Dataset Information

An Inverse Relation between Hyperglycemia and Skeletal Muscle Mass Predicted by Using a Machine Learning Approach in Middle-Aged and Older Adults in Large Cohorts.

ABSTRACT:

Background

Skeletal muscle mass (SMM) and fat mass (FM) are essentially required for health and quality of life in older adults.

Objective

To generate the best SMM and FM prediction models using machine learning models incorporating socioeconomic, lifestyle, and biochemical parameters and the urban hospital-based Ansan/Ansung cohort, and to determine relations between SMM and FM and metabolic syndrome and its components in this cohort.

Methods

SMM and FM data measured using an Inbody 4.0 unit in 90% of Ansan/Ansung cohort participants were used to train seven machine learning algorithms. The ten most essential predictors from 1411 variables were selected by: (1) Manually filtering out 48 variables, (2) generating best models by random grid mode in a training set, and (3) comparing the accuracy of the models in a test set. The seven trained models' accuracy was evaluated using mean-square errors (MSE), mean absolute errors (MAE), and R² values in 10% of the test set. SMM and FM of the 31,025 participants in the Ansan/Ansung cohort were predicted using the best prediction models (XGBoost for SMM and artificial neural network for FM). Metabolic syndrome and its components were compared between four groups categorized by 50 percentiles of predicted SMM and FM values in the cohort.

Results

The best prediction models for SMM and FM were constructed using XGBoost (R2 = 0.82) and artificial neural network (ANN; R2 = 0.89) algorithms, respectively; both models had a low MSE. Serum platelet concentrations and GFR were identified as new biomarkers of SMM, and serum platelet and bilirubin concentrations were found to predict FM. Predicted SMM and FM values were significantly and positively correlated with grip strength (r = 0.726) and BMI (r = 0.915, p < 0.05), respectively. Grip strengths in the high-SMM groups of both genders were significantly higher than in low-SMM groups (p < 0.05), and blood glucose and hemoglobin A1c in high-FM groups were higher than in low-FM groups for both genders (p < 0.05).

Conclusion

The models generated by XGBoost and ANN algorithms exhibited good accuracy for estimating SMM and FM, respectively. The prediction models take into account the actual clinical use since they included a small number of required features, and the features can be obtained in outpatients. SMM and FM predicted using the two models well represented the risk of low SMM and high fat in a clinical setting.

SUBMITTER: Wu X

PROVIDER: S-EPMC8156777 | biostudies-literature |

REPOSITORIES: biostudies-literature

ACCESS DATA

Similar Datasets

Project description:Little is known about the association between dietary patterns and hyperglycemia incidence among Korean adults. Hence, we aimed to prospectively investigate the major dietary patterns associated with hyperglycemia among middle-aged and older Korean adults. In total, 55,457 adults (18,292 men and 37,165 women) aged 40 to 79 years, who were previously enrolled in the Health Examinee Study of the Korean Genome and Epidemiology Study and had no history of type 2 diabetes mellitus (T2DM) or cancer at baseline, were included. Dietary patterns were identified by a factor analysis based on dietary data, which were assessed at baseline using a validated food-frequency questionnaire. Participants were classified as having hyperglycemia if fasting blood glucose levels were ?126 mg/dL or physician diagnosed T2DM during follow-up. Multivariable Cox proportional hazard models were used to examine the associations between each dietary pattern and future hyperglycemia risk after adjusting for potential confounders. After a mean follow-up of 4.9 years, 2574 new cases of hyperglycemia were identified. Using a factor analysis, four distinct dietary patterns were identified: "prudent;" "fatty fish, meat, and flour-based food;" "coffee and sweets;" and "whole grain (men)" or "white rice (women)." The "prudent" pattern was inversely associated with hyperglycemia risk only in women (hazard ratio [HR], 0.75; 95% confidence interval [CI], 0.63-0.89; p for trend = 0.0003). Conversely, women in the highest quintile of the "fatty fish, meat, and flour-based food" pattern showed an increased risk of hyperglycemia (HR, 1.22; 95% CI, 1.03-1.44; p for trend = 0.0210) compared with those in the lowest quintile. The "coffee and sweets" and "white rice" patterns were not associated with hyperglycemia risk in women. The dietary patterns observed in men had no associations with hyperglycemia incidence. Our findings suggest that a diet rich in vegetables, mushrooms, seaweeds, fruits, and soy products and low in fatty fish and high-fat meat may potentially play a protective role in T2DM development with sex differences in middle-aged and older Korean adults.

