Early detection for elderly people with musculoskeletal aging related diseases based on artificial intelligence model

Early detection for elderly people with musculoskeletal aging related diseases based on artificial intelligence model | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Early detection for elderly people with musculoskeletal aging related diseases based on artificial intelligence model Minjuan Li, Shuai Lu, Cheng Cheng, Kaiyuan Cheng, Maoqi Gong, and 2 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6124947/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Late-diagnosis is one of the main bottlenecks in musculoskeletal aging-related diseases prevention, and it is urgent to build early detection model. Twenty-two features were included to build early detection models based on binary and multiple classification respectively by XGBoost. In testing, the accuracy rate (63.74%~92.40%) and AUC (0.74 ~ 0.96) of binary-classification models were higher than the accuracy rate (61.40% ~85.96%) and AUC (0.63 ~ 0.86) of multiple-classification models. The optimal binary-classification model had an accuracy rate of 87.13% and an AUC of 0.92 in testing, including cooking, drinking milk, electronic devices use time, dental implant, dental decay, professional oral cleaning, falls in the past year, life satisfaction, the degree of pain or discomfort, indoor air improvement, drinking, body mass index, time spent indoors, grip grouping, SARC-F grouping, calf girth grouping and bone density examination. In elderly, musculoskeletal aging-related diseases can be early detected by model based on epidemiological factors. Health sciences/Biomarkers Health sciences/Risk factors Osteoporosis Sarcopenia Osteosarcopenia Risk factor Machine learning Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 INTRODUCTION With aging, the prevalence of osteoporosis and sarcopenia is increasing, which is associated with an increased risk of fractures, reduced quality of life, disability, and early death. 1 Osteoporosis, characterized by bone mass reduction and bone microstructure damage, leads to fragile bone and fracture. It also affects muscle condition, causing atrophy and function decline. Sarcopenia, characterized by reduced muscle strength, function, and mass with aging or disease, accelerates osteoporosis and increases the risk of falling. Based on the similar pathophysiology and health influence, osteosarcopenia is proposed to describe the overlap between osteoporosis and sarcopenia. 2 Elderly individuals with osteosarcopenia are more prone to fall, with a higher risk of fracture, disability, and death. Recent data indicates that the prevalence of osteosarcopenia is 18.5% that increases with aging. 3 The number of people aged 65 and over will increase to 1.5 billion in 2050, a three-fold increase from 2010. In worldwide, the disease burden of musculoskeletal aging-related diseases is heavy. Aging is an inevitable pathophysiological process caused by many factors that lead to a progressive reduction in the ability to resist stress and contribute to a variety of aging-related diseases. 4 The prevention and treatment of musculoskeletal aging-related diseases are promising but challenging. One of the challenges is late-diagnosis due to the silent progression and lack of early symptoms. Most patients are diagnosed when they visit a doctor for pain, fracture, or severe physical limitations. Hence, early detection of individuals at high risk for musculoskeletal aging-related diseases, particularly osteosarcopenia, is a significant public health and clinical concern. Previous studies have indicated that female, higher fat mass, low bone mass, early-life tobacco smoking, malnutrition, and multiple factors are the risk factors of osteoporosis or sarcopenia. 5 – 7 Additionally, clinicians have access to various tools for assessing osteoporosis and sarcopenia, such as FRAX© and SARC-F. 8 , 9 However, there is no early detection tool validated for osteosarcopenia. It is still necessary to explore the risk factors of musculoskeletal aging-related disease to provide a scientific basis for the early detection of musculoskeletal aging-related disease in elderly. Factors contributing to disease risk vary among individuals, making heterogeneous data and complex interactions challenging to evaluate using regression methods. Machine learning (ML) methods, however, can process complex data, including both linear and non-linear information, and can identify hidden relationships between variables and outcomes. Generally, ML methods can be classified into supervised and unsupervised algorithms. Supervised ML is suitable for annotated data, whereas unsupervised ML can process datasets that lack class labels. Supervised ML includes random forest, decision trees and, eXtreme Gradient Boosting (XGBoost). Recently, the XGBoost algorithm has been widely used in the risk stratification and early detection of several diseases, including diabetic retinopathy, 10 acquired immune deficiency syndrome 11 , endometrial injury, 12 cervical cancer, 13 and et al, which indicates its feasibility in the musculoskeletal aging-related diseases. Therefore, this study aims to identify risk factors associated with musculoskeletal aging-related diseases and establish early detection models based on XGBoost algorithm for different