Introduction

Chronic kidney disease (CKD) is a major global noncommunicable disease, projected to become the fifth leading cause of death by 20401. In China, the prevalence of CKD reaches 10.8%, affecting approximately 82 to 132.3 million people2,3. Maintenance hemodialysis (MHD) is the mainstay therapy for end-stage kidney disease but does not cure the disease. MHD patients often suffer from comorbidities and nutritional issues, including reduced muscle protein synthesis and increased catabolism, which contribute to skeletal muscle wasting and a high prevalence of sarcopenia4.

Sarcopenia is a progressive disorder marked by a decline in muscle mass, strength, and function5,6, and is associated with adverse outcomes in MHD patients, including higher mortality, cardiovascular events, and medical costs7. With the global rise in the MHD population due to population aging, early identification and intervention for sarcopenia have become increasingly important. Although diagnostic criteria proposed by the Asian Working Group for Sarcopenia (AWGS)8 and the European Working Group on Sarcopenia in Older People (EWGSOP)9 are widely used, their applicability in MHD patients is limited due to fluid shifts and electrolyte imbalances that may compromise muscle mass assessments. Therefore, exploring novel biomechanical diagnostic tools may enhance screening accuracy and reduce the burden of sarcopenia in this population.

Sarcopenia leads to muscle weakness and functional decline, resulting in reduced physical performance and daily life abilities, impacting body balance, gait stability, and substantially increasing the risk of falls and mortality9,10.Numerous studies have confirmed a strong correlation between sarcopenia and fall risk11,12,13. A meta-analysis of 33 studies involving over 45,000 individuals found that sarcopenia increased the risk of falls and fractures14, especially in men15, and this risk correlated with sarcopenia severity16.

Plantar pressure analysis has become a widely used tool for assessing balance and fall risk. Among its parameters, the center of pressure (COP) trajectory is one of the most informative. COP displacement reflects postural control and balance capacity, offering insights into how the body maintains stability during standing and movement. Research has shown that greater COP trajectory length and atypical pressure distribution are associated with higher fall risk17,18,19.

Several studies have investigated the relationship between sarcopenia and plantar pressure distribution. One study involving 239 postmenopausal women found that those with sarcopenic obesity exhibited higher peak pressure in the metatarsal, midfoot, and heel regions20. Some researchers conducted gait speed and plantar pressure tests in 20 older individuals aged 65 and above. They found that all participants had a gait speed lower than 1.0 m/s, with greater pressure in the right foot compared to the left foot, and greater pressure in the heel compared to the forefoot21. Exercise interventions in sarcopenic obese individuals have been shown to significantly reduce the area and velocity of the COP22. Furthermore, sarcopenic individuals tend to have uneven bilateral foot pressure distribution, with notable asymmetries in the toes and metatarsal regions, a reduced COP stability limit, and increased lateral sway23. Despite these findings, few studies have specifically explored plantar pressure in MHD patients with sarcopenia, and existing evidence remains primarily descriptive, lacking in-depth mechanistic analysis.

Previously, we developed gender-specific sarcopenia screening tools for MHD patients using machine learning, based on laboratory and physical measurements24. However, these models did not incorporate indicators of postural balance. Building on this foundation, the present study adopts AWGS 2019 criteria8 and applies plantar pressure assessments combined with machine learning techniques to develop an auxiliary screening model for sarcopenia in MHD patients. The study also investigates gender differences, aiming to provide biomechanical insights into sarcopenia screening and improve fall prevention strategies in this population.

Methods

Study design and participants

This cross-sectional study enrolled patients living on MHD who visited to the Wenjiang Hemodialysis Center in the Department of Nephrology in West China Hospital, Sichuan University, Chengdu, China between September and December 2023. The inclusion criteria were (1) patients receiving MHD, (2) at least 12 weeks of MHD treatment (2–3 sessions/week) and plan to continue MHD treatment during the study period, and 3) ≥ 18 years of age. The exclusion criteria were (1) have skeletomuscular system deformity, (2) dyskinesia, (3) cardiac pacemakers/ICD installed, or (4) psychiatric disorders/single-leg amputation. Sarcopenia diagnosis was carried out via AWGS 20198.

Ethical and legal considerations

This study received approval from the Ethics Committee of Sichuan University (ethical approval number: 2023 [1425]) and was performed in accordance with the principles of the Declaration of Helsinki. Written informed consent was obtained from all study participants and participants were informed that they could refuse to participate at any stage.

