INTRODUCTION
Physical disability is characterized by permanently limited physical function resulting from congenital or acquired neurological, musculoskeletal, or traumatic causes. Physical disabilities are subcategorized into disabilities due to amputation, joint disabilities, physical function disabilities, and disabilities due to deformity [1]. According to the Ministry of Health and Welfare, there are approximately 1.15 million individuals with physical disabilities in South Korea, which is the largest proportion of registered individuals with disabilities [2]. Approximately 87.6% of individuals with physical disabilities have at least one chronic disease, highlighting the need for countermeasures at national and social levels [3].
Although differences exist based on the level of disability, most individuals with physical disabilities use assistive measures or wheelchairs, which make them highly sedentary and restrict their activities [4]. Additionally, paralysis of the injured area or long-term non-use of leg muscles leads to decreased activation of the sympathetic nervous system and muscle atrophy, resulting in impaired metabolism at rest and during activity. This condition renders individuals with disabilities more susceptible to obesity and inadequate physical activity, thereby increasing their risk of metabolic diseases [5,6]. According to the latest national public health statistics for individuals with disabilities from 2022, more than half of the individuals with physical disabilities meet the criteria for dyslipidemia and hypertension, and approximately one-third meet the criteria for type 2 diabetes [3]. These metabolic diseases have been identified as major risk factors for premature death in individuals with disabilities [7,8].
As in the general population, physical activity and fitness are closely related to the incidence of metabolic disease in individuals with disabilities, and participation in regular physical activity and high fitness reportedly exerts a protective role against the risk of metabolic disease [9-12]. For example, in a study involving patients with paraplegic spinal cord injury, upper body-focused moderate-intensity exercise reduced the risk of cardiometabolic disorders by improving metabolic regulation [9]. Additionally, long-term regular exercise was shown to enhance metabolic health by improving insulin sensitivity and reducing inflammatory cytokines [10]. In a study investigating the relationship between lifestyle factors and metabolic syndrome in 8,237 South Koreans with physical disabilities, the group that met the recommendations for physical activity exhibited a 0.8-fold reduction in the risk of metabolic syndrome compared with those who did not [11]. Furthermore, a study involving 835 patients with spinal cord injury in South Korea revealed that high muscle strength and endurance could substantially reduce the risk of hypertension [12].
Predictive models have been extensively used for early detection and prevention of chronic diseases in individuals without disabilities [13,14]. For example, prediction models developed using the Framingham risk score (FRS) or atherosclerotic cardiovascular disease risk assessment have been effectively used in clinical and epidemiological research [13,14]. These predictive models are non-invasive and convenient methods for regularly monitoring disease risk and have excellent predictive ability [13,14]. Recent reports have revealed that improving physical activity and fitness can protect against chronic diseases, underscoring the need to develop predictive models that incorporate these factors. Furthermore, models that include these factors reportedly show better predictive abilities [15,16]. Gander et al. reported that the addition of physical activity and cardiopulmonary fitness to the FRS improved its predictive ability for coronary artery disease risk [15]. Kunutsor et al. incorporated grip strength into two previously reported diabetes prediction models, thereby noting improvements in their predictive abilities [16]. Given that physical activity and fitness are markedly associated with metabolic diseases in both individuals with and without physical disabilities, developing a predictive model based on physical activity and fitness could aid in the prevention and risk assessment of metabolic diseases. However, although the prevalence of metabolic disease is higher among individuals with physical disabilities than among those without, the risk of disease in this group cannot be overlooked, given that activity restrictions make access to healthcare facilities more difficult. Moreover, no predictive models have yet been developed. Given that metabolic syndrome is a common denominator of metabolic diseases, the objective of this study was to develop a predictive model of metabolic syndrome.
METHODS
Participants
Initially, the study comprised 156 individuals (males: n = 101, 75.4%; females: n = 33, 24.6%) aged ≥ 30 years with severe physical disabilities, specifically those diagnosed with poliomyelitis or spinal cord injury, who had been officially classified as having a severe level of disability under the Korean National Disability Rating System. All participants had been using a wheelchair for over 1 year following a physician’s diagnosis and were utilizing welfare and fitness facilities for individuals with disabilities in Provinces C or G. Subsequently, 22 individuals were excluded during data processing owing to difficulties in measuring physical fitness (n = 8) or collecting blood (n = 14); the remaining 134 participants were included in the final data analysis. All participants received a detailed explanation of the study’s purpose, methods, and procedures and provided consent to participate in the study. This study was approved by the institutional review board of C University (7001066-202401-HR-009).
