Association Between BMI and Neurocognitive Functions Among Middle-aged Obese Adults: Preliminary Findings Using Machine-learning (ML)-based Approach

Abstract

Background

Studies suggest that obesity predisposes individuals to developing cognitive dysfunction and an increased risk of dementia, but the nature of the relationship remains largely unexplored for better prognostic predictors.

Purpose

This study, the first of its kind in Indian participants with obesity, was intended to explore the use of quantification of different neurocognitive indices with increasing body mass index (BMI) among middle-aged participants with obesity. Additionally, machine-learning models were used to analyse the predictive performance of BMI for different cognitive functions.

Methods

In the cross-sectional analytical study, a total of 137 (n = 137) participants were included. Out of the total, 107 healthy obese (BMI = 23.0–30.0 kg m⁻²; age between 36 and 55 years of both genders) were recruited from the out-patient department of the Department of Endocrinology and General Medicine, and 30 participants were recruited as the control group, between March 2023 to February 2024. The participants underwent neuropsychological assessments, including mini-mental state examination (MMSE), Montreal cognitive assessment (MoCA) and serum levels of brain-derived neurotrophic factor (BDNF).

Results

Significant (p < .05) differences were observed for neurocognitive functions for the obese group versus the control group. According to the correlation heatmaps, BMI was significantly (p < .05) negatively associated with BDNF. Multivariate linear regression analysis revealed a substantial (p < .05) decline in BDNF with a change in BMI, accenting its significant impact on cognitive ageing. Additionally, consistent decreasing trends were observed across the MoCA and MMSE, confirming the robustness of the findings across diverse analytical methodologies. Furthermore, the linear regression model and super vector machine model contributed additional evidence to the consistency of the trends in cognitive decline linked to BMI variations.

Conclusion

The preliminary results of the present study support that increased BMI is an important physiological indicator that influences neurocognition and neuroplasticity in individuals with obesity.

Keywords

Healthcare body mass index cognitive decliner brain-derived neurotrophic factor Montreal cognitive assessment mini-mental state examination machine-learning models

Research Highlights

This is India’s first preliminary report to explore the scientific evidence to quantify a decrease in neurocognitive markers with increasing body mass index (BMI) among middle-aged participants with obesity using both aspects of subjective and objective assessments of psycho-neurocognitive parameters.

Additionally, different machine-learning models were used to analyse the predictive performance of BMI for different cognitive functions. This research contributes valuable insights for scientific understanding and practical applications in the ever-evolving intersection of health and technology.

Introduction

The prevalence of overweight and obesity is increasing worldwide,¹ and the major causative factors are related to lifestyle changes occurring due to rapid socioeconomic transitions. This situation has been exaggerated in the present scenario of the COVID-19 pandemic.² Reduced physical activity at work due to mechanisation, improved motorised transport, preferences for viewing television and video games during leisure time and consumption of fast food and sweetened carbonated drinks have resulted in a positive energy balance in most Asian countries.^{1, 3–5} As youths progress through adolescence, their participation in physical activity declines markedly.^3–7 The immediate risks of obesity to physical health are well established.⁷ However, available evidence^8–13 has shown that obesity predisposes individuals to developing cognitive dysfunction and an increased risk of dementia, but the nature of the relationship remains largely unexplored for better prognostic predictors.^14–19 However, assessing how rapidly cognitive decline progresses has important implications in clinical practice since the rate of disease progression may be the most important factor in determining prognosis.¹⁹ A biological marker for rapid cognitive decline would be valuable because at-risk patients could be targeted early for pharmacological, medical and psychosocial interventions designed to slow deterioration. Brain-derived neurotrophic factor (BDNF) is a member of the neurotrophins family of small secreted proteins that play major roles in nervous system development and function and has been studied as a biological marker for cognitive decline.^{14,19,20–24}

Currently, researchers are using different machine-learning models (MLMs) to explore cognitive performances.²⁵ To the best of our knowledge, only a few studies have reported the effect of BMI on changes in cognitive functions, and concomitant results have been obtained. These inconsistencies in their study findings may be due to the statistical methods used, methodological paradigms, population sampling methods and/or the setup of machine-learning (ML)-based algorithms, which have underestimated the potential efficacy of these methods in cognitive science. Therefore, the present study aimed to quantify the changes in specific psycho-neurocognitive functions (as cognitive ageing markers), including serum levels of BDNF, as well as questionnaires Montreal cognitive assessment (MoCA) and mini-mental state examination (MMSE) scores with increasing BMI among middle-aged participants with obesity. To achieve these goals, the authors have comprehensively explored the association between BMI and neurocognitive measures to increase clinical utility for future diagnostic and prognostic purposes. Therefore, the present study, part of a larger study, is intended to explore the association of quantification of different neurocognitive indices with increasing body mass index (BMI) in middle-aged participants with obesity without any comorbidities. Additionally, different MLM were used to analyse the predictive performance of BMI for different cognitive functions. This research might contribute valuable insights for scientific understanding and practical applications in the ever-evolving intersection of health and technology using conventional as well as model-based algorithms methods.

