Association and Predictive Capability of Body Composition and Diabetes Mellitus Using Artificial Intelligence: A cohort study

doi:10.21203/rs.3.rs-1675052/v1

PDF

Article

Association and Predictive Capability of Body Composition and Diabetes Mellitus Using Artificial Intelligence: A cohort study

https://doi.org/10.21203/rs.3.rs-1675052/v1

This work is licensed under a CC BY 4.0 License

You are reading this latest preprint version

This study aimed to investigate the association between regional body fat distribution and the prevalence of Diabetes mellitus (DM) in adult populations using machine learning. We applied machine learning methods to data from a cohort study to analyze the relationship between fat in different body areas and diabetes. All measurement was done by "Tanita Segmental Body Composition Analyzer BC-418 MA Tanita Corp, Japan". The correlation between the used parameters and DM was measured using some machine learning algorithms i.e. SVM, SGD, KNN, MLP, Adaboost and EDINet. A total of 4661 participants were included. The top features that reported higher importance in classification models were age, fat mass, and percentages in legs, arms, and trunk area. Fat-free mass in the legs, arm, and trunk area was reversely associated with diabetes. Our proposed method significantly outperformed the others. It has the best performances in Accuracy, Precision, Recall-0, Recall-1, and F1-score, which were 93.57, 93.67, 96.11, 74.55 and 93.62, respectively. Our machine learning models showed that regional body fat could have specific impacts on diabetes based on the location of the fat accumulation. The most predictor values of diabetes were age, fat mass, and percentages in arms, legs, and trunk area. Further studies on different ages, gender, ethnic groups, and races are recommended.

Diabetes Mellitus

Body Composition

Artificial Intelligence

Machine Learning

Diabetes is one of the critical global public health concerns. As people's life expectancies improve, type 2 diabetes mellitus (DM) has augmented globally in developed and developing nations (1). Furthermore, diabetes has developed into a major health care concern, resulting in increased mortality and morbidity, a huge worldwide impact on public health, and high social and financial expenses (2).

Glucose and lipid metabolism are inextricably linked. When lipid metabolism is harmed, glucose metabolism is also affected (3). Obesity is a known risk factor associated with type 2 DM. However, these studies measured obesity using the body mass index (BMI). BMI, waist circumference, waist-to-height, waist-to-hip, and waist-to-thigh ratios are obesity meters. The BMI measures overall body mass, while the waist circumference and waist-to-height ratio measure abdominal obesity. Likewise, it is recognized that central rather than lower body fat distribution for a given BMI leads to a greater risk of metabolic problems of obesity (4). New studies show that having even a little extra fat in the wrong places in the body can also increase the disease's probability (5). A meta-analysis including data from 10 longitudinal studies showed a strong connotation between abdominal obesity and the progress of type 2 DM (6).

Though adipose tissue placed in different body parts may influence glucose tolerance, only a few researchers have studied the association between distinct regional body fat and the incidence of diabetes mellitus. According to several studies, having a bigger hip and thigh circumference is related to a decreased risk of developing diabetes, most likely due to muscle mass's preventive benefits in these regions (7, 8). One study showed that fat in the neck area correlates with insulin sensitivity in the whole body (9). Moreover, different ethnic groups have varying fat distributions, which have varying effects on the development of diabetes (10).

Existing literature about the correlation between body fat and DM uses conventional attributes like demographic features (e.g., age, gender, race) or anthropometrics(e.g., waist circumference, and so on). Artificial intelligence has become an essential tool in biomedical research and has been engaged to predict DM in a few studies (11).

To our knowledge, there are no previous studies on the regional fat distribution or its contributions to the Iranian population using artificial intelligence and machine learning method. The principal purpose of the present study was to determine whether the prevalence of DM is influenced by site-specific adipose tissue accumulation in relatively healthy adult populations in Fasa, south of Iran.

In this section recruitment of participants, body composition analysis, dataset, evaluation metrics and the proposed method will be described in ditails. The abbreviations used in this research are listed in Table 1.

Table 1

Abbreviations used in this study.
Terms	Abbreviation
Support Vector Machine	SVM
Stochastic Gradient Descent	SGD
K-Nearest Neighbors	KNN
Multilayer Perceptron	MLP
Adaptive Boosting	Adaboost
Ensemble Diabetes Network	EDINet

2.1 Recruitment of participants

The Fasa cohort study was intended to assess the relationship between fat in different body areas and Diabetes in Fasa's rural region residents, a city in the eastern portion of the Fars province in southwest Iran, with a population of around 250,000 (12). The rural health care worker (Behvarz), who works as the representative of the primary health care system in each health home in villages and small towns, encourages the public to participate in the survey (13). All individuals were chosen from a group of qualified volunteers to participate in this study. The dataset used in this study is a subset of the Fasa cohort, containing 4661 models' records. Participants who reported having diabetes (571 diabetic models) were separated from those who did not (4090 nondiabetics models). The participants' self-declaration did this.

