Abstract

Background:

Cognitive decline is frequently reported in individuals with type 2 diabetes mellitus (T2DM), and working memory impairment represents a key feature of diabetes-related cognitive changes. Previous studies have primarily compared T2DM patients with healthy controls and have typically examined only a single working memory domain. The present study aimed to classify T2DM patients based on Mini-Mental State Examination (MMSE) scores and to examine verbal and visuospatial working memory performance under different cognitive load conditions in order to characterize the cross-sectional association between working memory and global cognitive status.

Methods:

Between November 2023 and June 2024, T2DM patients were recruited from the Department of Endocrinology at the Affiliated Hospital of Yan’an University. Based on MMSE scores, participants were classified into a higher cognitive status group (T2DM-HC, n = 29, MMSE ≥ 27) and a lower cognitive status group (T2DM-LC, n = 25, MMSE 21–26). Working memory performance was assessed using verbal and visuospatial N-back tasks (0-back to 2-back). Reaction time and accuracy were recorded under all task conditions.

Results:

Compared with the T2DM-HC group, the T2DM-LC group had a longer duration of diabetes, higher HbA1c levels, and a greater number of comorbidities (all p < 0.05). In both verbal and visuospatial N-back tasks, the T2DM-LC group exhibited significantly longer reaction times (all p < 0.01), and accuracy declined more markedly under high cognitive load (2-back) conditions (all p < 0.001). Partial correlation analyses indicated that MMSE scores were significantly associated only with 2-back accuracy in both verbal and visuospatial conditions (r = 0.461, p < 0.01; r = 0.659, p < 0.001). Hierarchical regression analyses showed that inclusion of verbal 2-back accuracy increased the explained variance by 18.9% (ΔR2 = 0.189), whereas inclusion of visuospatial 2-back accuracy increased the explained variance by 38.6% (ΔR2 = 0.386).

Conclusion:

At the cross-sectional level, verbal and visuospatial working memory performance was significantly associated with global cognitive status in patients with T2DM, with group differences most pronounced under high cognitive load conditions. In this sample, visuospatial 2-back performance demonstrated a numerically stronger association with global cognitive status. This finding is exploratory in nature and warrants further investigation in future studies.

1 Introduction

Type 2 diabetes mellitus (T2DM) is a metabolic disease characterized by reduced insulin sensitivity and relative insulin deficiency, accounting for more than 90% of all diabetes cases (1). According to the International Diabetes Federation (IDF) (2), approximately 537 million adults aged 20–79 years were living with diabetes worldwide in 2021, and this number is expected to increase to 783 million by 2045.

Mild cognitive decline, considered a transitional cognitive state between normal aging and dementia, has attracted increasing attention in populations with type 2 diabetes mellitus (T2DM). Diabetes-related cognitive dysfunction, also referred to as diabetes-associated cognitive decline (3), primarily describes cognitive changes observed in individuals with diabetes. Cognitive alterations associated with T2DM may manifest along a continuum ranging from mild cognitive decline to dementia (4). Previous studies have shown that mild cognitive decline occurs at a measurable rate in the general population, and this risk may be further elevated in the context of T2DM (5). Although the prevalence of mild cognitive decline in the general population is approximately 11.9%, it increases to 21.8% among individuals with diabetes (6). Furthermore, an international meta-analysis reported that the prevalence of mild cognitive decline among Asian patients with diabetes reached 46.4%, exceeding that observed in European populations (7). In China, individuals with diabetes are characterized by earlier β-cell dysfunction and a younger age at onset, posing greater challenges compared with Western populations (8, 9). Therefore, elucidating the patterns of cognitive changes in patients with T2DM and exploring objective and sensitive behavioral measures at the research level are of considerable importance.