Methods

Study Population

The present study was a cross-sectional analytical research study conducted in the Department of Physiology, Psychiatry along with Endocrinology & Department of Medicine, All India Institute of Medical Sciences (AIIMS), Bathinda, India, from March 2023 to February 2024. After obtaining approval from the scientific advisory committee, the clinical trial registry (CTRI/2022/10/046935) and the Institutional Ethics Committee (IEC/AIIMS/BTI/370), a total of 137 middle-aged participants aged between 36 and 55 years of both genders were enrolled. Out of the total, 107 participants were overweight/obese (BMI values between 23.0 and 30.0 kg m⁻²; as per ICMR, 2014)²⁶; and 30 were control participants (BMI of 18.5–22.9 kg m⁻²; as per ICMR, 2014).²⁶ The aim of including a control group was only to compare the natural variation in the cognitive performances of normal BMI participants versus higher BMI participants and was therefore not age- and sex-matched. Patients with known cases of diabetes or hypertension, neurological disorders, other endocrinology disorders, taking drugs for any chronic illness, history of uncontrolled smoking, alcoholism or who were currently engaged in any weight-loss programme were excluded from the study. Written informed consent was obtained from all the participants before they participated in the study.

Anthropometric Measurements

The height was recorded while the participants stood and had bare feet using a wall-mounted stadiometer (Holtain Ltd, UK). Body weight was measured using a digital weighing balance (Seca, Germany) to the nearest 0.1 kg. Waist circumference (WC) was measured at the widest diameter between the xiphoid process of the sternum and the iliac crest. Obesity indices, including BMI,^{26, 27} were calculated as weight (kilograms) divided by height (metres) squared and kg m⁻².

Cognitive Assessment Tools

All participants in the present study underwent cognitive assessment using the English and Hindi versions of the MMSE and the MoCA questionnaires. Both tools are valid and reliable among Indians by considering cultural and linguistic differences. Under the supervision of an expert clinician in the Psychiatry Department, administered the MMSE and MoCA within 5–10 min and 10–15 min, respectively.

MMSE: The MMSE is a 30-point questionnaire used extensively in clinical and research settings to measure cognitive impairment, including simple tasks in a number of areas: the test of time and place, the repeating lists of words, arithmetic such as serial subtractions of seven, language use and comprehension and basic motor skills. The total score was 30; this score provides a picture of an individual’s cognitive performance, and a score of <24 indicated cognitive impairment.²⁸

MoCA: The MoCA is another 30-point test covering eight cognitive domains for a rapid screening instrument for mild cognitive dysfunction that includes attention and concentration, executive functions, memory, language, visuoconstructional skills, conceptual thinking, calculations and orientation. MoCA score categories were defined as severe deficit (<10), moderate deficit (10–17), mild deficit (18–25) and normal (>26).²⁹

Serum levels of BDNF: An assessment tool to capture the complex nature of cognitive health measuring the serum levels of BDNF, a molecular marker that is essential for neurotrophic support and synaptic plasticity, is important. Estimating BDNF levels provides researchers with a better understanding of the neurobiological factors affecting cognitive function by illuminating the molecular details underlying cognitive processes. The concentration of serum BDNF levels was estimated using a commercially available sandwich enzyme-linked immunosorbent assay kit (Immunotag, Geno Technology, Inc, USA). Fasting venous blood samples (5 ml) were collected, allowed to clot for 10 min at room temperature, and centrifuged at 2,000–3,000 rpm for 10 min. The serum was stored at −80 °C until analysis. The detection limit for BDNF levels ranged between 31.25 and 2,000 pg ml⁻¹. Quality control assays for biomarkers and validation were performed. Fasting blood glucose levels were estimated using an enzymatic method using commercially available kits (Randox Laboratories Ltd, UK). All the assessments were carried out in the morning hours between 8:00 and 10:00 am.^{19, 21}

MLM-based Algorithms

Applying MLMs²⁵ involves utilising advanced algorithms to analyse data in studying BMI and cognitive assessments. Specifically, ML algorithms, including linear regression (LR), support vector machine (SVM) and random forest (RF) and XGBoost (XGB), are harnessed to reveal intricate patterns and forecast cognitive trajectories. Integrating conventional testing methods with advanced MLM-based analysis aims to comprehensively understand the relationship between BMI and cognitive function. This synergistic approach allows for developing ML-based algorithm-assisted applications in healthcare domains.

LR: LR^{30, 31} is a statistical method used to model the relationship between two variables. In this model, the algorithm assumes a linear relationship between the predictor variable (BMI) and the response variable (cognitive functions), allowing researchers to quantify and predict the impact of BMI on cognitive performance.

SVM/networks: This model is associated with learning algorithms that analyse data for classification and regression analysis for the predictor variable (BMI) and the response variable (cognitive functions). In this scenario, SVM might segment the study population based on various factors related to body mass (BMI) and cognitive function. It can help identify critical thresholds or conditions under which BMI significantly impacts cognitive abilities.³²

RF: RF^{33, 34} is a ML algorithm combining multiple decision trees’ output to reach a single result. In the context of BMI and cognitive function, the RF model can capture a broader range of patterns and relationships by combining the strengths of individual decision trees. This approach helps mitigate overfitting and provides a more robust analysis by considering diverse perspectives within the data. The RF model enhances the model’s predictive power and generalisability.

XGBoosting: XGB, or eXtreme Gradient Boosting,^{35, 36} is a ML algorithm that sequentially builds decision trees, correcting errors made by preceding trees and enhancing overall predictive accuracy. Its gradient-boosting framework optimises the model by minimising errors, contributing to a highly accurate predictive tool. With regularisation techniques, XGB prevents overfitting, ensuring robust generalisation to new data. Additionally, its ability to handle missing data enhances flexibility. When applied to the BMI-cognitive function context, XGB excels in predicting cognitive outcomes, capturing complex relationships and nonlinear patterns within the data and offering a nuanced understanding of how BMI indicators influence various cognitive abilities.