musculoskeletal states, including bone abnormality (osteoporosis), muscle abnormality (sarcopenia) or bone-muscle abnormality (osteosarcopenia). Based on the early detection model built in this study, we anticipate to quickly distinguish the high-risk population of musculoskeletal aging-related diseases, so as to diagnosis and treat them as early as possible, which is beneficial to the rational allocation of medical resources and the promotion of elderly health. RESULTS Baseline demographics of enrolled subjects 853 subjects were enrolled in this study, including 419 males (49.12%) and 434 females (50.88%). There was a significant difference in bone-muscle condition between males and females. ( P < 0.05) 80% of the subjects were randomly selected for the training dataset, and the remaining 20% for the testing dataset. The baseline characteristic of subjects in the training dataset was similar to it in the testing dataset. More details of enrolled subjects were shown in Supplementary Materials Combine Logistic regression, LASSO regression and XGBoost to select features In the univariate logistic regression analysis, a total of 88 features (70.4%) showed statistically significant differences with a P -value less than 0.05, including age, education level, drinking, weight, dental decay, pain in back and et al. In the LASSO regression analysis, a total of 51 features (40.8%) were included, involving gender, height, BMI, dental decay, smoking, drinking, and et al. Specifically, 14 features were associated with bone or muscle abnormality, such as gender, height, oral ulcer, pain in back and et al. Only the variable “the degree of pain or discomfort” was associated with at least one abnormality in bone or muscle. Considering univariate logistic regression and LASSO results together, only 42 features (33.6%) were overlapped. Based on the above 42 features, this study built a simple primary screening model based on the XGBoost algorithm. ( Figure-1a ) Then, the top 22 features of overall importance were included, involving 13 general features and 9 grouping features. ( Figure-1b ) General features included cooking at home, drinking milk, electronic devices use time, dental implant, dental decay, professional oral cleaning, falls in the past year, life satisfaction, the degree of pain or discomfort, indoor air improvement, drinking, BMI, and time spent indoors. Grouping features included calf girth grouping, grip grouping, SARC-F grouping, BMD, bone density examination, grip, BMD T value, calf girth, and SARC-F score. ( Supplementary Materials ) Simple primary screening model based on binary-classification In the binary-classification, the abnormal group indicated the subjects with muscle abnormalities, bone abnormalities, or muscle and bone abnormalities. Through the simple early detection model based on binary-classification, subjects classified as the abnormal need the further diagnosis of osteoporosis, sarcopenia, or osteosarcopenia. ( Figure-2a ) As Supplementary Materials shown, ten early detection models were constructed by combining the general features and grouping features as model_1 to model_10. Over the whole, the accuracy rate of model was from 95.90–100.00%, of which the AUC was all greater than 0.99 based on the training datasets. About the testing dataset, the accuracy rate of model was from 63.74–92.40%, the AUC of model was from 0.74 to 0.96. Model_1 included all general features and all grouping features, and the model_2 only included all general features. The accuracy rate and AUC were significantly higher in model_1 than in model_2 both in the training dataset and testing dataset. ( Figure-2b ) Model_3 (grip grouping), model_5 (calf girth grouping), and model_6 (SARC-F grouping) were all composed of general features and grouping features related to muscle condition. Based on the training dataset, the performance of model_5 was much better than model_3 and model_6. However, the performance of model_3 was more robust than model_5 and model_6 based on the testing dataset. ( Figure-2c ) Regarding the grouping features related to the bone condition, the accuracy rate (96.77%) and AUC (1.00) of model_4 were the same as the model_7 based on the training dataset. In the testing dataset, the accuracy rate and AUC of model_4 (accuracy rate: 70.76%, AUC:0.77) were much higher than model_7 (accuracy rate: 63.74%, AUC:0.74). ( Figure-2d ) Taken together, as Figure-2e shown, the accuracy rate and AUC of model_9 (accuracy rate: 87.13%, AUC:0.92) based on the testing dataset was higher than model_8 (accuracy rate: 79.53%, AUC:0.85) and model_10 (accuracy rate: 83.63%, AUC:0.91). Moreover, the false positive rate and false negative rate of model_9 were 15.11% (13/86) and 10.59% (9/85) respectively. The explanation of the optimal binary-classification model SHAP was used to measure the contributions to the outcome from each feature separately. 