This review adhered to the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) 2020 statement [33] and was registered on PROSPERO (CRD42024585459; Checklist 1).

We searched PubMed, Embase, Web of Science, PsycINFO, Scopus, CNKI, and Wanfang databases, from the inception of the databases to August 31, 2024. Only eligible full texts in English or Chinese were considered for review. Two thematic blocks of keywords were used: “Aged” and “Digital exclusion.” The search terms were (aged or “middle aged” or “older adult*” or aging or “elder*”) AND (“digital exclusion” or “digital inclusion” or “digital divide” or “Internet access” or “Internet use*” or “Internet non-use” or “web usage” or “web use” or “Internet usage” or “digital engagement”). In terms of developing search strategies, we first identified standardized terminology by consulting the Medical Subject Headings (MeSH) database in PubMed. This step ensured the inclusion of entry terms. Subsequently, we examined key published literature to incorporate nonstandardized terms that are conceptually related to digital exclusion, such as “digital divide,” “digital inclusion,” and “digital engagement” [27,34,35]. The detailed search strategies are shown in Multimedia Appendix 1.

The inclusion criteria for studies were as follows: (1) middle-aged and older adults as research participants, explicitly aged 45 years and older [36]; (2) digital exclusion was clearly specified as the exposure variable, without redefinition based on dose or frequency; and (3) reporting at least one measure of physical or cognitive function. Specifically, physical function was measured using standardized assessments of functional independence and mobility (eg, activities of daily living [ADLs] and Fried frailty phenotype). Cognitive function was assessed using standardized screening tools (eg, Mini-Mental State Examination [MMSE]) or based on a clinical diagnosis of cognitive impairment or dementia; (4) provision of the odds ratios (ORs), hazard ratios (HRs), risk ratios (RRs), and their corresponding 95% CIs for the associations between digital exclusion and health outcomes; and (5) observational studies.

This study excluded (1) editorials, letters, comments, meeting abstracts, and newspaper articles; (2) studies unable to provide the full text; (3) studies that only reported correlation metrics, such as Pearson r or Spearman ρ, without providing data that could be used to calculate ORs, RRs, HRs, and their 95% CIs; and (4) studies that were not published in English or Chinese.

Study selection was conducted independently in 2 stages by 2 researchers. A third researcher was invited to resolve the disagreement. Initially, the titles and abstracts of the studies were evaluated in terms of eligibility criteria. In the subsequent stage, full texts of preselected studies were reviewed to confirm their eligibility.

The selected studies were extracted by 2 researchers using a form to extract data on the setting, study design, sample size, basic participant information, digital exclusion assessment tools, categories of digital exclusion, outcome assessment tools, categories of assessment, covariates, and effect sizes of outcomes. Discrepancies between the researchers regarding data extraction were resolved through discussion and consensus with the assistance of a third researcher.

Two researchers independently assessed the methodological quality of each study using the Newcastle-Ottawa Scale (NOS), a quality assessment tool for both cohort and cross-sectional studies [37]. The tool comprises 8 items with a score range of 0‐9 for cohort studies, and 7 items from 0 to 8 for cross-sectional studies [38], as detailed in Multimedia Appendix 2 [17,18,30,32,39-47] and Multimedia Appendix 3 [19,24,31,48-50]. Initially, we conducted a pilot evaluation of 4 articles (2 cohort studies and 2 cross-sectional studies) to standardize the assessment criteria. Discrepancies between the researchers were discussed, and a third researcher was invited to resolve them collectively.

Descriptive analysis was conducted to summarize the characteristics of the studies. For meta-analysis, the associations between digital exclusion and health outcomes were quantitatively synthesized, and effect sizes were measured using RR, OR, HR, and 95% CIs. Separate meta-analyses were conducted for ORs, RRs, and HRs. Effect sizes adjusted for potential confounders were extracted to ensure comparability across studies, allowing them to be regarded as standardized estimates [51]. When the original studies used “digital exclusion” as the reference group, the reported effect sizes were inverted to represent the impact of digital exclusion versus nonexclusion. This conversion was justified by the mathematical principle that inverting a ratio measure was equivalent to swapping the reference and exposure groups [52]. And all conversions were performed with double-checking for accuracy.

The random-effects model was used if significant heterogeneity (I²>50% or P<.1) was found among the included studies; otherwise, the fixed-effect model was used. Due to the inclusion of fewer than 10 studies in the meta-analysis, publication bias analysis was not conducted. The meta-analysis and forest plots were performed using R software (R Core Team) with the main package of “metafor.”

