The bidirectional relationship between vision and cognition: a systematic review and meta-analysis

Authors: Vu TA, Fenwick EK, Gan ATL, Man REK, Tan BKJ, Gupta P, et al.

Geographical coverage: Worldwide

Sector: Burden of disease

Sub-sector: Epidemiology

Equity focus: Older population (people aged 40)

Study population: People aged 40+

Review type: Other review

Quantitative synthesis method: Meta-analysis

Qualitative synthesis method: Not applicable

Background: Visual impairment (VI) and cognitive impairment (CIM) are prevalent age-related conditions that impose substantial burdens on society. Findings on the hypothesised bidirectional association of VI and CIM remain equivocal. Sixty per cent risk of CIM has not been well elucidated in the literature. A bidirectional relationship between VI and CIM may support the development of strategies for early detection and management of risk factors for both conditions in older people.

Objectives: To examine the bidirectional relationship between VI and CIM, to confirm the association of CIM as a risk factor for VI and vice versa.

Main findings:

Overall, the review suggests that VI is a risk factor of CIM and there is some evidence that CIM is a risk factor for VI (although the latter relationship could not be confirmed through meta-analysis).

43 articles were included in the review (28 cross-sectional, 14 cohort, and 1 case-control study). Of the 28 cross-sectional articles, most (90%) reported moderate to high NOS scores, with 15 graded as high quality (8 stars) and 10 graded as moderate quality (5-7 stars). The remaining three studies were classified as poor quality (0-4 stars). Of the 14 cohort studies, 100% reported moderate to high NOS scores, with 12 graded as high quality and two graded as moderate quality. The case-control study was graded as moderate quality. The three articles classified as poor quality were excluded, leaving 40 articles for inclusion. Cross-sectional and 1 cohort investigated the relationship of VI and CIM bidirectionally. The total number of participants was 47,913,570; nine and 31 studies reported on Asian and White populations, respectively. Among the 40 studies in this systematic review, 31 had adequate data to be included in the meta-analyses.

Meta-analyses confirmed that people with VI were more likely to have CIM, with significantly higher odds of: (1) any CIM (cross-sectional: OR, 2.38 [95% CI, 1.84e3.07]; longitudinal: OR, 1.66 [95% CI, 1.46e1.89]) and (2) clinically diagnosed dementia (cross-sectional: OR, 2.43 [95% CI, 1.48e4.01]; longitudinal: OR, 2.09 [95% CI, 1.37e3.21]) compared with people without VI. Significant heterogeneity was explained partially by differences in age, sex and follow-up duration. Some evidence suggested that individuals with CIM, relative to cognitively intact persons, were more likely to have VI, with most articles (8/9 [89%]) reporting significantly positive associations; however, meta-analyses on this association could not be conducted because of insufficient data.

Authors report strategies for early detection and management of both conditions (VI and CIM) in older people may minimise individual clinical and public health consequences. Furthermore, authors note the need to confirm the association of CIM as a risk factor for VI.

Methodology:

The review included adults aged 40 or older and observational studies with VI or CIM as exposures or outcomes, and participants without VI and CIM as comparators. Exclusions were reviews, qualitative studies, case reports/series, conference abstracts, animal and in vitro/vivo studies, interventional studies, non-English studies, studies without clear exposure/outcome definitions, studies of special risk groups (e.g., people with diabetes, cancer, Down syndrome), and studies with insufficient data for evaluation or conclusion drawing.

PubMed, EMBASE and Cochrane Central registers were searched for observational studies, published from inception until 6 April  2020, in adults 40 years of age or older reporting objectively measured VI and CIM assessment using clinically validated cognitive screening tests or diagnostic evaluation. The core keywords included “visual impairment” AND “cognitive impairment” AND “adult”. The bibliographies of included articles were hand-searched to identify other relevant records.

Two authors assessed the titles and abstracts of identified articles independently. Data extraction was performed by the first author and checked for accuracy by two coauthors. Two authors independently assessed the risk of bias of observational studies using the Newcastle-Ottawa scale.

Statistical analysis was performed by one author and reviewed by another. Separate meta-analyses were conducted on the association between VI and CIM, stratified by study design and CIM definition. Significant heterogeneity was considered if the p-value for the Q test was less than 0.10 or if the I² statistic was 50% or more. Due to substantial heterogeneity, a random-effects model was used to synthesize study effects. A univariate random-effects meta-regression analysis was performed to identify potential study heterogeneity. Subgroup analysis was conducted on a vision-related characteristic: presenting versus best-corrected Visual Acuity (VA). Sensitivity of overall results to the exclusion of unadjusted estimates was examined. Funnel-plot asymmetry was assessed visually and using Egger’s bias test. If publication bias was suspected, the trim-and-fill method was used to re-estimate the pooled OR. Final pooled ORs were reported with 95% CIs, and a 2-sided p-value of less than 0.05 was considered statistically significant.

Applicability/external validity: The discovery that VI can predict cognitive decline aligns with the results of earlier studies. The authors also explore theoretical explanations for this connection. The predictive influence of VI on CIM is a more recent finding, and the data supporting it is less robust, as it couldn’t be analysed through meta-analysis. However, the authors propose some theoretical reasons for this potential association.

Geographic focus: Although no search restrictions were applied, authors mostly included studies from Asia.

Summary of quality assessment:

While the approaches undertaken to identify, include and critically appraise studies were generally robust, with all of these stages being undertaken by multiple areas, the review was restricted to studies published in English, which may have limited the pool of studies included. The approach to data analysis was very robust although, as the authors note, it was not possible to subject the relationship between VI and CIM to more detailed analysis, due to the low quality of the studies involved. For these reasons, we attributed medium confidence in the findings of this review.