Plasma glucose is not associated with performance on standard cognitive screening tests in a mixed memory clinic cohort—An observational cross‐sectional study

It has been shown under experimental conditions that cognitive performance, especially working memory, is impaired in patients with type I and type II diabetes mellitus during hyperglycemic and hypoglycemic conditions, perhaps due to altered cerebral glucose metabolism. It is not known if patients with neurodegenerative diseases, who also exhibit pathological cerebral glucose metabolism, are affected in a similar manner by their plasma glucose levels.


| INTRODUCTION
Cognitive performance can be influenced by various factors. 1 Sociodemographic factors, especially educational level and age, are known to affect the test score on widely used cognitive tests such as the Mini-Mental State Examination (MMSE). 2,3 Less extensively studied is the role of physiological variables such as glucose levels and their immediate effect on cognitive performance, 4 especially in settings such as memory clinics where patients with cognitive impairment are assessed, and where cognitive testing is routinely done.
In patients with type I and type II diabetes mellitus (T1DM and T2DM) it has been shown experimentally that a negative effect of hypo-and hyperglycemia can be observed for memory tasks and global cognition. [5][6][7][8] Glucose is transported across the blood-brain barrier by glucose transporters using facilitated diffusion and it is the primary energy source for the brain. 9 Altered cerebral glucose metabolism and glucose transport across the blood-brain barrier (BBB) may thus account for impaired cognition in hypo-and hyperglycemic states in patients with T1DM and T2DM. 10,11 Patients with neurodegenerative disorders often exhibit impaired cerebral glucose metabolism and glucose transport across the BBB is possibly altered in Alzheimer's disease (AD). [12][13][14][15] This could imply that patients with neurodegenerative disorders might be cognitively affected by peripheral glucose levels even within what is considered non-diabetic ranges (our laboratory reference: 4.2-6.3 mmol/L). To our knowledge, no studies have examined the possible impact of glucose levels when assessing cognitive performance in a clinical setting such as in relation to diagnostic assessment, but results point to a deleterious effect of long-term higher glucose levels in healthy individuals, 1,16 although this cannot consistently be shown. 17 For review of studies on glucose levels in healthy individuals see Kirvalidze et al 2022. 18 Clinicians working in memory clinics are often faced with the task of evaluating test scores on cognitive tests to ascertain whether a patient may meet criteria for dementia. In the assessment of patients with cognitive complaints, test scores of individual patients are essential for reaching a diagnosis. 19 The potential influence of reversible conditions and concomitant disease (e.g., myxedema or hypercalcemia) are usually taken into consideration when planning diagnostic investigations. Factors that may contribute to the cognitive impairment should be taken into consideration, before requesting extensive cognitive testing and making a final diagnosis.
To which extent plasma glucose might be a factor to be considered is largely unknown. Establishing which physiological factors influence cognitive performance is important as this can aid the clinician in interpreting test results to make a more precise diagnosis. An obvious concern is a diagnostic situation where an abnormal cognitive test result may be due to a high or low glucose level. This could lead to unnecessary supplemental testing or invasive procedures. To address this issue, studies in cohorts of patients undergoing diagnostic evaluation for neurodegenerative disorders are needed.
In this observational, cross-sectional study we aimed to investigate the association between plasma glucose and cognitive performance in a consecutive, routine cohort of memory clinic patients with a high frequency of neurodegenerative disorders. As studies in healthy individuals and patients with T1DM and T2DM have shown an inverse U-shaped relationship between glucose and cognition we hypothesized that glucose (predominantly within normal ranges) and cognitive performance were related in a similar way in our cohort of memory clinic patients.

| Glucose measurement
Patients underwent venipuncture by laboratory technicians according to standard operating procedures on the same day as their cognitive testing was performed, referred to as baseline, in most cases (95% of patients). The remaining 5% of the patients had blood drawn up to approx. One week after cognitive testing. Patients were not instructed to fast prior to venipuncture. Non-fasting glucose was enzymatically determined in plasma using a Cobas 8000 system (Roche Diagnostics, Basel) in a routine lab setting. As a part of standard operating procedures, patients who underwent lumbar puncture as a part of their diagnostic assessment also had an additional plasma glucose measurement at this time. This was usually approximately 3-5 weeks (but no more than 3 months) after the baseline visit. This measurement was used as a follow-up measurement in the analysis.

