The UK Biobank is a comprehensive longitudinal population-based cohort that enrolled 502,376 participants from 2006 to 2010 in the United Kingdom. In this study, we included participants who had baseline electronic device usage data and covariate data. The exclusion criteria were a history of the following diseases at baseline: dementia, stroke, PD, multiple sclerosis, demyelinating disorders, and other neurodegenerative diseases.

The UK Biobank was approved (number 21/NW/0157) by the North West Multi-centre Research Ethics Committee (MREC). Consent were acquired from participants at the recruitment center for the study. Compensation and consent details are available on the UK Biobank website [15]. This manuscript was approved for publication by the UK Biobank, and data in the UK Biobank were accessible for health-related research with the approval of the UK Biobank Ethics Advisory Committee. All the data used in the study were anonymized, and no identifiable features were included.

Data on the use of electronic devices were extracted from the UK Biobank database [16] and transferred into category variables: plays computer games (3 levels: never/rarely, sometimes, often) and length of mobile phone use (5 levels: never used a mobile phone at least once per week, 1 year or less, 2-4 years; 5-8 years, >8 years). For those who used mobile phones, the following data were extracted: weekly usage of mobile phones in the past 3 months (2 levels: <5 minutes, ≥5 minutes), hands-free device/speakerphone use with mobile phones in the past 3 months (2 levels: never/almost never, used), differences in mobile phone use compared to the previous 2 years (3 levels: no change, more frequent, less frequent), and usual side of the head for mobile phone use (3 levels: left, right, equally left and right). For details of the answer to each question and their form in the models, see Multimedia Appendix 1.

The details of the magnetic resonance imaging (MRI) performed can be found on the UK Biobank website and in previous studies [17-19]. In brief, participants were invited to attend the research centers for follow-up and further neuroimaging assessment from 2014. Imaging scans were conducted in 3 imaging centers with identical scanners and fixed platforms to minimize heterogeneity. Those who had a neurological disease before the imaging scan were excluded from the analysis.

Brain imaging was acquired with 3T Siemens Skyra (software platform VD13) using a 32-channel receive head coil. T1 images were collected at 1 mm isotropic resolution using 3D MPRAGE acquisition, with the superior-inferior field of view being 256 mm. Several structural metrics were provided by the UK Biobank using Freesurfer. The most segmented brain imaging metrics were selected to obtain the most detailed changes. We used metrics processed by Destrieux (a2009s) parcellation in Freesurfer to assess the area, mean thickness, and volume of the brain in 148 areas [20].

Diffusion tensor imaging (DTI) is an advanced MRI technology that can detect microstructural abnormalities in white matter (WM) before the visual injury emerges. Diffusion data were collected with 2 b-values (1000 and 2000 s/mm²) at 2 mm spatial resolution, with a multiband acceleration factor of 3. Fifty distinct diffusion-encoding directions were acquired for each diffusion-weighted shell. The diffusion preparation was a standard (“monopolar”) Stejskal-Tanner pulse sequence. We used 5 DTI metrics provided by the UK Biobank: fractional anisotropy (FA), mean diffusion (MD), orientation dispersion (OD), intracellular volume fraction (ICVF), and isotropic (free) water volume fraction (ISOVF) [17]. FA and MD were calculated by fitting a voxel-wise diffusion tensor model through DTIFIT in FSL. The OD, ICVF, and ISOVF were calculated by fitting neurite orientation dispersion and density imaging models using accelerated microstructure imaging via convex optimization [21,22]. Low FA or high MD indicate a higher overall deficit in WM fiber integrity. Specifically, OD represents the overall coherence of fibers, the ICVF represents axonal/neurite density, and the ISOVF indicates the free-water fraction [22].

We used the diagnosis of neurodegenerative diseases (dementia outcomes and parkinsonism outcomes) from category 47 and category 50: all-cause dementia (ACD), AD, vascular dementia (VD), all-cause parkinsonism (ACP), and PD. These outcomes were defined through algorithmic combinations of self-reported medical conditions, hospital admissions, and death registries [23].