Furthermore, the dataset covers data on 2155 males and 2506 females. Individuals were aged between 35 and 70 years. They were nominated as the cohort's target group because they are of appropriate age to have been exposed to health themes.

2.2 The field office

A field office was established in the region's principal town. All necessary equipment is given in the field, including computers, body Composition Analyzer device, etc.

2.3 Personnel resources

Every day, a field staff member works at the field office. A field supervisor, two physicians, interviewers, nurses, and an office boy complete the team. The field supervisor is responsible for supervising ongoing interviews and ensuring that data collecting regulations are adhered to.

2.4 Registration

Each individual's national ID code is used in the registration procedure. The field supervisor registers participants by scanning their identification cards and assigns each one a unique number Personal Computer Id (PCID), which will be visible on all forms from that point on. During registration, one of the evaluators assists the field supervisor. Throughout the cohort study's several stages, the PCID number may be accessed using a search engine. Simultaneously, a cohort ID card with a digital bar code, picture, and the cohort's name will be issued. A separate notebook is also used to conduct a registry. The person must be registered. Behvarz's list of invited participants was compared to the previously compiled list. In each case, before registering, an written informed consent form is completed. All surveys are electronic and are administered over a secure web portal—a web-based application. We exploited the electrical edge. Variable types are indicated; data input is halted unless the registrar performs a recheck and the submission is accepted.

2.5 Body composition analysis

Body composition analysis was performed on all participants. Members stand on the device and hold its handles without shoes and by light clothes.

The name of the device was "Tanita Segmental Body Composition Analyzer BC-418 MA Tanita Corp, Japan". This device is one of the first Segmental Body Composition analyzers that afford measurement for the trunk and each member thanks to 8 electrodes.

The device measures: Weight, Basal Metabolic Rate (BMR), Fat Percentage: \(\frac{Fat Mass}{Weight}\) × 100 (FATP), Fat Mass (FATM), Fat-Free Mass (FFM), Total body water (TBW), Desirable body fat ranges, and Segmental body fat information.

Segmental reading separates into:

Fat percentage (FATP): Right Leg Fat Percentage (RLFATP), Left Leg Fat Percentage (LLFATP), Right Arm Fat Percentage (RAFATP), Left Arm Fat Percentage (LAFATP), Trunk Fat Percentage (TRFATP).
Fat mass (FATM): Right Leg Fat Mass (RLFATM), Left Leg Fat Mass (LLFATM), Right Arm Fat Mass (RAFATM), Left Arm Fat Mass (LAFATM), and Trunk Fat Mass (TRFATM).
Fat-free mass: Right Leg Fat-Free Mass (RLFFM), Left Leg Fat-Free Mass (LLFFM), Right Arm Fat-Free Mass (RAFFM), Left Arm Fat-Free Mass (LAFFM), Trunk Fat-Free Mass (TRFFM)
Predicted muscle mass for the right arm, right leg, left arm, left leg.

The device has a Goal Setter (GS) function, which analyses the fat mass to be lost to attain a particular target. GS focuses attention on actual fat mass rather than weight.

2.6 Control of quality

The value of surveillance systems cannot be overstated. At the first enrollment stage, the quality control team has a checklist that has been empirically tested to address some data collecting quality indicators. The team supervisor is a non-affiliated epidemiologist. They take into account the three dimensions of data and anthropometric measurements. If any flaws are discovered, they will be corrected and considered for the subsequent cleanup process. The quality control manager, who is also one of the principal investigators, entered to the database and classified content. If data cleaning was required, another team member regained the codified data and performed the clear out. The FACS program is intended to conduct routine checkups. Likewise, automatic outliers were a problem in data fields.

The investigators have established, and the software alerts the operators to validate their morals.

They are incoming before the completion of the last registration. Incomplete data is in terms of varied kinds, lengths, and scales of amount are as well recognized, and ultimate agreement is contingent upon the different stages in the PI validation process.

