshold (N ≥ 240–250) was achieved, resulting in a final analytic sample of N = 246 (high-risk = 141; low-risk = 105).
Analyses were organized into prespecified and exploratory tiers. Prespecified primary analyses assessed group differences in major clinical sleep scales (e.g., ISI, PSQI-K, K-BRIAN) using ANCOVA models adjusted for age, sex, and BMI. Prespecified secondary analyses examined associations involving digital sleep and smartphone-use parameters through linear and correlational tests.
Exploratory analyses evaluated categorical or binary sleep-related outcomes (e.g., short sleep, daytime dysfunction) using logistic regression.
Sensitivity analyses included re-estimating models after covariate exclusion, applying log transformations to skewed variables, and using bootstrapped standard errors for robustness checks. Detailed assumptions, analytic structure (prespecified vs. exploratory), and power thresholds are summarized in Supplementary Table 2.
Statistical analysis
Statistical analyses were performed using JMP®, version 17 (SAS Institute Inc., Cary, NC, USA). First, an unadjusted analysis was performed, without accounting for covariates. For continuous variables such as scores, normality was tested using the Shapiro-Wilk test.
If normality was rejected at the 5% level, the Wilcoxon test was used to compare values between the two groups; otherwise, a one-way analysis of variance (ANOVA) was applied. As a result, HR CR goodness of fit, mean HR day week, mean HR day holiday, max HR night week, max HR night holiday, mean HR night week, and max exercise intensity night week were conducted by ANOVA, and other continuous variables were conducted by the Wilcoxon test (Supplementary Tables 3–7). The chi-square test was used for categorical variables.
If JMP Pro 17 indicated that 20% of the expected frequencies were below 5, Fisher's exact test was applied. Therefore, Marriage Status, Life Pattern, PSQIK, BPS index, IRLS index, PHQ-9 index, and GAD-7 index were conducted by Fisher's exact test (Supplementary Tables 3–6). Additionally, to account for potential confounding, we performed covariate-adjusted analyses: continuous outcomes were analyzed using analysis of covariance (ANCOVA) with age, sex, and body mass index (BMI) as fixed covariates, and categorical outcomes were analyzed using multivariable logistic regression including the same covariates.
Adjusted mean differences are presented as least-squares means (LS-means) with 95% confidence intervals (CIs), and associations for categorical outcomes as adjusted odds ratios (aORs) with 95% CIs. Two-sided p values < 0.05 were considered statistically significant.
Ethics
This study was approved by the Institutional Review Board of Korea University Anam Hospital (IRB No. 2022AN0587), and written informed consent was obtained from all participants prior to data collection.
Results
Based on the SOS-Q score, the study participants were classified into SOHG and SOLG. The two groups comprised 141 and 105 participants, respectively. The smartphone high-risk group was significantly younger than the low-risk group, with an average age of 28.42 years, compared to 34.80 years in the low-risk group (Hodges–Lehmann difference = 5.00 years (95% CI 3.00–7.00), Z = 4.456, p < .0001, effect size r = 0.284).
Additionally, there were significantly more females in the high-risk group (102/141 vs 63/105, 0R = 1.74 (95% CI 1.02–2.98), p = 0.042, φ = 0.130). Moreover, there were significant differences between the groups in terms of marital status (df = 3, aOR = 0.35 (95% CI 0.20–0.61), p < .0001), employment status (68/140 vs 70/105; aOR = 0.42 (95% CI 0.24–0.74), p = 0.003), sleep environment (df = 6, aOR = 0.31 (95% CI 0.18–0.54), p < .0001), weight (M = 60.78 kg, SD = 12.52 kg vs. M = 64.45 kg, SD = 15.03 kg, Hodges–Lehmann difference = 4.00 kg (95% CI 0.00, 7.00), Z = 2.185, p = 0.029, effect size r = 0.139), and BMI (M = 22.03, SD = 3.73 vs.
M = 22.77, SD = 3.32, Hodges–Lehmann difference = 1.00 kg/m2 (95% CI 0.00–1.97), Z = 2.00, p = 0.0456, effect size r = 0.127) (Supplementary Table 3, Table 1).
Based on the ecological momentary assessment, caffeine intake, naptime, and stress did not differ significantly between the groups. However, a significant difference was observed in alcohol consumption (LS mean 19.87 (95% CI 11.01–28.72) vs 34.01 (95% CI 24.19–43.83), LS-mean difference = −14.14 (95% CI −27.34, −0.95), p = 0.036, partial η2 = 0.018.; Supplementary Table 4, Table 2).
