Relationship between retinal neurodysfunction and cognitive impairment in type 2 diabetes: results of the RECOGNISED cross-sectional study

Abstract

Aims/hypothesis: There are no robust, reliable and easy to administer tests to screen for mild cognitive impairment (MCI) in people living with diabetes. Since the retina is ontogenically brain-derived, we hypothesised that retinal biomarkers could be used, alone or in combination with other simple tests, to screen for MCI in people with diabetes.

Methods: Baseline data from participants screened for RECOGNISED, a Horizon 2020-funded European project, were analysed. Main eligibility criteria for RECOGNISED included age ≥65 years, type 2 diabetes of over 5 years standing, no previous history of stroke or neurodegenerative disease, and no overt diabetic retinopathy or only mild-to-moderate non-proliferative diabetic retinopathy. Baseline characteristics of participants, including scores from the Montreal Cognitive Assessment test (MoCA) and Self-Administered Gerocognitive Examination, the Diabetes Specific Dementia Risk Score (DSDRS) and ophthalmological endpoints gathered from standardised seven field colour fundus photography, spectral domain optical coherence tomography, microperimetry and a hand-held portable electroretinography device (RETeval), were obtained and used in the work presented here as potential screening predictors for presence of MCI. MCI and normocognition (NC) were determined based on a full neuropsychological test battery and the Clinical Dementia Rating score. A stepwise selection of variables, based on Akaike's information criterion, and logistic regression models for predicting MCI were undertaken. Area under the receiver-operating characteristic curve analyses were used to predict the probability of the presence of MCI as well as sensitivity and specificity cut-off points.

Results: A total of 313 people living with diabetes (128 with NC and 185 with MCI) were included. People with diabetes with MCI were older (p=0.006) and had fewer years of education (p<0.001), lower retinal sensitivity (p=0.01) and less capacity of gaze fixation (p≤0.001) than those with NC. Statistically significant differences in pupillary area ratio (p=0.002) and photopic b-wave amplitude (p=0.03) were detected between people with diabetes with NC and with MCI. Multivariable logistic regression showed that the best model to identify people with diabetes with MCI was that combining retinal sensitivity, gaze fixation, photopic b-wave amplitude and pupillary size change following stimulation, years of education, DSDRS and MoCA score, with an AUC of 0.84 (sensitivity 79.9, specificity 79.0). The visuo-construction domain was the most affected in people with diabetes with MCI and its impairment was independently related to retinal sensitivity and gaze fixation.

Conclusions/interpretation: The assessment of retinal neurodysfunction in combination with simple clinical variables appears useful to identify people with diabetes with MCI. This strategy could optimise current screening of MCI in people living with diabetes.

Keywords: Cognitive impairment; Diabetic retinopathy; Electroretinography; Microperimetry; Mild cognitive impairment; Pupillary responses; Retinal neurodegeneration; Retinal neurodysfunction; Type 2 diabetes; Visuo-construction.

Fig. 1

RECOGNISED flow chart showing the screening process with numbers of people included and excluded in the RECOGNISED cross-sectional and longitudinal studies. A total of 313 out of 633 screened individuals were included in the cross-sectional study. The 70 ‘screening failures’ accomplished the inclusion criteria and, in fact, were not included in the longitudinal study because they did not sign the informed consent to participate in the longitudinal study or the recruitment was already completed

Fig. 2

AUC values and calibration plots for predicting MCI in five models: (a) univariate MoCA; (b) univariate DSDRS; (c) multivariate model I, which includes MAIA parameters, years of education and DSDRS; (d) full multivariate model; (e) multivariate model I + MoCA. CITL, calibration in the large; O:E, observed:expected ratio

Fig. 3

AUC values and calibration plots for predicting MCI in three models: (a) model I, which includes RETeval and MAIA parameters, years of education and DSDRS; (b) univariate model including MoCA; (c) model II, which includes the variables of model I plus MoCA. CITL, calibration in the large; O:E, observed:expected ratio

Fig. 4

AUC values and calibration plots for predicting MCI in: (a) a model including RETeval parameters, years of education and DSDRS (model I), and (b) a model including the variables of model I plus MoCA. CITL, calibration in the large; O:E, observed:expected ratio

Fig. 5

AUC values and calibration plots for predicting abnormal visuo-constructional skills in: (a) a model including MAIA and SD-OCT parameters (model I), and (b) a model including the variables of model I plus years of education. CITL, calibration in the large; O:E, observed:expected ratio