This project involved a secondary quantitative data analysis using data from the NHATS collected between 2015 and 2022 (Rounds 5-12). NHATS, supported through a cooperative agreement with the National Institute on Aging (U01AG032947), is a nationally representative longitudinal study of Medicare beneficiaries aged 65 years and older [21]. Annual interviews have been conducted since 2011 (Round 1), with sample replenishment in 2011, 2015, and 2022 (Rounds 1, 5, and 12 [21]). A detailed cohort profile is available in Freedman and Kasper [21], published by the NHATS study team.

The sample consisted of community-dwelling individuals aged 65 years or older without a diagnosis of dementia or mild cognitive impairment at baseline [5]. The original analytic sample included 5596 community-dwelling individuals aged 65 years and older without dementia. We identified 30 out of 5596 participants who were classified as having no dementia but lacked cognitive test scores. After consulting with the NHATS team, they indicated that these participants, or their proxy respondents, refused to complete the cognition test. In such cases, NHATS relied solely on self-reports of dementia instead of the tests. Since cognition scores were the main outcome measure, the level of missingness was minimal (0.5%), and these cases were excluded from the analysis. The final sample size in 2015 (Round 5) included 5566 older adults, with 31,038 observations from 2015 to 2022 (Rounds 5-12). The median number of follow-up observations per round was 3631.

In our study, we applied analytic weights using the weighting variables from the sample person file [27]. Since the 2015 cohort was chosen as the foundation of our target population, we tackled the problem of multiple (repeated) observations through a multilevel modeling strategy that incorporated the NHATS sample design. The weights for each round were used to account for differential probabilities of selection and nonresponse, as well as for clustering and stratification variables. Adhering to the guidelines and code detailed in NHATS Technical Paper 23 [28], we used round-specific analytic weights and developed a new weight variable to represent individual-level weights within the multilevel models [28].

We conducted generalized linear mixed-effects models separately for the 3 cognitive domains, both unadjusted and adjusted for covariates. The predictors were various types of technology activities, and the outcomes were the different domains of cognition, including episodic memory, executive function, and orientation.

General linear mixed-effects models are particularly suited for analyzing longitudinal or hierarchical data, where repeated measurements are nested within individuals. Given the nature of cognitive function may change over time due to aging and may also be influenced by exposure to different types of technology, the use of general linear mixed-effects models provided an appropriate statistical framework for this analysis.

Two models were specified: the main effect model (Model A), which assessed the association between transitions in technology use and cognitive function, and the moderation effect model (Model B), which indicated the effects of technology use on the rate of change in cognitive functioning. For example, the main effect model (Model A) for episodic memory and online shopping was specified as follows:

The moderation effects model (Model B) added interaction terms for transitions and time, as follows:

The index i represents the individual sample ID, and j denotes the time point (ie, j=2015, 2016, ..., 2022).

We first estimated whether the start (transition in) and stop (transition out) of technology activities were associated with cognitive functioning at a given time (Model A). We then added an interaction term between transitions and time to explore whether starting or stopping technology activities could change the rate of cognitive decline over time (Model B).

Since we evaluated multiple outcomes, we applied the Benjamini-Hochberg false discovery rate correction to reduce the risk of type I errors [29]. To minimize the possibility of a reverse causal relationship between digital activities and cognitive domains, we conducted sensitivity analyses using 1-year lagged data for both Model A and Model B to ensure the associations hold over time.

This study was exempted by the Johns Hopkins Medicine Institutional Review Board (IRB00460714), where informed consent is not required. The NHATS public use data is de-identified.

The study sample characteristics at baseline are provided in Table 2. At baseline, the study sample (N=5566; representing an estimated 32,929,570 older adults living in communities in the United States) had an average age of 76.48 (SD 7.04) years and no dementia. The sample was predominantly female (3190, 57.31%), with well-distributed racial and ethnic groups. Most participants had no college degree (3738, 67.16%), lived with someone else (3803, 68.33%), and resided in metropolitan areas (4476, 80.42%). Furthermore, most participants owned one or more digital devices (eg, cell phone, computer, or tablet), and around a quarter of them had performed some technology activities in the past month.

Table 3 shows the beta coefficients for the initiation and cessation of each technology activity across various cognitive domains. These findings stem from Model A and address research question 1—the associations between technology activities (online shopping, banking, medication refills, social media use, and checking health information) and cognitive domains (episodic memory, executive function, and orientation). P values are marked after the false discovery rate correction using the Benjamini-Hochberg method. The P values before and after adjustment are provided in Multimedia Appendix 2.

