Skip to main content
NCBI home page
Brain and Behavior logoLink to Brain and Behavior
. 2024 Mar 7;14(3):e3456. doi: 10.1002/brb3.3456

Predictive model for identifying mild cognitive impairment in patients with type 2 diabetes mellitus: A CHAID decision tree analysis

Rehanguli Maimaitituerxun 1, Wenhang Chen 2, Jingsha Xiang 3, Yu Xie 1, Fang Xiao 4, Xin Yin Wu 1, Letao Chen 5, Jianzhou Yang 6, Aizhong Liu 1, Wenjie Dai 1,✉
PMCID: PMC10918605  PMID: 38450963

Abstract

Background

As the population ages, mild cognitive impairment (MCI) and type 2 diabetes mellitus (T2DM) become common conditions that often coexist. Evidence has shown that MCI could lead to reduced treatment compliance, medication management, and self‐care ability in T2DM patients. Therefore, early identification of those with increased risk of MCI is crucial from a preventive perspective. Given the growing utilization of decision trees in prediction of health‐related outcomes, this study aimed to identify MCI in T2DM patients using the decision tree approach.

Methods

This hospital‐based case–control study was performed in the Endocrinology Department of Xiangya Hospital affiliated to Central South University between March 2021 and December 2022. MCI was defined based on the Petersen criteria. Demographic characteristics, lifestyle factors, and T2DM‐related information were collected. The study sample was randomly divided into the training and validation sets in a 7:3 ratio. Univariate and multivariate analyses were performed, and a decision tree model was established using the chi‐square automatic interaction detection (CHAID) algorithm to identify key predictor variables associated with MCI. The area under the curve (AUC) value was used to evaluate the performance of the established decision tree model, and the performance of multivariate regression model was also evaluated for comparison.

Results

A total of 1001 participants (705 in the training set and 296 in the validation set) were included in this study. The mean age of participants in the training and validation sets was 60.2  ±  10.3 and 60.4  ±  9.5 years, respectively. There were no significant differences in the characteristics between the training and validation sets (p > .05). The CHAID decision tree analysis identified six key predictor variables associated with MCI, including age, educational level, household income, regular physical activity, diabetic nephropathy, and diabetic retinopathy. The established decision tree model had 15 nodes composed of 4 layers, and age is the most significant predictor variable. It performed well (AUC = .75 [95% confidence interval (CI): .71–.78] and .67 [95% CI: .61–.74] in the training and validation sets, respectively), was internally validated, and had comparable predictive value compared to the multivariate logistic regression model (AUC = .76 [95% CI: .72–.80] and .69 [95% CI: .62–.75] in the training and validation sets, respectively).

Conclusion

The established decision tree model based on age, educational level, household income, regular physical activity, diabetic nephropathy, and diabetic retinopathy performed well with comparable predictive value compared to the multivariate logistic regression model and was internally validated. Due to its superior classification accuracy and simple presentation as well as interpretation of collected data, the decision tree model is more recommended for the prediction of MCI in T2DM patients in clinical practice.

Keywords: decision tree, mild cognitive impairment, predictor variable, type 2 diabetes mellitus

  

graphic file with name BRB3-14-e3456-g002.jpg

1. INTRODUCTION

Mild cognitive impairment (MCI) and type 2 diabetes mellitus (T2DM) are highly prevalent and often coexist in older adults (Srikanth et al., 2020). MCI is a transitional stage between normal ageing and dementia that was characterized by cognitive dysfunction with minimal impairment in instrumental activities of daily living (Petersen, 2004; Petersen et al., 1999). Accumulated evidence has consistently shown that the presence of MCI in the general population was associated with increased risk of dementia (Davis et al., 2018; Mitchell & Shiri‐Feshki, 2009; Zhang et al., 2021). In addition to that, MCI in T2DM could lead to reduced treatment compliance, medication management, and self‐care ability (Kim & Fritschi, 2021; Verma et al., 2021), which may be explained by the decrements in working memory, learning, executive function, and processing speed observed in MCI (Christman et al., 2010; Santos et al., 2018), though the function of daily activities is essentially preserved. Therefore, the management of MCI in T2DM patients is crucial, and early identification of those with MCI is a starting point.

The decision tree is one of the most commonly used machine learning models in a wide range of medical situations requiring decision‐making that provides high classification accuracy with a simple representation of gathered knowledge (Bae, 2014; Podgorelec et al., 2002). Compared to other machine learning models, the decision tree has the following advantages: (1) it can be visualized and is simple to understand and interpret in clinical practice; (2) it provides a remarkably transparent decision‐making process, allowing deep exploration of features; and (3) due to its high transparency, the decision‐making process can be easily validated by an expert that greatly enhances its utility in situations containing high uncertainty (Amendolara et al., 2023; Bae, 2014; Plante et al., 1986; Podgorelec et al., 2002; Ting Sim et al., 2023). A growing body of literature has demonstrated the effectiveness of decision trees in predicting the occurrence as well as the prognosis of health‐related outcomes (Toyoda et al., 2023; Yang et al., 2021; Zhou et al., 2023), and some studies directly compared decision tree model with common machine learning models to solve prediction problems (Hu et al., 2022; Langenberger et al., 2023). Previous work on comparison of logistic regression and decision tree models found comparable predictive values (Wentzlof et al., 2019; Zhang et al., 2022). However, as the interpretation and understanding of logistic regression model are more difficult than that of decision tree model, especially for those without experience with this particular model type, decision tree model is more recommended in clinical practice (Fu et al., 2022). In the diabetic population, few studies to date have used the decision tree models to predict diabetic comorbidities or complications (Kasbekar et al., 2017; Rinkel et al., 2020; Zhou et al., 2022), and to our knowledge, there is no study on prediction of MCI in patients with T2DM using the decision tree model. Given the growing utilization of decision trees in prediction of health‐related outcomes and the negative effects of MCI on the prognosis of T2DM patients, this study aimed to identify MCI in T2DM patients using the decision tree approach to help better identify MCI.

2. METHODS

This study adhered to the principles of the Declaration of Helsinki. The protocol of this study was reviewed and approved by the Ethics Committee of the Xiangya School of Public Health, Central South University. Before enrollment, a written informed consent was obtained from each participant.