14 Figure-3a showed the summary of the XGBoost model explainability with the SHAP in all cases based on the optimal binary-classification model (model_9). SHAP > 0 indicated a higher probability of bone-muscle abnormality, while SHAP < 0 indicated a higher chance of bone and muscle normal. A dot indicated a case, and the colors indicated feature values ranging from low to high. Some features have distributions between positive and negative SHAP values, such as falls in the past year, life satisfaction, drinking milk, cooking at home, and dental decay. Among them, some features showed a more obvious aggregation in positive or negative SHAP values. For example, the lower feature values of drinking milk and dental decay and the higher feature value of cooking at home, have a positive effect on the model prediction. In model_9, life satisfaction, falls in the past year, BMI, time spent indoors and drinking milk were important factors in the prediction of bone-muscle abnormality. ( Figure-3b ) Further, considering that each feature has different values, we explained the one case in detail. As Figure-3c shown, no falls in the past years, no drinking and professional oral cleaning may be the protective features of bone-muscle abnormality. Simple primary screening model based on multiple-classification In the multiple-classification, there were three groups: normal group (normal in muscle and bone condition), abnormal_1 group (abnormality in muscle or bone condition), and abnormal_2 group (abnormality in muscle and bone condition). ( Figure-4a ) As Supplementary Materials shown, ten models were constructed by combining the general features and grouping features as model_11 to model_20. ( Figure-4b ) Over the whole, the accuracy rate of model was from 92.08–94.72%, of which the AUC was all greater than 0.98 based on the training datasets. About the testing data, the accuracy rate of the model was from 61.40–85.96%, the AUC of the model was from 0.63 to 0.86. Further in each model, there was three kinds of comparisons including normal vs abnormal_1 (①), normal vs abnormal_2 (②), and abnormal_1 vs abnormal_2 (③). The top 3 models in terms of the AUC of comparisons_① were model_11 (0.92), model_19 (0.85), and model_20 (0.84). Consistently, the top3 models in terms of the AUC of comparisons_② were model_11 (0.97), model_19 (0.93), and model_20 (0.92). About the comparisons_③, the three models with the highest AUC were model_14 (0.74), model_11 (0.70), and model_18 (0.69). Except for model_13, the AUC of comparison_② was the highest among all other models. On the whole, the accuracy rate and AUC of model_19 were 78.95% and 0.80 respectively in testing dataset, of which the performance was better than other models. Moreover, the false positive rate and false negative rate of model_19 were 17.44% (15/86) and 11.76% (10/85) respectively. The explanation of the optimal multiple-classification model Figure-5a shows the summary of the XGBoost model explainability with the SHAP in all cases based on the optimal multiple-classification model (Model_19). SHAP > 0 indicated a higher probability of abnormality, while SHAP < 0 indicated a higher chance of normal. A dot indicated a case, colors indicated the feature values ranging from low to high. About the color distribution of the dots, we can infer the effect of this feature on the classification of bone-muscle condition. For example, drinking and indoor air improvement were significantly important for the abnormal_1 group compared to the normal group. Grip grouping, electronic devices use time, falls in the past year, life satisfaction, drinking milk, SARC-F grouping, dental implant, and calf girth grouping were more worthy of attention for the abnormal_2 group compared to the normal group. Then, considering that each feature had different values, we further explained the one case in detail. As Figure-5b shown, indoor air quality improved (indoor air improvement = 1), no dental decay (dental decay = 0), and no drinking (drinking = 0) may be beneficial to reduce bone and muscle abnormalities. The comparison of XGBoost with other machine learning models Model_9 and model_19 were the optimal binary-classification model and multiple-classification model respectively. Based on the accuracy rate (model_9: 87.13%; model_19: 78.95%), false positive rate (model_9: 15.11%; model_19: 17.44%), false negative rate (model_9: 10.59%; model_19: 11.76%) and AUC (model_9: 0.92; model_19: 0.80) of the testing dataset, model_9 performed better than model_19. Based on features included in the model_9, further comparison of XGBoost with random forest and decision tree showed that the accuracy rate of all three models was 87.13% (149/171), but the confusion matrix was different. The false negative rate of XGBoost, random forest, and decision tree were 10.59% (9/85), 12.94% (11/85), and 14.12% (12/85) respectively. The false positive rate of XGBoost, random forest, and decision tree were 15.12% (13/86), 12.79% (11/86), and 11.63% (10/86) respectively. DISCUSSIONS To early detect and concentrate the high-risk population of musculoskeletal aging-related diseases, this study combined logistic regression, LASSO, and XGBoost algorithms to explore the risk factors of musculoskeletal aging-related diseases, then constructed early detection models based on binary-classification and multiple-classification respectively, which provided an important basis for the early diagnosis and early treatment of musculoskeletal aging-related diseases such as osteosarcopenia and sarcopenia. Previous studies have indicated the prevalence of osteoporosis among those aged 40 years or older was 5.0% among men and 20.6% among women. 