Individuals with diabetes may exhibit mild cognitive decline, particularly in tasks requiring temporary information maintenance and simultaneous processing (10). Working memory is a core cognitive system supporting such complex cognitive activities and consists of relatively independent verbal and visuospatial subsystems (11). Impairment in working memory has been identified as a prominent feature of diabetes-related cognitive decline, manifested as difficulties in short-term information storage and manipulation, which may subsequently lead to slowed information processing and impaired executive function (12). These deficits not only affect daily functioning but may also complicate glycemic management and increase the risk of falls in individuals with T2DM (13). Neuroimaging studies further demonstrate significant structural atrophy and functional abnormalities in brain regions closely associated with working memory, including the prefrontal cortex, hippocampus, and amygdala. More widespread disruptions in higher-order cognitive networks have been observed in individuals with T2DM who exhibit cognitive decline (14). Collectively, these findings suggest that working memory dysfunction occupies a central position in T2DM-related cognitive changes.

The N-back task is a commonly used paradigm for assessing working memory (15). It requires participants to determine whether the current stimulus matches one of the stimuli presented N items previously, thereby providing a sensitive measure of working memory across different cognitive load levels. Previous studies utilizing the N-back task have suggested that individuals with T2DM may experience working memory impairments; however, the findings have been inconsistent. Regarding verbal working memory, Chen et al. (16) reported no significant differences in accuracy between individuals with T2DM and healthy controls on the 0-back, 1-back, and 2-back digit tasks. In contrast, Cansino et al. (17) using a letter-based N-back task, found higher accuracy rates for T2DM patients under the 1-back and 2-back conditions, with no group differences observed under the 0-back condition. Similar inconsistencies have also been reported in visuospatial working memory research (18, 19). These discrepancies may reflect variations in participant characteristics (such as age and baseline cognitive ability), as well as differences in experimental materials (e.g., digits, letters, or shapes) and task parameters (including stimulus duration and interstimulus intervals). This methodological heterogeneity complicates direct comparisons across studies, making it unclear under which cognitive load conditions differences in N-back performance are most pronounced within T2DM populations with varying cognitive status.

Moreover, verbal and visuospatial working memory rely on partially distinct neural mechanisms. Verbal working memory predominantly involves left-hemispheric networks and maintains information through rehearsal processes that actively refresh temporary representations, thereby preventing decay (20). In contrast, visuospatial working memory depends more heavily on right-hemispheric networks and is primarily maintained through selective attention mechanisms that require inhibition of task-irrelevant information. When attentional resources are limited and competition for cognitive resources occurs, visuospatial working memory may be particularly vulnerable (21). Given that individuals with T2DM who exhibit cognitive decline may have constrained attentional resources and vulnerable neural networks, performance across different types of working memory tasks may not be uniform and warrants further empirical investigation.

Although research in this area has gradually shifted from simple cross-sectional comparisons with healthy controls to a greater focus on within-group differences among individuals with T2DM (e.g., varying HbA1c levels) (22), studies examining T2DM populations stratified by cognitive functioning remain limited. In addition, most existing research has examined verbal and visuospatial working memory separately, with relatively few studies integrating both domains within a single experimental framework.

To address these gaps, the present study employed the N-back paradigm to systematically examine verbal and visuospatial working memory performance under different levels of cognitive load in T2DM participants categorized into a lower cognitive status group (T2DM-LC) and a higher cognitive status group (T2DM-HC) based on MMSE scores. Furthermore, we analyzed the relationship between N-back performance and MMSE scores to explore, from a cross-sectional behavioral perspective, the correspondence between objective working memory task performance and global cognitive functioning.

This study primarily aimed to address two questions: (1) Under which cognitive load conditions do performance differences between the T2DM-LC and T2DM-HC groups emerge in the N-back task? and (2) Are accuracy rates in verbal and visuospatial N-back tasks significantly associated with MMSE scores? We hypothesized that the T2DM-LC group would demonstrate poorer performance under higher cognitive load conditions and that N-back accuracy would be significantly correlated with MMSE scores. If supported, these findings would help characterize working memory performance patterns across different levels of cognitive functioning in T2DM and provide a basis for further investigation into the relationship between objective cognitive task performance and global cognitive status indicators.