2.1. Study design and setting

This was a hospital‐based case–control study that was performed in the Endocrinology Department of Xiangya Hospital affiliated to Central South University between March 2021 and December 2022, which is a Grade III general hospital and located in the capital city of Hunan Province in China with patients across the country.

2.2. Participants

The inclusion criteria for the participants were as follows: (1) a clinical diagnosis of T2DM by the established Chinese Guideline (Chinese Diabetes Society, 2021); (2) hospitalized due to T2DM; (3) aged ≥40 years; and (4) voluntarily participated in this study and signed informed consent. The exclusion criteria for the participants were as follows: (1) repeated hospitalizations during the study period; (2) those with dementia; and (3) those could not speak or understand Mandarin. Eligible participants who met the diagnostic criteria for MCI were considered cases, whereas those unfulfilled were considered controls with normal cognitive function.

2.3. Outcome of interest

The outcome of this study is MCI, and it was defined based on the Petersen criteria by the physicians (Petersen, 2004): (1) chief complaint of memory loss confirmed by families or caregivers; (2) objective cognitive impairment measured by the mini–mental state examination score (≤19 for illiterate individuals, ≤22 for individuals with elementary school education, and ≤26 for individuals with middle school education or above) (Fan et al., 2021; Li et al., 2016); (3) essentially preserved function of daily activities; and (4) absence of dementia.

2.4. Data collection

Demographic characteristics and lifestyle factors were obtained retrospectively through face‐to‐face interviews by well‐qualified investigators. T2DM‐related information was extracted from medical records. All investigators underwent unified training and were blinded to the cognitive status of the participants.

2.5. Predictor variables

The demographic characteristics included age (40–59 vs. ≥60), sex (male vs. female), marital status (married vs. unmarried), educational level (middle school or below vs. high school or above), household income (≤5000 RMB vs. >5000 RMB), location of residence (urban vs. rural), primary caregiver (self‐care vs. others), and current work status (employed vs. unemployed).

Lifestyle factors included information on smoking, drinking, and physical activity, which were all regarded as dichotomous variables (current smoker [Yes vs. No], current drinker [Yes vs. No], and regular physical activity [Yes vs. No]). Specifically, a current smoker was defined as one who smoked at least one cigarette per day in the past month; a current drinker was defined as one who consumed at least one alcohol per day in the past month; and regular physical activity was defined as performing at least one activity such as walking, square dancing, or cycling for at least 150 min per week in the past month (Umpierre et al., 2011; World Health Organization [WHO], 2020).

T2DM‐related factors included information on body mass index (<24 vs. ≥24 kg/m2), duration of diabetes (<10 years vs. ≥10 years), family history of diabetes (Yes vs. No), glycated hemoglobin A1c (≤7% vs. >7%), diabetic comorbidities (stroke [Yes vs. No], hypertension [Yes vs. No], coronary heart disease [Yes vs. No], and fatty liver [Yes vs. No]), and diabetic complications (diabetic nephropathy [Yes vs. No], diabetic retinopathy [Yes vs. No], and diabetic foot [Yes vs. No]).

2.6. Statistical analyses

The study sample was randomly divided into the training and validation sets in a 7:3 ratio. The training set was used to develop the decision tree model and the validation set was used to validate the decision tree model internally. Based on the fact that excluding all observations with missing values could induce substantial bias as well as lack of efficiency, missing data were filled using random forest interpolation (Salgado, et al., 2016; Fox‐Wasylyshyn & El‐Masri, 2005).

Continuous variables were reported as mean ± standard deviation or median and interquartile range as appropriate, and categorical variables were reported as frequency (n) and proportion (%). Chi‐square test or Fisher's exact test was used to compare the categorical predictor variables in the training set as appropriate, and predictor variables that differed significantly were entered into the multivariate logistic regression model.

The chi‐square automatic interaction detection (CHAID) algorithm, whose main purpose was to identify key factors related to the outcomes of interest, was used to develop the decision tree model (Kass, 1980). In this algorithm, homogenous groups could be constructed by any possible combination of the known values of a predictor variable. The number of predictor variables for creating the decision tree model depends on the values of chi‐square test and whether the differences are statistically significant or not. There are three types of nodes in decision tree: (1) a root node, representing a choice that will induce the subdivision of all records into two or more mutually exclusive subsets; (2) internal nodes, representing one of the possible choices that are available at that point in the tree structure; and (3) leaf nodes, representing the final result of a combination of decisions (Song & Lu, 2015). The node containing all the cases is considered the root node in the CHAID algorithm, and the predictor variable with the largest chi‐square value divides the entire sample into at least two subgroups, which are subsequently split by the next most significant predictor variable. The analysis continues in this stepwise way to choose the next most significant predictor variable until there are no more significant predictor variables.

The performance of the decision tree model and multivariate logistic regression model was evaluated by the receiver operating characteristic (ROC) curves and the area under the curve (AUC). All statistical analyses were two‐sided and performed using the IBM SPSS software (version 26.0) or R software (version 4.2.1; https://www.r‐project.org/). A p value of <.05 was considered statistically significant.

3. RESULTS

3.1. Characteristics of the study participants

A total of 1001 participants (274 cases and 727 controls) were included in this study, 705 (191 cases and 514 controls) and 296 (83 cases and 213 controls) of which were randomly assigned into the training and validation sets, respectively. The characteristics of the study participants are shown in Table 1. The mean age of participants in the training and validation sets was 60.2  ±  10.3 and 60.4  ±  9.5 years, respectively, and there were 432 (61.3%) and 181 (61.1%) male participants in the training and validation sets, respectively. There were no significant differences in the characteristics between the training and validation sets (p > .05).

TABLE 1.

Characteristics of the study participants.

