Alzheimer’s disease (AD) is associated with brain neurodegeneration, particularly in the medial temporal area, which leads to cognitive decline and dementia. These lesions occur several years before the clinical phase of dementia (1, 2). Detection of this neurodegenerative process at an early stage could allow prediction of cognitive decline in subsequent years.

The retina and brain are intimately linked as a result of common embryonic origin. The eye is a sensory organ that is truly part of the central nervous system, with neuronal cells that could be susceptible to degeneration, directly or indirectly. Thus, the condition of nerve fibers in the eye could reflect the condition of nerve fibers in the brain (3). Spectral domain optical coherence tomography (SD-OCT) allows easy and accurate evaluation of the peripapillary retinal nerve fiber layer (RNFL), by measuring the thickness of this layer containing ganglion cell axons in a circle around the optic nerve. This RNFL could thus enable evaluation of neurodegeneration in mild cognitive impairment (MCI) and AD pathology (4).

Abnormality in RNFL thickness has been described in several neurodegenerative conditions, in particular multiple sclerosis, AD, and Parkinson’s disease (5–7). Regarding MCI and/or AD, several previous studies have demonstrated reduced RNFL thickness compared to that of a healthy control population, initially in histological studies (8) and then in OCT studies (4, 9, 10); these studies were, however, mostly cross-sectional case–control studies. Additionally, we previously found an association between glaucoma pathology, which has reduced RNFL thickness, and the occurrence of dementia 3 years later (11). However, except for one study with a small sample (12), the association between RNFL and the evolution of cognitive function has not yet been studied in a general elderly population.

Our study aims to explore the relationship between peripapillary RNFL thickness measured by SD-OCT and the evolution of cognitive function in several cognitive domains measured over a 2-year period in the elderly population.

Population characteristics of the 427 participants are described in Table 1. The mean age was 81.6 (SD: 4.2) years and 62.3% were women. The mean global RNFL thickness was 86.8 μm (SD: 13.7). As expected, the inferotemporal (122.3 μm) and superotemporal (115.9 μm) segments were thickest. Mean MMSE score at baseline was 27.8 (SD: 1.9). Cognitive performance significantly declined over the 2-year follow-up, irrespective of the test considered (Table S1 in Supplementary Material).

The global RNFL was not associated with initial cognitive performance regardless of cognitive test considered (Table 2). Global RNFL was not associated with cognitive evolution based on the MMSE or the Isaacs’ set test at follow-up (RNFL × time term). However, global RNFL was associated with the two FCSRT scores of delayed recall. Specifically, a thicker global RNFL was associated with a lower decline of delayed free recall (β = 0.015; 95% CI, 0.005 to 0.025; p = 0.004 for 1 μm RNFL thickness) and delayed free plus cued recalls (β = 0.007; 95% CI, 0.002 to 0.011; p = 0.004) (Table 2, model A).

These associations remained unchanged after additional adjustment for axial length in the subset which had this measure (Table 2, model B). Mean delayed FCSRT scores, at baseline and at the follow-up, are graphically represented in Figure S1 in Supplementary Material according to baseline RNFL value.

For FCSRT scores showing significant association with global RNFL, the relationships with each of the six RNFL segments were analyzed (Table 3). The temporal segments had the strongest association with the two FCSRT delayed recall scores. Thicker RNFL of the temporal, inferotemporal, and superotemporal segments was significantly associated with a lower decline in the delayed free recall score during follow-up (β = 0.013; 95% CI, 0.003 to 0.023; p = 0.011; β = 0.007; 95% CI, 0.002 to 0.012; p = 0.005; and β = 0.007; 95% CI, 0.002 to 0.013; p = 0.013, respectively) in model A. The same trends were observed in model B after additional adjustment for axial length; although, the associations were only borderline significant for the temporal and inferotemporal segments. Surprisingly, the temporal segment also showed an association with a lower initial score (β = −0.029; 95% CI, −0.054 to −0.005; p = 0.020), but this association was no longer significant after adjustment for axial length.