| Statistical analysis
Spearman's rho was calculated for cohort characteristics variables to test correlations. Linear regressions were carried out with MMSE and ACE total scores as the outcome and plasma glucose as the explaining variable. Models were adjusted for age, sex, and diabetes. The DDBB does not contain data on diagnoses of DMI or DMII nor the presence of diabetic symptoms. As such, we defined a diagnosis of DMI or DMII as a quantitative variable of having a baseline plasma glucose value >11.1 mmol/L, which is the threshold defined by the WHO as diagnostic for DMII (with the presence of diabetic symptoms). 22 The linearity of regression models was checked by visual inspection of Q-Q plots of residuals and homoscedasticity was checked by visual inspection of scale-location plots. Multicollinearity of models was checked by evaluating variance inflation factors (VIFs), where multicollinearity was considered problematic if the VIF was >2.5. We did not identify any errors in the data nor did we remove outliers before conducting the statistical analyses.
To test for curvilinear relationships a quadratic term of plasma glucose was included in each model. We explored models which introduced centering of the plasma glucose variable in the polynomial regressions. Centering was done by subtracting the mean of plasma glucose from the plasma glucose variable (5.814 mmol/L). Further, we carried out a cubic spline regression with knots at values 4.2 and 6.3 mmol/L plasma glucose, which reflected the reference range at the hospital lab to investigate non-linear associations. Low and high glucose was defined as outside the reference range at the hospital lab (4.2-6.3 mmol/L) but below the diabetic threshold of 11.1. mmol/L. To assess the model validity, we performed a 10-fold cross-validation using the caret package. R (ver. 4.2.0) was used for all analyses. 23 A pvalue < 0.05 was considered significant and only two-tailed tests were applied. Table 1 gives the characteristics of the cohort stratified according to plasma glucose levels. In total, 2714 patients had their plasma glucose measured in conjunction with cognitive testing with MMSE, and 1978 of these patients were also tested with ACE. A total of 612 patients had a follow-up plasma glucose measurement. Patients differed with regard to sex (67.3% of patients with a glucose level > 11.1 mM were men vs. 48.8% of euglycemic patients) and there was a trend towards increasing age in patients with higher glucose levels (p = 0.07). The cohort reflected our usual frequency of diagnoses with most patients receiving a diagnosis of Alzheimer's disease (33.2%). Diagnoses did not differ between glucose level groups.

| RESULTS
MMSE total score and ACE total score did not differ significantly between groups, but there was a trend towards lower ACE scores in patients with higher glucose levels (p = 0.094). ACE total score The results of the linear regression on the relationship between plasma glucose and cognitive test scores are shown in Table 2.  In a separate model, we added a quadratic term of plasma glucose, to see if a curvilinear relationship existed between plasma glucose and cognitive performance as hypothesized.
In this model ( Table 2, Model 1) the adjusted R-squared only increased minimally, not indicating a better fit, although the estimates were both significant. This curvilinear relationship seemed to be driven by few extreme values at the lower and upper end of the spectrum (see Figure 3). Also, the relationship was counter to what we hypothesized, which was an inverse U-shaped curve.
When we adjusted for age, sex, and diabetes, the linear association between plasma glucose and cognitive performance was not significant in both the model with and without the quadratic term, and thus plasma glucose did not influence the ACE score when adjusting for these possible confounders (Table 2 and Table 3. In none of the patient groups was plasma glucose significantly associated with cognitive performance.
To further explore possible non-linear relationships between plasma glucose and the ACE total score we fitted a cubic spline regression model with knots at 4.2 and 6.3 mmol/L plasma glucose, which was the reference range of plasma glucose at our hospital lab.
T A B L E 2 Linear regression on cognitive tests (MMSE and ACE) and plasma glucose levels.   Table 4). Further, we tested if plasma glucose levels were associated with cognitive performance differentially in certain diagnostic groups, owing to their specific pathophysiology.
We found no indication that plasma glucose levels were associated with cognitive performance specifically in patients diagnosed with Alzheimer's disease nor in patients with cerebrovascular disease (Supplementary Tables 5 and 6).