Variables Category Total sample (n = 1001, %) Training set (n = 705, %) Validation set (n = 296, %) p Value
Age (year) 40–59 520 (51.9) 365 (51.8) 155 (52.4)
≥60 481 (48.1) 340 (48.2) 141 (47.6) .864
Sex Male 613 (61.2) 432 (61.3) 181 (61.1)
Female 388 (38.8) 273 (38.7) 115 (38.9) .970
Marital status Married 924 (92.3) 649 (92.1) 275 (92.9)
Unmarried 77 (7.7) 56 (7.9) 21 (7.1) .650
Educational level Middle school or below 542 (54.1) 387 (54.9) 155 (52.4)
High school or above 459 (45.9) 318 (45.1) 141 (47.6) .464
Household income (RMB) ≤5000 638 (63.7) 450 (63.8) 188 (63.5)
>5000 363 (36.3) 225 (36.2) 108 (36.5) .924
Location of residence Urban 595(59.4) 474 (67.2) 198 (66.9)
Rural 406 (40.6) 231 (32.8) 98 (33.1) .916
Primary caregiver Self‐care 438 (43.8) 314 (44.5) 124 (41.9)
Others 563 (56.3) 391 (55.5) 172 (58.1) .441
Current work status Employed 326 (32.6) 229 (32.5) 97 (32.8)
Unemployed 675 (67.4) 476 (67.5) 199 (67.2) .929
Current smoker No 561 (56) 565 (80.1) 235 (79.4)
Yes 440 (44) 140 (19.9) 61 (20.6) .878
Current drinker No 626 (62.5) 611 (86.7) 263 (88.9)
Yes 375 (37.5) 94 (13.3) 33 (11.1) .343
Regular physical activity No 393 (39.3) 274 (38.9) 119 (40.2)
Yes 608 (60.7) 431 (61.9) 177 (59.8) .693
BMI (kg/m2) <24 524 (52.3) 362 (51.3) 162 (54.7)
≥24 477 (47.7) 343 (48.7) 134 (45.3) .328
HbA1c (%) a ≤7 323 (32.3) 218 (30.9) 105 (35.5)
>7 678 (67.7) 487 (69.1) 191 (64.5) .160
Duration of diabetes (year) <10 462 (46.2) 322 (45.7) 140 (47.3)
≥10 539 (53.8) 383 (54.3) 156 (52.7) .640
Family history of diabetes No 563 (56.2) 386 (54.8) 177 (59.8)
Yes 438 (43.8) 319 (45.2) 119 (40.2) .142
Stroke No 876 (87.5) 619 (87.8) 257 (86.8)
Yes 125 (12.5) 86 (12.2) 39 (13.2) .670
Hypertension No 363 (36.3) 252 (35.7) 111 (37.5)
Yes 638 (63.7) 453 (64.3) 185 (62.5) .600
Coronary heart disease No 786 (78.5) 547 (77.6) 239 (80.7)
Yes 215 (21.5) 158 (22.4) 57 (19.3) .267
Fatty liver No 739 (73.8) 517 (73.3) 222 (75.0)
Yes 262 (26.2) 188 (26.7) 74 (25.0) .584
Diabetic nephropathy No 493 (49.3) 347 (49.2) 146 (49.3)
Yes 508 (50.7) 358 (50.8) 150 (50.7) .976
Diabetic retinopathy No 613 (6.2) 434 (61.6) 179 (60.5)
Yes 388 (38.8) 271 (38.4) 117 (39.5) .747
Diabetic foot No 922 (92.1) 647 (91.8) 275 (92.9)
Yes 79 (7.9) 58 (8.2) 21 (7.1) .544
Open in a new tab

Abbreviations: BMI, body mass index; HbA1c, glycated hemoglobin A1c; RMB, renminbi.

a

indicated there were 45 missing values of this variable, which was filled using the random forest interpolation.

3.2. Univariate analyses of factors associated with MCI in the training set

The univariate analyses of factors associated with MCI in the training set are shown in Table 2. Age, sex, marital status, educational level, household income, location of residence, primary caregiver, current work status, current smoker, current drinker, regular physical activity, duration of diabetes, stroke, hypertension, coronary heart disease, fatty liver, diabetic nephropathy, diabetic retinopathy, and diabetic foot differed significantly between the cases and controls in the training set (p < .05).

TABLE 2.

Univariate analyses of factors associated with MCI in the training set.

Variables Category Cases (n = 191, %) Controls (n = 514, %) χ2 Value p Value
Age (year) 40–59 57 (29.8) 308 (59.9)
≥60 134 (70.2) 206 (40.1) 50.46 <.001
Sex Male 105 (55.0) 327 (63.6)
Female 86 (45.0) 187 (36.4) 4.39 .036
Marital status Married 162 (84.8) 487 (94.7)
Unmarried 29 (15.2) 27 (5.3) 18.78 <.001
Educational level Middle school or below 124 (64.9) 263 (51.2)
High school or above 67 (35.1) 251 (48.8) 10.64 .001
Household income (RMB) ≤5000 154 (80.6) 296 (57.6)
>5000 37 (19.4) 218 (42.4) 32.02 <.001
Location of residence Urban 117 (61.3) 357 (69.5)
Rural 74 (38.7) 157 (30.5) 4.25 .039
Primary caregiver Self‐care 104 (38.0) 334 (54.1)
Others 170 (62.0) 393 (12.8) 5.76 .016
Current work status Employed 71 (37.2) 243 (47.3)
Unemployed 120 (62.8) 271 (52.7) 42.53 <.001
Current smoker No 167 (87.4) 398 (77.4)
Yes 24 (12.6) 116 (22.6) 4.86 .028
Current drinker No 180 (94.2) 614 (84.5)
Yes 11 (5.8) 113 (15.5) 11.00 .016
Regular physical activity No 99 (51.8) 175 (34.0)
Yes 92 (48.2) 339 (66.0) 18.54 <.001
BMI (kg/m2) <24 100 (52.4) 262 (51.0)
≥24 91 (47.6) 252 (49.0) .11 .744
HbA1c (%) ≤7 62 (32.5) 156 (30.4)
>7 129 (67.5) 358 (69.6) .29 .590
Duration of diabetes (year) <10 67 (35.1) 255 (49.6)
≥10 124 (64.9) 259 (50.4) 11.85 <.001
Family history of diabetes No 108 (56.5) 278 (54.1)
Yes 83 (43.5) 234 (45.9) .71 .968
Stroke No 154 (80.6) 465 (90.5)
Yes 37 (19.4) 49 (9.5) 12.59 <.001
Hypertension No 50 (26.2) 202 (39.3)
Yes 141 (73.8) 312 (60.7) 10.44 .001
Coronary heart disease No 129 (67.5) 418 (81.3)
Yes 62 (32.5) 96 (18.7) 15.22 <.001
Fatty liver No 157 (82.2) 360 (70.0)
Yes 34 (17.8) 154 (30.0) 10.53 .001
Diabetic nephropathy No 67 (35.1) 280 (54.5)
Yes 124 (31.4) 234 (50.0) 20.96 <.001
Diabetic retinopathy No 97 (50.8) 337 (65.6)
Yes 94 (49.2) 177 (34.4) 12.58 <.001
Diabetic foot No 167 (87.4) 480 (93.4)
Yes 24 (12.6) 34 (6.6) 6.53 .011
Open in a new tab

Abbreviations: BMI, body mass index; HbA1c, glycated hemoglobin A1c; MCI, mild cognitive impairment; RMB, renminbi.