Comprehensive physiological indicators data collection

The collected data includes the patient’s basic information, body measurements, and laboratory test results. Basic information was obtained from patients’ medical archives, while body measurement results and laboratory findings were derived from the latest centralized examination at the hemodialysis center before data collection. Laboratory findings included routine blood examination, hepatic and renal function indicators, serum inorganic salts, and parathyroid hormone (PTH). Since urea and creatinine were measured again after hemodialysis, participants’urea and creatinine after hemodialysis were also collected.

Assessments of sarcopenia

Sarcopenia was assessed by two trained investigators according to the updated 2019 consensus of the AWGS 20198, including measurements of muscle strength, physical performance, and body composition. (1) Muscle strength was evaluated by handgrip strength using an electronic grip dynamometer (Zhongshan Camry Electronic Co., Ltd.). Participants stood upright with arms naturally hanging down and elbows fully extended. To avoid complications, measurements were performed on the arm without vascular access. Grip strength was measured three times at one-minute intervals, with the highest value recorded. (2) Physical performance was assessed using the Short Physical Performance Battery (SPPB), which includes balance tests, a 4-meter usual gait speed test, and a five-time chair stand test. The total score ranges from 0 to 12, with lower scores indicating worse physical function. Both handgrip strength and SPPB were measured before dialysis on a mid-week dialysis day to account for interdialytic weight changes and minimize the influence of fluid fluctuations. (3) Body composition was measured using a bioelectrical impedance analysis (BIA) device (InBody S10, InBody Co., Ltd., Cheonan, Republic of Korea) within two hours after the mid-week dialysis session to minimize variability due to fluid shifts. Measurements were taken with participants barefoot, standing on an insulated mat in light clothing, with arms positioned at approximately a 15° angle from the torso and legs shoulder-width apart. Contact electrodes were attached to the thumb and middle finger, including in patients with arteriovenous fistulas. The skeletal muscle mass index (SMI) was calculated by dividing appendicular skeletal muscle mass by height squared (kg/m²).

Plantar pressure testing

Despite the visual system’s contribution to static balance control, given that the human body is typically in an open-eyed state during most physical activities, visual input is present. Consequently, this study considered two distinct stances, feet-together and feet-separated by 10 cm, to investigate the utility of plantar pressure indicators in sarcopenia screening.

1) Feet-together stance test: Participants assumed a feet-together stance, with arms naturally at their sides and maintaining an upright posture. The waveforms illustrating pressure fluctuations in both feet were observed, and upon achieving pressure stability, the test data were recorded. The test duration was 30 s.

2) Feet-separated by 10 cm stance test: Participants adopted a feet-separated stance with a 10 cm separation, ensuring parallel alignment of the foot arch axis. Arms remained naturally at the sides, maintaining an upright posture. The waveforms illustrating pressure fluctuations in both feet were observed, and upon achieving pressure stability, the test data were recorded. The test duration was 30 s.

The plantar surface is segmented into 10 distinct zones (see Fig. 1) for data acquisition, each with specific nomenclature and abbreviations as detailed below: hallux (H), toe 2 to toe 5 (T), metatarsal 1 (M1), metatarsal 2 (M2), metatarsal 3 (M3), metatarsal 4 (M4), metatarsal 5 (M5), mid foot (MF), heel medial (HM), and heel lateral (HL).

Fig. 1
figure 1

The plantar pressure partition diagram23.

Full size image

Plantar pressure was measured using a pressure plate manufactured by Zebris Medical GmbH (see Fig. 2). The plate measures 710 mm in length, 400 mm in width, and 15 mm in thickness, with a working frequency of 200 Hz. The measurement system includes distributed force sensors and a corresponding computer software, capable of measuring dynamic balance during walking and running as well as static balance while standing. Prior to testing, the test environment underwent a preliminary inspection to ensure the tablet is positioned on a flat surface, devoid of any objects beneath it and free from obstructions within the sensor’s operational area on the plate’s upper surface. Plantar pressure assessments were conducted before the commencement of the patients’ hemodialysis treatment. After the test, the data was archived and labeled in accordance with the test stance. Utilizing the Zebris FDM software, comprehensive reports were generated, enabling the scrutiny of plantar pressure distributions, the trajectory length, mean movement velocity and deviation distance of COP.

Fig. 2
figure 2

The sample graph of plantar pressure plate and dimensional labeling (mm).