Data were described as mean (SD) values for continuous variables and counts (percentages) for category variables. Continuous variables were compared using the t test, and category variables were compared using the chi-square test or the Fisher exact test. We used Cox regression analysis to evaluate the risk of dementia. Participants were considered at risk for dementia or parkinsonism from baseline (2006-2010) and were followed up to the date of the first diagnosis, death, loss of follow-up, or the last hospital admission.

The covariates included age, sex, the BMI, smoking (ever or never), drinking (ever or never), hypertension, diabetes, education (college/university degree or lower degree), the polygenic risk score, socioeconomic status (Townsend deprivation index, a score calculated according to the output area in which participants’ postcodes were located and the preceding national census output areas), the average total household income before tax, the mental health score, and physical activity. Using scaled Schoenfeld residuals to test proportional hazards, there was no indication of a violation of the hypothesis. In the sensitivity analysis, we further conducted several models: (1) model 1 excluded those diagnosed with outcome diseases within 5 years from baseline to exclude the potential reverse causality, (2) model 2 only included diagnoses that were from the hospital or death registration; (3) model 3 was used to replace the polygenic risk score (PRS) with the status of the apolipoprotein ε4 gene, and (4) model 4 divided patients into 2 groups according to age (60 years old) at baseline to exclude the influence of age.

Linear regression models were used with variables using electronic devices as independent variables and neuroimaging metrics as dependent variables. Sex, age, ethnicity, the BMI, smoking, drinking, hypertension, diabetes, education, socioeconomic status, the WM hyperintensity volume, the whole WM volume, handedness (only for the usual use of the side of the mobile phone), and the average total household income before tax were included as covariates [24]. In the sensitivity analysis, we further adjusted the mental health score and physical activity. The variance inflation factor was tested and controlled under 5 to avoid multicollinearity. We used Bonferroni-corrected P values to determine the statistical significance: .05/5 for disease risk analysis, .05/(48 tracts×5 metrics) for DTI, and .05/148 regions for the surface, thickness, and volume in linear regression models. All analyses were performed using R version 4.1.2 (R Foundation for Statistical Computing).

The inclusion flowchart is displayed in Figure 1 (detailed version in Multimedia Appendix 1), and the demographic characteristics of the participants are shown in Table 1. A total of 277,020 and 276,665 participants were included in the Cox regression model for dementia and parkinsonism, respectively. The mean age of the dementia and parkinsonism analysis groups at baseline was 55.85 (SD 8.07) years and 55.86 (SD 8.07) years, respectively. Half of the included participants (dementia group: n=139,166, 50%; parkinsonism group: n=139,276, 50%) were female. The mean duration of follow-up was 13.90 (SD 1.99) years and 13.89 (SD 2.01) years for the dementia and parkinsonism groups, respectively. We further compared the characteristics of the included and excluded participants (Tables S1 and S2 in Multimedia Appendix 1). The participants in the included risk analysis were younger, had higher education, a lower Townsend deprivation index, and a higher income compared to those excluded from the analysis.

In the follow-up, we included 3007 (1.1%) participants with ACD, 1231 (0.4%) with AD, and 631 (0.2%) with VD. Lengthy mobile phone use (≥2 years) at baseline was related to a lower risk of ACD and VD compared to never using a mobile phone at least once per week (Table 2 and Table S3 in Multimedia Appendix 1). Additionally, participants who had used mobile phones for 5-8 years had a significantly lower risk of AD.

To exclude the influence of reverse causality, several sensitivity analyses were conducted. After excluding participants who developed outcomes within 5 years after baseline, we found that lengthy mobile phone use was related to a lower risk of ACD (mobile phone use≥2 years), VD (mobile phone use=2-4 years), and AD (mobile phone use=5-8 years), as shown in Table S4 in Multimedia Appendix 1. After excluding participants whose source of outcomes was not from the hospital or death registration, we found that lengthy mobile phone use (≥2 years) was related to a lower risk of ACD and VD (Table S5 in Multimedia Appendix 1). To analyze the influence of apolipoprotein ε4, we adjusted the status of apolipoprotein ε4 instead of the PRS and found that lengthy mobile phone use was related to a decreased risk of ACD (mobile phone use≥2 years) and VD (mobile phone use=2-4 years), as shown in Table S6 in Multimedia Appendix 1.