Furthermore, the different MLMs, including LR, SVM, RF and XGB, were used to analyse the predictive performance of BMI for different cognitive functions for scientific understanding and practical applications in the ever-evolving intersection of health and technology. Using single-time point research data of the present study, the authors have used error metrics to assess the model’s accuracy and stability. This provides valuable information on how well the model performs with the available data and can guide us in improving or choosing better models. The error metrics will help us to understand how accurate our model’s prediction is compared to the actual data. The overall decline measures the average difference between predicted and actual values; it indicates how reliable the model is for predicting future outcomes. On the other hand, the percentage decline measures the proportion of decline, making us understand the magnitude of the change in a relative sense; it also gives insight into the model’s performance and stability. A negative rate indicates that the predictions are lower than the actual values, while a positive rate means the predictions are higher.

Statistical Analysis

The data were analysed using SPSS software Version 20.0 (IBM, Chicago, IL, USA, October 2020) for conventional data analysis and Python programming (version Jupyter Notebook 6.5.4) software for ML-based algorithms. The normality of the data was analysed using the Kolmogorov–Smirnov test. The data is presented as the mean ± standard deviation and interquartile range for parametric and non-parametric variables, respectively. The mean differences between different sociodemographic and neuropsychological parameters were measured using an ‘Independent two t-test with equal variance’ for parametric variables at a confidence interval (CI) of 95% and a ‘Mann–Whitney test’ for non-parametric variables and also for sub-domain analysis for different psychological parameters. An ‘Independent sample t-test’ was used to determine gender differences in neuropsychological variables for the obese group. The ‘Multivariate Linear regression’ analysis with an adjusted model (β-coefficient) with all co-founders, including age, gender and education among the obese group, to explore the quantitative association between BMI and BDNF, MoCA and MMSE. The correlation heat maps between BMI and various neurocognitive parameters were assessed using Pearson’s correlation analysis. Different MLMs, including LR, SVM, RF and XGB were used to analyse the predictive performance of BMI for different cognitive functions for scientific understanding and practical applications in the ever-evolving intersection of health and technology. The p-value < .05 was considered to indicate statistical significance.

Results

Associations Between BMI and Different Neurocognitive Functions

A total of 107 participants with obesity (F = 54; M = 53), age 44.47 ± 7.44 years, and 30 control participants, age 43.96 ± 6.69, were included in the analysis. The flowchart of the sample selection process in the study is mentioned in Figure 1. Using Independent two-t test with equal variance, the significant mean differences for weight (p = .002), BMI (p = .001), WC (p = .01), fasting blood sugar levels (p = .02), systolic blood pressure (p = .04), MoCA scores (p = .03) and MMSE scores (p = .01) were observed for the obese group versus control group. On the other hand, using the Mann–Whitney test, the mean difference for the serum BDNF levels was significantly (z = −4.19; p = .001) reduced for the obese group compared with the control group (Table 1). The present study did not find significant results for BMI and different neurocognitive assessments for males and females, as shown in Table 2. The cognitive sub-domain analysis for MoCA scores and MMSE scores for obese and control groups was carried out, and it has been observed that the recall domain (p = .04) of MoCA and orientation domain (p = .01) for MMSE were found to decrease significantly for the obese group when compared to the control group as shown in Table 3.

Figure 1.

Flowchart of the Sample Selection Process in this Study.

Table 1.

Sociodemographic Parameters and the Mean Differences for Neuropsychological Parameters for Obese Group Versus Control Group.

Sociodemographic Parameters	Obese Group	Control Group	Independent Two-sample T-test with Equal Variance
Sociodemographic Parameters	Obese Group	Control Group	Mean Difference, CI 95%	p-Value
Number of subjects (n)	107	30	–	–
Gender, n (%)
Female	54 (50.46)	14 (46.66)	–	–
Male	53 (49.54)	16 (53.34)
Age (years)	44.47 ± 7.44	43.96 ± 6.69	0.51 (0.43, 0.67)	.73
Married, n (%)
Yes	78.0 (72.85)	22.0 (73.33)	–	–
No	26.0 (24.28)	7.0 (23.33)
Divorced	3.0 (2.87)	1.0 (3.34)
Education, n (%)
Higher education	36 (51.4)	19 (63.33)
Secondary education	15 (21.4)	08 (26.66)	–	–
Primary education	19 (27.1)	03 (10.01)
Weight (kg)	80.12 ± 7.59	62.16 ± 5.33	17.96 (16.59, 18.91)	.002
Body mass index (kg m⁻²)	26.54 ± 2.12	22.68±1.57	3.86 (3.74, 5.35)	.001
Waist circumference (WC, cm)	95.60 ± 6.47	82.51 ± 6.47	13.09 (12.73, 14.71)	.01
Fasting blood sugar (mg dl⁻¹)	95.5 ± 7.12	87.71 ± 7.12	7.79 (6.08, 8.71)	.02
Systolic blood pressure (SBP, mmHg)	127.63 ± 10.45	118.53 ± 7.89	9.1 (7.43,10.51)	.04
Diastolic blood pressure (DBP, mmHg)	84.57 ± 9.62	81.92 ± 7.28	2.65 (1.78, 3.98)	.07
Neuropsychological Parameters
Serum brain-derived neurotrophic  factor (BDNF) levels (pg ml⁻¹)	196.72 (34.93–905.75)	502.76 (103.59–962.09)	z = −4.195	.001*
Montreal cognitive assessment (MoCA) scores	22.70 ± 2.46	28.23 ± 5.73	−5.53 (−6.87, −4.18)	.03
Mini-mental state examination (MMSE) scores	23.46 ± 1.46	27.56 ± 1.25	−4.1 (−2.55, −4.97)	.01