2 Methods2.1 Participants

From November 2023 to June 2024, patients with type 2 diabetes mellitus (T2DM) were recruited from the inpatient ward of the Department of Endocrinology at the Affiliated Hospital of Yan’an University. The required sample size was estimated using G*Power 3.1.9.2 (Cohen’s f = 0.25, α = 0.05, power = 0.95). For the 2 (T2DM-HC vs. T2DM-LC) × 3 (0-back, 1-back, 2-back) mixed design analysis, the minimum required sample size was calculated to be 44 participants. A total of 60 patients were recruited for the study. Among them, 4 were excluded due to insufficient education levels, which prevented them from completing the cognitive tasks; 1 patient was excluded due to a history of stroke; and 1 participant withdrew during the course of the study. Ultimately, 54 patients completed the entire study protocol and were included in the final analysis. This study was approved by the Ethics Committee of the Affiliated Hospital of Yan’an University (approval number: YA-L20240004) and was conducted in accordance with the Declaration of Helsinki. Prior to participation, the study procedures and objectives were explained to all participants and their families, and written informed consent was obtained after confirming their understanding. For participants with lower MMSE scores or suspected cognitive difficulties, additional consent was obtained from their families.

2.2 Inclusion and exclusion criteria

Inclusion criteria were as follows: (1) age ≥ 18 years; (2)sufficient educational level and comprehension ability to complete the cognitive task(primary school education or above); (3) a diagnosis of type 2 diabetes mellitus in accordance with the 2020 Chinese Guidelines for the Prevention and Treatment of Type 2 Diabetes, defined by the presence of any one of the following criteria: (a) typical diabetic symptoms with a random plasma glucose level ≥ 11.1 mmol/L; (b) fasting plasma glucose ≥ 7.0 mmol/L; or (c) a 2-h plasma glucose level ≥ 11.1 mmol/L during an oral glucose tolerance test (OGTT); and (4) provision of written informed consent by the participant and their family members, with the ability to complete all study procedures.

Exclusion criteria included: (1) significant visual, auditory, speech, or motor impairments that could interfere with task performance or button responses; (2) a documented history of central nervous system disorders, such as stroke, traumatic brain injury, or neurodegenerative diseases; (3) a history of severe psychiatric disorders, including major depressive disorder, bipolar disorder, or schizophrenia; (4) use of medications known to affect cognitive function within a recent period prior to assessment (e.g., within 4 weeks); (5) presence of acute diabetic complications, such as diabetic ketoacidosis or hyperosmolar hyperglycemic state; (6) comorbid malignant tumors; (7) significant dysfunction of major organs, including the heart, liver, or kidneys; (8) a history of alcohol dependence; and (9) occurrence of hypoglycemic episodes, severe hyperglycemia, or active infection within 24 h prior to cognitive assessment.

To minimize the potential impact of neurological or psychiatric conditions on cognitive performance, participants were screened using a multi-step procedure. First, two trained researchers independently reviewed complete inpatient electronic medical records to confirm the absence of relevant neurological or psychiatric diagnoses, hospitalization records, or medication histories. Second, structured interviews were conducted by trained researchers in accordance with the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-5) to screen for psychiatric disorders. Participants with clinically significant depressive symptoms or those currently receiving antidepressant treatment were excluded. Depressive symptom severity was not quantitatively assessed in the present study, as depression was considered solely as an exclusion criterion rather than as a continuous variable in the statistical analyses.

2.3 Grouping criteria

The MMSE was used to assess overall cognitive function and served as the basis for cognitive status grouping. Developed by Folstein et al. (23), the MMSE consists of 30 items, covering five domains: orientation, attention and calculation, memory, recall, and language, with a total score range of 0–30; higher scores indicate better overall cognitive function. Previous studies have shown that the MMSE has good internal consistency (Cronbach’s α = 0.81) and is widely used for the quantitative assessment of overall cognitive function (24).

In this study, MMSE scores were used solely as an operational criterion for group classification. Participants with MMSE scores of 21–26 were categorized into the T2DM-LC group (n = 25), whereas those with MMSE scores ≥27 were categorized into the T2DM-HC group (n = 29). It should be emphasized that group classification in this study was based exclusively on MMSE cut-off values for research purposes and does not constitute a formal clinical diagnosis of cognitive impairment according to comprehensive clinical criteria.

Furthermore, the MMSE primarily reflects global cognitive functioning and has limited sensitivity to subtle changes in specific cognitive domains, such as working memory. MMSE total scores may also be influenced by demographic factors, including age, sex, and years of education. Therefore, age, sex, and years of education were included as covariates in subsequent statistical analyses to control for potential confounding effects.