15 In clinical practice, the diagnosis and treatment of osteoporosis are relatively well established. However, sarcopenia has only been classified internationally as a disease in the last ten years, which needs more attention. During aging, muscle and bones are intricately connected tissues displaying marked co-change. 16 The risk of hip fracture (HR: 2.67), major osteoporotic fracture (HR: 2.04), and death (HR: 1.91) were significantly higher in patients with osteosarcopenia than in those without osteoporosis or sarcopenia. 7 Besides, Yoo et al. indicated the one-year mortality rate of elderly hip fracture patients with osteosarcopenia (15.1%) was significantly higher than that of patients with osteoporosis (5.1%) or sarcopenia (10.3%). 17 It can be seen that elderly patients with osteosarcopenia are more likely to fall and have a significantly increased risk of fracture, disability, and death after a fall. A meta-analysis shows that the overall prevalence of osteosarcopenia is 18.5%. 3 Under limited conditions, osteosarcopenia is the key to the protection of bone-muscle health in the elderly. Of course, osteoporosis and sarcopenia are also important. 18 Consequently, this study set up the multiple-classification and binary-classification to suit the different demands of early detection. In this study, multiple-classification included normal (bone and muscle condition are both normal), one abnormality (bone or muscle condition is abnormal), and two abnormalities (bone and muscle condition are both abnormal). Based on this kind of classification, the early detection model built in this study was aimed to solve the problem of risk stratification of musculoskeletal aging-related disease, then subjects with different risks should receive differentiated further processing. There was interaction between muscle and bone at the molecular and cellular level. 19 Combined with public literature, we speculate that the two abnormalities should be paid more attention than the one abnormality. The AUC of comparison_② (normal vs abnormal_2) is generally better than that of comparison_① (normal vs abnormal_1), further suggesting that the difference gap between normal and bone-muscle abnormal was more obvious. However, the effect of comparison_① (AUC: 0.79) and comparison_② (AUC: 0.77) is similar and are both higher than that of comparison_③(AUC: 0.51) in model_13. In model_14, the effect of comparison_② (AUC:0.84) is better than that of comparison_③ (AUC:0.74) and comparison_① (AUC:0.67). The BMD result is the diagnostic criteria for osteoporosis and osteosarcopenia. 19 20 In this study, the only difference between model_13 and model_14 is that model_14 includes whether accept BMD examination, while model_13 includes gripping. Nowadays, the uptake of BMD examination remains low. Besides, there may be recall bias during the questionnaire about the BMD results. Fu et al. found osteoporosis was diagnosed in 86.2% of women and half of men among those willing to uptake BMD examination. 21 It suggests that even the acceptance of receiving BMD examination may play a critical role in the primary screening of osteoporosis and osteosarcopenia, which explains the better performance of whether to accept a BMD examination in this study. However, the performance of comparison_③ was not well among all multiple-classification models. And the performance of model based on binary-classification was better than model based on multiple-classification. The more categorical the model, the more complex it is to construct. Gripping and calf dimensions were related to the muscle condition of the upper and lower limbs, respectively. In this study, the AUC of model included griping was higher than that of model that included calf dimension, but both were better than the SARC-F score. Similarly, calf circumference demonstrated greater diagnostic accuracy than SARC-F in Thia older adults. 22 However, SARC-F questionnaire has been recommended as a screening tool for sarcopenia. SARC-F has shown reasonable diagnostic accuracy for sarcopenia screening and were demonstrated low sensitivity but high specificity, which may partly explain the poor performance in this study. 23 Despite its low sensitivity of SARC-F, it proves to be a useful tool for identifying severe cases in early detection taking advantage of its simplicity. Moreover, the SARC-F questionnaire also can be used to predict the presence of mild cognitive impairment in postmenopausal women. 24 Concerning bone condition, whether or not accept BMD examination can still be used to some extent as a substitute for BMD diagnosis for primary screening of musculoskeletal aging-related disease. Based on the optimal binary-classification XGBoost model, we further compared the performance of XGBoost, random forest and decision tree models. In the training dataset, the accuracy rate of the above three models were the same, but XGBoost model had a lower false negative rate and was better appropriate to early detection demands that decreased the missed diagnosis as much as possible. However, there is still a need to further explore the muscle-related diagnostic tools, including the development of new tools and the adjustment of the cut-off value of the existing tools. In addition, the performance of identifying bone or muscle abnormal (abnormal_1 group) from bone and muscle abnormal (abnormal_2 group) is limited, which indicates more tools that can be used to distinguish between osteoporosis