Notes: The values are in mean ± SD for parametric variables and median (interquartile range) for non-parametric variables. p-Value for between groups (obese versus control) change obtained by independent t-test with equal variance for parametric variables at confidence interval (CI) at 95%. The remaining variables are presented in frequency (%). Significant at p < .05. *p-Value for between groups (obese versus control) change obtained by Mann–Whitney test for non-parametric variables.

Table 2.

Body Mass Index (BMI) and Different Neurocognitive Assessments for Males and Females of Obese Group.

Variables	Male Obese Group	Female Obese Group	p-Value
Total number of participants (n)	53	54	–
Age (years)	44.07 ± 8.05	44.87 ± 6.84	.58
Body mass index (kg m⁻²) (mean ± SD)	26.34 ± 1.94	26.73 ± 2.28	.35
Serum brain-derived neurotrophic factor (BDNF) levels (pg ml⁻¹) (IQR)	277.93 ± 18.66	293.76 ± 34.80	.71
Mini-mental state examination (MMSE) scores (mean ± SD)	24.24 ± 1.60	24.68 ± 1.28	.12
Montreal cognitive assessment (MoCA) scores (mean ± SD)	22.58 ± 1.94	22.81 ± 2.43	.59

Notes: All values are in mean ± SD; p-value between gender differences obtained by independent sample t-test; significance at p < .05.

Table 3.

Cognitive Sub-domain Analysis for MoCA Scores and MMSE Scores for Obese and Control Groups.

Domains	Items/Maximum Scores	MoCA (Mean ± SD)		p-Value
Domains	Items/Maximum Scores	Obese Group (n = 107)	Control Group (n = 30)	p-Value
Orientation	Orientation to time and place/6	6.0 (4.0–6.0)	6.0 (5.0–6.0)	.33
Executive function	Trail-making test/1	1.0 (0.00–1.0)	1.0 (0.00–1.00)	.14
	Digit span test/2	2.0 (0.00–2.00)	2.0 (1.00–2.00)	.18
	Verbal fluency/1	1.0 (0.00–1.0)	1.0 (0.00–1.0)	.10
	Abstraction test/2	2.0 (0.00–2.00)	2.0 (0.00–2.00)	.83
Calculation	Serial 7 subtractions/3	3.0 (0.00–3.00)	3.0 (1.00–3.00)	.79
Naming	Naming (lion, giraffe and camel)/3	3.0 (1.00–3.00)	3.00 (2.00–3.00)	.74
Repetition	2 longer sentences/2	2.0 (0.00–2.00)	2.0 (1.00–2.00)	.32
Visuoconstructional skills	Copy cube/1	0.00 (0.00–1.00)	1.00 (0.00–1.00)	.07
Visuoconstructional skills	Draw clock face/3	2.00 (1.00–3.00)	3.00 (1.00–3.00)	.07
Attention	Vigilance test for number ‘1’/1	1.00 (0.00–1.00)	1.00 (0.00–1.00)	.07
Recall	Recall 5 words/5	2.00 (0.00–5.00)	4.00 (2.00–5.00)	.04
Domains	Items/Maximum Scores	MMSE (Mean ± SD)		p-Value
Domains	Items/Maximum Scores	Obese Group (n = 107)	Control Group (n = 30)	p-Value
Orientation	Orientation to time and place/10	8.00 (7.00–10.00)	10.00 (9.00–10.00)	.01
Executive function	Repeat 3 words/3	3.00 (1.00–3.00)	3.00 (2.00–3.00)	.47
Executive function	Reading command test/1	1.00 (0.00–1.00)	1.00 (0.00–1.00)	.76
Calculation	Serial 7 subtraction/5	4.00 (0.00–5.00)	5.00 (3.00–5.00)	.55
Naming	Naming (pencil and watch)/2	2.00 (0.00–2.00)	2.00 (1.00–2.00)	.06
Registration	Repeat 3 words/3	3.00 (1.00–3.00)	3.00 (2.00–3.00)	.47
Recall	Recall 3 words/3	2.00 (0.00–3.00)	3.00 (1.00–3.00)	.01
Visuoconstructional skills	Copy intersecting pentagons/1	1.00 (0.00–1.00)	1.00 (1.00–1.00)	.41
Writing	Write a sentence/1	1.00 (0.00–1.00)	1.00 (1.00–1.00)	.08
Repetition	1 short sentence/1	1.00 (0.00–1.00)	1.00 (1.00–1.00)	.08

Notes: All values are in the median (interquartile range); Mann–Whitney test between the obese and control groups. The differences were considered statistically significant for p < .05. Bold values are statistically significant.