2.4 Data collection2.4.1 Clinical data collection

Data for all participants were collected through the electronic medical record system, including the following:

Demographic Information: age, sex, educational level, economic status, marital status, smoking and drinking history.

Clinical Characteristics: height, weight, body mass index (BMI), disease duration, systolic blood pressure (SBP), diastolic blood pressure (DBP), fasting plasma glucose (FPG), hemoglobin A1c (HbA1c), number of comorbidities, and treatment regimen (insulin, metformin, insulin combined with metformin).

The glycemic treatment during hospitalization was determined and adjusted by the attending clinicians based on individual circumstances. This study did not intervene in the treatment regimen; relevant medication information was used solely for sample description and, when necessary, included in the analysis as a potential confounding factor.

2.4.2 Working memory assessment

Working memory was assessed using verbal and visuospatial N-back tasks programmed in E-Prime 3.0. The N-back paradigm, originally developed by Kirchner (25), requires participants to determine whether the current stimulus matches the stimulus presented N trials earlier. Stimuli were presented on a 15.3-inch Lenovo ThinkBook laptop computer (resolution: 1024 × 768) with a white background and black text. The viewing distance was fixed at 50 cm. All tasks were administered in a quiet environment, and standardized instructions were provided to all participants.

2.4.2.1 Verbal N-back task

In the verbal N-back task, Arabic numerals were used as stimuli, and participants were required to determine whether the currently presented number matched the number presented N trials earlier. The task included three levels of cognitive load (0-back, 1-back, and 2-back). Following previous research (16), each trial began with a 500 ms fixation point, followed by a 1,000 ms stimulus presentation, with a 2,000 ms interstimulus interval and a maximum response window of 3,000 ms. The practice phase consisted of 10 trials, while the formal experimental phase included 30 trials, with a target-to-nontarget trial ratio fixed at 1:2. Each cognitive load condition lasted approximately 7 min, with a 1-min rest interval between different load conditions, resulting in a total task duration of about 10 min (Figure 1). The verbal and visuospatial N-back tasks were presented in a fixed order for all participants. Within each task, cognitive load conditions (0-back, 1-back, 2-back) were presented in a fixed sequence from low to high to reduce the initial task load for older participants.

Flowchart of the verbal and visuospatial N-back tasks: (A) Flowchart of the verbal N-back task; (B) Flowchart of the visuospatial N-back task.

2.4.2.2 Visuospatial N-back task

The visuospatial N-back task used a gray cross and a red circle appearing at pseudorandom locations on the screen. Participants were required to determine whether the location of the current stimulus matched the location presented N trials earlier. Trial timing parameters, number of trials, and rest intervals were identical to those used in the verbal N-back task.

2.4.2.3 Experimental design and data processing

A single-blind design was employed, such that researchers administering the N-back tasks were unaware of participants’ group assignments (T2DM-LC or T2DM-HC). Behavioral data, including accuracy and reaction time, were automatically recorded by E-Prime. Individual trials were excluded according to predefined criteria: accuracy <50%, reaction time <200 ms, or values exceeding ±2.5 standard deviations from the participant’s mean (26, 27), to ensure data quality.

2.5 Statistical analysis

Statistical analyses were performed using SPSS version 26.0. All tests were two-tailed, and the significance level was set at p < 0.05. The normality of continuous variables was assessed before analysis. Normally distributed variables were expressed as mean ± standard deviation and compared between groups using independent t-tests. Non-normally distributed variables were expressed as median (P25, P75) and compared using the Mann–Whitney U test. Categorical variables were presented as counts (percentages) and analyzed using the chi-square test, with Fisher’s exact test used when appropriate.