Transitions into online shopping were associated with an increase in the score of episodic memory at a given wave (𝛽=.046, P=.02; 95% CI 0.013-0.080; Model A), as well as an increase in the score of executive function (𝛽=.041, P=.043; 95% CI 0.007-0.075; Model A) and orientation at a given wave (𝛽=.091, P<.001; 95% CI 0.055-0.126; Model A).

Starting online banking was associated with an increase in executive function at a given wave (𝛽=.061, P=.043; 95% CI 0.010-0.112; Model A) and orientation (𝛽=.077, P=.004; 95% CI 0.032-0.123; Model A). Stopping online banking was associated with decreasing scores of episodic memory (𝛽=−.078, P=.01; 95% CI −0.129 to −0.026; Model A).

Transitions into online medication refills were associated with increasing episodic memory (𝛽=.073, P<.001; 95% CI 0.038-0.109; Model A) and increasing orientation at a given wave (𝛽=.097, P<.001; 95% CI 0.054-0.140; Model A).

Starting to visit social networking sites online was associated with increasing episodic memory (𝛽=.065, P=.01; 95% CI 0.022-0.109; Model A) and orientation at a given wave (𝛽=.080, P<.001; 95% CI 0.041-0.119; Model A). Stopping visits to social network sites online was associated with decreasing episodic memory (𝛽=−.066, P=.004; 95% CI −0.105 to −0.028; Model A). Transitions into checking health information online were associated with increasing orientation (𝛽=.098, P<.001; 95% CI 0.068-0.127; Model A) at a given wave.

Among all the technology activities examined, online shopping and online banking showed the widest impact across various areas of cognitive function. Combining our main findings with results from lagged models in sensitivity analyses (Multimedia Appendix 3), older adults who started using online shopping and social media were related to better orientation; those who began managing prescriptions digitally showed cognitive benefits in episodic memory; and those who stopped using online banking were linked to a decline in episodic memory. Notably, the effects were statistically significant but small in magnitude.

In Model B, we added an interaction term between transitions of technology activities and time to assess their effect on the rate of cognitive decline. Table 4 shows the beta coefficients for the interaction terms between shifts in technology activities and time, reflecting the rate of cognitive decline. A positive value indicates a reduction in cognitive decline, while a negative score suggests an increase in cognitive decline.

Stopping online shopping was associated with a faster rate of cognitive decline in episodic memory (𝛽=−.023, P=.047; 95% CI −0.041 to−0.005; Model B). The transition into online shopping was associated with a slower cognitive decline in the orientation domain (𝛽=.049, P<.001; 95% CI 0.035-0.063; Model B). The onset of online banking was associated with a mitigated rate of orientation decline (𝛽=.049, P<.001; 95% CI 0.035-0.063; Model B). The onset of online medication refill was associated with a slower orientation decline (𝛽=.052, P<.001; 95% CI 0.036-0.068; Model B). The onset of social media use was associated with a slower decline in the orientation domain (𝛽=.039, P<.001; 95% CI 0.025-0.053; Model B). The transition into checking health information online was associated with a mitigated rate of orientation decline (𝛽=.036, P<.001; 95% CI 0.022-0.051; Model B). When incorporating findings from lagged models in sensitivity analyses (Multimedia Appendix 3), regarding the interaction with time, beginning to use any of the studied digital technologies was linked to significantly better but small preservation of orientation.

The summary of all models, stratified by technology activities, is provided in Multimedia Appendix 4. Several noteworthy confounders emerged in the analysis. A higher number of device owners was consistently associated with better cognitive functioning across all technology activities. Similarly, individuals with higher educational attainment (college degree or beyond) demonstrated superior performance across all cognitive domains. Females exhibited stronger effects specifically in episodic memory, regardless of the technology activity.

This study highlights the distinct effects of technology use on specific cognitive domains among older adults. Compared with other digital activities, the use of online shopping and online banking services showed the most comprehensive influence on the 3 cognitive domains. Similar to findings from a previous study, online shopping and banking involve a series of actions such as memorizing accounts and passwords, thinking about recipes and corresponding products to purchase, or financial plans to arrange, and remaining oriented to dates, time, and their own names to process transactions [18,30].

Older adults who began using online shopping and social media showed a small improvement in orientation. These activities may enhance spatial and temporal awareness by requiring users to follow delivery schedules, navigate websites, and engage with time-stamped social content. Starting to manage prescriptions online was linked to better episodic memory. This task likely activates memory systems through routine recall of medications and schedules, reinforcing memory through repeated use and the integration of new and existing information. In contrast, stopping online banking was associated with a decline in episodic memory. Online banking involves complex memory tasks such as recalling passwords and managing finances. Disengaging from such cognitively demanding activities may reduce mental stimulation, contributing to memory decline. However, all observed effects were small in magnitude, and their clinical significance remains uncertain. Future trials are needed to evaluate the clinical relevance of targeted digital engagement interventions on specific cognitive domains.