Leave-one-out sensitivity analysis was conducted to evaluate the influence of each study on the pooled association estimate. In this procedure, the meta-analysis was repeated multiple times, each time omitting one study and recalculating the pooled association using the remaining studies. When between-study heterogeneity (I²) exceeded 50%, a random-effects model was applied for the recalculated estimate [53]. Specifically, a study was considered to have a substantial influence if its exclusion caused the new point estimate to fall outside the 95% CI of the original pooled estimate. A study was also deemed influential if the effect size changed noticeably compared to the primary analysis. Noticeable changes were defined as either a reduction of 25‐50 percentage points or a shift from high heterogeneity (I^²>50%) to low or moderate heterogeneity (I^²≤50%) [54-56]. Due to the limited number of available studies, subgroup analysis based on different definitions, frequencies, or ages could not be performed. All results were presented visually using forest plots, along with corresponding CIs. A statistical significance threshold of P<.10 was adopted.

A total of 31,414 articles were searched from the databases. After excluding duplicate publications (n=10,110) and ineligible articles (n=21,198), 19 studies were included in this review (Multimedia Appendix 4) [17-19,24,30-32,39-50]. The selection process of the studies was summarized in the PRISMA flow diagram (Figure 1).

Tables 1 and 2 provide the characteristics of the included cohort and cross-sectional studies, respectively. Three studies focused on middle-aged and older adults aged 45 years or older [24,42,45], while the remaining articles specifically concentrated on older adults aged 60 years or older [17-19,30-32,39-41,43,44,46-50]. Thirteen cohort studies [17,18,30,32,39-47] and 6 cross-sectional studies [19,24,31,48-50] were from different regions, including Spain, England, Japan, the United States, China, Europe, Mexico, Brazil, Sweden, and the Netherlands. These studies were conducted between 2012 and 2024, with the sample sizes ranging from 409 to 155,695. The follow-up duration of the cohort studies ranged from 3 to 17.1 years.

The assessment tools for digital exclusion showed diversity among the included studies. None of the studies used a validated scale; instead, they all used single-item measures to assess digital exclusion. Most studies (n=14) [17,18,24,30,31,39-43,45-48] defined digital exclusion as nonuse of the internet, while other studies defined digital exclusion as nonuse of computers (n=2) [32,44], nonuse of information and communication technology (n=1) [50], nonuse of mobile payment (n=1) [49], and inability to send or receive online messages (n=1) [19].

Seven studies focused exclusively on physical function [17,19,24,30,32,39,50], while 11 examined cognitive function [18,40-49]. Only 1 study assessed both physical and cognitive outcomes [31]. Physical function was assessed by ADLs (n=3) [19,24,30], Frailty index (n=1) [17], and Fried frailty phenotype (n=2) [32,50], with 1 study obtaining physical function outcomes by linking to data from local government systems [39]. The cognitive function was assessed using numerous cognitive tests, including MMSE (n=5) [40,41,47-49], orientation, memory, and executive function tests (n=1) [18], Modified Telephone Interview for Cognitive Status (n=2) [42,43], short-form Informant Questionnaire on Cognitive Decline in the Elderly (IQCODE) questionnaire (n=1) [45], and Brief Community Screening Instrument for Dementia (n=1) [31], with some studies obtaining cognitive function outcomes by linking to data from local government systems.

The overall risk of bias assessment of cohort and cross-sectional studies is shown in Multimedia Appendices 2 and 3.

The overall risk of bias score of cohort studies had a low risk of bias, varying from 7 to 9 points. Regarding the selection bias, all studies included the selection of a nonexposed cohort and the ascertainment of exposure. Similarly, in the comparability and outcome domains, all studies ensured the comparability of cohorts based on design or analysis, assessment of outcomes, and an adequate follow-up period for outcomes to occur. However, 6 studies failed to demonstrate that the outcome of interest was not present at the start of the study [17,18,30,41,43,46], and 4 studies had inadequate cohort follow-up [18,40,41,47].

The overall risk of bias score of cross-sectional studies ranged from 3 to 7 points. All studies clearly stated the study population, had a sample size of more than 300, and assessed the outcome. However, none of the studies reported whether the studied sample had a participation rate of at least 70% (7/10) of the invited individuals.