A mixed-design repeated measures analysis of variance (ANOVA) was conducted, with group (T2DM-LC vs. T2DM-HC) as the between-subjects factor and cognitive load (0-back, 1-back, 2-back) as the within-subjects factor, to examine the main effects and interaction effects of verbal and visuospatial N-back task performance. Greenhouse–Geisser correction was applied when the sphericity assumption was violated. Simple effects analysis was performed only when the interaction was significant, and Bonferroni correction was applied in post-hoc comparisons between groups under different cognitive load conditions. Additionally, partial correlation analysis was conducted to examine the relationship between N-back accuracy and MMSE scores, controlling for age, gender, and years of education. Further, hierarchical regression analysis was performed, with demographic variables entered in the first step and N-back accuracy in the second step, to test the association between working memory task performance and MMSE scores after controlling for demographic factors. The above correlation and regression analyses were exploratory in nature and addressed different research questions, and no global multiple comparison correction was applied, which may increase the risk of Type I error.

3 Results3.1 Baseline characteristics of the study population

Among the 54 T2DM participants included in the analysis, 25 were classified into the T2DM-LC group and 29 into the T2DM-HC group based on the MMSE operational grouping criteria. Compared to the T2DM-HC group, participants in the T2DM-LC group had a longer duration of diabetes (13.0 vs. 10.0 years), higher HbA1c levels (9.0% vs. 8.0%), more comorbidities (2.0 vs. 1.0), and lower MMSE scores (24 vs. 28). All group differences were statistically significant (p < 0.05) (Table 1).

CharacteristicsT2DM-HC (n = 29)T2DM-LC (n = 25)t/Z/χ2pAge (y)57.72 ± 4.8658.84 ± 4.90−0.838a0.406Gender1.862c0.172Male, n (%)17 (58.6)10 (40.0)Female, n (%)12 (41.4)15 (60.0)Duration of education2.658b0.447Primary school, n (%)5 (17.2)7 (28.0)Junior high school, n (%)14 (48.3)14 (56.0)High school, n (%)8 (27.6)3 (12.0)College degree or above, n (%)2 (6.9)1 (4.0)Economic situation (yuan/month)0.739b0.691<3,0008 (27.6)9 (36.0)3,000 ~ 5,00014 (48.3)12 (48.0)>5,0007 (24.1)4 (16.0)Marital status1.182c0.463Married, n(%)29 (100.0)24 (96.0)Widowhood and Other, n(%)0 (0.00)1 (4.0)Treatment regimen1.293c0.524Insulin, n (%)12 (41.4)7 (28.0)Metformin, n (%)8 (27.6)7 (28.0)Insulin plus metformin, n (%)9 (31.0)11 (44.0)Smoking, n (%)14 (48.3)8 (32.0)2.241c0.524Drinking, n (%)18 (62.1)15 (60.0)1.359c0.715BMI(kg/m2)23.41 ± 1.9224.48 ± 2.88−1.571a0.124SBP (mmHg)127.34 ± 14.75126.92 ± 12.680.113a0.911DBP (mmHg)79.45 ± 8.2677.32 ± 7.940.962a0.341Duration of diabetes (y)10.00 (8.00, 13.00)13.00 (11.00, 17.00)−2.525b0.012FPG (mmol/L)8.00 (7.00, 9.50)9.00 (8.00, 10.50)−1.470b0.142HbA1c (%)8.00 (8.00, 9.00)9.00 (8.00, 10.50)−2.469b0.014Number of complications1.00 (0.00, 2.00)2.00 (1.00, 2.00)−2.461b0.017MMSE28 (27, 29)24 (22.5, 24.5)−6.362b<0.001

Baseline characteristics of patients with T2DM-LC and T2DM-HC.

Data are presented as mean ± standard deviation (SD), median (interquartile range) [M (P25, P75)], or as number (percentage).

BMI, body mass index; SBP, systolic blood pressure; DBP, diastolic blood pressure; FPG, fasting plasma glucose; HbA1c, glycosylated hemoglobin; MMSE: Mini-Mental State Examination; aIndependent samples t-test. bMann–Whitney U test (Z value reported). cChi-square test (χ2).

During hospitalization, two patients (one from the T2DM-LC group and one from the T2DM-HC group) had received sedative medications, and one patient (from the T2DM-HC group) was admitted through the emergency department. These patients were excluded from the sensitivity analysis.