Full size image

Grouping and statistical analysis

Similar to Hassler’s work25, this study amalgamates MHD patients’ exhibiting varying degrees of sarcopenia into a single sarcopenia group, encompassing possible sarcopenia, sarcopenia, and severe sarcopenia. Conversely, patients devoid of sarcopenia constitute the control group. This methodology is designed to construct a binary classifier, facilitating the discernment of sarcopenia in newly MHD patients. The integration of these groups not only streamlines the model but also diminishes the requisite data volume, thereby augmenting the model’s screening efficacy.

Statistical analysis was performed using SPSS version 26.0 software (IBM Corp., Armonk, NY, USA). The normality of continuous variables was assessed using the Shapiro-Wilk test. Data with a normal distribution are expressed as mean ± standard deviation (SD), while non-normally distributed data are presented as median (Q1, Q3). Categorical variables are summarized as frequency (N) and percentage (%). Independent t-tests were used for normally distributed variables, and Mann-Whitney U tests were applied to non-normally distributed variables to compare the plantar pressure data between the control groups and sarcopenia groups. A p-value of < 0.05 was considered statistically significant.

All categorical outcomes were converted into binary numeric values for further analysis, with sarcopenia cases coded as “1” and control cases as “0”. The initial dataset of plantar pressure metrics comprises 49 attributes (including gender but not the sarcopenia diagnostic outcome). Owing to discomfort or other factors encountered by certain patients during the assessment, several test postures were left incomplete, leading to 25 features and 40 instances in the original dataset containing missing entries, aggregating to 552 (6.12%) missing data. Consequently, to address these gaps, this study extrapolated missing values from the extant data to maximally retain potentially informative content. For instances exhibiting a single missing value within a feature, comparable instances aligned with the same gender and sarcopenia condition were identified. Subsequently, these missing values were interpolated using the mean of the corresponding feature absent of missing data from the selected instances. After interpolation, the data format was standardized to mirror the original format.

Feature selection

Feature importance was determined using both the ranking of importance scores and the absolute values of the Lasso regression weights. The top features were selected based on their average ranking across these two metrics. Feature importance calculation and lasso regression programs were all performed via Python 3. To assess the utility of plantar pressure indicators in the auxiliary screening of sarcopenia among MHD patients, the methodology employed in this study mirrors that utilized in our previous study24. Throughout the model’s developmental phase, this research adhered to the application of ten prevalent binary classification machine learning algorithms, encompassing K-Nearest Neighbor (KNN), Naive Bayes (NB), Logistic Regression (LR), Support Vector Machine (SVM), Multi-layer Perceptron (MLP), and an array of tree-based methodologies: Decision Tree (DT), Random Forest (RF), Adaptive Boosting (AdaBoost), Gradient Boosting Decision Tree (GBDT), and Light Gradient Boosting Machine (LGBM). The execution of these machine learning model development protocols was facilitated through the utilization of Python 3, with algorithm implementations drawn from the sklearn and lightgbm libraries.

Subsequent to the modeling process confined to the optimal plantar pressure indicators identified for sarcopenia screening within the MHD patient cohort, this investigation further scrutinized the screening efficacy of plantar pressure indicators across varying stances. Concurrently, we developed a gender-independent auxiliary screening model for sarcopenia by integrating the newly established optimal set of comprehensive physiological indicators, which were not included in our previous study24, thereby evaluating the synergistic contribution of plantar pressure indicators in conjunction with comprehensive physiological indicators to the auxiliary screening of sarcopenia.

Results

We conducted a statistical analysis of the demographic variables and body measurements of all 184 participants included in this study, stratified by mixed-gender, male, and female groups (Table 1). In the mixed-gender group, the proportion of female patients was higher in the sarcopenia group than in the control group (p = 0.003). Patients in the sarcopenia group were consistently older than controls across all stratifications (p < 0.001). In the mixed-gender group, sarcopenic patients had lower height and weight than controls (p = 0.001 and p = 0.017, respectively), although these differences were not significant in the sex-stratified analyses. No significant differences were observed in the prevalence of hypertension, while diabetes was more prevalent in sarcopenic patients in the mixed-gender and male groups (p = 0.001 and p < 0.001, respectively), but not in females. Sarcopenia-related indicators, including handgrip strength, SPPB scores, and SMI, were lower in the sarcopenia group across all stratifications. In addition, pre- and post-dialysis creatinine levels and total body water (TBW) were lower in sarcopenic patients compared with controls (all p < 0.05). Gender-stratified analyses showed consistent trends, although TBW differences did not reach statistical significance in males (p = 0.114) or females (p = 0.073).