For delayed free plus cued recalls, the decline over the follow-up was also associated with RNFL thickness in the “temporal side”; this association was significant only for the inferotemporal segment in model A (β = 0.003; 95% CI, 0.0003 to 0.005; p = 0.025). It was also significant for the three temporal segments in the subset with axial length measured (Model B; β = 0.009; 95% CI, 0.002 to 0.016; p = 0.010; β = 0.005; 95% CI, 0.002 to 0.008; p < 0.001; and β = 0.005; 95% CI; 0.001 to 0.009; p = 0.010, for temporal, inferotemporal and superotemporal segments, respectively). Additionally, thicker RNFL of the nasal segment was also associated with lower decline of the cognitive score (Models A and B). Surprisingly, in model B, an association was also demonstrated between inferotemporal segment thickness and a lower initial score (β = −0.009; 95% CI, −0.017 to −0.002; p = 0.011).

Results were globally unchanged in sensitivity analyses excluding participants with glaucoma.

Similar results were also observed in complementary analyses. Participants classified as borderline (12.6%) or outside normal limits (17.8%) for their global RNFL value (i.e., with thin RNFL) according to the Spectralis software classification had a higher decline for the two FCSRT scores of delayed recall (β = −0.40; 95% CI, −0.71 to −0.10; p = 0.009 for delayed free recall and β = −0.20; 95% CI, −0.34 to −0.05; p = 0.007 for delayed free plus cued recalls).

We found that, whereas global RNFL thickness was not associated with initial cognitive scores, a thicker RNFL was associated with a lower decline of cognitive performances based on the delayed recall scores of the FCSRT. None of the other cognitive functions were associated with RNFL thickness. These findings suggest that reduced RNFL thickness may reflect a higher risk of episodic memory deterioration over time.

The reduction of RNFL thickness in patients with MCI and/or AD, compared to controls, has been previously reported in several studies (4, 9, 10). These previous studies support a systematic and cumulative reduction of the RNFL thickness with the progression of the cognitive impairment. The association between RNFL and cognition at an earlier stage is largely unknown; yet, cognitive decline is a very long process (25). Thus, our results may fit into a very early stage of cognitive impairment. Moreover, our study takes place within a general elderly population, using a longitudinal design, thus adding complementary knowledge compare to previous studies, which are mostly cross-sectional in clinical settings.

The association shown in our results was specific to episodic memory, which is a cognitive domain strongly implicated in AD process. A previous study evaluating the relationship between RNFL and cognition also found an association between RNFL and episodic memory (word list learning and story recall) among participants with normal cognition (26). However, this was a cross-sectional study with a small sample size (n = 52).

No association was found with MMSE score, which reflects global cognition. However, our population is mainly a well-functioning population without dementia at baseline; the MMSE test may be not sensitive enough to detect subtle changes in the early phase of cognitive deterioration (27). Similarly, no correlations between MMSE and peripapillary thickness were found in most previous studies comparing peripapillary RNFL in AD patients and healthy controls (28, 29). Only 1 previous study comparing 35 AD, 35 MCI, and 35 healthy controls reported a significant correlation between peripapillary RNFL and MMSE in an analysis combining all the participants (30).

More surprisingly, we also found no association with Isaacs’ verbal fluency test, even though this test has detected changes during very early cognitive decline, as many as 15 years before the clinical diagnosis of Alzheimer’s dementia (25). However, among elderly people, dementia is the result of several processes including neurodegeneration and vascular damage (31). Cognitive impairment due to vascular processes is believed to preferentially manifest through impaired attention and executive function rather than memory function (32). Therefore, verbal fluency could reflect the global vascular process more than the neurodegenerative process. This may explain the lack of association with RNFL thickness, which may be more strongly related to neurodegeneration.

Our SD-OCT device automatically calculates RNFL thickness both globally and in each of the six segments. This offers the advantage of providing data for the superotemporal and inferotemporal segments, which are the thickest zones in the peripapillary area and the most sensitive zones to early age-related defects (20, 33). As previously reported, the highest RNFL values in our data were found in the inferotemporal and superotemporal segments, which, together with the temporal segment, are also associated with cognitive decline (20).

Our results showing cognitive test alterations reflecting episodic memory changes are consistent with previous reports stating that the primary degenerative process leading to MCI and/or AD is marked by an early deterioration of episodic memory (27, 34, 35). In normal elderly subjects, the evaluation of this deterioration can even predict the time of progression to MCI as well as the time of progression from MCI to AD (36). Furthermore, even though “recall measures” are cited more often as good cognitive markers of future AD, “delayed recall measures” are excellent predictors of progression to AD (37, 38) and can even increase the sensitivity of recall scores used in AD identification (39). Therefore, our findings may be relevant to the specific brain changes found in the hippocampus and temporal regions of people in the preclinical stage of AD (40–42). Previous imaging data showing a correlation between RNFL and cerebral volumes in the temporal and occipital lobes are concordant with these findings (43). In addition to RNFL, future studies should explore the ganglion cell-body layer in relationship with cognitive performance and brain alteration.