Measured variables
Body composition and anthropometry
With the patient lying supine, height was measured in centimeters using a tape measure from the soles of the feet to the vertex of the head. Body weight was measured in kilograms using wheelchair scales (AD-6105NP; AND, Seoul, Korea) by weighing the participants in the wheelchair and then subtracting the weight of the chair. Body mass index (BMI) was calculated using the following formula: weight (kg)/height (m²). Body fat percentage was measured using bioelectrical impedance analysis with a body composition analysis device (S10; Inbody, Seoul, Korea), with the participant lying down and removing all metal objects. Finally, neck circumference was measured in centimeters at the widest part, located below the laryngeal prominence, with the participant sitting in a wheelchair [17].
Physical activity
Physical activity was measured using the Korean version of the Physical Activity Scale for Individuals with Physical Disabilities, a questionnaire designed to assess physical activity in individuals with physical disabilities and adapted to the cultural characteristics of South Korea [18,19]. The questionnaire comprised 13 items for assessing leisure-time physical activity and physical activity related to occupation (e.g., household chores and work). Each item was measured based on the weekly frequency and time per session. The results were calculated in units of metabolic equivalents per hour per day, and the results of leisure-time physical activity (LTPA), occupational physical activity, and total physical activity were analyzed.
Physical fitness
Physical fitness was measured using the same methods employed at fitness testing centers for individuals with disabilities, which are part of sports welfare services aimed at increasing fitness and improving health among individuals with disabilities. The measurements were divided into muscular strength, muscular endurance, cardiorespiratory endurance, and flexibility. Muscular strength was measured twice per hand using a dynamometer (TKK-5401; Takei, Tokyo, Japan), and relative grip strength was calculated by inserting the maximum measurement into the formula grip strength (kg)/BMI (kg/m²). Muscular endurance was measured as the number of arm curl repetitions performed in 2 min using dumbbells (4 kg for males and 2 kg for females). Flexibility was measured in centimeters as the distance between the two hands in Apley’s scratch test. To measure cardiorespiratory endurance, a 20 m straight-line interval was set up, and the distance traveled by the wheelchair in 5 min up and down the course was measured in meters.
Definition and measurement of risk factors for metabolic syndrome
Waist circumference was measured using a tape measure at the midpoint between the inferior margin of the ribs and the superior margin of the iliac crest. The mean of the two measurements was calculated. Blood pressure (BP) was measured using an automatic sphygmomanometer (HEM-7156; Omron, Kyoto, Japan) after at least 10 min of rest. After two measurements, mean systolic BP (SBP) and diastolic BP (DBP) were calculated. To measure blood glucose, triglycerides (TG), and high-density lipoprotein cholesterol (HDL-C) levels, a lancet was used to collect 70 μL of whole blood from the thumb on the non-dominant hand after at least 10 h of fasting. Blood samples were placed in a Biochemistry Test 9 cartridge and analyzed using a spectrophotometric blood lipid analyzer (Exida PT 10; Precision Biosensor, Daejeon, Korea). Based on the diagnostic criteria for metabolic syndrome proposed by the National Cholesterol Education Program Adult Treatment Panel III and the Korean Society for the Study of Obesity, which includes waist circumference thresholds, metabolic syndrome was defined as meeting at least three of the following criteria [20,21].
1) Waist circumference: males ≥90 cm, females ≥85 cm
2) BP: SBP ≥130 mmHg, or DBP ≥85 mmHg, or taking medications for blood pressure regulation
3) Fasting blood glucose: ≥100 mg/dL or taking medications for blood glucose regulation
4) TG: ≥150 mg/dL, or taking medications for lipid regulation
5) HDL-C: males <40 mg/dL, females <50 mg/dL
Definitions of covariates
Factors investigated as covariates included disability characteristics, such as duration of injury and type of disability, and health-related factors, such as smoking status, drinking, menopause, and number of medications. The duration of injury was calculated as the time between the first diagnosis of the disability and the time of the survey. Disability was categorized as either spinal cord injury or poliomyelitis based on the medical diagnosis. Smoking status was categorized in terms of current smoking status and lifetime smoking history of at least five packs [22], while binge drinking was defined as having at least seven drinks in a sitting position for males and at least five drinks for females [23].
Data processing
All continuous variables are expressed as the mean and standard deviation; all categorical variables are expressed as frequencies (n) and percentages (%). To develop a prediction model for metabolic syndrome, participants were randomly allocated to either the derivation set (70%) or the validation set (30%). Independent sample t-tests and chi-square tests were used to verify the homogeneity of the two groups. In the derivation set, differences between variables were analyzed between those with and without metabolic syndrome. To develop an equation to predict the number of risk factors for metabolic syndrome, the derivation set was subjected to stepwise multiple linear regression analysis; a Bland-Altman plot was used to verify the criterion validity of the prediction model developed from the derivation set; and cross-validation was performed using the validation set. Finally, the area under the curve (AUC) was calculated using receiver operating characteristic (ROC) curve analysis to assess the predictive ability of the model in identifying metabolic syndrome. Data analyses were performed using SPSS-PC ver. 28.0 (IBM Corp., Armonk, NY, USA), and the statistical significance level for hypothesis testing was α = 0.05.
