Two screen behaviors, opposite directions
The paper asks whether screen time is one cognitive exposure or several. It separates passive television viewing from interactive computer use and analyzes them in parallel across four cohorts: NHANES (n = 2,928), HRS-HCAP (n = 3,033), ELSA-HCAP (n = 958), and SHARE (n = 2,077), totaling 8,996 participants with mean ages from 69.2 to 76.2 years.
In fully adjusted models, the two behaviors pointed in opposite directions. Every additional hour per day of television viewing was associated with a lower global cognitive z-score in all four cohorts: -0.047 (95% CI -0.071 to -0.022) in NHANES, -0.034 (-0.049 to -0.020) in HRS-HCAP, -0.035 (-0.064 to -0.005) in ELSA-HCAP, and -0.036 (-0.054 to -0.018) in SHARE. Every additional hour per day of computer use was associated with a higher score: 0.059, 0.072, 0.076, and 0.047 respectively, all with P < 0.001.
The unit matters here. These are z-score changes per hour per day, not raw test points, and they come from observational models that adjust for measured covariates but cannot by themselves rule out reverse causation. That is why the paper adds a genetic design rather than resting on the cohort associations.
Genetic instruments reproduce the split
Mendelian randomization uses genetic variants associated with an exposure as instruments, so that the comparison is less easily reversed by early cognitive decline changing someone's viewing habits. The paper applied it to genetically predicted television viewing time and computer use time, with sensitivity checks for heterogeneity, horizontal pleiotropy, and causal direction.
The genetic estimates matched the cohort pattern. Genetically predicted television viewing was associated with lower cognitive performance (β = -0.214, 95% CI -0.269 to -0.159) and lower cognitive function (β = -0.327, 95% CI -0.494 to -0.160), both surviving multiple-testing correction. Genetically predicted computer use was associated with better cognitive performance (β = 0.501, 95% CI 0.411 to 0.591), though its association with the separate cognitive function measure was not statistically significant.
That asymmetry is worth keeping. The computer-use signal is strong for one overall cognitive measure and absent for the other, and domain-specific results were not uniform across every trait. The paper reports Bonferroni-supported positive associations for fluid intelligence, prospective memory, maximum digits remembered, correct matches in a round, and direction-harmonized completion time on an alphanumeric path, while spelling was not significantly associated.
Memory follows TV, processing speed follows computers
The cohort analyses break global cognition into domains, and the two behaviors do not map onto the same ones. The inverse television association was most consistent for immediate recall, significant in all four cohorts (β range -0.038 to -0.026). Delayed recall showed inverse associations driven by HRS-HCAP, ELSA-HCAP, and SHARE (β range -0.045 to -0.026), and processing speed/orientation inverse associations were concentrated in NHANES, HRS-HCAP, and SHARE (β range -0.035 to -0.027).
Computer use was more consistently positive for processing speed/orientation (β range 0.050 to 0.090) and delayed recall (β range 0.044 to 0.061), both significant across cohorts. So the same domain, delayed recall, carries an inverse signal for television and a positive signal for computer use, which is the clearest within-paper comparison of how the two activities diverge.
The authors interpret this as a difference in cognitive demand: television as a passive sedentary exposure, computer use as involving information seeking, communication, problem solving, and task management. That framing is the authors' explanation, not a measured property of the activities in this dataset.
The computer benefit plateaus above moderate use
The dose-response curves do not support a simple more-is-better reading. Television viewing was largely linear and inverse in NHANES, ELSA-HCAP, and SHARE (P for nonlinearity > 0.10), while HRS-HCAP showed a flatter curve at low-to-moderate viewing and a clearer decline at longer viewing times (P for nonlinearity = 0.003).
Computer use was significantly nonlinear in all four cohorts (all P < 0.001). Global cognitive z-scores rose up to roughly 2 to 3 hours per day and then plateaued or attenuated. In categorical models, the largest estimates appeared in the 2 to 3 hours per day group in NHANES, HRS-HCAP, and ELSA-HCAP, while SHARE showed its largest estimate for more than 0 to 1 hour per day; the 4 or more hours per day group generally showed attenuated estimates.
This is the practical threshold the paper actually supports: the positive computer association is clearest at low-to-moderate use, not at the top of the range. It also means the cohort estimates for television are not driven by a threshold effect in most cohorts, since the inverse pattern was approximately linear.
The left insula as a candidate link
The neuroimaging arm asked which brain structures might sit between screen behavior and cognition. For genetically predicted television viewing, the strongest result was lower left insula volume (β = -0.220, 95% CI -0.321 to -0.120, P = 1.78 × 10⁻⁵), which survived Bonferroni correction. Right insula volume survived FDR correction but not Bonferroni (β = -0.184).
A two-step Mendelian randomization mediation analysis suggested that lower left insula volume accounted for approximately 14.8% of the association between genetically predicted television viewing and poorer cognitive performance. The authors connect the insula to salience-network switching between internally directed cognition and external attention, and note exploratory signals of weaker insula/anterior cingulate to frontal connectivity.
The mediation figure is a proportion of a genetic association, not a demonstrated clinical pathway, and the connectivity findings did not survive multiple-testing correction. For computer use, no regional brain volume association survived Bonferroni or FDR correction; nominal positive signals involved the left supramarginal gyrus and left posterior cingulate, with exploratory signals of higher parietal-frontal connectivity.
No dementia signal and a European-ancestry scope
Despite the cognitive associations, the paper found no convincing evidence that genetically predicted television viewing time or computer use time was associated with the risk of four dementia subtypes: Alzheimer's disease, vascular dementia, frontotemporal dementia, and Lewy body dementia. That null result separates performance on cognitive tests from diagnosed dementia risk and should temper any claim that these screen behaviors shift dementia incidence.
Two scope conditions limit how far the results travel. The genetic data were largely from individuals of European ancestry, so generalization to other populations is uncertain. And the neuroimaging Mendelian randomization analyses were designated exploratory: main figures show inverse-variance-weighted estimates with unadjusted P values, and nominal associations without correction-level significance are labeled exploratory signals rather than confirmed effects.
Stratify by activity, not total screen hours
The paper's own conclusion is that public health guidance should distinguish passive screen exposure from purposeful, cognitively engaging digital activity rather than relying only on total screen-time thresholds. The evidence for that priority is the consistent divergence: television viewing was inversely associated with global cognition in all four cohorts and in the genetic analyses, computer use was positively associated in both, and the computer association flattened rather than grew at the highest use levels.
For readers weighing the results, the most decision-relevant split is therefore by activity type and by dose, not by total hours in front of a screen. The television signal was approximately linear and concentrated in memory-related outcomes; the computer signal was strongest at low-to-moderate use and clearest for processing speed and delayed recall. The left insula mediation, at roughly 14.8%, is a hypothesis-generating candidate mechanism rather than a settled explanation, and the absence of a dementia-subtype association marks where the cognitive-test findings stop.
Sources used in this answer
Screen time modalities and neurocognition in older adults: multi-cohort and Mendelian randomization evidence
Television viewing was associated with lower global cognition and computer use with higher global cognition across four cohorts, with Mendelian randomization supporting opposing directions and about 14.8% of the television association mediated by left insula volume.