Glaucoma was taken into account in our analyses by adjusting for this pathology and by performing sensitivity analyses excluding participants with glaucoma. Glaucoma is a degenerative disease of the optic nerve, and we previously found an increased risk of dementia among participants with glaucoma (11). The inferotemporal and superotemporal segments are the first segments damaged in glaucoma. These segments (and the temporal segment) are also the ones associated with cognitive decline. Therefore, similar neurodegenerative processes could lead to both glaucoma and cognitive decline. Axial length was also adjusted for in our analyses. However, supplementary analyses comparing results with and without adjustment for axial length in the subsample of 204 participants with available axial length data showed that this adjustment did not materially change our results (data not shown); thus, even if axial length is associated with RNFL, it seems not confound the relation between RNFL and cognition.

Our study has several limitations. First, our follow-up period of 2 years may be too short to demonstrate larger effects on other cognitive functions. However, to our knowledge, this is the first longitudinal study with a large cohort in this field. Second, our study included well-functioning subjects who consented to eye examination and who therefore may have a lower decline of cognitive performance over time than the general population. Thus, despite our large study population compared to previous case-control studies in this field, we cannot exclude a possible lack of power. Third, as race was not available in our cohort, we could not adjust for in the analyses; however, our population was mostly from a Caucasian origin. Fourth, data on refractive error, another ocular variable known to influence RNFL (44) was not included in our analyses; but given our findings on axial length it is unlikely to have affected results. Finally, as RNFL measurement was not available at the baseline time of the Alienor study, the effect of the RNFL evolution over time on cognition could not be studied; further follow-ups of our cohort will allow studying this.

Our study has some strengths. It is based on a longitudinal design with a large population-based cohort including nearly 430 elderly participants. It includes a systematic eye examination that allows the identification and the exclusion of ocular pathologies likely to influence RNFL measurement (vitreomacular traction, myelinated retinal nerve fibers, peripapillary choroidal neovascularization, or myopic chorioretinopathy). The OCT assessments were performed on the same OCT machine by the same experienced technician following a standardized protocol. Moreover, the spectral domain technique provided the best quality and reproducibility currently available (45). Cognitive functions were systematically assessed by trained psychologists blinded to ophthalmologic data using standardized procedures and evaluating several cognitive domains. The association between RNFL and delayed episodic memory was found with both RNFL as a continuous variable and using the threshold provided by the device to classify participants as borderline or outside normal limits for RNFL. Finally, numerous potential confounding factors were taken into account.

Our findings enrich the literature on the potential utility of RNFL in cognitive deterioration. Although there are still gaps in detailed knowledge of the early association between neurodegeneration in the retina and in the brain, our findings could lead to the next steps of knowledge in that domain. However, these results should be corroborated with larger populations and longer follow-ups. Moreover, cerebral imaging studies should be implemented to better understand the association between retinal and brain neurodegeneration.

JM-G analyzed the data and draft the first version of the manuscript. CH, CD, J-FD, J-FK, and JM-G designed the study. M-BR, J-FK, CS, M-ND, HA, J-FD, CD, and CH acquired the data. All the co-authors contributed to the interpretation of the results, critically revised the manuscript, and approved the final version.

JM-G, LT, and HA disclose no conflict of interest. M-BR is consultant for Allergan, Bausch & Lomb. She has received funding for conferences from Biogen, Thea, Novartis. CS has received funding for travel from Alcon, Allergan, AMO. M-ND has received funding from laboratoires Théa (advisory board, travel funding) and Novartis (advisory board, travel funding). J-FK is consultant for Alcon, Allergan, Bayer, Carl Zeiss Meditec, Novartis, Théa. J-FD serves on a scientific advisory board for and has received funding for travel from Janssen-Cilag; received a gift worth more than US$500 from Novartis; holds a corporate appointment with Merck Serono; and has received research support from Novartis and Ipsen. CD has received funding from laboratoires Théa (grant, advisory board, travel funding), Novartis (consultant), and Bausch & Lomb (consultant). CH has received honoraries from Novartis.