This study is a component of the General Lifestyle and AD (GLAD) study, an ongoing prospective cohort study launched in 2020. By January 2022, the study had recruited 225 adults for eligibility assessment. Out of these, 196 non-demented individuals aged between 65 and 90 years were enrolled in the initial GLAD cohort. The exclusion of 29 participants was based on the following criteria: 7 were disqualified due to either existing comorbid conditions—medical, psychiatric, or neurological—that could influence mental functioning (5 individuals), severe communication or behavioral issues complicating clinical assessments (1 individual), or concurrent participation in another clinical trial involving an investigational product (1 individual). Additionally, 2 participants were excluded for not fitting any diagnostic categories, and 20 were removed from the study due to either withdrawal of consent (17 individuals) or loss of contact (3 individuals). The composition of the baseline GLAD cohort included 113 cognitively normal (CN) adults and 83 individuals diagnosed with mild cognitive impairment (MCI). Eligible participants were identified through the dementia screening program conducted at the memory clinic of Hallym University Dongtan Sacred Heart Hospital in Hwaseong, South Korea. Moreover, community volunteers were enlisted through recommendations from existing participants, family members, friends, or acquaintances. The CN individuals had Clinical Dementia Rating (CDR) (Morris, 1993) score of 0 and no diagnosis of MCI or dementia. The MCI individuals had CDR of 0.5 and met the inclusion criteria outline in the core clinical criteria for diagnosis of MCI, as per the recommendations of the NIA-AA guidelines (Albert et al., 2011). Regarding objective memory impairment, the age-, education-, and sex-adjusted z-score was < −1.0 for at least one of four episodic memory tests included in the Korean version of the CERAD-K neuropsychological battery: word list memory (WLM), word list recall (WLR), word list recognition (WLRc), and constructional recall (CR) tests (Morris et al., 1989; Lee et al., 2002, 2004). The exclusion criteria included the existence of a major psychiatric disorder, a notable neurological or medical condition, or any coexisting health issue capable of influencing mental function. Additionally, factors such as illiteracy, visual or hearing impairments, severe communication or behavioral issues that could hinder clinical examinations, and the utilization of an investigational drug were considered grounds for exclusion.

Each participant underwent standardized clinical assessments by trained psychiatrists following the clinical assessment of the GLAD study protocol. This protocol included the Korean version of the CERAD (CERAD-K) (Morris et al., 1989; Lee et al., 2002). Additionally, trained neuropsychologists administered the GLAD neuropsychological assessment protocol, which incorporated the CERAD-K neuropsychological battery (Lee et al., 2004) to all participants. To evaluate the relationship between individual items of depression and overall cognitive function, we utilized the total score (TS) of the CERAD (Seo et al., 2010). This score, indicative of overall cognitive ability, is derived by totaling the scores from seven tests within the CERAD neuropsychological battery, including verbal fluency, the Boston Naming Test, WLM, constructional praxis, WLR, WLRc, and CR tests. In comparing the diagnostic accuracy of individual depression items for MCI, we utilized the MMSE (Nakata et al., 2009), one of the most widely used scales in clinical settings, either alone or in combination with individual depression items. The evaluation of vascular risk factors (VRF), such as hypertension, diabetes mellitus, dyslipidemia, coronary heart disease, transient ischemic attack, and stroke, involved systematic interviews of participants and their family members, with data collection performed by trained researchers. The vascular risk score (VRS) was determined by calculating the percentage of present VRF (DeCarli et al., 2004). To assess the severity and components of depressive symptoms, the 30-item Geriatric Depression Scale (GDS) was employed, with detail available at https://web.stanford.edu/~yesavage/GDS.english.long.html (Yesavage et al., 1982; Kim et al., 2008). The body mass index (BMI) was computed by dividing the body weight in kilograms by the square of the height in meters, following the guidelines of the World Health Organization (WHO) available at https://www.euro.who.int/en/health-topics/disease-prevention/nutrition/a-healthy-lifestyle/body-mass-index-bmi. Physical activity was measured using the Korean-version of the Physical Activity Scale for the Elderly (PASE) (Washburn et al., 1993; Choe et al., 2010). The PASE score, indicating the level of physical activity, was calculated by summing weighted scores from leisure, household, and occupational activity subscales. Participants’ activity levels were scored and divided into low and high categories, with higher scores indicating greater activity. The study divided participants into three income-based groups: those earning less than the minimum cost of living (MCL), those earning between the MCL and twice the MCL, and those earning at least twice the MCL, as defined by the Ministry of Health and Welfare, Republic of Korea, in November 2012.¹ The MCL for a single-person household was set at 572,168 Korean Won per month (approximately US$ 507.9), with an additional 286,840 Korean Won (about US$ 254.6) per month added for each extra member in the household. Lifetime alcohol intake status (never/former/drinker) and smoking status (never/ex-smoker/smoker) were assessed through interviews conducted by trained researchers and a review of medical records. To ensure the accuracy of information, reliable informants were interviewed.