Table 1 Statistical analysis results of basic characteristics of study group.
Full size table

Statistical analysis of plantar pressure indicators in three gender groups

Descriptive statistical analyses were performed on the average plantar pressure and COP-related indicators for each plantar region of two stances, feet-together and feet-separated by 10 cm. The findings for patients of mixed-gender group are presented in Supplementary Table 1, for male patients in Supplementary Tables 2, and for female patients in Supplementary Table 3. The unadjusted statistical analysis indicated that the gender differences and nine plantar pressure indicators among mixed-gender group were statistically significant (P < 0.05). Specifically, three plantar pressure indicators showed statistically significant differences (P < 0.05) among male MHD patients and eleven indicators among female MHD patients.

Feature selection and ranking of three gender groups

The feature importance and absolute weight of lasso regression for 49 features in each gender group were calculated. Based on features’ importance and weight values, the average ranking was calculated and sorted in descending order (see Supplementary Table 4). Among the top-ranked plantar pressure indicators, there were several average plantar pressure indicators under different stances and foot regions, but no COP-related indicators. To establish a simple and efficient machine learning screening model, the combination of the top 2 to 5 features and the top 2 to 5 features with statistically significant differences between the groups (P < 0.05) in Table 2 were selected. By comparing the performance of the developed models, the optimal set of plantar pressure indicators for screening sarcopenia in MHD patients was found. Additionally, modeling with all the features was also included to compare and validate whether using only a few top-ranked features is superior to using all feature data.

Table 2 Optimal plantar pressure feature sets used in voting classifiers and their evaluation results of three gender groups.
Full size table

The comparison of the effectiveness of voting classifiers using different feature sets in three gender groups

For the mixed-gender group, after discarding overfitting or underfitting models, the evaluation results of voting classifiers using the top 2 to 5 features, statistically significant top 2 to 5 features (P < 0.05), and all plantar pressure features are presented in Supplementary Table 5. Notice that the top 3 features in Supplementary Table 4 had statistically significant differences between the two groups, thus this study compared seven sets of plantar pressure indicator features. The sensitivity, specificity, F1 score, and the area under the receiver operating characteristic curve (AUC) of the voting classifiers using these seven feature sets were plotted into a box plot, as shown in Fig. 3.

Fig. 3
figure 3

The box plot of evaluation results for each voting classifier using 7 plantar pressure feature sets after feature selection of the mixed-gender group.

Full size image

The results indicate that the average accuracy difference of the classifiers using the top 3 to 5 features (P < 0.05) was within 10%, while it exceeded 10% for the other classifiers. The classifier using all features performed the worst, while the classifier using the top 2 features achieved the highest precision but lower sensitivity. Classifiers using the top 3 to 5 features balanced sensitivity and specificity effectively. Overall, the classifier using the top 5 features in Supplementary Table 4 is most suitable for assisting in sarcopenia screening in mixed-gender MHD patients due to its minimal accuracy difference, high F1 score, and AUC.

For the male group, after discarding overfitting or underfitting models, Supplementary Table 6 presents the evaluation results of voting classifiers utilizing various feature sets, with Fig. 4 illustrating their sensitivity, specificity, F1 score, and AUC through the box plot.

Fig. 4
figure 4

The box plot of evaluation results for each voting classifier using 9 plantar pressure feature sets after feature selection of the male group.

Full size image

For male MHD patients, the generalization capabilities of the classifiers decreased, with some instances of overfitting observed. The classifiers using all features showed an average accuracy difference exceeding 35%, whereas those using the top 2 features maintained this difference within 10%. The classifier using the top 3 features (P < 0.05) demonstrated the best performance in terms of average precision, specificity, F1 score, and AUC, despite a reduced generalization capacity likely due to the relatively small sample size. Thus, this classifier is optimal for assisting in sarcopenia screening in male patients.

For the female group, after discarding overfitting or underfitting models, Supplementary Table 7 presents the evaluation results of each voting classifier utilizing various feature sets, with Fig. 5 illustrating their sensitivity, specificity, F1 score, and AUC through the box plot.

Fig. 5
figure 5

The box plot of evaluation results for each voting classifier using 9 plantar pressure feature sets after feature selection of the female group.

Full size image

For female MHD patients, the selection and modeling of plantar pressure characteristics yielded better results. Except for the classifier using all features, the average accuracy difference of the remaining classifiers was maintained within approximately 10%. The classifier using the top 3 features achieved the optimal balance in sensitivity and specificity, and its overall performance was notably high. Therefore, the classifier using the top 3 features in Supplementary Table 4 is the best choice for assisting in sarcopenia screening in female MHD patients.