In sleep-related clinical assessments, the ISI score was significantly higher in the high-risk smartphone overuse group (LS Mean = 15.57 (95% CI 14.79–16.35)) compared to the low-risk group (LS Mean = 13.42 (95% CI 12.56–14.29), LS-mean difference = 2.14 (95% CI 0.98–3.09), p = 0.0004, partial η2 = 0.052). Regarding the ISI classification, the proportion of participants classified as having severe insomnia was higher in the high-risk group (n = 13) than in the low-risk group (n = 5). The prevalence of mild insomnia was lower in the high-risk group (n = 60) than in the low-risk group (n = 68), whereas moderate insomnia was more frequent in the high-risk group (n = 68) than in the low-risk group (n = 32).
A significant difference was observed across insomnia severity categories (df = 3, aOR = 2.63 (95% CI 1.52–4.55), p = 0.0005). Moreover, significant differences were found across all other sleep-related scales, including the DBAS, K-BRIAN, PSQI-K, BERLIN, MEQ, BPS, and FIRST (Supplementary Table 5, Table 3).
A significant difference was observed in non-sleep-related clinical assessments, such as PHQ-9, GAD-7, KMDQ, KRQ-53, and MFS (Supplementary Table 6, Table 4).
On comparing the data collected by SOMDAY with Fitbit, the high-risk group exhibited a lower daytime minimum heart rate on both weekdays (LS Mean 61.75 (95% CI 60.40–63.10) vs 64.17(95% CI 62.68–65.66), LS-mean difference = −2.42 (−4.42 to −0.41), p = 0.0182, partial η2 = 0.023) and holiday (LS Mean 58.72 (95% CI 57.47–59.97) vs 60.83(95% CI 59.44–62.23), LS-mean difference = −2.11 (95% CI -3.98 to −0.24), p = 0.0268, partial η2 = 0.020). On holidays, exercise intensity differed between groups during both daytime and nighttime (daytime: LS Mean 0.20 (95% CI 0.17–0.23) vs 0.13 (95% CI 0.10–0.17), LS-mean difference = 0.06 (95% CI 0.02–0.11), p = 0.0052, partial η2 = 0.032; night time: LS Mean 0.11 (95% CI 0.09–0.23) vs 0.08 (95% CI 0.06–0.10), LS-mean difference = 0.03 (95% CI 0.00–0.06), p = 0.0451, partial η2 = 0.017 (Supplementary Table 7, Table 5).
Discussion
This study represents the first comprehensive investigation integrating digital phenotyping with extensive clinical assessments to examine smartphone overuse and insomnia relationships while controlling for key demographic confounders. Our multidimensional approach reveals profound insights into the independent associations between problematic smartphone use and sleep-mental health disturbances, while simultaneously demonstrating the value of combining objective digital biomarkers with traditional clinical measures (Cho, 2023).
The integration of continuous wearable monitoring with ecological momentary assessment represents a methodological advancement in understanding real-world smartphone-sleep interactions. Unlike previous studies relying solely on self-report measures, our approach captured objective physiological and behavioral patterns across four weeks of naturalistic observation, providing temporal resolution of smartphone overuse effects (Torous, Staples, & Onnela, 2015).
The convergence of subjective clinical scales with objective digital biomarkers revealed distinct patterns that would be impossible to detect through either modality alone. This multimodal approach enables more comprehensive characterization of smartphone overuse effects, moving beyond simple screen time metrics toward behavioral-physiological profiles (Dagum, 2018; Onnela & Rauch, 2016).
The key finding of this research—elevated K-BRIAN scores that persist after rigorous covariate adjustment (LS mean 47.85 (95% CI 45.92–49.79) vs 41.00 (95% CI 38.85–43.15), LS-mean difference = 6.86 (95% CI 3.97–9.74), p < .0001, partial η2 = 0.083)—implicates fundamental circadian system disruption as a central mechanism linking smartphone overuse to sleep pathology. This goes beyond simple sleep-timing preferences, pointing to deeper biological rhythm dysfunction affecting multiple physiological systems (Ashman et al., 1999; Giglio, Magalhães, Kapczinski, Walz, & Kapczinski, 2010). In parallel, the Insomnia Severity Index (ISI) also differed significantly (aOR: 2.63 (95% CI 1.52–4.55), p = 0.0005), with the high-risk group showing greater severity of insomnia, corroborating clinically meaningful sleep disturbance.