MMSE: Mini-Mental State Examination, SD: Standard Deviation.Using a multimodal approach, notable significant differences were observed in the correlation between BMI and different neurocognitive parameters among middle-aged participants with obesity. Using Pearson’s correlation heatmaps, a significant negative correlation was observed between the BMI of overweight/obese participants and the serum BDNF concentration (r = −0.27; p = .024) (Figure 2). However, the MMSE and MoCA scores were negatively correlated with BMI, but the differences were not statistically significant (r = −0.083; p = .49; r = −0.099; p = .41, respectively) (Figure 1). For sub-domain analysis for MoCA and MMSE, it has been observed that the recall domain (r = −0.189; p = .04) of MoCA and serial subtraction (calculation) (r = −0.12; p = .013) domain of MMSE were negatively associated with increasing BMI (Figures 3 and 4). Furthermore, to determine the strength of the association between BMI and BDNF levels, the multivariate LR-adjusted and unadjusted models were used for all potential confounders, including age, gender and education. A significant negative association with the β-coefficient of −20.50, with CI at 95% is −40.49, −0.52; p = .04, between BDNF levels and BMI, has been observed. Also, a significant negative association with the β-coefficient of −69.05, with CI at 95% is −126.41, −11.70; p = .01, between BDNF levels and BMI, has been observed. These negative associations signify that BDNF levels tend to decrease with a magnitude of −20.54 as BMI increases with each unit and vice versa, also affecting the education strata, as shown in Table 4.

Figure 2.

Notes: BMI – body mass index; BDNF – brain-derived neurotrophic factor; MoCA – Montreal cognitive assessment; MMSE – mini-mental state examination.

Figure 3.

Notes: BMI – body mass index; MoCA – Montreal cognitive assessment.

Figure 4.

Notes: BMI – body mass index; MMSE – mini-mental state examination.

Table 4.

Association Between Body Mass Index (BMI) with Different Neurocognitive Parameters Using Multivariable Linear Regression Among Obese Group.

Variables	Adjusted Model			Unadjusted Model
Variables	β¹-coefficient	p-Value	95% Confidence Interval (CI)	β²-coefficient	p-Value	95% Confidence Interval (CI)
BDNF levels
Body mass index (BMI)	−20.50	.04	−40.49 −0.52	−21.47	.03	−41.65 −1.28
Age	1.28	.65	−4.41 6.98	2.11	.47	−3.74 7.98
Gender	−18.28	.66	−102.77 66.20	−15.83	.71	−102.84 71.17
Education	−69.05	.01	−126.41 11.70	−73.95	.01	−131.28 16.62
Constant	898.65	.04	274.98 1522.32	855.84	.00	318.36 1393.32
MMSE scores
Body mass index (BMI)	−0.05	.40	−0.18 0.07	−0.04	.49	−0.17 0.08
Age	−0.02	.22	−0.06 0.01	−0.02	.27	−0.05 0.01
Gender	−0.47	.09	−1.03 0.08	−0.43	.12	−0.99 0.11
Education	−0.08	.65	−0.46 0.29	−0.08	.65	−0.46 0.29
Constant	27.85	.00	23.70 32.01	25.70	.00	22.15 29.24
MoCA scores
Body mass index (BMI)	−0.09	.37	−0.29 0.11	−0.09	.36	−0.29 0.10
Age	0.03	.19	−0.02 0.09	0.04	.16	−0.01 0.09
Gender	−0.22	.60	−1.07 0.62	−0.22	.59	−1.07 0.61
Education	−0.15	.58	−0.73 0.41	−0.21	.45	−0.78 0.35
Constant	23.99	.00	17.70 30.28	25.11	.00	19.79 30.42

Notes: Multiple linear regression analysis with adjusted model with all co-founders age, gender and education; linear regression analysis with an unadjusted model. In models, exposure is body mass index (BMI), and outcomes are BDNF, MoCA and MMSE; co-founders are age, gender and education. β¹ = beta-coefficient for adjusted model. β² = beta-coefficient for unadjusted model. p < .05 is significant at CI = 95%. Bold values are statistically significant.

Model Performance Comparison

Furthermore, the associations between BMI and different neurocognitive parameter datasets were analysed by using innovative ML-based algorithms, including LR, super vector machine, RF and XGB methodologies, to quantify the interpretations behind the associations between cognitive decline and increasing BMI (Figure 5). For this purpose, the authors divided the datasets into two categories, the training dataset and the testing dataset, to extract algorithm-based insights into the associations between BMI and BDNF and between the MoCA and MMSE scores. The training dataset comprises 80% of the data, while the remaining 20% is allocated to the testing dataset. The root mean square error (RMSE) was calculated to assess the efficiency of the MLMs considered in this study (Figure 6). From the present datasets, LR and super vector machine models were shown to have lower RMSEs than did the RF and XGB models; hence, large datasets are needed to validate these models.

Figure 5.

Notes: BMI – body mass index; BDNF – brain-derived neurotrophic factor; MoCA – Montreal cognitive assessment; MMSE – mini-mental state examination.

Figure 6.

Notes: BMI – body mass index; BDNF – brain-derived neurotrophic factor; MoCA – Montreal cognitive assessment; MMSE – mini-mental state examination.

On the other hand, LR and SVM analysis revealed a particularly noteworthy finding, revealing a substantial (−) 46.81 (−16.27%) and (−) 21.59 pg ml⁻¹ (−7.50%) increase in the incidence of cognitive decline in association with BMI and BDNF, respectively (Figure 5). Moreover, the investigations extended to the impact of BMI on the MOCA and MMSE scores, revealing a more moderate cognitive decline of −0.20 (−0.86%) and −0.414 (−1.62%), respectively (Figure 5). Additionally, the outcomes obtained from the super vector machine, RF and XGB models revealed consistent patterns of cognitive decline with increasing BMI. These results reinforce the initial LR and super vector machine findings and underscore the robustness and coherence of the observed trends in cognitive decline, reinforcing the study’s implications for understanding the impact of BMI variations on cognitive health across diverse analytical approaches and methodologies. The present study is ongoing to obtain more datasets that can be generated to increase the MLM’s performance for better application development for the healthcare industry to assess the changes in the cognitive functions in obesity.