or sarcopenia and osteosarcopenia are still to be explored and developed. To this, further work is required to identify biomarkers, which, in turn, may increase the accuracy rate of diagnosis, risk stratification, and targeted treatments to improve health outcomes. Specifically, this study indicated some potential risk factors were interpreted concretively by the SHAP algorithm, providing us with new clues. Most risk factors were modifiable. Addressing these modifiable risk factors can prevent, or at least delay, the onset of musculoskeletal aging-related diseases. Previous studies have indicated that menopause, and current smoking are independent risk factors of osteoporosis in females and males respectively, leisure screen time, and coffee intake played significantly causal roles in sarcopenia, and physical inactivity and poor nutrition are two major risk factors for osteosarcopenia. 19 , 25 , 26 In this study, multi-dimensional risk factors were paid attention to, such as indoor air quality and cooking at home. As visualized results show, the improvement of indoor air quality, and cook less frequently at home were potential protective factors against musculoskeletal aging-related diseases. Indoor air pollution from cooking at home has been associated with several diseases, including cognitive decline, cardiovascular diseases, and et al. 27 , 28 Besides, solid fuel for cooking the number of solid fuel use potentially facilitates the onset and progression of muscle loss and sarcopenia. 29 And chronic exposure to biomass smoke increased the risk of bone resorption and consequent osteoporosis. 30 Actually, both infiltration of outdoor air pollution into the indoor space and indoor sources (such as cooking or heating practices and household materials) contribute to unique exposure mixtures influencing indoor air quality. 31 Regardless of the source of indoor air pollution, it is recommended to actively improve indoor air quality as suggested by the results of this study. However, more validation studies about the risk factors proposed in this study are still needed in the future. Combined with statistical analysis and machine learning algorithms, this study provided the potential early detection clues for high-risk populations of musculoskeletal aging-related diseases. Subjects enrolled in this study came from different regions in China and were diagnosed with a variety of fracture types, which has a certain representativeness and lays a population foundation for the generalization of the early detection model. However, there still have some limitations. Firstly, the calf circumference, griping and SARC-F score included in this study are simpler and easier to perform in the early detection process. However, more objective muscle-related indexes based on DXA or bioelectrical impedance analysis were still needed. Secondly, some risk factors are proposed by this study, but the relevant investigation content needs to be in-depth, and then accurate intervention recommendations are put forward. Lastly, the sample size is relatively small. In the future, it is still needed to expand the sample size and include multi-center data as much as possible. In the elderly, musculoskeletal diseases are common and significantly increase the risk of falls, fractures, and disability, which are also associated with multiple comorbidities. This study mainly explored the risk factors for musculoskeletal aging-related diseases and built the early detection model of high-risk populations. Through early detection models, we can simply and quickly identify high-risk populations of musculoskeletal aging-related diseases, and then carry out diagnostic examinations, so as to promote early diagnosis and early treatment, further reducing its disease burden. Moreover, compared to osteoporosis, efficacious biomarkers for sarcopenia and osteosarcopenia are still currently lacking. It is anticipated that those in the field of aging will figure out answers to these questions and in turn, promote the development of precision medicine for elderly health. METHODS Subject enrollment All subjects were hospitalized patients with fractures at Beijing Jishuitan Hospital affiliated with Capital Medical University in China from January 2024 to December 2024, aged 50 years and above. The study strictly adhered to the Declaration of Helsinki, and all subjects were required to sign informed consent before participating in the study. This study has been approved by the Ethics Committee of Beijing Jishuitan Hospital affiliated to Capital Medical University. (No. K2024-273-00) Feature collections In this study, a musculoskeletal aging-related disease questionnaire was designed by the project team to collect the related epidemiological factors. This questionnaire comprised 125 features divided into two sections (general features and grouping features). General features included 12 parts as follows: baseline, cognitive function, covid-19 infection, diet habit, disease history, family history, oral health, physical activity, EQ-5D survey, lifestyle, mental health and pharmacohistory. Grouping features included grip strength, calf girth, SARC-F score, bone mineral density (BMD) T value, BMD Z value, grip grouping, calf girth grouping, SARC-F grouping, and bone density examination. On the day of hospitalization, this questionnaire was administered to the enrolled subjects by trained staff using a tablet computer. All subjects were capable of completing the questionnaire independently. Bone and muscle condition The classification of subjects enrolled in this study was mainly based on the condition of bone and muscle. The bone condition was assessed based on the BMD results from dual energy x-ray absorptiometry (DXA) examination. If the subject was diagnosed as osteoporosis (T value ≤ -2.5) or osteopenia (-2.5 < T value ≤ -1.0), it would be classified as the abnormal of bone condition. The muscle condition was assessed based on calf girth, griping and SARC-F score. 