Clinically, such circadian disruption is substantial: biological rhythm instability is a core vulnerability factor for psychiatric morbidity and metabolic dysregulation, suggesting that smartphone overuse may contribute to broader health complications (McClung, 2007; Scheer, Hilton, Mantzoros, & Shea, 2009). Notably, the magnitude of K-BRIAN elevation in the smartphone overuse group approximates levels reported in seasonal affective disorder and delayed sleep–wake phase disorder, reinforcing the interpretation of clinically consequential circadian dysfunction (Chang, Reid, Gourineni, & Zee, 2009; Giglio et al., 2009).
The robust persistence of depression (PHQ-9 index, aOR: 2.77 (95% CI 1.66–4.62), p = 0.0001) and anxiety (GAD-7 index, aOR: 1.59 (95% CI 0.86–2.92), p = 0.0059) associations after covariate adjustment reveals genuine neuropsychiatric vulnerability pathways rather than demographic artifacts. These findings align with emerging neuroimaging evidence demonstrating brain changes in regions critical for mood regulation among individuals with problematic smartphone use (Lin et al., 2012).
The emergence of significant body sensation awareness differences (BSQ, LS mean 38.21 (95% CI 35.84–40.58) vs 35.32 (95% CI 32.69–37.94), LS-mean difference = 2.89 (95% CI −0.63–6.42), p = 0.1075, partial η2 = 0.011) only after covariate adjustment represents a novel clinical discovery. This pattern suggests that smartphone overuse may be associated with heightened interoceptive sensitivity—the ability to perceive internal bodily signals—which plays a crucial role in emotional regulation and anxiety disorders (Garfinkel, Seth, Barrett, Suzuki, & Critchley, 2015; Khalsa et al., 2018). Enhanced interoceptive awareness in the context of smartphone overuse may reflect chronic physiological arousal states, potentially contributing to anxiety symptoms.
The persistence of mood disorder screening elevations (KMDQ, LS mean 6.09 (95% CI 5.46–6.72) vs 4.97 (95% CI 4.27–5.67), LS-mean difference = 1.12 (95% CI 0.18–2.06), p = 0.0199, partial η2 = 0.022) suggests subclinical mood instability that may represent early-stage mood difficulties. Given the well-established relationship between circadian disruption and mood disorders, our findings support the consideration of sleep-wake cycle assessment in evaluating smartphone-related mental health concerns (Allison G. Harvey, 2008; Crouse et al., 2021).
The robust bedtime procrastination signal (BPS index, aOR: 1.96(1.12–3.41), p = 0.0173) highlights a specific self-regulatory mechanism linking smartphone use to sleep disturbance. Importantly, after ANCOVA adjustment for age, sex, and BMI, there was no significant between-group difference in sleep-onset time (weekday: LS mean 10.51 (95% CI 10.23–10.79) vs 10.63 (95% CI 10.32–10.94), LS-mean difference = −0.12 (95% CI −0.53–0.30), p = 0.5802, partial η2 = 0.001; holiday: LS mean 10.96 (95% CI 10.61–11.31) vs 10.92 (95% CI 10.53–11.31), LS-mean difference = 0.04 (95% CI −0.48–0.56), p = 0.8917, partial η2 = 0.000), indicating that procrastination reflects intentional bedtime delay rather than a systematic shift of the habitual sleep phase. This aligns with prior work framing bedtime procrastination as postponing sleep in the absence of external constraints, a pattern increasingly mediated by digital engagement (Kroese, De Ridder, Evers, & Adriaanse, 2014).
Elevated stress reactivity to insomnia (FIRST index, aOR: 1.67 (0.99–2.83), p = 0.0574) reveals heightened vulnerability to sleep disruption during stressful periods. This finding suggests that smartphone overuse populations may experience greater sleep difficulties during times of stress, indicating a potential area for targeted intervention (Carney et al., 2017; Harvey, 2011)
The persistence of lower minimum daytime heart rate differences after covariate adjustment presents an intriguing physiological marker. While lower resting heart rate typically indicates better cardiovascular fitness, in the context of smartphone overuse, this pattern may reflect altered autonomic nervous system function that warrants further investigation (Kemp & Quintana, 2013; Thayer, Åhs, Fredrikson, Sollers, & Wager, 2012).