Project description:BackgroundAge-related loss of skeletal muscle mass contributes to poor outcomes including sarcopenia, physical disability, frailty, type 2 diabetes, and mortality. Vitamin C has physiological relevance to skeletal muscle and may protect it during aging, but few studies have investigated its importance in older populations.ObjectivesWe aimed to investigate cross-sectional associations of dietary and plasma vitamin C with proxy measures of skeletal muscle mass in a large cohort of middle- and older-aged individuals.MethodsWe analyzed data from >13,000 men and women in the European Prospective Investigation into Cancer and Nutrition-Norfolk cohort, aged 42-82 y. Fat-free mass (FFM), as a proxy for skeletal muscle mass, was estimated using bioelectrical impedance analysis and expressed as a percentage of total mass (FFM%) or standardized by BMI (FFMBMI). Dietary vitamin C intakes were calculated from 7-d food diary data, and plasma vitamin C was measured in peripheral blood. Multivariable regression models, including relevant lifestyle, dietary, and biological covariates, were used to determine associations between FFM measures and quintiles of dietary vitamin C or insufficient compared with sufficient plasma vitamin C (<50 μmol/L and ≥50 μmol/L).ResultsPositive trends were found across quintiles of dietary vitamin C and FFM measures for both sexes, with interquintile differences in FFM% and FFMBMI of 1.0% and 2.3% for men and 1.9% and 2.9% for women, respectively (all P < 0.001). Similarly, FFM% and FFMBMI measures were higher in participants with sufficient than with insufficient plasma vitamin C: by 1.6% and 2.0% in men, and 3.4% and 3.9% in women, respectively (all P < 0.001). Associations were also evident in analyses stratified into <65-y and ≥65-y age groups.ConclusionsOur findings of positive associations, of both dietary and circulating vitamin C with measures of skeletal muscle mass in middle- and older-aged men and women, suggest that dietary vitamin C intake may be useful for reducing age-related muscle loss.

Project description:BackgroundWith urbanization and aging increasing in coming decades, societies face the challenge of keeping aging populations active. Land use mix (LUM) has been associated with cycling and walking, but whether changes in LUM relate to changes in cycling/walking is less known.ObjectivesOur objective was to study the effect of LUM on cycling/walking in two Dutch aging cohorts using data with 10 years of follow-up.MethodsData from 1183 respondents from the Health and Living Conditions of the Population of Eindhoven and Surroundings (GLOBE) study and 918 respondents from the Longitudinal Aging Study Amsterdam (LASA) were linked to LUM in 1000-m sausage network buffers at three time-points. Cycling/walking outcomes were harmonized to include average minutes spent cycling/walking per week. Data was pooled and limited to respondents that did not relocate between follow-up waves. Associations between LUM and cycling/walking were estimated using a Random Effects Within-Between (REWB) model that allows for the estimation of both within and between effects. Sensitivity analyses were performed on smaller (500-m) and larger (1600-m) buffers.ResultsWe found evidence of between-individual associations of LUM in 1000-m buffers and walking (β: 11.10, 95% CI: 0.08; 21.12), but no evidence of within-associations in 1000-m buffers. Sensitivity analyses using 500-m buffers showed similar between-associations, but negative within-associations (β: -35.67, 95% CI: - 68.85; - 2.49). We did not find evidence of between-individual associations of LUM in any buffer size and cycling, but did find evidence of negative within-associations between LUM in 1600-m buffers and cycling (β: -7.49, 95% CI: - 14.31; - 0.66).DiscussionOur study found evidence of positive associations between LUM and average walking time, but also some evidence of negative associations between a change in LUM and cycling/walking. LUM appears to be related to cycling/walking, but the effect of changes in LUM on cycling/walking is unclear.