RESULTS
Comparison of variables between the derivation and validation sets
Table 1 shows a comparison of variables used to verify the homogeneity between the derivation and validation sets. Except for age (p = 0.033), the two groups did not differ significantly in any variable, indicating that the overall characteristics of the two groups were similar.
Comparison of variables using metabolic syndrome status
Table 2 provides a comparison of variables in the derivation set depending on metabolic syndrome status. Significant differences were observed between the groups in terms of body weight (p < 0.001), BMI (p < 0.001), body fat percentage (p < 0.001), neck circumference (p < 0.001), number of medications (p = 0.016), LTPA (p = 0.039), and muscular strength (p = 0.035).
Development of an equation to predict the number of risk factors for metabolic syndrome
Table 3 provides the results of the equation predicting the number of risk factors for metabolic syndrome using stepwise multiple linear regression analysis in the derivation set. In addition to age, neck circumference was selected as an explanatory variable among the body composition indices that showed significant differences based on metabolic syndrome status, given the ease in measurement and ability to account for multicollinearity. Regarding other factors, the number of medications, LTPA, and muscular strength were included, as they differed significantly depending on metabolic syndrome status. The multiple correlation coefficient was 0.630, and the explanatory power of the model was 39.7% (p < 0.001). The Durbin-Watson statistic was 1.936, confirming independence between the residuals, and the variance inflation factors between the variables did not exceed the criterion, indicating that there were no problems with multicollinearity. The final equation for estimating the number of risk factors for metabolic syndrome was as follows:
Number of metabolic syndrome risk factors: -3.523 + 0.188 (Neck circumference [cm]) + 0.211 (Number of medication) - 0.019 (LTPA [METs-h/day) - 0.613 (Muscular strength [kg/BMI]).
Validity testing of the equation
Figure 1 illustrates the results of criterion validity testing in the derivation and validation sets using a Bland-Altman plot and cross-validation. The mean difference between the predicted and actual values in the derivation set was close to zero, indicating that the predictions were stable and that there was no systematic bias. Most of the observed values were within 95% limits of agreement, indicating that the model demonstrated high agreement with the actual values. Similar trends were observed in the validation set; the mean difference was small, and overall, the distribution remained within the limits of agreement, indicating an excellent predictive ability.
Validation of the performance of the equation for predicting metabolic syndrome
Figure 2 illustrates the results of the ROC analysis for the derivation and validation sets, assessing the ability to predict the presence or absence of metabolic syndrome based on an equation estimating the number of risk factors for metabolic syndrome. The AUC in the derivation set was 0.867 (95% confidence interval [CI] = 0.796−0.937; p < 0.001), demonstrating excellent predictive ability, while the AUC in the validation set was 0.765 (95% CI = 0.617−0.913; p = 0.009), showing good predictive ability. The results were statistically significant for both sets. These results suggested that the equation developed in this study could be used effectively to predict metabolic syndrome in individuals with physical disabilities.
DISCUSSION
In the current study, we aimed to develop a predictive model for metabolic syndrome based on physical activity and fitness using data from 134 individuals aged ≥ 30 years with severe physical disabilities for at least 1 year. After randomly dividing the participants into derivation and validation sets, an equation was derived from the derivation set to predict the number of risk factors associated with metabolic syndrome. The criterion validity of the equation was analyzed in both sets, and cross-validation was performed. Both sets exhibited a strong agreement between the actual measured number of risk factors and the number of risk factors derived from the equation. Upon analyzing the performance of the equation for metabolic syndrome screening, the AUCs for the derivation and validation sets were 0.867 and 0.765, respectively, indicating excellent and good performance.
Rapid lifestyle changes owing to advances in science and technology have led to a sharp increase in the prevalence of chronic diseases. Accordingly, various predictive models have been developed and utilized for the early identification and prevention of chronic diseases in clinical practice [13,14]. However, these predictive models were limited to individuals without disabilities. Individuals with physical disabilities are susceptible to chronic diseases because their disabilities increase their risk of developing obesity and inadequate physical activity, and access to healthcare facilities is restricted [4]. Physical activity and fitness are closely related to chronic diseases, and models that include these factors have shown improved predictive ability [9-12]. Based on these findings, we performed a stepwise linear regression analysis on individuals with physical disabilities to establish a prediction model for estimating the number of risk factors associated with metabolic syndrome, which is known to be a common denominator for chronic diseases. Regression analysis identified neck circumference, the number of medications, LTPA, and muscular strength as explanatory variables. These results can be interpreted in light of previous studies, which reported correlations between these variables and metabolic syndrome.