2.7 Dataset

The Fasa cohort included 4661 participants (571 diabetics and 4090 healthy samples), including 2155 Males and 2506 Females and 22 input features age (Between 35 and 70), gender ID (1: male, 2: female), Basal Metabolic Rate (BMR), Fat Mass (FATM), Fat Percentage (FATP) which is (Fat Mass)/Weight × 100, Fat-Free Mass (FFM), Total Body Water (TBW), Right Leg Fat Percentage (RLFATP), Right Leg Fat Mass (RLFATM), Right Leg Fat-Free Mass (RLFFM), Left Leg Fat Percentage (LLFATP), Left Leg Fat Mass (LLFATM), Left Leg Fat-Free Mass (LLFFM), Right Arm Fat Percentage (RAFATP), Right Arm Fat Mass (RAFATM), Right Arm Fat-Free Mass (RAFFM), Left Arm Fat Percentage (LAFATP), Left Arm Fat Mass (LAFATM), Left Arm Fat-Free Mass (LAFFM), Trunk Fat Percentage (TRFATP), Trunk Fat Mass (TRFATM), and Trunk Fat-Free Mass (TRFFM).

Institutional approval was granted for the use of the patient datasets in research studies for diagnostic and therapeutic purposes. Approval was granted on the grounds of existing datasets. Informed consent was obtained from all of the patients in this study. All methods were carried out in accordance with relevant guidelines and regulations. All experimental protocols were approved by Research Ethical Committees of School of Medicine-Shiraz University of Medical Science.

2.8 Evaluation metrics

By following previous studies in the literature (14–17), the performance of the machine learning algorithms was estimated by accuracy, precision, recall, and F1-score, according to the following equations:

Accuracy =\(\frac{\text{T}\text{P}+\text{T}\text{N}}{\text{T}\text{P} + \text{T}\text{N} + \text{F}\text{P} + \text{F}\text{N}}\)

Weighted Average Precision =\(\frac{\sum _{i=0}^{n-1}\left|\text{n}\text{u}\text{m}\text{b}\text{e}\text{r} \text{o}\text{f} \text{e}\text{l}\text{e}\text{m}\text{e}\text{n}\text{t}\text{s} \text{i}\text{n} \text{c}\text{l}\text{a}\text{s}\text{s} \text{i}\right|\frac{{TP}_{i}}{{TP}_{i}+{FP}_{i}}}{\sum _{i=0}^{n-1}\left|\text{n}\text{u}\text{m}\text{b}\text{e}\text{r} \text{o}\text{f} \text{e}\text{l}\text{e}\text{m}\text{e}\text{n}\text{t}\text{s} \text{i}\text{n} \text{c}\text{l}\text{a}\text{s}\text{s} \text{i}\right|}\)

Recall =\(\frac{\text{T}\text{P}}{\text{T}\text{P}+\text{F}\text{N}}\)

F1-score =\(\frac{2 \text{P}\text{r}\text{e}\text{c}\text{i}\text{s}\text{i}\text{o}\text{n} \times \text{R}\text{e}\text{c}\text{a}\text{l}\text{l} }{\text{P}\text{r}\text{e}\text{c}\text{i}\text{s}\text{i}\text{o}\text{n}+ \text{R}\text{e}\text{c}\text{a}\text{l}\text{l}}= \frac{\text{T}\text{P}}{\text{T}\text{P}+\frac{1}{2}\left(\text{F}\text{P}+\text{F}\text{N}\right)}\)

Weighted Average F1-score =\(\frac{\sum _{i=0}^{n-1}\left|\text{n}\text{u}\text{m}\text{b}\text{e}\text{r} \text{o}\text{f} \text{e}\text{l}\text{e}\text{m}\text{e}\text{n}\text{t}\text{s} \text{i}\text{n} \text{c}\text{l}\text{a}\text{s}\text{s} \text{i}\right|{F1-score}_{i}}{\sum _{i=0}^{n-1}\left|\text{n}\text{u}\text{m}\text{b}\text{e}\text{r} \text{o}\text{f} \text{e}\text{l}\text{e}\text{m}\text{e}\text{n}\text{t}\text{s} \text{i}\text{n} \text{c}\text{l}\text{a}\text{s}\text{s} \text{i}\right|}\)

Where TP is the number of test results that are correctly classified as positive. TN is the number of test results that are correctly classified as negative. FP is the number of test results that are incorrectly classified as positive and FN is the number of test results that are incorrectly classified as negative.

2.9 Proposed Model

In this section, we present the overall view of the proposed methodology. At first, split the dataset into Training (80%) and Testing (20%) datasets. Since our dataset was imbalanced, we proposed oversampling technique on the Training part, using SVMSMOTE. Afterward, we implemented the base classifiers: (linear) SVM, Decision tree, Stochastic Gradient Descent, Logistic regression, Gaussian naïve Bayes, K-Nearest Neighbors (k = 3 and k = 4), and Multi-Layer Perceptron neural network, to train machine learning models.