To exclude the influence of age, we conducted a separate analysis based on participants’ age at baseline. We found that lengthy mobile phone use was related to a decreased risk of ACD (mobile phone use≥2 years) and AD (mobile phone use≥5 years) in participants younger than 60 years (Table S7 in Multimedia Appendix 1). Lengthy mobile phone use was also related to a decreased risk of ACD (mobile phone use≥2 years), AD (mobile phone use≥5 years), and VD (mobile phone use≥2 years) in participants older than 60 years (Table S8 in Multimedia Appendix 1).

In the follow-up, we included 1578 (0.6%) participants with ACP and 1415 (0.5%) with PD. Often playing computer games was related to a lower risk of ACP (hazard ratio [HR] 0.604, 95% CI 0.425-0.859; P=.005) and PD (HR 0.574, 95% CI 0.392-0.841; P=.004) compared to never or rarely playing computer games (Table S9 in Multimedia Appendix 1). However, after excluding those who developed outcomes within 5 years, none of the above results remained statistically significant (Table S10 in Multimedia Appendix 1). We excluded participants whose source of outcomes was not from the hospital or death registration and reached a similar result that often playing computer games was related to a lower risk of ACP and PD (Table S11 in Multimedia Appendix 1). After age stratification, in participants younger than 60 years, we found that the association between often playing computer games and a lower risk of ACP did not reach statistical significance (HR 0.393, 95% CI 0.175-0.844; P=.02; Table S12 in Multimedia Appendix 1). However, lengthy mobile phone use (≥5 years) was related to a decreased risk of ACP and PD in participants older than 60 years (Table S13 in Multimedia Appendix 1).

The demographic characteristics of participants who underwent neuroimaging analysis are displayed in Table S14 in Multimedia Appendix 1. A total of 35,643 participants were included, 48% (n=17,109) of them were male, and the mean age at baseline was 54.77 (SD 7.51) years. The mean duration between baseline and the neuroimaging scan was 8.97 (SD 1.73) years.

Areas where the brain structure metrics at follow-up were related to the use of electronic devices at baseline are displayed in Tables 3 and 4. Using mobile phones for a long duration (≥2 years) was related to a thicker cortex in the left superior temporal sulcus. In addition, lengthy mobile phone use (≥8 years) was related to a significantly higher volume of the parahippocampal gyrus and thicker left orbital gyri, left precuneus, left superior temporal sulcus, right posterior-dorsal part of the cingulate gyrus, right middle temporal gyrus, and right subparietal sulcus (Table 3). Lengthy mobile phone use (2-4 years, ≥8 years) was related to higher mean FA in the pontine crossing tract on the FA skeleton (Table 4). Often playing computer games was related to a higher volume in the left anterior part of the cingulate gyrus and sulcus and the left triangular part of the inferior frontal gyrus, while sometimes playing computer games was related to a lower volume of the parahippocampal gyrus (Table 3). In addition, often holding a mobile phone on the right side of the head was related to larger areas in the bilateral posterior division of the middle temporal gyrus, bilateral anterior division of the cingulate gyrus, and a larger area of several regions compared to those who usually held their phones on the left side of the head (Table 3). Results of sensitivity analyses of neuroimaging are displayed in Tables S15-S17 in Multimedia Appendix 1.

Overall, the findings of this study showed that electronic device use is associated with a reduced risk of neurodegenerative diseases (Table 5). This table summarizes the results of the main models but not the results of sensitivity analyses.

In our study, we found evidence suggesting that the use of electronic devices is associated with a reduced risk of neurodegenerative diseases. Specifically, we observed that using mobile phones for 2 or more years is linked to a decreased risk of ACD and VD compared to never/rarely using mobile phones. Upon conducting sensitivity analyses, we confirmed the stability of the association between long mobile phone use and a lower risk of ACD and VD. Moreover, individuals aged 60 years and older appeared to benefit from extended duration of mobile phone use with a reduction in the risk of AD, ACP, and PD. Furthermore, our neuroimaging analysis revealed that various electronic device use habits, including the duration of mobile phone use, the frequency of computer use, and the preferred side of the head for holding mobile phones, were associated with distinct characteristics in the gray matter and WM structures of several brain regions over the follow-up period.