Following an overnight fast, blood samples were collected via venipuncture in the morning (08,00–09:00). Albumin, glucose, high-density lipoprotein (HDL)-cholesterol, and low-density lipoprotein (LDL)-cholesterol were determined using a COBAS c702 analyzer and dedicated reagents (Roche Diagnostics, Mannheim, Germany). Apolipoprotein E (apoE) was genotyped using a Seeplex ApoE ACE genotyping kit (Seegene, Seoul, Korea). The positivity for the apoE ε4 allele (APOE4) was defined as the presence of at least one ε4 allele.

Between-group comparisons for continuous data, encompassing demographic and clinical information, were conducted using two-tailed t-tests. Categorical data were analyzed through the chi-square test. Logistic regression analysis was performed to identify specific items within the 30-item GDS that could predict MCI. Stepwise logistic regression analyses were executed to pinpoint the GDS items minimizing the probability of misclassification between subjects with and without MCI, utilizing the likelihood ratio statistic for item entry and removal at p < 0.05 and p < 0.10. To assess the diagnostic accuracy of components of depression along with MMSE, multiple logistic regression analyses were performed with GDS-items and MMSE as the independent variables and MCI status as the dependent variable. Given potential confounders such as VRS, BMI, PASE, alcohol intake, smoking, and blood nutritional markers, including albumin, glucose, HDL- and LDL-cholesterol, all participants were systematically evaluated. Two models were tested, adjusting for covariates in a stepwise fashion. The first model included age, sex, education, APOE4, and VRS as covariates. The second model encompassed these covariates plus BMI, PASE, alcohol intake, smoking, albumin, glucose, and HDL- and LDL-cholesterol. We utilized the differences in −2 log likelihood (−2LL) for the statistical comparison of the predictive ability among various models with different numbers of independent variables (Lee et al., 2006). The −2LL is a measure derived from logistic regression analysis and is directly correlated with the contribution of variables to the distinction between groups; a smaller −2LL value suggests a superior predictive ability of the model. The probability distribution of −2LL difference between a simple (model 1) and a more complex model (model 2) can be approximated by a chi-square distribution with the difference between degrees of freedom ( Δ df = df2 − df1), where df1 and df2 represent the degrees of freedom of models 1 and 2, respectively. Consequently, the difference in −2LL enables a direct comparison of prediction models with different complexities (Hosmer, 1989). As part of sensitivity analyses, the same analyses were conducted for subjects exhibiting no moderate to extreme severe depressive symptoms (GDS score ≤ 23), employing a cutoff score derived from prior research (Yi, 2016). This approach was adopted as depression might exhibit greater discriminatory effects in individuals with MCI. To examine the relationship between the individual depression items, which were deemed significant in the aforementioned analyses, and overall cognitive function, we conducted regression analysis. In this analysis, each individual depression item was used as an independent variable and overall cognitive function as the dependent variable. Covariates were controlled using the two-step model approach as described above. Moreover, we explored the moderation effects of age, sex, APOE4, education, GDS, VRS, BMI, and PASE status on the relationships between the individual depression items and the overall cognitive function. We included two-way interaction term between the individual depression item and any one of the factors, as well as the individual depression item itself, as an independent variable in the regression model. For significant interactions, subsequent subgroup analyses were performed using an additional regression model for each subgroup divided by the moderation variable. All statistical analyses were carried out using SPSS Statistics software ver. 28 (IBM, Armonk, NY, United States).