Finally, to improve the performances of our models, we propose some ensemble learning algorithms: Gradient boosting, Adaboost, Stacking (for top3 and top4), and Voting (for top3 and top4). A framework of the proposed models for diagnosis of diabetes is shown in Fig. 1.

In this section, we discuss the main findings obtained after applying several well-known machine learning algorithms. In this regard, we first discuss the obtained results after applying individual classifiers. After that, the optimized results will be presented by applying different ensemble learning techniques.

3.1 Individual Classifiers

We proposed some of the most states of the art machine learning models on the dataset, including SVM (with the linear kernel), decision tree, stochastic Gradient descent, logistic regression, Gaussian naïve Bayes, KNN (for K = 3 and K = 4) and neural network (MLP). The goal feature of the dataset, i.e., Has-Diabetes, has two values: 0 and 1. The value 1 means the sample is diabetic, whereas the value 0 means the sample is non-diabetic.

The results of applying classification algorithms to our dataset are shown in Table 2. Accordingly, KNN for K = 4 has the best accuracy of 88.64%, while Gaussian naïve Bayes has the worst with the accuracy of 56.81%. Other performance parameters, Precision, Recall, and F1-score, are shown in Table 2.

Table 2

The performance of different classification algorithms proposed on the dataset.
Classification algorithm	Accuracy (%)	Weighted average Precision (%)	Recall-0(%)	Recall-1(%)	Weighted average F1-score (%)
SVM (linear)	74.71	84.26	77.76	51.82	78.32
Decision tree	80.28	85.38	84.20	50.91	82.33
Stochastic gradient descent (SGD)	75.03	84.33	78.13	51.82	78.55
Logistic regression	73.10	84.09	75.82	52.73	77.17
Gaussian naïve Bayes	56.81	82.40	56.38	60.00	64.41
KNN (K = 3)	85.96	91.08	86.51	81.82	87.60
KNN (K = 4)	88.64	90.95	90.64	73.64	89.49
Neural network (MLP)	80.17	85.36	84.08	50.91	82.26

3.2 Ensemble of Classifiers

To improve the performance of the proposed classification algorithms, in this sub-section, we discuss the main results after applying several ensemble learning methods, including gradient boosting, Adaboost, Stacking ensemble, and voting ensemble. The results of applying these ensemble-learning models to our dataset are shown in Table 3.

Table 3

The performance of different classification algorithms after applying several ensemble learning techniques on the dataset.
Classification algorithm	Accuracy (%)	Weighted average Precision (%)	Recall-0(%)	Recall-1(%)	Weighted average F1-score (%)
Gradient boosting	83.92	84.91	89.91	39.09	84.39
Adaboost	79.53	84.56	83.96	46.36	81.60
EDINet (Stacking-Top3)	93.57	93.67	96.11	74.55	93.62
EDINet (Stacking-Top4)	92.50	93.10	94.65	76.36	92.74
EDINet (Voting-Top3)	87.14	90.51	88.82	74.55	88.33
EDINet (Voting-Top4)	88.53	89.04	92.83	56.36	88.77

According to the previous studies (18, 19) and based on the obtained results in Table 3, we can see that applying the stacking ensemble learning method could improve the performance of individual classifiers. The results in Table 3 show that EDINet (Stacking-Top3) significantly outperformed the other applied methods. Moreover, we can see that EDINet (Stacking-Top3) and EDINet (Stacking-Top4) have the best performances in Accuracy, Precision, Recall-0, Recall-1, F1-score.

3.3 Confusion matricx and Correlations

Information extracted from confusion matrix of the proposed algorithms and the correlations between features and being diabetic are shown in Table 4 and Fig. 2, respectively.

Table 4

Information extracted from confusion matrix of proposed algorithms
	TP	TN	FP	FN
SVM	57	640	183	53
Decision tree	56	693	130	54
SGD	643	57	180	53
Logistic regression	624	58	199	52
Gaussian naïve Bayes	464	66	359	44
KNN (K = 3)	712	90	111	20
KNN (K = 4)	746	81	77	29
Neural network (MLP)	692	56	131	54
Gradient boosting	740	43	83	67
Adaboost	691	51	132	59
EDINet (Stacking-Top3)	791	82	32	28
EDINet (Stacking-Top4)	779	84	44	26
EDINet (Voting-Top3)	731	82	92	28
EDINet (Voting-Top3)	764	62	59	48

To our knowwledge, this is the first study to use site-specific adipose tissue accumulation to predict DM in a cohort population using machine learning. DM inevitably produces considerable quantities of clinical data which can be complex and difficult to utilise in routine practice (11). Artificial intelligence offers enormous potential in medicine and prediction of diabetes at an early stage.