Discussion

In the present study, the authors aimed to quantify the association between BMI and neurocognitive markers, including the serum BDNF concentration and MMSE and MOCA scores, employing ML-based models in obesity. The present study found significantly decreased cognitive task performances with increasing BMI in obesity without comorbidities. These decreases in BDNF levels may be due to low-grade inflammation, increased oxidative stress, lifestyle factors and mutations in the BDNF gene or tyrosine kinase (Trk)-B receptor in individuals with obesity.^{14, 37, 38} Our study findings contrast with those of Monteleone et al,³⁷ who recorded increased serum BDNF levels with increasing BMI and weight. This inconsistency in the research findings may be due to differences in ethnicity, population age group, study design, methodology and sample size. Thus, to generalise these findings to the entire population with obesity, we validated our findings with present datasets generated using ML-based algorithms. Furthermore, we found greater BDNF levels in females than males, but we did not find any significant difference between them. Our study findings are supported by Marcos-Pasero et al,³⁸ who did not find a significant correlation between sex differences and cognitive functions. Lower education strata for obese group participants also exhibited more cognitive decline, possibly due to limited resources such as access to education that has affected healthcare disparities, lifestyle factors and social interactions. These findings are in line with the previous study, which showed that cognition is affected by sociodemographic parameters.²⁶

The present study observed decreased MoCA and MMSE scores with mild deficiency and a significant negative association between MoCA and MMSE scores with an increasing BMI, which was positively supported by decreased BDNF levels in obesity. This decrease in cognitive performance with a decrease in the serum BDNF concentration indicates the impairment of recall memory, serial subtraction (calculation) and orientation to time and place in an individual, thus revealing diminished cognitive processing. Consistent decreasing trends were observed across the MOCA and MMSE scores, confirming the robust findings highlighting the impact of BMI on cognitive health. This decline in cognitive functions among the middle-aged obese group may be due to decreased physical activity; psychological dysfunction with negative mood; consumption of a high-fat diet that may modulate the brain ageing, zinc finger proteins downregulation that further affects the expressions of the BDNF genes or its Trk-B receptor; and limited resources to access the education that may affect the health domains.^39–41 The present research findings concur with previous studies showing decreased neural circuits for sensory and motor activity using ERPs-P300 and decreased BDNF levels along with various neurocognitive assessments, including MoCA and MMSE.^42–46 Moreover, peripheral BDNF levels could parallel central BDNF levels, as BDNF has both neural and nonneural origins and can aid in developing neural circuits in the brain and cognitive function via neuroplasticity after crossing the blood–brain barrier. These neural circuits help in basic learning, memory and other cognitive functions. Our findings concur with those of Marqués-Iturria et al,⁴⁷ who found a correlation between lower BDNF levels and obesity, resulting in lower cortical thickness, associated with more cognitive performance task errors.

Importantly, for the first time, India has proposed a potential application for quantifying cognitive decline using the model’s algorithms. Understanding how BMI affects cognitive health can contribute to diagnostic and prognostic purposes in clinical settings.^{48, 49} Recently, Du et al,⁴⁹ have tried to develop an obesity risk prediction system for better health management plans and achieving obesity management with few shortcomings. The present study also shows the trend to bridge the gap between BMI and cognitive decline using MLM-based algorithms, offering valuable insights for scientific understanding and practical applications. The LR and super vector models obtain superior results for the present research study compared to other techniques. Body size (BMI) is linked to cognitive decline, and various methods, including traditional and algorithm models, have been used to gain deeper insights.⁴⁹ The strengths of this study lie in its comprehensive approach, diverse use of artificial intelligence-based algorithms, conventional analysis and potential for cognitive problem detection.

Although conducted with a relatively sound methodology, this study has a few limitations. First, due to the small datasets, the findings of this study with learning algorithms need to be corroborated with larger datasets because the authors presented preliminary datasets with the algorithms-based approach of an ongoing research study. Second, the present study did not analyse the possible effects of different menstrual cycle phases on females’ cognitive functions. Third, diet assessment could not be carried out for obese patients to provide a better explanation of cognitive decline. Despite these limitations, the present findings provide scientific evidence with algorithm-based solutions for overweight/obese participants to be aware of their cognitive status, which will guide them to undergo lifestyle modification programmes. Nevertheless, the present study findings provide valuable insights into the intricate relationship between cognitive functions and obesity within the Indian context. Furthermore, the algorithms-based sub-domain cognitive domain analysis for overweight and obese individuals is not shown in the present study and will be covered in a future research paper.

Conclusion

The present findings scientifically affirm the quantitative decline in cognitive functions in terms of recall memory domain, serial subtraction domain and orientation to time and place in obesity, affecting their quality of life in day-to-day activities. Therefore, these patients should undergo lifestyle modification programmes to improve cognitive performance. Furthermore, the use of ML-based algorithms helps to provide feasibility to quantify the cognitive decline in obese participants and are used to predict obesity levels and associated cognitive decline. These model analyses also help assess the risk of cognitive decline in obese patients by identifying key risk factors and predictive biomarkers.