3.3. Multivariate analysis of factors associated with MCI in the training set

The multivariate analysis of factors associated with MCI in the training set is shown in Table 3. Age (adjusted odds ratio (aOR) = 2.75, 95% confidence interval [CI]: 1.68–4.50, p < .001), marital status (aOR = 2.59, 95% CI: 1.36–4.93, p = .004), household income (aOR = .50, 95% CI: .30–.85, p = .010), regular physical activity (aOR = .59, 95% CI: .40–.87, p = .008), diabetic nephropathy (aOR = 1.68, 95% CI: 1.11–2.56, p = .015), and diabetic retinopathy (aOR = 1.55, 95% CI: 1.02–2.35, p = .042) were predictor variables independently associated with MCI by the multivariate logistic regression analysis.

TABLE 3.

Multivariate analysis of factors associated with mild cognitive impairment (MCI) in the training set.

Variables Category B SE Adjusted OR (95% CI) p Value
Age (year) 40–59 vs. ≥60 1.01 .25 2.75 (1.68–4.50) <.001
Sex Male vs. female −.15 .22 .86 (.56–1.32) .483
Marital status Married vs. unmarried .95 .33 2.59 (1.36–4.93) .004
Educational level Middle school or below vs. high school or above .03 .24 1.03 (.65–1.64) .901
Household income (RMB) ≤5000 vs. >5000 −.69 .27 .50 (.30–.85) .010
Location of residence Urban vs. rural .27 .22 1.31 (.85–2.03) .217
Primary caregiver Self‐care vs. others −.08 .20 .92 (.62–1.38) .688
Current work status Employed vs. unemployed .39 .30 1.47 (.82–2.63) .193
Current smoker No vs. yes −.31 .30 .74 (.41–1.34) .316
Current drinker No vs. yes −.66 .39 .52 (.24–1.12) .093
Regular physical activity No vs. yes −.53 .20 .59 (.40–.87) .008
Duration of diabetes (year) ≥10 vs. <10 .14 .21 1.15 (.77–1.73) .490
Stroke No vs. yes .50 .27 1.65 (.97–2.80) .063
Hypertension No vs. yes .08 .23 1.09 (.69–1.70) .721
Fatty liver No vs. yes .34 .22 1.41 (.91–2.18) .129
Coronary heart disease No vs. yes −.38 .24 .68 (.43–1.10) .117
Diabetic nephropathy No vs. yes .52 .21 1.68 (1.11–2.56) .015
Diabetic retinopathy No vs. yes .44 .21 1.55 (1.02–2.35) .042
Diabetic foot No vs. yes .46 .32 1.58 (.84–2.96) .157
Open in a new tab

Abbreviations: CI, confidence interval; OR, odds ratio; RMB, renminbi; SE, standard error.

3.4. Development of the decision tree model

A decision tree model was developed based on the 19 significant predictors variables (including age, sex, marital status, educational level, household income, location of residence, primary caregiver, current work status, current smoker, current drinker, regular physical activity, duration of diabetes, stroke, hypertension, coronary heart disease, fatty liver, diabetic nephropathy, diabetic retinopathy, and diabetic foot) found in the univariate analyses. According to the CHAID algorithm, age, educational level, household income, regular physical activity, diabetic nephropathy, and diabetic retinopathy were included in the decision tree model. The established decision tree model is shown in Figure 1. It had 15 nodes composed of 4 layers, and age is the most significant predictor variable.

FIGURE 1.

FIGURE 1

Open in a new tab

The established decision tree model.

3.5. Validation of the decision tree model

The ROC curves for the decision tree models and multivariate logistic regression models in the training and validation sets are shown in Figure 2. The AUC value of the decision tree models was .75 (95% CI: .71–.78) and .67 (95% CI: .61–.74) in the training and validation sets, respectively, and that of the multivariate logistic regression models was 0.76 (95% CI: .72–.80) and 0.69 (95% CI: .62–.75) in the training and validation sets, respectively.

FIGURE 2.

FIGURE 2

Open in a new tab

The receiver operating characteristic curves for the decision tree and multivariate logistic regression models: a—Multivariate logistic regression model in the training set; b—decision tree model in the training set; c—multivariate logistic regression model in the validation set; d—decision tree model in the validation set; e—reference line.

4. DISCUSSION

This study developed a decision tree model to help identify MCI in patients with T2DM by comprehensively taking demographic characteristics, lifestyle factors, and T2DM‐related information into account. To the best of our knowledge, this was the first study attempting to use the CHAID decision tree analysis to identify MCI in patients with T2DM. The principle finding of this study was that compared to the multivariate logistic regression model, the established decision tree model had comparable predictive value and was well internally validated.

Previous studies that directly compared the performance of conventional logistic regression and decision tree models to solve prediction problems found comparable results. For example, Kuang et al. (2021) found that both logistic regression and decision tree models performed well at predicting the transition from MCI to Alzheimer's disease with ideal stability; Yu et al. (2024) found that the AUC value was .868 (95% CI: .821–.916) and .863 (95% CI: .814–.912) for the logistic regression and decision tree models to identify suicidal ideation in schizophrenia patients, respectively; and similar prediction accuracy in these two approaches was also observed in a study using the self‐reported clinical symptoms of dengue fever to predict potential dengue infection (Khosavanna et al., 2021). Compared to logistic regression, the main advantages of decision tree model are easy visualization and simplicity; the results are transformed to a set of decision rules that are similar to clinical reasoning; and there is an intuitive and straightforward explanation about how the decision process was made. This study added significantly to the existing body of knowledge by indicating that both logistic regression and decision tree models performed well at predicting MCI in patients with T2DM. Additionally, the decision tree model established in this study identified 6 key predictor variables with 15 nodes composed of 4 layers. It was simple and easy to understand without issues on multiple levels and nonrelevant splits. Therefore, the utilization of decision tree model to identify MCI in T2DM patients was more suggested in clinical practice considering its superior classification accuracy and simple presentation as well as interpretation of collected data.