The combined average approximate ROC curves of voting classifiers for the three gender groups were integrated into a single chart to facilitate comparison of model performance, as shown in Fig. 6.

Fig. 6
figure 6

Combined average approximate ROC curves of voting classifiers of three gender groups. The solid line represents the mean value of sensitivity, and the light area can be regarded as SD values.

Full size image

The optimal set of plantar pressure indicator features and the evaluation results of the voting classifier for the three gender groups are shown in Table 2. By comparing the results of the three gender groups, it can be observed that the auxiliary screening model for female patients performs the best, followed by the mixed-gender model, while the results for male patients are the worst.

Results of the sarcopenia auxiliary screening model based on plantar pressure indicators under different stances

During plantar pressure test, this study set two stances, feet-together and feet-separated by 10 cm, to test the distribution of plantar pressure. Therefore, in this study, the plantar pressure indicators under these two stances were combined with the gender feature for feature selection and model evaluation, to explore the effect of different stances on the auxiliary screening of sarcopenia in MHD patients, as shown in Table 3. Comparing the results in Table 3, it is evident that considering only the three plantar pressure indicators of feet-together stance, the performance of the optimal model surpasses that of using features from both stances. In contrast, considering only the plantar pressure indicators of feet-separated by 10 cm stance and gender leads to a decline in the model’s classification performance.

Table 3 Optimal features used and evaluation results of voting classifiers that combine plantar pressure indicators of different stances with the gender feature.
Full size table

Results of the model using the optimal comprehensive physiological indicators combined with plantar pressure indicators

This study utilized previously collected data in our previous study to recalculate the optimal sarcopenia auxiliary screening model using comprehensive physiological indicators without gender distinction24. Although both pre- and post-dialysis blood test data were collected in the initial dataset, only pre-dialysis creatinine and age were retained in the final model. This is because these two variables were selected through feature selection as the most relevant comprehensive physiological indicators for sarcopenia screening. The model achieved an average AUC of 79%. We further explored the effect of combining these two selected physiological indicators with plantar pressure features of the feet-together stance (mean pressure on M3, M4, and MF of the left foot). Through machine learning modeling and evaluation, the results of the auxiliary sarcopenia screening model and comparisons with models using either set of features alone are presented in Table 4. The integrated model demonstrated enhanced performance in terms of reduced accuracy difference, improved generalization, higher average evaluation metrics, lower standard deviations, and an average AUC approaching 90%.

Table 4 Evaluation and comparative results of the auxiliary sarcopenia screening model using optimal composite physiological indicators combined with plantar pressure indicators.
Full size table

Discussion

This study investigated the relationship between sarcopenia and plantar pressure distribution in MHD patients. The results (see Supplementary Table 2) revealed that male patients with sarcopenia had significantly higher COP path length and average movement speed in the feet-together stance, as well as higher average pressure in the midfoot of the right foot during the feet-separated by 10 cm stance, indicating poorer balance conditions. For female patients with sarcopenia, the study found that (see Supplementary Table 3) during feet-together stance, there was an increase in average pressure in the midfoot of both feet and a decrease in average pressure in heel medial compared to control group, suggesting a transfer of pressure from heel medial to midfoot. During feet-separated by 10 cm stance, female patients exhibited a decrease in average pressure in heel medial and heel lateral of the left foot, but an increase in average pressure in the midfoot of the right foot, as well as increased path length and movement speed of COP. These results indicate that female patients with sarcopenia have poorer balance conditions and higher risk of falls compared to males. This may be attributed to weaker muscle strength in females26, slower reaction times27, poorer performance in functional balance tests28, and higher fall rates29. Overall, patients with sarcopenia had higher path length and average movement speed of COP during standing compared to non-sarcopenia patients, indicating poorer balance ability and the need for greater adjustments to maintain balance.