32 Abnormal calf girth was defined as less than 34 cm for men and 33 cm for women. Abnormal grip strength was defined as less than 28 kg for men and 18 kg for women. A SARC-F score of 4 points or higher was considered abnormal. If subjects exhibited at least one abnormality in calf girth, grip strength, or SARC-F score, they were classified as abnormal muscle condition. Subject grouping methods Abnormalities in muscle or bone are both common among the elderly. In particular, the co-existence of muscle and bone abnormalities deserves more attention. Based on the assessment of bone and muscle condition, there were two grouping methods (binary-classification and multiple-classification) in this study. (1) Binary classification. Subjects enrolled in this study were classified as the normal group (subjects with normal bone condition and normal muscle condition) and the abnormal group (subjects with abnormal bone condition and/or abnormal muscle condition). (2) Multiple classification. Subjects enrolled in this study were classified as the normal group (subjects with normal bone condition and normal muscle condition), abnormal_1 group (subjects with abnormal bone condition or abnormal muscle condition), and abnormal_2 group (subjects with abnormal bone condition and abnormal muscle condition). Statistical analyses Categorical variables were evaluated using the chi-square (X 2 ) or Fisher’s exact test. According to the type of feature, mice package (V.3.17.0) was used for data interpolation primary, specifically including “polr”, “polyreg”, ‘logreg’, and ‘pmm’ methods. Univariate logistic regression analysis (rms package: V.6.7-1) was performed to determine independent risk factors associated with musculoskeletal aging-related diseases. Next, least absolute shrinkage and selection operator regression (LASSO) was also performed to select potential risk factors associated with musculoskeletal aging-related diseases. (glmnet package: V.4.1-8) The above statistical analyses were performed in R Studio (2023.09.0 + 463). Statistical significance was defined as a two-tailed P -value < 0.05. Machine learning models Subjects were randomly assigned to an 80% training dataset (n = 682) and a 20% testing dataset (n = 171). Combining the results of univariate logistic regression and LASSO regression, then this study built a XGBoost model (xgboost package: V.1.7.8.1) to further select the potential risk factors of musculoskeletal aging-related diseases based on the importance of each feature. A total of 22 characteristic features (13 general features and 9 grouping features) was included to construct the early detection models based on XGBoost algorithm. ( Figure-1b ) And there were ten models by combining general features (n = 13) and grouping features (n = 9) based on binary-classification (model_1 to model_10) and multiple-classification (model_11 to model_20) respectively. ( Figure-1a ) “Multi:softmax” and “binary:logistic” were the main objective parameters of multi-classification models and binary-classification models respectively. The eta value, num_class, max_dep, and nrounds were 0.1, 3, 4, and 200 respectively. The performances of the XGBoost models were assessed by the area under the curve (AUC) and accuracy rate (%). (pROC package: V.1.18.5) Then, shapley additive explanations (SHAP) algorithm (shapviz package: V.0.9.7) was used to interpret and extract SHAP values for each feature based on the optimal multiple-classification model and the optimal binary-classification model respectively. Random forest (randomForest package: V.4.7–1.1) and decision tree (rpart package: V.4.1.24) algorithms were used to compare with the optimal XGBoost model, and the accuracy rate (%), false negative rate (%), and false positive rate (%) were the evaluation index. The above statistical analyses were performed in R Studio (2023.09.0 + 463). Statistical significance was defined as a two-tailed P -value < 0.05. Declarations DATA AVAILABILITY Data not presented in the manuscript will be provided upon reasonable request to the corresponding authors. The data are not publicly available due to privacy or ethical restrictions. CODE AVAILABILITY The underlying code for this study [and training/testing datasets] is not publicly available but may be made available to qualified researchers on reasonable request from the corresponding author. CONFLICTS OF INTEREST The authors declare no competing interests. Author Contribution M.L, M.G, Y.Z and X.J conceived this study. M.L, S.L and C.C initiated the reported analysis. M.L, S.L, C.C and K.C performed figure 1-5. M.L and C.C drafted this manuscript, and all authors contributed to data interpretation and critical revision of this manuscript. All authors have read and approved the manuscript. ACKNOWLEDGEMENTS This study was supported by the Beijing Municipal Public Welfare Development and Reform Pilot Project for Medical Research Institutes (NO. JYY2023-8, JYY2023-11), the Project supported by Beijing Jishuitan Research Funding (NO. KYYC202301), the Natural science research Beijing Jishutian Hospital (No. ZR-202402), and the National Key Research and Development project (2024YFC3044700). References Faidra Laskou et al., Associations of osteoporosis and sarcopenia with frailty and multimorbidity among participants of the Hertfordshire Cohort Study. J Cachexia Sarcopenia Muscle. 