In terms of the key predictor variables identified by the decision tree model, age, educational level, household income, and regular physical activity were well‐known factors that were frequently observed by previous work in the general population (Biessels et al., 2008; Dominguez et al., 2021; Jia et al., 2020; Zhang et al., 2019), whereas diabetic nephropathy and retinopathy were factors specifically limited to the diabetic population. As the most significant predictor variable for the identification of MCI in T2DM patients, the increased risk of MCI in those aged ≥60 observed in this study could be explained by the brain structure changes and decreased brain functioning that occur with increased age (Mankovsky et al., 2018). Educational level has been believed to be the strongest noncognitive factor affecting cognitive function (Fan et al., 2021; Pedraza et al., 2017), and the protective effects of high household income against MCI in T2DM may be attributed to higher social engagement and more social resources (Coughlin, 2020), as well as the ability to tolerate higher levels of neuropathology, which would be beneficial for the maintenance of cognitive function (Pais et al., 2021). Additionally, the role of exercise in maintaining cognitive function has been well documented (Donnelly et al., 2016; Houston et al., 2018), and this study contributed to the existing knowledge by supporting the protective effects of regular physical activity against MCI in patients with T2DM.

Diabetic nephropathy and retinopathy were the most prevalent microvascular complications of T2DM (Seewoodhary, 2021). Previous work has linked these two complications with a wide range of poor health‐related outcomes including reduced quality of life and increased risk of depression, anxiety, and bipolar disorder (Chen et al., 2023; Edalat‐Nejad et al., 2014; Khoo et al., 2019; Mahobia et al., 2021; Valluru et al., 2023). In addition to that, this study found diabetic nephropathy and retinopathy were both associated with higher risk of MCI. Potential mechanisms underlying the association between diabetic nephropathy and cognitive impairment included vascular dysfunction, lymphatic dysfunction, decreased clearance of uremic toxins, and hemodynamic changes during dialysis, which could lead to cognitive decline (Drew et al., 2019). Additionally, the impacts of diabetic retinopathy on cognitive function may be explained by the processes of shared pathways, including neuroinflammation and degeneration, vascular degeneration, and glial activation (Little et al., 2022). Therefore, the findings of this study strongly highlighted the importance of strengthening the management of diabetic complications including diabetic nephropathy and retinopathy for maintaining cognitive function of T2DM patients in clinical practice.

In conclusion, this study developed and validated a decision tree model to help identify MCI in T2DM patients in clinical practice by employing a large sample size. The established decision tree model based on age, educational level, household income, regular physical activity, diabetic nephropathy, and diabetic retinopathy performed well with comparable predictive value compared to the multivariate logistic regression model and was internally validated. However, some limitations need to be acknowledged. First, this study was hospital‐based. Therefore, whether the findings of this study can be generalized into those from community settings remains unclear. Second, though internally well validated, there was still a lack of external validation for the findings of this study. Third, this study only compared the performance of decision tree model with logistic regression, and it remains a possibility that other machine learning models, such as random forest, gradient boosting machine, and artificial neural network, could perform better than the decision tree model for the identification of MCI in T2DM patients. Therefore, future community‐based studies with comparisons between various machine learning models are warranted.

AUTHOR CONTRIBUTIONS

Rehanguli Maimaitituerxun: Methodology; software; funding acquisition; formal analysis; visualization; writing—review and editing; writing—original draft. Wenhang Chen: Methodology; software; formal analysis; visualization; writing—review and editing; writing—original draft. Jingsha Xiang: Investigation; data curation; validation; writing—review and editing. Yu Xie: Data curation; validation; investigation; writing—review and editing. Fang Xiao: Data curation; validation; writing—review and editing; investigation. Xin Yin Wu: Data curation; validation; writing—review and editing; investigation. Letao Chen: Investigation; validation; writing—review and editing; data curation. Jianzhou Yang: Validation; writing—review and editing; methodology; supervision. Aizhong Liu: Validation; writing—review and editing; supervision; methodology. Wenjie Dai: Writing—review and editing; conceptualization; project administration; resources; methodology; funding acquisition.

CONFLICT OF INTEREST STATEMENT

The authors have no conflicts of interest to declare.

PEER REVIEW

The peer review history for this article is available at https://publons.com/publon/10.1002/brb3.3456.

ACKNOWLEDGMENTS

The authors are grateful to all participants enrolled in this study.

Maimaitituerxun, R. , Chen, W. , Xiang, J. , Xie, Y. , Xiao, F. , Wu, X. Y. , Chen, L. , Yang, J. , Liu, A. , & Dai, W. (2024). Predictive model for identifying mild cognitive impairment in patients with type 2 diabetes mellitus: A CHAID decision tree analysis. Brain and Behavior, 14, e3456. 10.1002/brb3.3456

Rehanguli Maimaitituerxun and Wenhang Chen contributed equally to the work.

DATA AVAILABILITY STATEMENT

The data that support the findings of this study are available from the corresponding author, Wenjie Dai, upon reasonable request.