Remarkably, fat mass and fat percentage were related factors of being diabetic in our analysis. The fat-free group was inversely associated with diabetes in our results. In line with the study of Pramyothin et al., higher fat-free mass is associated with type 2 diabetes (20). Also, Lagacé JC et al. indicated that the prevalence of Metabolic syndrome decreased with increasing quartiles of whole-body FFM percentage. It was also indicated that metabolic syndrome prevalence increases with fat-free mass (21). Fat mass and fat percentage in arms, legs, and trunk are among the most significant features correlated with diabetes. The severity of obesity, and specific fat distribution is a risk factor for diabetes (22, 23). Prior articles have shown that the prevalence of diabetes increases with a higher waist-to-hip girth ratio and abdominal adiposity (24, 25). In conclusion, fat mass and percentage are related to diabetes, and fat-free mass is inversely associated with diabetes. Our study is consistent with previous studies that specific fat distribution patterns are powerful predictors of diabetes.

4.1 Trunk fat mass and diabetes

Importantly, TRFATM and TRFATP are among the significant features related to diabetes in our study. In line with the study of Yoshimi et al. that percentage of trunk fat was positively associated with the onset of DM, we identify TRFATP and TRFATM as strong features to predict DM (26). Also, our findings are in good agreement with Horejsi et al. that indicated subcutaneous body fat in the upper trunk increases with type 2 diabetes especially (27). It is noteworthy to mention that TRFFM is reversely related to diabetes. In addition, in the study of Wang et al., a higher fat-to-muscle ratio in the trunk was associated with a higher risk of developing type 2 diabetes (28). In conclusion, TRFATM and TRFATP should be considered as probable factors to predict diabetes, and TRFFM is reversely related to diabetes.

4.2 Leg fat mass and diabetes

We have found that fat mass and fat percentages in both lower extremities are features related to diabetes. Our results contrast with Choi et al., which have found that having a larger calf fat mass was connected with a decreased risk of developing Diabetes in Korean people (29). Snijder et al. have concluded that leg fat mass can have some protective effects against a disturbed glucose metabolism (30). In the study of Heshka et al., there was an association between less total leg fat and type 2 diabetes (8). Also, Yoshimi et al. indicated that the percentage of leg fat was negatively associated with the development of diabetes (26). A popular explanation for this difference is that fat distribution varies in different ethnic groups and may influence the onset of DM (10).

4.3 Arm fat mass and diabetes

We have found a strong association between fat mass and fat percentages in upper extremities and diabetes. It is interesting to indicate that RAFFM and LAFFM are not related to diabetes. Zhu et al. reported that a mid-upper arm circumference is valuable for insulin resistance in type 2 diabetes (31). In a study by Hou et al., mid-upper arm circumference significantly interacted with diabetes and insulin resistance (32). Also in the study of Zhu et al., mid-upper arm circumference was associated with insulin resistance, and it can be used as a simple tool for central obesity in type 2 diabetes patients (31). To sum up, higher fat mass in upper extremities is related to diabetes, and fat-free mass in both upper extremities can decrease the chance of being diabetic.

4.4 Age and diabetes

In our study, age was reported as one of the most decisive factors predicting DM. Age is a significant basis of diabetes since blood glucose concentrations rise with age (33). The results of our study were in line with the results of the previous studies about the correlation of age and type 2 DM. Identifying the patients at earlier ages will provide more significant opportunities for interventions to reduce the risk of complication-associated morbidity and mortality for patients developing type 2 DM (34). The incidence of diagnosed diabetes continues to growth with age in non-Hispanic whites, non-Hispanic blacks, and Mexican American men and women, as revealed in the Third National Health and Nutrition Examination Survey, 1988–1994 (NHANES III) and these result are lined with our study (35). Thus, age can be considered one of the most significant features to predict diabetes.

4.5 Gender and Diabetes

We found that female gender is associated with DM. Gender influences the pathogenesis of many diseases, including metabolic disorders such as DM. Type II diabetes is equally frequent in men and women in most populations, with certain nations demonstrating a male predominance in early middle life (36). In the study of Kautzky-Willer et al., it was indicated that men had a greater risk of having diabetes at a younger age and a lower body mass index (BMI) compared to women. Still, women had a dramatic increase in the risk of diabetes-associated cardiovascular diseases after menopause (36). In a study by Sattar et al., it was concluded that adult men are at more significant risks of having diabetes than women because of having higher waist circumferences and more insulin resistance in men (37). This contrast may be because our data was gathered in a rural area where women had less physical activity than men.