| DISCUSSION
In the present study, we evaluated the association between plasma glucose and cognitive performance measured with two cognitive tests widely used to assess patients with cognitive impairment, namely the MMSE and ACE, utilizing data from a consecutive routine cohort. We found no significant linear association between plasma glucose and cognitive performance on either test, although the glucose levels of our cohort mainly reflected those of patients without diabetes (predominantly within normal ranges). Surprisingly, we found a U-shaped curvilinear relationship between plasma glucose and ACE total score, but the association was weak, and it was not significant when we adjusted for possible confounders. We found no specific differential association between plasma glucose levels and cognitive performance in patients with MMSE < 22 nor in individual diagnostic groups (Alzheimer's disease and cerebrovascular disease).
Our results indicate that plasma glucose levels experienced by the majority of patients seen in a mixed memory clinic cohort do not influence cognitive performance on the MMSE and ACE tests.
Our cohort encompasses some cognitively healthy individuals (including persons with subjective cognitive complaints), but the majority of our cohort (54%) were diagnosed with a dementia disorder.
We have not identified other studies that directly address the question of whether plasma glucose can impact cognitive performance in patients with dementia or in memory clinic cohorts in general.
Recently, Edlund et al reported that plasma glucose levels were associated with cognitive outcomes in a cognitively healthy cohort, although their results were not statistically significant when a stratification on sex was done, which is in line with our results. 24 Burns et al reported that longitudinal changes in glucose levels associate with visual memory but not verbal memory scores in healthy carriers and non-carriers of the APOE ε4 allele. 25 We did not test visual memory, but both the MMSE and ACE incorporate verbal memory tasks. Our results also point to a null association with verbal memory, but we did not investigate sub-scores on the cognitive tests, as we did not have data on this.
Our hypothesis was in part based on studies done in patients with DMI and DMII, which showed that these patient groups' cognitive abilities may be negatively influenced by fluctuations in glucose levels. 6,7 In these studies, the glucose level was controlled by the investigator using a glucose clamp. This allows the investigator to test the direct effect of glucose. Gejl    It is possible that the adverse effect of lowering and increasing glucose levels is only or especially relevant for patients with DMI or DMII, perhaps owing to an altered cerebral glucose metabolism and uptake. 10 In our study we tried to adjust for the presence of diabetes as we were mainly interested in assessing the association between normal-high range (below the diagnostic threshold of 11.1 mmol/L) glucose and cognitive performance. In DMII especially, significant comorbidity and several negative socioeconomic factors -which could confound the cognitive test scores-are also prevalent. 26 We also investigated whether a possible association between glucose and cognitive performance was specific to certain dementia aetiologies, namely patients with cerebrovascular disease and AD. As blood-brain barrier leakiness is present to varying degrees in these disease entities, we hypothesized a priori that this could influence brain glucose utilization. 27 Surprisingly, we found no specific association between glucose and cognitive performance in these subgroups. As our cohort encompasses a wide spectrum of disease severity, it could be that a possible weak association is washed out by larger groups with less severe disease. As we did not have data on disease severity on an individual level, we could not investigate this aspect further.

Test-day normal (<6.3 mmol/l)-follow-up normal (<6.3 mmol/l)
The strengths of our study are the large, well-characterized cohort and the heterogeneous diagnostic composition. We have explored the influence of glucose levels in a variety of diagnostic groups and thus limited bias in selecting only one diagnostic category.
We also adjusted for potential confounders, which limits the possibility of bias.
We are aware of certain limitations of our study. Firstly, in most cases, plasma glucose was measured within a timespan of a few hours from cognitive testing and for a minor part of patients up to a week after cognitive testing. This allows for after-test fluctuations (for example due to eating) which are not meaningful for the period of cognitive testing, although we expect these values to correlate highly.
While this is a substantial limitation of our study, we have also considered patients with consistently high glucose and found no association in this sub-group, which we find reassuring. Further, we performed a 10-fold cross-validation which showed a high validity of our model, which theoretically limits the influence of sub-groups in the data. Secondly, our sample was a convenience sample not particularly collected for the objectives of this study. This increases the chances of spurious findings. However, our study did not support an association between plasma glucose levels and cognitive performance. We did not have data on diabetes medication, other medical comorbidities (e.g., hypertension), or educational level. This might have impaired our model, although when we adjusted for age, sex, and diabetes the association between plasma glucose and ACE disappeared. We also did not investigate sub-scores on the MMSE and ACE test, which would elucidate if the association between glucose and cognitive performance pertained to certain areas of the test such as memory or fluency tasks. The cognitive tests were global and did not focus on the potentially most relevant performance on executive function. Finally, we did not have information on previous testing with the MMSE and ACE which could induce practice effects, which we cannot account for.
Our results indicate that the performance on the MMSE and ACE is not associated with non-fasting, plasma glucose levels (within normal range). This means that clinicians working in memory clinics do not need to account for this aspect when interpreting the results of the MMSE or ACE, although our data mainly relates to normal range plasma glucose and caution should be taken, when ascertaining whether extreme glucose levels may indeed affect the rest results.
Rather, as have been confirmed by others earlier, age and gender seem to influence the test score more, although our study was not designed to address this issue directly. 2 Further studies should look at the pre-test condition of glucose levels to see if levels measured closer to cognitive testing might reveal a potential association.