First, neck circumference is an index that reflects subcutaneous fat in the upper body and is closely related to risk factors for metabolic syndrome, such as visceral fat and insulin resistance [24]. Clinical studies have demonstrated that neck circumference is an independent predictor of metabolic syndrome, both in individuals with and without physical disabilities [25,26]. Second, although only a few studies have directly examined the relationship between the number of medications and metabolic syndrome, increased medication use reportedly correlates with a worse prognosis for metabolic syndrome [27]. Finally, several clinical and epidemiological studies have demonstrated that LTPA and fitness are independent protective factors, reducing the risk of metabolic syndrome by improving insulin sensitivity and metabolic function [26,28,29]. These reports support the importance of the variables included in the equation derived in the current study. Our findings suggest that considering these variables in conjunction could play an important role in predicting metabolic syndrome in individuals with physical disabilities using convenient and non-invasive methods.
The study participants were divided in a 7:3 ratio into derivation and validation sets, respectively, and the derived equation was applied to each set. After examining its criterion validity, cross-validation was performed, followed by ROC analysis to verify the performance of the equation for predicting metabolic syndrome. In the validity analysis, we observed a strong agreement between the actual values and values calculated from the equation in both sets. Upon testing the performance in predicting metabolic syndrome, the AUCs for the two sets were 0.867 and 0.765, indicating excellent and good discrimination, respectively.
These findings are consistent with those previously reported. For example, Tavares et al. developed a non-invasive predictive model for metabolic syndrome using machine learning based on data from 18,182 Brazilian adults. The authors reported excellent discrimination using a model that included anthropometric factors and physical activity [30]. Additionally, Kubutsor et al. incorporated grip strength into two previous models for predicting diabetes and reported an improved predictive ability [16]. In the current study, by similarly testing the validity and performance of a model based on anthropometric factors, physical activity, and grip strength in individuals with physical disabilities, we extended the results of previous studies on individuals without disabilities, reporting excellent discrimination by models that incorporated anthropometric factors, physical activity, and fitness to include individuals with physical disabilities. By demonstrating the validity and performance of the predictive model, we believe this study will contribute to the identification of risk factors for metabolic syndrome in individuals with physical disabilities using convenient and non-invasive methods.
Recently, machine learning and artificial intelligence approaches have garnered attention for their potential to identify complex patterns and improve predictive accuracy in disease risk modeling [31]. However, these methods are frequently prone to overfitting when applied to small datasets, which can limit their reliability [32,33]. Conversely, regression-based models are grounded in well-established statistical theory, offer greater interpretability, and are considered more suitable for studies involving populations such as individuals with physical disabilities, where data availability is often limited and clinical applicability is crucial [33]. Given the relatively small sample size and the need for practical interpretation, a stepwise multiple linear regression model was deemed appropriate. The model demonstrated a strong predictive performance and extended the findings of previous studies on non-disabled populations to individuals with physical disabilities. Future research should utilize larger datasets and apply machine learning techniques to enhance predictive accuracy, as well as to explore potential nonlinear relationships among variables in this population.
This study has several limitations. First, individuals with physical disabilities who participated in this study were limited to those with spinal cord injury or poliomyelitis who used a wheelchair. Therefore, our findings cannot be generalized to individuals with other types of physical disabilities, including amputations. Second, because the validity of the derived model was tested using a relatively small sample size, statistical stability may be limited. Therefore, additional validity testing based on external data with a larger sample size is required. Third, because this was a cross-sectional survey, it has limited utility in terms of establishing long-term predictions on the risk of metabolic syndrome or analyzing causal relationships. Therefore, additional investigations are needed to overcome these limitations and use a longitudinal design to examine various disability-related characteristics in a larger sample of individuals with physical disabilities. Fourth, the regression model developed in this study explained approximately 40% of the variance in the number of metabolic syndrome risk factors (R2 = 0.397), indicating moderate explanatory power. This suggests that the model may not fully capture all the relevant factors impacting the risk of metabolic syndrome. Fifth, several potentially important variables, such as dietary habits, psychological stress, and side effects of medications, were not included in the analysis and may have influenced the outcomes. Future studies should consider incorporating such variables to enhance the accuracy and comprehensiveness of the predictive models.
PDF Links
PubReader
ePub Link
XML Download
Full text via DOI
Full text via PMC
Download Citation