Footnotes

Acknowledgement

The authors thank the participants for cooperating to make this research possible. Thanks to the committed research team, collaborators and institutions (All India Institute of Medical Sciences, Bathinda, Punjab and All India Institute of Medical Sciences, New Delhi) for their invaluable support.

Author Contributions

Dipti Magan was involved in the study conceptualisation, study design, acquiring and interpreting the data, ML-based datasets analysis, data interpretation, manuscript writing and manuscript review. Raj Kumar Yadav was involved in datasets interpretation and manuscript review. Jitender Aneja was involved in psychological questionnaire analysis and data interpretation. Shivam Pandey was involved in data analysis and data interpretation.

Clinical Trial Registry

The study was registered at the Clinical Trial Registry India (CTRI), CTRI/2022/10/046935.

Copyright

The copyright (registration number: L-147090/2024) for ML-based algorithms for the present research data has been approved by the Government of India.

Data Availability Statement

The data underlying this study will be made available upon reasonable request.

Ethical Approval

All study procedures and protocols involving human participants were approved by the Institutional Ethics Committee (IEC/AIIMS/BTI/205) and adhere to the tenets of the Declaration of Helsinki.

Patient Consent

Informed consent was obtained from all participants included in the study.

Declaration of Conflicting Interests

The authors declare no potential conflicts of interest concerning the research, authorship and/or publication of this article.

Funding

The authors received no financial support for the research, authorship and/or publication of this article.

ORCID iD

Dipti Magan

References

World Health Organization. Obesity and overweight [Internet], https://www.who.int/news-room/fact-sheets/detail/obesity-and-overweight (2020, accessed 31 March 2021).

Karlsson

, Gatz

, Arpawong

, . The dynamic association between body mass index and cognition from midlife through late-life, and the effect of sex and genetic influences. Sci Rep 2021; 11(1): 7206.

, Xu

and Shi

Association between obesity, physical activity, and cognitive decline in Chinese middle and old-aged adults: a mediation analysis. BMC Geriatr 2024; 24(1): 54.

Parizkova

, Chin

, Chia

, . An international perspective on obesity, health and physical activity: current trends and challenges in China and Asia. J Exerc Sci Fit 2007; 5: 7–23.

Cheng

TO.

Fast food and obesity in China. J Am Coll Cardiol 2003; 42: 773.

Kimm

, Glynn

, Kriska

, . The decline in physical activity in Black girls and White girls during adolescence. N Engl J Med 2002; 347: 709–715.

Nilsson

LGR

, and Nilsson

Overweight and cognition. Scand J Psychol 2009; 50: 660–667.

Caspersen

, Pereira

, and Curran

KM.

Changes in physical activity patterns in the United States, by sex and cross-sectional age. Med Sci Sports Exerc 2000; 32: 1601–1609.

, Dai

, Jackson

, . Overweight is associated with decreased cognitive functioning among school-age children and adolescents. Obesity 2008; 16(8): 1809–1815.

10.

Agarwal

, Bhalla

, Kaur

, . Effect of body mass index on physical self-concept, cognition & academic performance of first-year medical students. Indian J Med Res 2013; 138: 515–522.

11.

Lance

, Richard

, and Victor

MH.

P300 and the stroop effect in overweight minority adolescents. Neuropsychobiology 2010; 61: 180–187.

12.

Kolker

IA.

Using the P300 potential for the analysis of cognitive functions in children. Neurophysiology 2002; 34(4): 307–312.

13.

Cournot

, Marquié

, Ansiau

, . Relation between body mass index and cognitive function in healthy middle-aged men and women. Neurology 2016; 67(7): 1208–1214.

14.

Katuri

, Gaur

, Sahoo

, . Association of circulating brain-derivedneurotrophic factor with cognition among adult obese population. JObes Metab Syndr 2021; 30(2): 163–172.

15.

Elias

, Elias

, Sullivan

, . Lower cognitive function in the presence of obesity and hypertension: the Framingham Heart Study. Int J Obes Relat Metab Disord 2003; 27(2): 260–268.

16.

Steenbergen

and Colzato

LS.

Overweight and cognitive performance: high body mass index is associated with impairment in reactive control during task switching. Front Nutr 2017; 4: 1–11.

17.

Best

JB.

Cognitive psychology 5th ed. New York City, United States:

Wiley

1999 pp. 15–17.

18.

Kraemer

, Tinklenberg

and Yesavage

JA.

, and ‘How far’

vs.

‘how fast’ in Alzheimer’s disease. The question revisited. Arch Neurol 1994; 51: 275–279.

19.

Sandrini

, Di Minno

, Amadio

, . Association between obesity and circulating brain-derived neurotrophic factor (BDNF) levels: systematic review of literature and meta-analysis. Int J Mol Sci 2018; 19: 2281.

20.

Kutas

, MacCarthy

and Donchin

Augmenting mental chronometry: the P300 as a measure of stimulus evaluation. Science 1977; 197: 792–795.

21.

Christoph

, Konstantinos

, Nadine

, . Higher BDNF serum levels predict slower cognitive decline in Alzheimer’s disease patients. Int J Neuropsychopharmacol 2011; 14: 399–404.

22.

Pirjo

, Maria

, Tuomo

, . BDNF is a novel marker of cognitive function in aging women: the DR’s EXTRA study. Neurobiol Learn Mem 2008; 90: 596–603.