3.2 Descriptive performance on the N-back tasks

In the verbal N-back task, the T2DM-LC group demonstrated longer reaction times than the T2DM-HC group across all cognitive load conditions (0-back: 882.3 vs. 787.6 ms; 1-back: 1179.2 vs. 995.9 ms; 2-back: 1577.0 vs. 1347.9 ms). With respect to accuracy, no clear group differences were observed under the 0-back or 1-back conditions; however, accuracy was lower in the T2DM-LC group under the 2-back condition (64.7% vs. 69.6%) (Table 2).

Load levelT2DM-HC (n = 29)T2DM-LC (n = 25)tMD (95% CI)Cohen’s dpVerbal N-back- Reaction time (ms)0-back787.56 ± 118.13882.32 ± 139.90−2.70−94.76 (−165.21, −24.31)0.740.0091-back995.89 ± 95.631179.15 ± 133.12−5.87−183.26 (−245.95, −120.57)1.60<0.0012-back1347.94 ± 109.041577.02 ± 116.15−7.47−229.09 (−290.63, −167.54)2.04<0.001Verbal N-back- Accuracy (%)0-back96.55 ± 3.1995.58 ± 3.521.060.97 (−0.87, 2.80)0.290.2961-back88.47 ± 6.4786.26 ± 5.301.352.20 (−1.06, 5.47)0.370.1822-back69.60 ± 5.1164.69 ± 4.373.764.91 (2.29, 7.529)1.03<0.001Visuospatial N-back-Reaction time (ms)0-back850.35 ± 123.96957.92 ± 153.13−2.85−107.57 (−183.25, −31.89)0.780.0061-back1109.18 ± 133.471347.25 ± 118.38−6.88−238.07 (−307.48, −168.67)1.88<0.0012-back1446.47 ± 121.541663.55 ± 118.43−6.62−217.08 (−282.86, −151.30)1.81<0.001Visuospatial N-back- Accuracy (%)0-back95.71 ± 3.2894.07 ± 4.281.601.65 (−0.42, 3.72)0.430.1151-back87.24 ± 6.1184.66 ± 5.821.582.58 (−0.69, 5.86)0.430.1202-back65.75 ± 4.1158.39 ± 4.616.217.37 (4.99, 9.75)1.69<0.001

Reaction time and accuracy of subjects in verbal and visuospatial N-back tasks.

Data are mean ± SD unless otherwise indicated; MD, mean difference; CI, confidence interval; Cohen’s d represents the standardized effect size for between-group comparisons.

In the visuospatial N-back task, a comparable pattern was observed. The T2DM-LC group showed longer reaction times across all load levels, and differences in accuracy were primarily observed under the 2-back condition (Table 2).

3.3 Verbal N-back task performance in patients with T2DM

Based on the previous analysis, the verbal and visuospatial N-back task performances of the T2DM-LC and T2DM-HC groups were further compared.

Reaction Time: Repeated measures analysis of variance (ANOVA) revealed a significant main effect of group [F(1, 52) = 66.926, p < 0.001, ηp2 = 0.563], a significant main effect of cognitive load [F(2, 104) = 426.878, p < 0.001, ηp2 = 0.891], and a significant group × load interaction [F(2, 104) = 4.991, p < 0.01, ηp2 = 0.088]. Simple effects analysis indicated that, under the 0-back, 1-back, and 2-back conditions, the reaction time of the T2DM-LC group was significantly longer than that of the T2DM-HC group (all p < 0.01). Additionally, reaction times for both groups significantly increased as task load increased (p < 0.001) (Table 3; Figure 2).

TaskEffectSSMSFpηp2Verbal N-back reaction time (ms)Groupa1150860.8831150860.88366.926<0.0010.563Load levelb10708515.795354257.896426.878<0.0010.891Group*Load levelc125193.85562596.9284.9910.0090.088Verbal N-back accuracy (%)Groupa291.908291.9089.1080.0040.149Load levelb23648.92011824.460629.601<0.0010.924Group*Load levelc525.989304.38815.353<0.0010.228Visuospatial N-back reaction time (ms)Groupa1417128.8301417128.83060.707<0.0010.539Load levelb11375473.515687736.757433.490<0.0010.893Group*Load levelc131857.50665928.7535.0250.0080.088Visuospatial N-back accuracy (%)Groupa602.206602.20615.738<0.0010.232Load levelb30924.00115462.001