REFERENCES

  1. Amendolara, A. , Pfister, D. , Settelmayer, M. , Shah, M. , Wu, V. , Donnelly, S. , & Bills, K. (2023). An overview of machine learning applications in sports injury prediction. Cureus, 15(9), e46170. 10.7759/cureus.46170 [DOI] [PMC free article] [PubMed] [Google Scholar]
  2. Bae, J. M. (2014). The clinical decision analysis using decision tree. Epidemiology and Health, 36, e2014025. 10.4178/epih/e2014025 [DOI] [PMC free article] [PubMed] [Google Scholar]
  3. Biessels, G. J. , Deary, I. J. , & Ryan, C. M. (2008). Cognition and diabetes: A lifespan perspective. Lancet Neurology, 7(2), 184–190. 10.1016/s1474-4422(08)70021-8 [DOI] [PubMed] [Google Scholar]
  4. Chen, C. , Lan, Y. , Wang, Z. , Yan, W. , Yan, X. , & Han, J. (2023). Causal effects of diabetic retinopathy on depression, anxiety and bipolar disorder in the European population: A Mendelian randomization study. Journal of Endocrinological Investigation, 00, 00–00. 10.1007/s40618-023-02176-3 [DOI] [PubMed] [Google Scholar]
  5. Chinese Diabetes Society . (2021). Chinese guidelines for the prevention and treatment of type 2 diabetes (2020 edition) in Chinese. Chinese Journal of Diabetes, 13(4), 315–409. 10.3760/cma.j.cn115791-20210221-00095 [DOI] [Google Scholar]
  6. Christman, A. L. , Vannorsdall, T. D. , Pearlson, G. D. , Hill‐Briggs, F. , & Schretlen, D. J. (2010). Cranial volume, mild cognitive deficits, and functional limitations associated with diabetes in a community sample. Archives of Clinical Neuropsychology, 25(1), 49–59. 10.1093/arclin/acp091 [DOI] [PMC free article] [PubMed] [Google Scholar]
  7. Coughlin, S. S. (2020). Social determinants of colorectal cancer risk, stage, and survival: A systematic review. International Journal of Colorectal Disease, 35(6), 985–995. 10.1007/s00384-020-03585-z [DOI] [PubMed] [Google Scholar]
  8. Salgado, C. M. , Azevedo, C. , Proença, H. , & Vieira, S. M. (2016). Missing data. In Secondary analysis of electronic health records (pp. 143–162). Springer. 10.1007/978-3-319-43742-2_13 [DOI] [Google Scholar]
  9. Davis, M. , O Connell, T. , Johnson, S. , Cline, S. , Merikle, E. , Martenyi, F. , & Simpson, K. (2018). Estimating Alzheimer's disease progression rates from normal cognition through mild cognitive impairment and stages of dementia. Current Alzheimer Research, 15(8), 777–788. 10.2174/1567205015666180119092427 [DOI] [PMC free article] [PubMed] [Google Scholar]
  10. Dominguez, L. J. , Veronese, N. , Vernuccio, L. , Catanese, G. , Inzerillo, F. , Salemi, G. , & Barbagallo, M. (2021). Nutrition, physical activity, and other lifestyle factors in the prevention of cognitive decline and dementia. Nutrients, 13(11), 4080. 10.3390/nu13114080 [DOI] [PMC free article] [PubMed] [Google Scholar]
  11. Donnelly, J. E. , Hillman, C. H. , Castelli, D. , Etnier, J. L. , Lee, S. , Tomporowski, P. , & Szabo‐Reed, A. N. (2016). Physical activity, fitness, cognitive function, and academic achievement in children: A systematic review. Medicine and Science in Sports and Exercise, 48(6), 1197–1222. 10.1249/mss.0000000000000901 [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. Drew, D. A. , Weiner, D. E. , & Sarnak, M. J. (2019). Cognitive impairment in CKD: Pathophysiology, management, and prevention. American Journal of Kidney Diseases, 74(6), 782–790. 10.1053/j.ajkd.2019.05.017 [DOI] [PMC free article] [PubMed] [Google Scholar]
  13. Edalat‐Nejad, M. , Jafarian, N. , & Yousefichaijan, P. (2014). Diabetic nephropathy: A strong predictor of sleep quality in hemodialysis patients. Saudi Journal of Kidney Diseases and Transplantation, 25(4), 774–780. 10.4103/1319-2442.134994 [DOI] [PubMed] [Google Scholar]
  14. Fan, R. , Zhao, L. , Ding, B. J. , Xiao, R. , & Ma, W. W. (2021). The association of blood non‐esterified fatty acid, saturated fatty acids, and polyunsaturated fatty acids levels with mild cognitive impairment in Chinese population aged 35–64 years: A cross‐sectional study. Nutritional Neuroscience, 24(2), 148–160. 10.1080/1028415x.2019.1610606 [DOI] [PubMed] [Google Scholar]
  15. Fox‐Wasylyshyn, S. M. , & El‐Masri, M. M. (2005). Handling missing data in self‐report measures. Research in Nursing & Health, 28(6), 488–495. 10.1002/nur.20100 [DOI] [PubMed] [Google Scholar]
  16. Fu, K. , Li, Y. , Lv, H. , Wu, W. , Song, J. , & Xu, J. (2022). Development of a model predicting the outcome of in vitro fertilization cycles by a robust decision tree method. Frontiers in Endocrinology (Lausanne), 13, 877518. 10.3389/fendo.2022.877518 [DOI] [PMC free article] [PubMed] [Google Scholar]
  17. Houston, D. K. , Neiberg, R. H. , Miller, M. E. , Hill, J. O. , Jakicic, J. M. , Johnson, K. C. , & Kritchevsky, S. B. (2018). Physical function following a long‐term lifestyle intervention among middle aged and older adults with type 2 diabetes: The look AHEAD study. Journals of Gerontology. Series A, Biological Sciences and Medical Sciences, 73(11), 1552–1559. 10.1093/gerona/glx204 [DOI] [PMC free article] [PubMed] [Google Scholar]
  18. Hu, P. , Li, Y. , Liu, Y. , Guo, G. , Gao, X. , Su, Z. , & Chen, Q. (2022). Comparison of conventional logistic regression and machine learning methods for predicting delayed cerebral ischemia after aneurysmal subarachnoid hemorrhage: A multicentric observational cohort study. Frontiers in Aging Neuroscience, 14, 857521. 10.3389/fnagi.2022.857521 [DOI] [PMC free article] [PubMed] [Google Scholar]
  19. Jia, L. , Du, Y. , Chu, L. , Zhang, Z. , Li, F. , Lyu, D. , & Jia, J. (2020). Prevalence, risk factors, and management of dementia and mild cognitive impairment in adults aged 60 years or older in China: A cross‐sectional study. Lancet Public Health, 5(12), e661–e671. 10.1016/s2468-2667(20)30185-7 [DOI] [PubMed] [Google Scholar]