4.6 BMR and Diabetes

Our findings demonstrate that BMR is not a significant predictor of DM. Although this finding is in contrast with the study of Kumar et al. that low BMR was considered a risk factor for the development of DM (38). In a study by Kumar et al., it is indicated that BMR was considered a risk factor that is increased in people who have type 2 diabetes (38). Significantly, BMR depends on multiple factors such as sex, ethnicity, age, physical activity, genetic factors, the presence of diabetes or obesity, body composition, and caloric intake (39). This difference may be because of that BMR is affected by such a variable factors that was not considered in our study.

4.7 TBW and Diabetes

As far as we know, it is for the first time that the association of TBW and Diabetes has been analyzed with machine learning. The most striking result from the data is that there is no association between Diabetes and TBW. Although in a study by Nayak et al., lower extracellular water was a predictor for pre-diabetic patients (40). Further studies about TBW and Diabetes are recommended.

4.8 Limitations and strengths

This research is not without limits. The first restriction is that anti-diabetic and anti-hypertensive medications may alter insulin resistance, affecting our findings. Including patients' entire medication histories in our investigation would be reasonable. However, since this is cross-sectional research rather than a randomized controlled trial, we could not intervene to discontinue anti-diabetic or other drugs as clinical studies need. Another limitation is that we did not measure the associations between ethnic groups with visceral body fat; we could not generalize our results to other ethnic groups. Finally, we couldn't count the visceral body fat composition and the DM complications. The strength of the present study was using the machine learning method as the new method for determining strong predictors of DM. Another strength is the use of an acceptable sample of cohort patients.

The present study's finding points us to a novel research area of measuring visceral fats as a biomarker to diagnosing DM. Although high levels of BMI and total body fat are risk factors for diabetes type 2, Regional body fat can have specific impacts on diabetes based on the location of the fat accumulation. The most related values of being diabetic were age, fat mass, and percentage in both upper and lower extremities and trunk area. Further studies on different ages, gender, ethnic groups, and races are recommended.

Data availability

All data included in this study are available from the corresponding authors and can be provided upon request as needed.

Author Contributions

Contributed to prepare the first draft: M. A. N., Ar.A., Am. A., M. S., M.M., S.S., M.B., M.F., R.H., B.P., M.A., M.R., R.A., and H.B.D. Contributed to editing the final draft: M. A. N., Ar.A., Am. A., M. S., M.M., S.S., M.B., M.F., R.H., B.P., M.A., and H.B.D. Contributed to all analysis of the data and produced the results accordingly: M. A. N., Ar.A., Am. A., M. S., M.M., S.S., M.B., M.F., R.H., B.P., and M.A. Searched for papers and then extracted data: M. A. N., Ar.A., Am. A., M. S., M.M., S.S., M.B., M.F., R.H., B.P., and M.A. Provided overall guidance and managed the project: H.B.D., and S.M.S.I.

Competing interests

Te authors declare no competing interests.