The emergence of exercise intensity differences during holiday daytime periods (LS Mean 0.20 (95% CI 0.17–0.23) vs 0.13 (95% CI 0.10–0.17), LS-mean difference = 0.06 (95% CI 0.02–0.11), p = 0.0052, partial η2 = 0.032) represents a novel behavioral signature suggesting qualitative rather than quantitative activity changes. This pattern may indicate fragmented attention and multitasking behaviors that compromise physical activity engagement (Lepp, Barkley, Sanders, Rebold, & Gates, 2013; Rebold, Lepp, Sanders, & Barkley, 2015)
Our findings suggest that assessment of smartphone use patterns may be valuable in clinical evaluation of sleep and mental health difficulties, particularly among younger adults. The robust associations with circadian disruption, depression, anxiety, and behavioral sleep patterns indicate that problematic smartphone use represents a legitimate clinical concern requiring attention in treatment planning.
The identification of specific measures that remain significant after demographic adjustment—including K-BRIAN, ISI, PHQ-9, GAD-7, BPS, and FIRST—provides guidance for clinical assessment protocols. These findings also suggest that interventions addressing circadian stabilization, mood symptoms, and behavioral sleep patterns may be particularly relevant for individuals with smartphone-related sleep difficulties.
Formal interaction tests between smartphone-risk group and age, sex, and occupation were nonsignificant after multiplicity control; accordingly, we offer clinical targeting rather than claims of effect modification. Briefly, younger adults warrant circadian screening and bedtime-procrastination interventions; and women merit assessment of stress-reactive insomnia and comorbid anxiety may benefit from coaching to limit phone-mediated multitasking during discretionary daytime activity. These pointers align with our adjusted findings (K-BRIAN, ISI, PHQ-9, GAD-7, BPS, FIRST) and with literature on eveningness/social jetlag, light-mediated circadian delay, media-related bedtime procrastination, and mood risk linked to insomnia (Chang et al., 2015; Exelmans & Van den Bulck, 2016; Roenneberg et al., 2007).
The methodological innovations demonstrated in this study establish a foundation for future smartphone-health research. Priority areas include longitudinal studies to establish causal relationships, intervention trials targeting the specific mechanisms identified in our study, and replication in diverse populations to establish generalizability (Firth et al., 2017; Huckvale, Venkatesh, & Christensen, 2019)
The integration of digital phenotyping with clinical assessment approaches should be further developed and validated, particularly regarding the optimal duration and frequency of monitoring needed to capture meaningful behavioral patterns. Additionally, the novel finding regarding interoceptive sensitivity warrants specific investigation in future studies.
While our primary inferences now derive from multivariable models (ANCOVA/logistic) adjusted for age, sex, and BMI, residual confounding cannot be excluded (e.g., work schedule, stimulant/caffeine use, device light spectra/content, depression/anxiety severity beyond screening). The cross-sectional design limits causal inference. Moreover, we did not include polysomnography (PSG) or circadian phase markers (e.g., DLMO); and although consumer-grade wearables enhance ecological validity, their measurement precision is lower than research-grade devices (De Zambotti, Cellini, Goldstone, Colrain, & Baker, 2019).
Finally, the Korean sample may constrain international generalizability given cultural differences in smartphone norms and sleep practices (Park & Lee, 2012).
To isolate effects attributable to smartphone use, future work should employ longitudinal or experimental designs (e.g., timing/content/light-restriction interventions; phone-free activity blocks), incorporate PSG (and/or DLMO) alongside consumer wearables/EMA in a validation subsample, and pre-specify covariate adjustment to strengthen confounding control. Analyses should report adjusted effect sizes with 95% CIs and include sensitivity checks (e.g., events-per-variable thresholds, multicollinearity diagnostics, and robustness to missingness), thereby advancing causal interpretation while retaining real-world measurement.
Conclusions
We integrated objective wearable-derived digital phenotyping with validated clinical assessments to examine links between problematic smartphone use and sleep–mental health outcomes among individuals with insomnia. After adjustment for age, sex, and BMI, the high-risk smartphone-use group showed clinically meaningful elevations in circadian rhythm disturbance (K-BRIAN) and greater insomnia severity (ISI), alongside higher odds of depressive and anxiety symptoms. These findings support the clinical relevance of assessing smartphone-use patterns when evaluating insomnia and related psychiatric concerns and motivate tailored behavioral strategies—particularly circadian stabilization and reducing bedtime procrastination—to mitigate symptoms.
Given the cross-sectional design and potential residual confounding, causal inference is limited. Future work should employ longitudinal and experimental designs, include polysomnography and circadian phase markers in validation subsamples, and pre-specify covariate adjustment to clarify mechanisms and strengthen generalizability.
Supplementary material