Four studies investigated the prospective association between digital exclusion and frailty, basic activities of daily living (BADLs), instrumental activities of daily living (IADLs), and incident disability [17,30,32,39]. García-Esquinas et al [32] and Li et al [17] investigated the OR value of the association between digital exclusion and frailty. The meta-analysis indicated that there was no statistically significant prospective association between them (OR 1.21, 95% CI 0.92‐1.59, I²=95.2%; Figure 2A). Lu et al [30] investigated the incidence rate ratio (IRR) value of the association between digital exclusion and BADLs or IADLs. The pooled associations indicated that digital exclusion was associated with decreased BADLs (IRR=1.35, 95% CI 1.12‐1.64, I²=94.7%; Figure 2B) and IADLs (IRR=1.46, 95% CI 1.13‐1.89, I²=96.2%; Figure 2C).

Regarding the cross-sectional association, Liu et al [31] and Wen et al [24] analyzed the association between digital exclusion and ADLs. Meta-analysis indicated that there was no significant pooled association between digital exclusion and ADLs (OR 1.23, 95% CI 0.41‐3.73, I²=91%; Figure 2D). Tomioka et al [39] found an association between internet use and a lower risk of incident disability. While García-Vigara et al [50] revealed that digital exclusion was associated with a higher risk of frailty.

The pooled analysis indicated that digital exclusion had prospective associations with decreased BADLs and IADLs, no statistical significance in the prospective association with frailty, or cross-sectional association with ADLs.

Concerning the cohort studies, Krug et al [40], Berner et al [41], and Wang et al [18] explored the OR value of the association between digital exclusion and decreased cognitive function. Meta-analysis of these cohort studies demonstrated that digital exclusion was significantly associated with higher odds of decreased cognitive function (OR 2.22, 95% CI 1.95‐2.53, I²=77.3%; Figure 3A). Williams et al [43] and Quialheiro et al [47] investigated the relationship between digital exclusion and cognitive impairment. However, meta-analysis of 2 studies [43,47] found no statistically significant association (RR 2.08, 95% CI 0.98‐4.44, I²=78.2%; Figure 3B). Cho et al [42] and d’Orsi et al [45] analyzed the HR value of the association between digital exclusion and dementia. Meta-analysis demonstrated a significant association between them (HR 1.78, 95% CI 1.43‐2.22, I²=0%; Figure 3C).

In terms of the cross-sectional studies, a meta-analysis of 2 studies [48,49] showed that digital exclusion was associated with a decline in MMSE scores (OR 2.90, 95% CI 2.07‐4.07, I²=0%; Figure 3D).

There were prospective associations between digital exclusion and dementia and decreased cognitive function, as well as cross-sectional associations with MMSE scores, but no statistically significant prospective association with cognitive impairment.

Due to the limited number of studies, sensitivity analysis for the association between digital exclusion and physical function was not performed. In the sensitivity analysis of the prospective association between digital exclusion and cognitive function, a leave-one-out approach was applied, and only minor changes were observed in the results, indicating the robustness of our findings (Figure 4).

The sensitivity analysis revealed only minor variations in the results, confirming the stability of our findings, as evidenced by the prospective association between digital exclusion and cognitive function.

To the best of our knowledge, this study presents the first systematic review and meta-analysis to comprehensively evaluate the associations of digital exclusion on both physical and cognitive functions across multiple countries. Synthesizing evidence from 13 cohorts [17,18,30,32,39-47] and 6 cross-sectional studies [19,24,31,48-50], we found that digital exclusion was prospectively associated with declines in IADLs and BADLs, as well as increased risks of incident disability, cognitive impairment, and dementia, and was cross-sectionally related to lower MMSE scores. However, no statistically significant prospective association was observed between digital exclusion and frailty.

Consistent with previous studies [30,39], the meta-analysis identified a negative prospective association between digital exclusion and physical function. Several underlying explanations may account for the associations observed in this study. First, digital exclusion was associated with a minimal chance of using eHealth [57] and limited access to acquire digital preventive and therapeutic health services [58,59]. Second, it might cause a decrease in physical exercise behavior among middle-aged and older adults [60], which can lead to a reduction in skeletal muscle mass, strength, and physical mobility [61], ultimately accelerating the deterioration of physical functions and creating a vicious cycle [62-64]. Additionally, middle-aged and older adults experiencing digital exclusion were prone to dopamine imbalance and induced negative emotions such as loneliness and social isolation [65,66], increasing the risk of incident disability [67].