  20. Kasbekar, P. U. , Goel, P. , & Jadhav, S. P. (2017). A decision tree analysis of diabetic foot amputation risk in Indian patients. Frontiers in Endocrinology (Lausanne), 8, 25. 10.3389/fendo.2017.00025 [DOI] [PMC free article] [PubMed] [Google Scholar]
  21. Kass, G. V. (1980). An exploratory technique for investigating large quantities of categorical data. Journal of the Royal Statistical Society, 29(2), 119–127. [Google Scholar]
  22. Khoo, K. , Man, R. E. K. , Rees, G. , Gupta, P. , Lamoureux, E. L. , & Fenwick, E. K. (2019). The relationship between diabetic retinopathy and psychosocial functioning: A systematic review. Quality of Life Research, 28(8), 2017–2039. 10.1007/s11136-019-02165-1 [DOI] [PubMed] [Google Scholar]
  23. Khosavanna, R. R. , Kareko, B. W. , Brady, A. C. , Booty, B. L. , Nix, C. D. , Lyski, Z. L. , & Messer, W. B. (2021). Clinical symptoms of dengue infection among patients from a non‐endemic area and potential for a predictive model: A multiple logistic regression analysis and decision tree. American Journal of Tropical Medicine and Hygiene, 104(1), 121–129. 10.4269/ajtmh.20-0192 [DOI] [PMC free article] [PubMed] [Google Scholar]
  24. Kim, M. J. , & Fritschi, C. (2021). Relationships between cognitive impairment and self‐management in older adults with type 2 diabetes: An integrative review. Research in Gerontological Nursing, 14(2), 104–112. 10.3928/19404921-20201117-01 [DOI] [PubMed] [Google Scholar]
  25. Kuang, J. , Zhang, P. , Cai, T. , Zou, Z. , Li, L. , Wang, N. , & Wu, L. (2021). Prediction of transition from mild cognitive impairment to Alzheimer's disease based on a logistic regression‐artificial neural network‐decision tree model. Geriatrics & Gerontology International, 21(1), 43–47. 10.1111/ggi.14097 [DOI] [PubMed] [Google Scholar]
  26. Langenberger, B. , Schulte, T. , & Groene, O. (2023). The application of machine learning to predict high‐cost patients: A performance‐comparison of different models using healthcare claims data. PLoS ONE, 18(1), e0279540. 10.1371/journal.pone.0279540 [DOI] [PMC free article] [PubMed] [Google Scholar]
  27. Li, H. , Jia, J. , & Yang, Z. (2016). Mini‐mental state examination in elderly Chinese: A population‐based normative study. Journal of Alzheimer's Disease, 53(2), 487–496. 10.3233/jad-160119 [DOI] [PubMed] [Google Scholar]
  28. Little, K. , Llorián‐Salvador, M. , Scullion, S. , Hernández, C. , Simó‐Servat, O. , Del Marco, A. , & Stitt, A. W. (2022). Common pathways in dementia and diabetic retinopathy: Understanding the mechanisms of diabetes‐related cognitive decline. Trends in Endocrinology and Metabolism, 33(1), 50–71. 10.1016/j.tem.2021.10.008 [DOI] [PubMed] [Google Scholar]
  29. Mahobia, A. , Sahoo, S. R. , Maiti, N. , Sathyanarayanan, R. , Aravinth, R. , Pandraveti, R. R. , & Tiwari, H. (2021). Diabetic retinopathy and its effect on quality of life: An original research. Journal of Pharmacy & Bioallied Sciences, 13(Suppl 2), S1365–S1368. 10.4103/jpbs.jpbs_199_21 [DOI] [PMC free article] [PubMed] [Google Scholar]
  30. Mankovsky, B. , Zherdova, N. , van den Berg, E. , Biessels, G. J. , & de Bresser, J. (2018). Cognitive functioning and structural brain abnormalities in people with type 2 diabetes mellitus. Diabetic Medicine, 35(12), 1663–1670. 10.1111/dme.13800 [DOI] [PubMed] [Google Scholar]
  31. Mitchell, A. J. , & Shiri‐Feshki, M. (2009). Rate of progression of mild cognitive impairment to dementia–meta‐analysis of 41 robust inception cohort studies. Acta Psychiatrica Scandinavica, 119(4), 252–265. 10.1111/j.1600-0447.2008.01326.x [DOI] [PubMed] [Google Scholar]
  32. Pais, R. , Ruano, L. , Moreira, C. , Fraga, S. , Carvalho, O. P. , & Barros, H. (2021). Social support and cognitive impairment: Results from a Portuguese 4‐year prospective study. International Journal of Environmental Research and Public Health, 18(16), 8841. 10.3390/ijerph18168841 [DOI] [PMC free article] [PubMed] [Google Scholar]
  33. Pedraza, O. L. , Montes, A. M. S. , Sierra, F. A. , Montalvo, M. C. , Muñoz, Y. , Díaz, J. M. , & Piñeros, C. (2017). Mild cognitive impairment (MCI) and dementia in a sample of adults in the city of Bogotá. Dementia & Neuropsychologia, 11(3), 262–269. 10.1590/1980-57642016dn11-030008 [DOI] [PMC free article] [PubMed] [Google Scholar]
  34. Petersen, R. C. (2004). Mild cognitive impairment as a diagnostic entity. Journal of Internal Medicine, 256(3), 183–194. 10.1111/j.1365-2796.2004.01388.x [DOI] [PubMed] [Google Scholar]
  35. Petersen, R. C. , Smith, G. E. , Waring, S. C. , Ivnik, R. J. , Tangalos, E. G. , & Kokmen, E. (1999). Mild cognitive impairment: Clinical characterization and outcome. Archives of Neurology, 56(3), 303–308. 10.1001/archneur.56.3.303 [DOI] [PubMed] [Google Scholar]
  36. Plante, D. A. , Kassirer, J. P. , Zarin, D. A. , & Pauker, S. G. (1986). Clinical decision consultation service. American Journal of Medicine, 80(6), 1169–1176. 10.1016/0002-9343(86)90680-7 [DOI] [PubMed] [Google Scholar]
  37. Podgorelec, V. , Kokol, P. , Stiglic, B. , & Rozman, I. (2002). Decision trees: An overview and their use in medicine. Journal of Medical Systems, 26(5), 445–463. 10.1023/a:1016409317640 [DOI] [PubMed] [Google Scholar]
  38. Rinkel, W. D. , van der Oest, M. J. W. , & Coert, J. H. (2020). Item reduction of the 39‐item Rotterdam diabetic foot study test battery using decision tree modelling. Diabetes Metabolism Research and Reviews, 36(4), e3291. 10.1002/dmrr.3291 [DOI] [PMC free article] [PubMed] [Google Scholar]