This study effectively differentiated between MHD patients with and without sarcopenia by comparing the distribution of plantar pressure among three gender groups. The results showed that the screening model for the female group had the best performance, followed by the mixed-gender model, while the model for the male group performed the worst. Both the male and female models used three features and one machine learning algorithm. However, the classification performance was significantly better for the female group. This phenomenon may be attributed to the smaller proportion of sarcopenia in male MHD patients recruited in this study. Moreover, considering the physiological differences between males and females, the impact of sarcopenia on plantar pressure in male patients may not be as pronounced as females, resulting in less satisfactory model performance compared to the other two gender groups. In addition, Supplementary Table 2 shows that the number of statistically significant features between the two male MHD patient groups was the lowest. When combining both male and female groups, the model required additional features and machine learning algorithms to achieve good generalization, but its specificity was lower compared to the other two models for single gender. However, in order to obtain a sarcopenia screening model applicable to the entire population of MHD patients rather than exclusively accurate screening for certain subgroups, this study finally selected the mixed-gender model to assist in screening. The study suggests that the top 5 plantar pressure features (the mean pressure on M3 and MF of left foot in a feet-together stance, and the mean pressure on HM of left foot and M1 and MF of right foot in a feet-separated by 10 cm stance) are considered most suitable for assisting in the screening of sarcopenia in MHD patients. However, due to the lack of numerical simulation and modeling on foot forces in sarcopenic patients, a detailed interpretation of the significance of each feature included in the optimal feature set cannot be provided at present.

The study on the auxiliary sarcopenia screening model based on the plantar pressure indicators under different stances shows that using the plantar pressure indicators of the feet-together stance provided the best screening effect, while the indicators of the feet-separated by 10 cm stance performed the worst. It is well known that a wide stance increases stability and can improve posture control30. This viewpoint has also been confirmed in diabetes patients31. Therefore, in this study, the stance with the feet-separated by 10 cm helped patients maintain balance better, and they are less likely to fall, resulting in biased effects of the auxiliary screening model. On the contrary, the feet-together posture made patients more prone to instability, thereby revealing the impact of sarcopenia on their body balance. In this case, machine learning models are more easily identified the relationship between various data features and sarcopenia from the patients’ plantar pressure data and classify them. Therefore, when using plantar pressure indicators for the auxiliary screening of sarcopenia in MHD patients, only feet-together stance needs to be considered, and there is no need to test the distribution of plantar pressure when the feet are separated by 10 cm. This can save testing costs and improve the convenience of the model.

In addition to plantar pressure features, comprehensive physiological indicators were considered in constructing the auxiliary sarcopenia screening model. As shown in Table 1, sarcopenic patients had significantly lower pre- and post-dialysis creatinine levels compared to controls, with similar trends across gender groups. TBW was also lower in sarcopenic patients; however, the differences were not statistically significant in male or female subgroups and TBW was not included in the final screening model. Based on our previous modeling work, age and pre-dialysis creatinine were identified as the most informative physiological features for sarcopenia screening, reflecting reduced muscle mass in affected patients. When combined with plantar pressure indicators, the auxiliary screening model demonstrated further improved effectiveness, highlighting the complementary role of physiological measurements in assisting sarcopenia detection. Therefore, under clinically permissible conditions, assessing plantar pressure in the feet-together stance together with age and pre-dialysis creatinine provides a practical and effective approach for sarcopenia screening in MHD patients.

Limitations

This study has several limitations. First, it was not a large-scale cross-sectional study, which may limit the generalizability and representativeness of the findings. Second, to facilitate the development of the screening model, we merged possible sarcopenia, sarcopenia, and severe sarcopenia into a binary classification. Although this approach improved model performance, it precluded the differentiation of sarcopenia severity, potentially introducing misclassification bias and limiting clinical interpretability. Third, we applied only a single diagnostic framework (AWGS 2019), which may have led to an underestimation of sarcopenia prevalence. Validation using multiple diagnostic criteria and across multi-center or multi-regional populations is needed to enhance the robustness of the findings. Fourth, diabetes, one of the common comorbidities of MHD patients, had a statistically significant difference between the sarcopenia group and control group in mixed-gender and male group (see Table 1). However, in our previous study, comorbidity indicators like diabetes were excluded after feature selection so we didn’t add diabetes indicator to our screening model24. Comorbidities influence the onset and progression of sarcopenia in MHD patients32. With a larger sample size and further model refinement, comorbidities may play a more significant role and should be investigated in future studies. Finally, this study assessed only static plantar pressure distribution. Since dynamic balance is more relevant to daily function, future research should include dynamic plantar pressure and gait analysis to enhance the ecological validity of sarcopenia screening tools.

Conclusion

Using only three plantar pressure indicators of feet-separated by 10 cm stance can help clinical healthcare professionals identify MHD patients with sarcopenia accurately. Moreover, under clinically permissible conditions, combining certain comprehensive physiological and plantar pressure indicators of feet-together stance can provide more accurate results in the auxiliary screening of sarcopenia.