The associations of digital exclusion on frailty remained debated, and the meta-analysis found no statistically significant prospective association. Frailty is influenced by numerous factors such as age, gender, physiological changes, and other factors [68-70]. Furthermore, the influence of digital exclusion on frailty is likely indirect. It may operate through social isolation, limited access to health information, and reduced physical or cognitive engagement, which accumulate gradually over time [32]. These factors could mask a true association in short-term analyses. In addition, the limited number of studies included in the meta-analysis provided only a modest evidence base. The association between digital exclusion and frailty requires further validation through standardized longitudinal studies with longer follow-up periods and consistent assessments of both digital exclusion and frailty.

Meanwhile, digital exclusion did not demonstrate a cross-sectional association with physical function. One explanation for this inconsistency was the influence of confounding factors. It was also possible that socioeconomic and health-related factors masked the cross-sectional relationship between digital exclusion and physical function. Wen et al [24] reported a cross-sectional association between digital exclusion and physical function without adjusting for confounders [71], whereas Liu et al [31] found no such association after adjustment. In addition, the cross-sectional design was not well-suited to detect the slower impact of digital exclusion on physical function, which emerged gradually through changes in health behaviors and disease management. These findings underscore the need for longitudinal designs to clarify temporal relationships and potential causal pathways.

Findings of this study regarding the relationship between digital exclusion and cognitive function broadly aligned with previous research [72-75]. Apart from the aforementioned impact of health care resources and negative psychological emotions, digital exclusion may influence cognitive function through several mechanisms. It not only reduces engagement in cognitively stimulating activities that build cognitive reserve but may also promote negative emotions, which can contribute to neurovascular dysfunction [73,76-78]. Additionally, it is linked to structural brain changes such as reduced volume of the globus pallidus [48].

However, it was noteworthy that no statistical prospective association was observed between digital exclusion and cognitive impairment. This discrepancy could be attributed to several methodological variations across studies. For instance, the studies by Williams et al [43] and Quialheiro et al [47] both investigated populations in the United Kingdom with similar follow-up durations. But the latter had a much smaller sample size (n=594 vs 3937), which could have limited its statistical power and affected the ability to detect significant associations. Furthermore, the 2 studies [43,47] used different assessment tools, which might contribute to the considerable heterogeneity in the meta-analysis.

Digital exclusion or excessive internet use both had negative impacts on health outcomes [79-81]. These findings revealed “two sides of the same coin” and emphasized the importance of the frequency of internet use. In terms of the dose-response relationships, previous studies have found an inverted U-shaped association between internet use and cognitive function [42]. In the large-scale prospective cohort study of Cho et al [42], using the internet almost weekly was most beneficial to cognitive function. Additionally, Williams et al [43] and Liao et al [82] have indicated that different types of internet use exert varying impacts on physical and cognitive functions. Previous research showed diverse internet activities benefit cognition more than single ones [83]. However, the optimal types, combinations, and intensity of digital engagement for enhancing physical and cognitive functions in middle-aged and older adults remain unclear. Future research should investigate these factors across different age groups and geographic regions to tailor digital interventions to the needs and capacities of diverse populations.

The prevalence of digital exclusion varies substantially across studies due to differences in national populations, as well as inconsistencies in the definition and assessment tools of digital exclusion among middle-aged and older adults. Some studies had suggested that digital exclusion constituted a part of social exclusion [30], as digital technology marginalized older adults from the digital world [84-86]. It might be associated with a perceived lack of knowledge, inability to access necessary technology and services, and barriers posed by mental health difficulties [85]. The absence of a unified definition and assessment tool not only reduced the reliability and generalizability of our findings due to high heterogeneity but also hindered the development of effective policies and interventions to address digital exclusion. Future research should focus on establishing a consensus-based conceptual framework and developing rigorously validated assessment instruments of digital exclusion. To ensure global applicability, these tools should be culturally and contextually adaptable, with careful attention to cross-cultural validation that ensures their relevance across diverse populations and settings.