13, 220–229 (2021). Binkley N, B. B, Beyond FRAX: it's time to consider "sarco- osteopenia". J Clin Densitom 12, 413–416 (2009). Shanping Chen et al., Global epidemiological features and impact of osteosarcopenia: A comprehensive meta-analysis and systematic review. J Cachexia Sarcopenia Muscle 15, 8–20 (2024). Jun Guo et al., Aging and aging-related diseases: from molecular mechanisms to interventions and treatments. Signal Transduct Target Ther. 7, 391 (2022). D. Di et al., Early-life tobacco smoke elevating later-life osteoporosis risk: Mediated by telomere length and interplayed with genetic predisposition. J Adv Res. S2090-1232, 00083–00083 (2024). B. Kirk, J. Zanker, G. Duque, Osteosarcopenia: epidemiology, diagnosis, and treatment—facts and numbers. J Cachexia Sarcopenia Muscle 11, 609–618 (2020). Tine Kolenda Paulin, Linnea Malmgren, Fiona E McGuigan, K. E. Akesson, Osteosarcopenia: Prevalence and 10-Year Fracture and Mortality Risk – A Longitudinal, Population-Based Study of 75-Year-Old Women. Calcif Tissue Int 114, 315–325 (2024). J A Kanis, O Johnell, A Oden, H Johansson, E. McCloskey, FRAX and the assessment of fracture probability in men and women from the UK. Osteoporos Int. 19, 385–397 (2008). Juergen Bauer et al., Sarcopenia: A Time for Action. An SCWD Position Paper. J Cachexia Sarcopenia Muscle. 10, 956–961 (2019). Dariusz Tarasewicz et al., Development and Validation of a Diabetic Retinopathy Risk Stratification Algorithm. Diabetes Care 46, 1068–1075 (2023). Dandan Tang et al., Study on the prediction performance of AIDS monthly incidence in Xinjiang based on time series and deep learning models. BMC Public Health 25, 780 (2025). Bohan Li, Hui Chen, Xiaona Lin, H. Duan, Multimodal learning system integrating electronic medical records and hysteroscopic images for reproductive outcome prediction and risk stratification of endometrial injury: a multicenter diagnostic study. Int J Surg 110, 3237–3248 (2024). Binhua Dong et al., Development, validation, and clinical application of a machine learning model for risk stratification and management of cervical cancer screening based on full-genotyping hrHPV test (SMART-HPV): a modelling study. Lancet Reg Health West Pac 55, 101480 (2025). Junyi Zhang et al., Insights into geospatial heterogeneity of landslide susceptibility based on the SHAP-XGBoost model. J Environ Manage 332, 117357 (2023). Linhong Wang et al., Prevalence of Osteoporosis and Fracture in China: The China Osteoporosis Prevalence Study. JAMA Netw Open 4, e2121106 (2021). K. M. Kim et al., Longitudinal Changes in Muscle Mass and Strength, and Bone Mass in Older Adults- Gender-Specific Associations Between Muscle and Bone Losses. J Gerontol A Biol Sci Med Sci 73, 1062–1069 (2018). Jun Il Yoo, Hyunho Kim, Yong Chan Ha, Hyuck Bin Kwon, K. H. Koo, Osteosarcopenia in Patients with Hip Fracture Is Related with High Mortality. J Korean Med Sci 33, e27 (2018). Oscar Rosas-Carrasco, Betty Manrique-Espinoza, J. C. López-Alvarenga., Beatriz Mena-Montes, I. Omaña-Guzmán, Osteosarcopenia predicts greater risk of functional disability than sarcopenia: a longitudinal analysis of FraDySMex cohort study. The Journal of nutrition, health and aging 28, 100368 (2024). Ben Kirk, Sarah Miller, Jesse Zanker, G. Duque, A clinical guide to the pathophysiology, diagnosis and treatment of osteosarcopenia. Maturitas 140, 27–33 (2020). Fan Yu, W. Xia, The epidemiology of osteoporosis, associated fragility fractures, and management gap in China. Archives of Osteoporosis 14, 32 (2019). S-H Fu et al., Screening of Fracture Risk and Osteoporosis Among Older Long-term Care Residents: A Prospective Study. J Nutr Health Aging. 27, 1255–1261 (2023). Ekasame Vanitcharoenkul et al., Evaluating SARC-F, SARC-CalF, and calf circumference as diagnostic tools for sarcopenia in Thai older adults: results from a nationwide study. BMC Geriatr 24, 1043 (2024). Jia-Yu Guo et al., The application of Chinese version of SARC-F and SARC-CalF in sarcopenia screening against five definitions: a diagnostic test accuracy study. BMC Geriatr 24, 883 (2024). María S Vallejo et al., Risk of sarcopenia: A red flag for cognitive decline in postmenopause? Maturitas 194, 108193 (2025). Xianxian Yang et al., Prevalence and risk factors associated with osteoporosis among residents aged above 20 years old in Chongqing, China. Arch Osteoporos 16, 57 (2021). Mingchong Liu et al., Causal Roles of Lifestyle, Psychosocial Characteristics, and Sleep Status in Sarcopenia: A Mendelian Randomization Study. J Gerontol A Biol Sci Med Sci 79, glad191 (2024). Tingting Xu et al., Association between solid cooking fuel and cognitive decline: Three nationwide cohort studies in middle-aged and