This study used observational data, which introduces the possibility of reverse causation. It is plausible that older adults with lower cognitive performance were less likely to adopt or continue using technology. However, findings from Choi et al [31], using a cross-lagged panel analysis of NHATS data, suggest that technology use predicts cognitive improvements over time, particularly in episodic memory and executive function. While episodic memory also predicted subsequent technology use, executive function did not, and orientation was not assessed in their study [31]. Similarly, Hartanto et al [32], using data from the Midlife in the United States study, found that computer use significantly predicted positive changes in executive function. Their analysis showed consistent associations between computer use and improvements across all executive function tasks, though no significant effect was observed for episodic memory. Collectively, these findings support a stronger directional relationship from technology use to cognitive function, although a bidirectional relationship cannot be ruled out.

Some limitations stemmed from the instrument used by NHATS. For example, the question about social media use simply asked whether the participant had gone online to visit social network sites (eg, Facebook and LinkedIn) in the last month. The specific platform, content on the sites, and the level of engagement were not asked. We recognize the heterogeneity of social media and the potential varied associations with cognitive function. In addition, digital activity questions were designed to help participants recall their technology use behaviors over the past month or year, depending on the activity. This study only assessed changes in technology use and cognition over 8 years, with the smallest time unit being a year. Thus, one instance of use within a given year was coded as “yes,” regardless of frequency. We acknowledge that this annual resolution limits our ability to capture short-term fluctuations in digital engagement or the precise timing of technology use relative to cognitive assessments.

Attrition in this study occurred through 3 primary mechanisms: death, nonresponse, and transitions from community to institutional settings. While the NHATS cohort maintained high response rates (>90%) with relatively low attrition per round [21], we used weighted analyses to account for potential attrition bias. Analysis of NHATS data from 2011 to 2018 revealed that 86.2% of older adults remained in community settings, 9% transitioned to residential care, and 4.9% to nursing homes [33]. Although only a small proportion of older adults changed settings, these transitions predominantly occurred among older adults with greater functional and cognitive limitations, potentially limiting our ability to capture technology access and use patterns among this more vulnerable subgroup.

Another limitation concerns the last few years of the study period, which coincided with the COVID-19 pandemic. Social isolation during this time was particularly acute for older adults [34]. Evidence from NHATS data shows that older adults who reduced their use of emails, texts, and phone calls during the pandemic, especially those with lower socioeconomic status, showed a limited increase in video calls to stay connected with family and friends [35]. The pandemic’s potential impact on technology use trends and its subsequent influence on cognitive function remains unclear, introducing uncertainty to the study’s results.

In considering generalizability, it is important to note that wired high-speed internet access remains more prevalent in urban than rural areas—available in 76% of noninstitutional households in urban settings compared with 62% in rural areas as of 2023 in the United States [36]. While our findings highlight the potential cognitive benefits of certain digital activities, real-world implementation may be limited by disparities in local infrastructure and household financial capacity. Without reliable internet access or the resources to obtain and maintain digital devices, many older adults, particularly in underserved areas, may be unable to engage in these beneficial activities.

The observed association between social media use and cognitive function aligns with previous research. For example, Byrne and Ghaiumy Anaraky [37] found that greater use of social technology was linked to enhanced cognitive performance among socially isolated older adults, though their outcome measure was a global cognitive score rather than disaggregated domains. Anaraky et al [19] reported that stopping the use of computers, tablets, the internet, texts, and emails predicted declines in episodic memory, executive function, and orientation. Similarly, Kim and Han [22] reported that initiating internet use was associated with improved cognitive functioning and a slower rate of cognitive decline over time, while ceasing internet use was linked to worse cognitive outcomes and accelerated decline. These findings align with this study’s results, which show that both initiating and maintaining technology use are crucial for cognitive health. Our study provides more granular insights, demonstrating the unique effects of various technology activities on episodic memory, executive function, and orientation.

Our findings suggest that interventions aiming to maintain episodic memory in older adults should focus on encouraging the adoption of online medication refill services and preventing disengagement from online banking, particularly among those already accustomed to these activities. To preserve orientation, efforts should prioritize supporting older adults in initiating and maintaining online shopping and social media use. More broadly, all types of digital engagement examined in this study—including instrumental, social, and informational activities—were associated with a slower decline in orientation over time. Teaching older adults new technology skills and providing resources to prevent disengagement from technology activities may help preserve cognitive health. Furthermore, when designing technology-based interventions, researchers can use these findings to target specific cognitive domains with appropriate technology activities. By tailoring interventions to match the cognitive benefits of different activities, resources can be allocated more effectively to maximize their impact on cognitive well-being in older adults.