Current assessments of physical and cognitive function in the included studies rely largely on scales that capture activities of daily living, memory, orientation, and related abilities [18,30]. While these tools provide valuable information on functional status, they are largely subjective and may be influenced by self-report bias, assessor variability, or ceiling and floor effects [30,87]. Moreover, evidence directly linking these functional scales to objective biomarkers remains limited. Gotti et al [88] found that inflammatory and immune biomarkers changed significantly after internet-based psychological interventions, demonstrating that digital engagement can trigger measurable biological changes. This evidence presented a promising avenue for exploring underlying health mechanisms through objective biological markers. Advances in digital health further extended this possibility [89]. Chan et al [90] developed Watch Walk wrist-worn algorithms that accurately quantify gait speed [90]. This precision allowed large-scale use and turned routine movements into scalable digital biomarkers. Fan et al [70] demonstrated that integrating wearable gait data with machine learning models can substantially improve frailty prediction [91]. Together, these developments illustrate an emerging capacity to collect continuous or longitudinal data in real-world settings, enabling earlier detection of functional decline and a more nuanced understanding of how digital behaviors may shape health trajectories over time.

Although growing evidence supports the use of digital technologies to improve physical and cognitive function [92,93], reconciling digital exclusion and digital adoption remains a major challenge. To bridge this divide, it is essential to implement strategies that enhance digital access, strengthen digital literacy, and support digital assimilation [94]. Facilitating the introduction and adoption of user-friendly technologies, particularly those tailored to the needs of older adults, can play a critical role in ensuring equitable digital participation. A growing body of research highlighted co-design as a key approach [95-97]. Chan et al [98] used co-design workshops to involve older adults, health care providers, and community staff in refining a mobile health prototype, ensuring better usability and engagement. Digital navigators can offer literacy training, resolve technical issues, and guide app use, thereby improving technology adoption and integration into care [99]. Beyond design, social support also plays a crucial role. Hänninen et al [100] found that younger family members often act as “warm experts,” offering practical, personalized support that helps older adults manage digital tools with greater confidence.

First, the generalizability of our findings is subject to certain constraints. Our meta-analysis was primarily restricted to studies published in English and Chinese, as these were the languages most accessible to the research team. Under the constraints of time and resources, conducting translations of studies in other languages was not feasible. This limitation may hinder the applicability of the results to regions where other languages are dominant or where cultural and socioeconomic contexts differ significantly. Additionally, the overall number of available studies focusing specifically on digital exclusion remains limited, particularly for middle-aged populations. Most included studies predominantly involved older adults, resulting in inadequate representation of middle-aged cohorts. Due to the limited number of included studies, it is hard to follow different age, sex, education level, income, or rural-urban residence groups to conduct subgroup analyses. This restricts the broader generalization of our conclusions across middle-aged and older adults. Future research should aim to incorporate more diverse geographical and linguistic sources, as well as prioritize well-defined studies involving middle-aged populations, to enhance the generalizability and age-specific understanding of digital exclusion.

Second, our meta-analysis is influenced by the substantial heterogeneity observed in some pooled estimates. This heterogeneity mainly arose from clinical diversity, including variations in population characteristics across different countries and income levels. Methodological diversity also contributed to heterogeneity, including differences in how digital exclusion was assessed across the large cohort studies, restriction to risk-based effect measures, variations in sample sizes, and inconsistent adjustment for covariates. By restricting quantitative synthesis to risk-based effect measures, our findings primarily reflect associations framed within an epidemiological risk perspective and may not fully capture evidence derived from correlation-based analyses using continuous measures. We performed sensitivity analyses to test the robustness of the findings and to identify key sources of heterogeneity. However, the limited number of studies in certain subgroups restricted more detailed investigations. If enough studies are available, conducting a subgroup analysis based on different definitions is informative. In the future, it is necessary to unify the definition of digital exclusion and develop recognized tools for assessment.

Third, the impact of the types and frequency of digital exclusion on physical and cognitive functions could not be explored. Future in-depth research could explore the dose-response relationships of various frequencies and types of digital exclusion on physical and cognitive functions.

Fourth, the exclusion of gray literature might lead to publication bias. It is advisable to search specialized gray literature databases when time, resources, and evidence permit.

Our findings emphasize the negative associations between digital exclusion and physical or cognitive functions. Although our results do not directly address optimal patterns of digital engagement, it is noted that moderate and diverse digital use may be more beneficial than either nonuse or overuse for maintaining these functions. In clinical practice, integrating digital tools into interventions can support physical and cognitive health when tailored to users’ abilities and needs. For policymakers, efforts should focus on expanding internet access, reducing financial barriers, and improving digital literacy through training and awareness programs. For future studies, the need for cautious interpretation and further validation are needed due to variations across studies, assessment tools, and populations.