Our findings suggest that using a mobile phone for more than 2 years is associated with a reduced risk of ACD and VD in the middle-aged population, with no significant impact of weekly mobile phone use. Previous studies using UK Biobank data have reported an association between mobile phone use and a lower risk of ACD in individuals over 60 years old [7]. Our study builds upon this knowledge by focusing on specific types of dementia and details of mobile phone use. We found that a long weekly time of mobile phone use (≥5 minutes) does not alter the risk of dementia compared to a shorter weekly time (<5 minutes). However, the duration of mobile phone use affects the association between mobile phone use and reduced risk of dementia. Using a mobile phone enriches the user’s social and mental activities, potentially leading to improved cognition in later life [25,26]. The association between mobile phone use and a decreased risk of dementia is partly mediated by rich social support and flourishing leisure activities [7]. Additionally, prior research has suggested that registering a mobile phone decreases the risk of PD [8]. Our study further revealed that this association is observed only in populations aged 60 years and older, where a long duration of mobile phone use decreases the risk of developing ACP and PD compared to those who rarely use mobile phones.

The benefits of using electronic devices can be reflected in improved brain structure. We found that long durations of mobile phone use are associated with a thicker cortex in multiple areas of the brain. Specifically, individuals who use mobile phones for more than 8 years exhibit a higher volume of the parahippocampal gyrus, a region crucial for memory function [27]. This finding is particularly significant as atrophy in the parahippocampal gyrus is observed in the early stages of AD, serving as an early pathological biomarker [28,29]. Previous studies have shown that daily mobile phone use for less than 1 hour is associated with lower functional connectivity in certain brain regions, such as the superior temporal sulcus, and higher functional connectivity in others, like the cingulate [14]. In this study, we observed that only weekly mobile phone use of 5 minutes or more is associated with a decreased volume and area in the right posterior transverse collateral sulcus compared to less than 5 minutes of weekly use. However, no significant results were found in other areas or the DTI analysis. These findings suggest that daily mobile phone use, without reaching problematic levels, has minimal impact on the user’s brain structure. Interestingly, we found inconsistent effects of occasional or frequent computer usage on the brain structure. Often playing computer games is associated with a higher brain volume, while sometimes playing computer games is related to a lower brain volume. For those who sometimes play computers, the related shortcomings, such as sedentariness, are more prominent and result in cortical atrophy [30].

This study has several strengths. First, we adjusted for covariates that may have influenced the results using comprehensive data from the UK Biobank, which is the largest population-based cohort in the world. This allowed us to provide important information on the use of electronic devices. Second, the cohort’s follow-up duration was notably extensive, allowing sufficient time for patients to reach the point at which diseases developed. Additionally, the duration between baseline and neuroimaging scans offered crucial insights into the long-term effects of electronic device use on brain health.

There was 1 limitation that should be mentioned. The mobile phone use in this study was limited to receiving calls. The rapid advancement and evolution of electronic devices in recent years have transformed the usage patterns of mobile phones, heightening the potential for dependency [31,32]. However, the evolution happened years after the launch of the UK Biobank cohort. Our study still provides important insights into the long-term influence of electronic device use. The time that people spend on mobile phones now is much longer than it was decades before [31,33]. However, this study found that there are no additional benefits of longer weekly usage. In addition, playing games on the mobile phone is popular nowadays [33]. We found an association between playing computer games and decreased neurodegenerative disease risk, but the association is unclear if game devices change from computers to mobile phones. Future studies should record the details of the daily time and the type of function of mobile phone use to analyze the impact of modern electronic devices on brain health.

In conclusion, our study revealed that individuals with long duration of mobile phone use are associated with a decreased risk of ACD and VD compared to those who rarely use mobile phones. The protective effect of mobile phone use is also observed in individuals older than 60 years, who experience a reduced risk of AD, ACP, and PD. However, spending more time on mobile phones per week may not further reduce the risk of neurodegenerative disease. Our findings also indicate that electronic device use can impact the brain structure. Specifically, using a mobile phone and often playing computer games are associated with a better brain structure, while sometimes playing computer games is linked to a poorer brain structure. These results highlight the complex relationship between electronic device use and neurodegenerative disease risk, as well as the influence on the brain structure. Future studies should concentrate on the mechanism underlying the influence of using mobile phones and computers and new electronic devices on health.