Chan JC, Malik V, Jia W, Kadowaki T, Yajnik CS, Yoon K-H, et al. Diabetes in Asia: epidemiology, risk factors, and pathophysiology. Jama. 2009;301(20):2129–40.
Guariguata L, Whiting D, Weil C, Unwin N. The International Diabetes Federation diabetes atlas methodology for estimating global and national prevalence of diabetes in adults. Diabetes research and clinical practice. 2011;94(3):322–32.
Zhu A, Cui B, Dang H, Yao D, Yu H, Jia H, et al. Correlation of abdominal fat distribution with different types of diabetes in a Chinese population. Journal of Diabetes Research. 2013;2013.
Jung SH, Ha KH, Kim DJ. Visceral fat mass has stronger associations with diabetes and prediabetes than other anthropometric obesity indicators among Korean adults. Yonsei medical journal. 2016;57(3):674–80.
Teufel F, Seiglie JA, Geldsetzer P, Theilmann M, Marcus ME, Ebert C, et al. Body-mass index and diabetes risk in 57 low-income and middle-income countries: a cross-sectional study of nationally representative, individual-level data in 685 616 adults. The Lancet. 2021;398(10296):238–48.
Freemantle N, Holmes Ja, Hockey A, Kumar S. How strong is the association between abdominal obesity and the incidence of type 2 diabetes? International journal of clinical practice. 2008;62(9):1391–6.
Snijder M, Zimmet PZ, Visser M, Dekker J, Seidell J, Shaw JE. Independent and opposite associations of waist and hip circumferences with diabetes, hypertension and dyslipidemia: the AusDiab Study. International journal of obesity. 2004;28(3):402–9.
Snijder MB, Dekker JM, Visser M, Bouter LM, Stehouwer CD, Kostense PJ, et al. Associations of hip and thigh circumferences independent of waist circumference with the incidence of type 2 diabetes: the Hoorn Study. The American journal of clinical nutrition. 2003;77(5):1192–7.
Thamer C, Machann J, Staiger H, Mussig K, Schwenzer N, Ludescher B, et al. Interscapular fat is strongly associated with insulin resistance. The Journal of Clinical Endocrinology & Metabolism. 2010;95(10):4736–42.
Nazare J-A, Smith JD, Borel A-L, Haffner SM, Balkau B, Ross R, et al. Ethnic influences on the relations between abdominal subcutaneous and visceral adiposity, liver fat, and cardiometabolic risk profile: the International Study of Prediction of Intra-Abdominal Adiposity and Its Relationship With Cardiometabolic Risk/Intra-Abdominal Adiposity. The American journal of clinical nutrition. 2012;96(4):714–26.
Nath T, Ahima RS, Santhanam P. Body fat predicts exercise capacity in persons with Type 2 Diabetes Mellitus: A machine learning approach. Plos one. 2021;16(3):e0248039.
Farjam M, Bahrami H, Bahramali E, Jamshidi J, Askari A, Zakeri H, et al. A cohort study protocol to analyze the predisposing factors to common chronic non-communicable diseases in rural areas: Fasa Cohort Study. BMC public health. 2016;16(1):1–8.
Soori H, Motlagh E. Iranian rural health workers [behvarz] and risk factors of childhood injury. EMHJ-Eastern Mediterranean Health Journal, 5 (4), 684–689, 1999. 1999.
Abdar M, Fahami MA, Chakrabarti S, Khosravi A, Pławiak P, Acharya UR, et al. BARF: A new direct and cross-based binary residual feature fusion with uncertainty-aware module for medical image classification. Information Sciences. 2021;577:353–78.
Abdar M, Salari S, Qahremani S, Lam H-K, Karray F, Hussain S, et al. UncertaintyFuseNet: Robust uncertainty-aware hierarchical feature fusion with ensemble Monte Carlo dropout for COVID-19 detection. arXiv preprint arXiv:210508590. 2021.
Abdar M, Samami M, Mahmoodabad SD, Doan T, Mazoure B, Hashemifesharaki R, et al. Uncertainty quantification in skin cancer classification using three-way decision-based Bayesian deep learning. Computers in biology and medicine. 2021;135:104418.
Joloudari JH, Azizi F, Nematollahi MA, Alizadehsani R, Hassannatajjeloudari E, Nodehi I, et al. GSVMA: A Genetic Support Vector Machine ANOVA Method for CAD Diagnosis. Frontiers in cardiovascular medicine. 2022;8:760178.
Abdar M, Acharya UR, Sarrafzadegan N, Makarenkov V. NE-nu-SVC: a new nested ensemble clinical decision support system for effective diagnosis of coronary artery disease. IEEE Access. 2019;7:167605–20.
Dong Y, Zhang H, Wang C, Zhou X. Wind power forecasting based on stacking ensemble model, decomposition and intelligent optimization algorithm. Neurocomputing. 2021;462:169–84.
Pramyothin P, Limpattanachart V, Dawilai S, Sarasak R, Sukaruttanawong C, Chaiyasoot K, et al. Fat-free mass, metabolically healthy obesity, and type 2 diabetes in severely obese asian adults. Endocrine Practice. 2017;23(8):915–22.
Lagacé JC, Marcotte-Chenard A, Paquin J, Tremblay D, Brochu M, Dionne IJ. Increased odds of having the metabolic syndrome with greater fat‐free mass: counterintuitive results from the National Health and Nutrition Examination Survey database. Journal of Cachexia, Sarcopenia and Muscle. 2021.