23.

Makhout

, Vermeiren

, Van

DMK

, . The role of brain-derived neurotrophic factor in obstructive sleep apnea and endothelial function in a pediatric population with obesity. Front Med (Lausanne) 2022; 8: 835515.

24.

Pandit

, Behl

, Sachdeva

, . Role of brain-derived neurotrophic factor in obesity. Obes Med 2020; 17: 100189.

25.

, Shen

and Xiao

Applications of artificial intelligence to obesity research: scoping review of methodologies. J Med Internet Res 2022; 24(12): e40589.

26.

Kaur

, Ranjan

, Kaloiya

, . Body mass index and cognitive functioning decline: exploring the relationship. J Educ Health Promot 2024; 13: 197.

27.

Misra

, Chowbey

, Makkar

, . Consensus statement for diagnosis of obesity, abdominal obesity and the metabolic syndrome for Asian Indians and recommendations for physical activity, medical and surgical management. J Assoc Physicians India 2009; 57: 163–170.

28.

Folstein

, Folstein

, and state”.

McHugh PR. “Mini-mental

A practical method for grading the cognitive state of patients for the clinician. J Psychiatr Res 1975; 12(3): 189–198.

29.

Kim

, Yu

, and Lee

BC.

Validity of the Montreal Cognitive Assessment (MoCA) index scores: a comparison with the cognitive domain scores of the Seoul Neuropsychological Screening Battery (SNSB). Dement Neurocogn Disord 2021; 20(3): 28–37.

30.

Baihaqi

, Dianingrum

, Ramadhan

KAN

, . Linear regression method to model and forecast the number of patient visits in the hospital. In: Third International Conference on Information Technology, Information System and Electrical Engineering (ICITISEE) , Yogyakarta, Indonesia, 2018, pp. 247–252. IEEE.

31.

Chi

, Wu

, Huang

, . Using linear regression to identify critical demographic variables affecting patient safety culture from viewpoints of physicians and nurses. Hosp Pract Res 2017; 2(2): 47–53.

32.

Cortes

and Vapnik

Support-vector networks. Mach Learn 1995; 20, 273–297.

33.

Prakash

, Bushra

, Subramanian

, . Random forest regression with hyperparameter tuning for medical insurance premium prediction. Int J Health Sci 2022; 6(6): 7093–7101.

34.

Zhao

, Gu

and McDermaid

Predicting outcomes of chronic kidney disease from EMR data based on Random Forest Regression. Math Biosci 2019; 310: 24–30.

35.

, Zhang

, Hu

, . Novel binary logistic regression model based on feature transformation of XGBoost for type 2 diabetes mellitus prediction in healthcare systems. Future Gener Comput Syst 2022; 129: 1–12.

36.

Zhang

, Yan

, Gao

, . Predicting missing values in medical data via XGBoost regression. J Health Inform Res 2020; 4: 383–394.

37.

Monteleone

, Tortorella

, Martiadis

, . Opposite changes in the serum brain-derived neurotrophic factor in anorexia nervosa and obesity. Psychosom Med 2004; 66: 744–748.

38.

Marcos-Pasero

, Aguilar-Aguilar

, Ikonomopoulou

, . BDNF gene as a precision skill of obesity management. Adv Exp Med Biol 2021; 1331: 233–248.

39.

Yarube

, Hassan

, Ahmad

, . Association between cognition and peripheral brain-derived neurotrophic factor in a sample of normal adults in Kano, Nigeria. Niger J Basic Clin Sci 2019; 16: 55–59.

40.

Lan

, Wang

, Li

, . The association between overweight/obesity and poor cognitive function is mediated by inflammation in patients with major depressive disorder. J Affect 2022; 313: 118–125.

41.

, Gong

, Cui

, . Body mass index and death by cognitive impairment. Aging Clin Exp Res 2023; 35(3): 689–698.

42.

Lentoor

, and Myburgh

Correlation between body mass index (BMI) and performance on the Montreal Cognitive Assessment (MoCA) in a cohort of adult women in South Africa. Behav Neurol 2022; 8994793.

43.

Tascilar

, Turkkahraman

, Oz

, . P300 auditory event-related potentials in children with obesity: is childhood obesity related to impairment in cognitive functions. Pediatr Diabetes 2011; 12: 589–595.

44.

Taha

, Al-Maqati

, Alnaam

, . The association between brain-derived neurotrophic factor (BDNF) protein level and body mass index. Medicina (Kaunas) 2022; 59(1): 99.

45.

Roy

Association between body mass index and cognitive function among older adults in India: findings from a cross-sectional study. Exp Aging Res 2024; 50: 102–116.

46.

Feng

, Fang

, Ma

, . Obesity-induced blood-brain barrier dysfunction: phenotypes and mechanisms. J Neuroinflammation 2024; 21(1): 110.

47.

Marqués-Iturria

, Garolera

, Pueyo

, . The interaction effect between BDNF val66met polymorphism and obesity on executive functions and frontal structure. Am J Med Genet Neuropsychiatr Genet 2014; 165B: 245–253.

48.

Nashwan

, Abdi

, and AlBarakat

MM.

Rethinking BMI and obesity management: the transformative role of artificial intelligence. Cureus 2024; 16(2): e54995.

49.

, Yang

, Zeng

, . Visualization obesity risk prediction system based on machine learning. Sci Rep 2024; 14: 22424.