  39. Santos, T. , Lovell, J. , Shiell, K. , Johnson, M. , & Ibrahim, J. E. (2018). The impact of cognitive impairment in dementia on self‐care domains in diabetes: A systematic search and narrative review. Diabetes Metabolism Research and Reviews, 34(6), e3013. 10.1002/dmrr.3013 [DOI] [PubMed] [Google Scholar]
  40. Seewoodhary, M. (2021). An overview of diabetic retinopathy and other ocular complications of diabetes mellitus. Nursing Standard, 36(7), 71–76. 10.7748/ns.2021.e11696 [DOI] [PubMed] [Google Scholar]
  41. Song, Y. Y. , & Lu, Y. (2015). Decision tree methods: Applications for classification and prediction. Shanghai Archives of Psychiatry, 27(2), 130–135. 10.11919/j.issn.1002-0829.215044 [DOI] [PMC free article] [PubMed] [Google Scholar]
  42. Srikanth, V. , Sinclair, A. J. , Hill‐Briggs, F. , Moran, C. , & Biessels, G. J. (2020). Type 2 diabetes and cognitive dysfunction‐towards effective management of both comorbidities. The Lancet Diabetes & Endocrinology, 8(6), 535–545. 10.1016/s2213-8587(20)30118-2 [DOI] [PubMed] [Google Scholar]
  43. Ting Sim, J. Z. , Fong, Q. W. , Huang, W. , & Tan, C. H. (2023). Machine learning in medicine: What clinicians should know. Singapore Medical Journal, 64(2), 91–97. 10.11622/smedj.2021054 [DOI] [PMC free article] [PubMed] [Google Scholar]
  44. Toyoda, H. , Terai, H. , Yamada, K. , Kato, M. , Suzuki, A. , Takahashi, S. , & Nakamura, H. (2023). A decision tree analysis to predict clinical outcome of minimally invasive lumbar decompression surgery for lumbar spinal stenosis with and without coexisting spondylolisthesis and scoliosis. The Spine Journal: Official Journal of the North American Spine Society, 23(7), 973–981. 10.1016/j.spinee.2023.01.023 [DOI] [PubMed] [Google Scholar]
  45. Umpierre, D. , Ribeiro, P. A. , Kramer, C. K. , Leitão, C. B. , Zucatti, A. T. , Azevedo, M. J. , & Schaan, B. D. (2011). Physical activity advice only or structured exercise training and association with HbA1c levels in type 2 diabetes: A systematic review and meta‐analysis. JAMA, 305(17), 1790–1799. 10.1001/jama.2011.576 [DOI] [PubMed] [Google Scholar]
  46. Valluru, G. , Costa, A. , Klawe, J. , Liu, B. , Deobhakta, A. , & Ahmad, S. (2023). Depression in individuals with diabetic retinopathy in the us national health and nutrition examination survey, 2005–2008. American Journal of Ophthalmology, 256, 63–69. 10.1016/j.ajo.2023.07.005 [DOI] [PubMed] [Google Scholar]
  47. Verma, S. , Agarwal, S. , Tashok, S. , Verma, A. , & Giri, R. (2021). Cognitive impairment in type 2 diabetes and its impact on daily living and self‐care: A case—Control study in Kanpur, North India. Diabetes & Metabolic Syndrome, 15(4), 102147. 10.1016/j.dsx.2021.05.020 [DOI] [PubMed] [Google Scholar]
  48. Wentzlof, C. A. , Kerka, J. E. , Albert, J. H. , Sprague, J. E. , & Maddox, L. O. (2019). Comparison of decision tree and logistic regression models for utilization in sexual assault kit processing. Journal of Forensic Science, 64(2), 528–533. 10.1111/1556-4029.13920 [DOI] [PubMed] [Google Scholar]
  49. World Health Organization (WHO) . (2020). WHO Guidelines on physical activity and sedentary behaviour: Web annex evidence profiles. WHO. https://www.who.int/publications/i/item/9789240015111 [Google Scholar]
  50. Yang, Q. , Li, J. , Zhang, Z. , Wu, X. , Liao, T. , Yu, S. , & Sun, J. (2021). Clinical characteristics and a decision tree model to predict death outcome in severe COVID‐19 patients. BMC Infectious Diseases [Electronic Resource], 21(1), 783. 10.1186/s12879-021-06478-w [DOI] [PMC free article] [PubMed] [Google Scholar]
  51. Yu, H. , Sun, Y. , Ren, J. , Qin, M. , Su, H. , Zhou, Y. , & Zhang, W. (2024). Factors related to suicidal ideation of schizophrenia patients in China: A study based on decision tree and logistic regression model. Psychology, Health & Medicine, 1–15. 10.1080/13548506.2023.2301225 [DOI] [PubMed] [Google Scholar]
  52. Zhang, Q. , Wu, Y. , Han, T. , & Liu, E. (2019). Changes in cognitive function and risk factors for cognitive impairment of the elderly in China: 2005–2014. International Journal of Environmental Research and Public Health, 16(16), 2847. 10.3390/ijerph16162847 [DOI] [PMC free article] [PubMed] [Google Scholar]
  53. Zhang, Q. , Zhang, Z. , Huang, X. , Zhou, C. , & Xu, J. (2022). Application of logistic regression and decision tree models in the prediction of activities of daily living in patients with stroke. Neural Plasticity, 2022, 9662630. 10.1155/2022/9662630 [DOI] [PMC free article] [PubMed] [Google Scholar]
  54. Zhang, Y. , Natale, G. , & Clouston, S. (2021). Incidence of mild cognitive impairment, conversion to probable dementia, and mortality. American Journal of Alzheimer's Disease and Other Dementias, 36, 15333175211012235. 10.1177/15333175211012235 [DOI] [PMC free article] [PubMed] [Google Scholar]
  55. Zhou, Y. Y. , Zhou, T. C. , Chen, N. , Zhou, G. Z. , Zhou, H. J. , Li, X. D. , & Yang, Y. (2022). Risk factor analysis and clinical decision tree model construction for diabetic retinopathy in Western China. World Journal of Diabetes, 13(11), 986–1000. 10.4239/wjd.v13.i11.986 [DOI] [PMC free article] [PubMed] [Google Scholar]
  56. Zhou, Z. , Chen, C. , Sun, M. , Xu, X. , Liu, Y. , Liu, Q. , & Sun, B. (2023). A decision tree model to predict liver cirrhosis in hepatocellular carcinoma patients: A retrospective study. PeerJ, 11, e15950. 10.7717/peerj.15950 [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Data Availability Statement

The data that support the findings of this study are available from the corresponding author, Wenjie Dai, upon reasonable request.

Articles from Brain and Behavior are provided here courtesy of Wiley

RESOURCES