Skyler JS, Oddo C. Diabetes trends in the USA. Diabetes/Metabolism Research and Reviews. 2002;18(S3):S21-S6.
Ohlson L-O, Larsson B, Svärdsudd K, Welin L, Eriksson H, Wilhelmsen L, et al. The influence of body fat distribution on the incidence of diabetes mellitus: 13.5 years of follow-up of the participants in the study of men born in 1913. Diabetes. 1985;34(10):1055–8.
Hartz AJ, Rupley Jr DC, Kalkhoff RD, Rimm AA. Relationship of obesity to diabetes: influence of obesity level and body fat distribution. Preventive medicine. 1983;12(2):351–7.
Koh-Banerjee P, Wang Y, Hu FB, Spiegelman D, Willett WC, Rimm EB. Changes in body weight and body fat distribution as risk factors for clinical diabetes in US men. American journal of epidemiology. 2004;159(12):1150–9.
Tatsukawa Y, Misumi M, Kim YM, Yamada M, Ohishi W, Fujiwara S, et al. Body composition and development of diabetes: a 15-year follow-up study in a Japanese population. European journal of clinical nutrition. 2018;72(3):374–80.
Horejsi R, Möller R, Pieber TR, Wallner S, Sudi K, Reibnegger G, et al. Differences of subcutaneous adipose tissue topography between type-2 diabetic men and healthy controls. Experimental Biology and Medicine. 2002;227(9):794–8.
Wang N, Sun Y, Zhang H, Chen C, Wang Y, Zhang J, et al. Total and regional fat-to‐muscle mass ratio measured by bioelectrical impedance and risk of incident type 2 diabetes. Journal of cachexia, sarcopenia and muscle. 2021;12(6):2154–62.
Choi SI, Chung D, Lim JS, Lee MY, Shin JY, Chung CH, et al. Relationship between regional body fat distribution and diabetes mellitus: 2008 to 2010 Korean National Health and Nutrition Examination Surveys. Diabetes & metabolism journal. 2017;41(1):51–9.
Heshka S, Ruggiero A, Bray GA, Foreyt J, Kahn SE, Lewis CE, et al. Altered body composition in type 2 diabetes mellitus. International journal of obesity. 2008;32(5):780–7.
Zhu Y, Lin Q, Zhang Y, Deng H, Hu X, Yang X, et al. Mid-upper arm circumference as a simple tool for identifying central obesity and insulin resistance in type 2 diabetes. PloS one. 2020;15(5):e0231308.
Hou Y, Jia X, Xuan L, Zhu W, Deng C, Wang L, et al. Association between mid-upper arm circumference and cardiometabolic risk in Chinese population: a cross-sectional study. BMJ open. 2019;9(9):e028904.
Alanazi NH, Alsharif MM, Rasool G, Alruwaili ABH, Alrowaili AMZ, Aldaghmi AS, et al. Prevalence of diabetes and its relation with age and sex in Turaif city, northern Saudi Arabia in 2016–2017. Electronic physician. 2017;9(9):5294.
Nanayakkara N, Curtis AJ, Heritier S, Gadowski AM, Pavkov ME, Kenealy T, et al. Impact of age at type 2 diabetes mellitus diagnosis on mortality and vascular complications: systematic review and meta-analyses. Diabetologia. 2021;64(2):275–87.
Chan JM, Rimm EB, Colditz GA, Stampfer MJ, Willett WC. Obesity, fat distribution, and weight gain as risk factors for clinical diabetes in men. Diabetes care. 1994;17(9):961–9.
Kautzky-Willer A, Harreiter J. Sex and gender differences in therapy of type 2 diabetes. Diabetes Research and Clinical Practice. 2017;131:230–41.
Sattar N, Rawshani A, Franzén S, Rawshani A, Svensson A-M, Rosengren A, et al. Age at diagnosis of type 2 diabetes mellitus and associations with cardiovascular and mortality risks: findings from the Swedish National Diabetes Registry. Circulation. 2019;139(19):2228–37.
Kumar AS, Maiya GA, Shastry B, Vaishali K, Maiya S, Umakanth S. Correlation between basal metabolic rate, visceral fat and insulin resistance among type 2 diabetes mellitus with peripheral neuropathy. Diabetes & Metabolic Syndrome: Clinical Research & Reviews. 2019;13(1):344-8.
Hagedorn T, Poggiogalle E, Savina C, Coletti C, Paolini M, Scavone L, et al. Indirect calorimetry in obese female subjects: factors influencing the resting metabolic rate. World Journal of Experimental Medicine. 2012;2(3):58.
Nayak VKR, Raghurama Nayak K, Vidyasagar S, Kamath A. Body composition analysis, anthropometric indices and lipid profile markers as predictors for prediabetes. PloS one. 2018;13(8):e0200775.

No competing interests reported.

PDF

Editor invited by journal
09 Jun, 2022
Submission checks completed at journal
09 Jun, 2022
First submitted to journal
19 May, 2022

You are reading this latest preprint version

Association and Predictive Capability of Body Composition and Diabetes Mellitus Using Artificial Intelligence: A cohort study

Status:

Version 1

Abstract

Figures

1 Introduction

2 Method

3 Experimental Results

4 Discussion

5 Conclusion

Declarations

References

Competing Interests

Status:

Version 1