The current study protocol underwent review and approval by the institutional review board of Hallym University Dongtan Sacred Heart Hospital and was conducted in accordance with the principles outlined in the current version of the Declaration of Helsinki. Participants provided informed consent to participate in the study.

The demographic and clinical characteristics of the study population are presented in Tables 1, 2. Among the 196 nondemented adult participants, 113 were classified as CN, and 83 were diagnosed with MCI.

Following the models obtained through stepwise logistic regression analyses, GDS item #10 (helplessness: “Do you often feel helpless?”; odds ratio = 4.531, 95% CI = 2.218 to 9.258, p < 0.001) and GDS item #14 (subjective memory problems: “Do you feel you have more problems with memory than most?”; odds ratio = 1.999, 95% CI = 1.040 to 3.840, p = 0.038) were extracted from the 30-item GDS as having significant contributions to the diagnostic accuracy for MCI (Table 3).

Initially, helpless item (GDS item #10) and subjective cognitive problem item (GDS item #14) (referencing the results from the aforementioned stepwise logistic regression analyses), along with MMSE (recognized as one of the best objective cognitive tools for predicting MCI diagnosis), were selected as candidate independent variables for logistic regression analyses to compare models. The analyses proceeded in three steps (Table 4). In the first step, three one-candidate models were tested: Model H (including helpless item); Model S (including subjective memory problems item); and Model M (including MMSE). Two one-candidate models, specifically Model H and Model M, demonstrated statistical significance, with Model M exhibiting the highest classification accuracy and the smallest −2LL among the three models. Additionally, Model H displayed the highest classification accuracy and the smallest −2LL among the two one-candidate models with a depression component. In the second step, the −2LL was compared between each one-candidate model and the corresponding two-candidate models, which included MMSE and either helpless item or subjective memory problems item. The MCI diagnostic accuracy of both Model HM (helplessness + MMSE) including helpless item and MMSE, and Model SM (subjective memory problems + MMSE) including subjective memory problems item and MMSE, was significantly superior to that of Model M (MMSE), which had the highest classification accuracy and smallest −2LL among the three one-candidate models. In the third step, the three-candidate model, Model HSM (helplessness + subjective memory problems + MMSE), including all three variables, was compared with Model HM or Model SM. Model HSM was significantly superior to Model SM, but not to Model HM. The ultimately selected logistic regression model for MCI diagnosis was Model HM (helplessness + MMSE) (Table 5). These findings were corroborated by a sensitivity analysis performed after excluding participants with no moderate to extremely severe depressive symptoms (Table 6).

The multiple linear regression analyses revealed that helplessness was associated with lower TS, with this association remaining significant even after adjusting for potential confounders (B = −5.300, SE = 1.899, β = −0.162, p = 0.006) (Table 7). Subjective memory problem was also associated with lower TS (B = −3.557, SE = 1.680, β = −0.109, p = 0.036) (Table 7, Model 1). However, this association ceased to be significant after adjusting for potential confounders (Table 7, Model 2).

The analysis revealed a significant interaction between helplessness and physical activity, indicating that physical activity influences the relationship between hKelplessness and TS (B = 0.090, SE = 0.042, β = 0.225, p = 0.034). On the other hand, interactions between helplessness and other factors such as age, sex, APOE4, VRS, and BMI was not significant (Table 8). Further subgroup analysis indicated that helpless was significantly linked to TS only among participants engaging in lower level of physical activity (B = −7.291, SE = 2.862, β = −0.216, p = 0.013). However, this relationship was not observed in participants with higher level of physical activity (Table 9).

Multiple logistic regression analyses revealed that overall GDS contributes to the diagnostic accuracy for MCI (odds ratio = 1.071, 95% CI = 1.024 to 1.121, p = 0.003) (Supplementary Table S1).

The Model G (incorporating overall GDS alone) exhibited high diagnostic accuracy for MCI (Supplementary Table S2). Furthermore, the MCI diagnostic accuracy of Model GM (combining overall GDS and MMSE) was significantly superior to that of Model M (Supplementary Table S2).

The Model H and Model T (incorporating TS alone) exhibited high diagnostic accuracy for MCI (Supplementary Table S2). Furthermore, the MCI diagnostic accuracy of Model HT (combining helplessness and TS) was significantly superior to that of Model T (Supplementary Table S3).