older population. Environ Int 173, 107803 (2023). Hsiao-Chi Chuang et al., Long-term indoor air conditioner filtration and cardiovascular health: A randomized crossover intervention study. Environ Int 106, 91–96 (2017). Zhigang Hu, Yufeng Tian, Xinyu Song, Fanjun Zeng, A. Yang, Associations between indoor air pollution for cooking and heating with muscle and sarcopenia in Chinese older population. J Cachexia Sarcopenia Muscle 14, 2029–2043 (2023). Hirak Saha, Bidisha Mukherjee, Banani Bindhani, M. R. Ray, Changes in RANKL and osteoprotegerin expression after chronic exposure to indoor air pollution as a result of cooking with biomass fuel. J Appl Toxicol 36, 969–976 (2016). Jared Radbel, Meghan E Rebuli, Howard Kipen, E. Brigham, Indoor air pollution and airway health. J Allergy Clin Immunol 154, 835–846 (2024). Liang-Kung Chen et al., Asian Working Group for Sarcopenia: 2019 Consensus Update on Sarcopenia Diagnosis and Treatment. J Am Med Dir Assoc 21, 300–307.e302 (2020). Additional Declarations No competing interests reported. Supplementary Files Supplementarymaterials.pdf Graphicabstract.jpg Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-6124947","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":443069090,"identity":"69970cb8-c593-4110-8bb8-99de9674882d","order_by":0,"name":"Minjuan Li","email":"","orcid":"","institution":"Beijing Jishuitan Hospital, Capital Medical University","correspondingAuthor":false,"prefix":"","firstName":"Minjuan","middleName":"","lastName":"Li","suffix":""},{"id":443069092,"identity":"6d7d2615-031d-4af8-b3f0-12f8e9c8b489","order_by":1,"name":"Shuai Lu","email":"","orcid":"","institution":"Beijing Jishuitan Hospital, Capital Medical University","correspondingAuthor":false,"prefix":"","firstName":"Shuai","middleName":"","lastName":"Lu","suffix":""},{"id":443069095,"identity":"de15aed5-5924-4023-8fc8-cdfb283e9165","order_by":2,"name":"Cheng Cheng","email":"","orcid":"","institution":"Beijing Jishuitan Hospital, Capital Medical University","correspondingAuthor":false,"prefix":"","firstName":"Cheng","middleName":"","lastName":"Cheng","suffix":""},{"id":443069097,"identity":"c5a308fa-19db-4492-9780-30ab5397609c","order_by":3,"name":"Kaiyuan Cheng","email":"","orcid":"","institution":"Beijing Jishuitan Hospital, Capital Medical University","correspondingAuthor":false,"prefix":"","firstName":"Kaiyuan","middleName":"","lastName":"Cheng","suffix":""},{"id":443069098,"identity":"2bf4326d-0f98-4090-b021-05cd5930a747","order_by":4,"name":"Maoqi Gong","email":"","orcid":"","institution":"Beijing Jishuitan Hospital, Capital Medical University","correspondingAuthor":false,"prefix":"","firstName":"Maoqi","middleName":"","lastName":"Gong","suffix":""},{"id":443069099,"identity":"056318ad-335d-446c-b47e-7a5245632a61","order_by":5,"name":"Yejun Zha","email":"","orcid":"","institution":"Beijing Jishuitan Hospital, Capital Medical University","correspondingAuthor":false,"prefix":"","firstName":"Yejun","middleName":"","lastName":"Zha","suffix":""},{"id":443069100,"identity":"e93535cf-b8d4-40d0-991b-bb042bf7cd8c","order_by":6,"name":"Xieyuan Jiang","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAyklEQVRIiWNgGAWjYBACxgYYi72x8eEHIrVAdfEcbjaWIM0iifQ2AR5i1DPPyDF/XFFzOJp/5sM2BgkGOzndBgJaGHvOGDaeOXY4d8btxLYHBQzJxmYHCGlp7zFsbGA7nNtwO7HdQILhQOI2glqaeYBa/h3OnX/zYJsED1FaQLY0th3O3XCDkVgtPccKZzb2peduPJMIDGQDIvxiOCN5w8eGb9a5844ff/jwQ4WdHGEtDRwGSFwDnAoRQJ6B/QERykbBKBgFo2BEAwDy/Uos4AC5KgAAAABJRU5ErkJggg==","orcid":"","institution":"Beijing Jishuitan Hospital, Capital Medical University","correspondingAuthor":true,"prefix":"","firstName":"Xieyuan","middleName":"","lastName":"Jiang","suffix":""}],"badges":[],"createdAt":"2025-02-28 03:38:06","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6124947/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6124947/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":80756338,"identity":"5b9a71e3-033a-475f-839f-fee5c414ddda","added_by":"auto","created_at":"2025-04-16 17:55:05","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":506669,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eThe selected features included in the XGBoost models. (a)\u003c/strong\u003e The flowchart of the feature selection. \u003cstrong\u003e(b)\u003c/strong\u003eThe importance of characteristic features enrolled in XGBoost models. Features with the upper character “a” were the grouping features, and the others were the general features.\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-6124947/v1/80b01ebe776d2e5ce5bb1095.png"},{"id":80756631,"identity":"537eb320-b61b-4fb2-b99c-a987901e891d","added_by":"auto","created_at":"2025-04-16 18:03:05","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":199977,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eThe performance of binary-classification early detection models based on XGBoost algorithm. (a)\u003c/strong\u003e The study design for model construction. \u003cstrong\u003e(b)\u003c/strong\u003e The receiver operating characteristic curve (ROC) of binary-classification models that combined general features and grouping features. General features included cooking at home, drinking milk, electronic devices use time, dental implant, dental decay, professional oral cleaning, falls in the past year, life satisfaction, the degree of pain or discomfort, indoor air improvement, drinking, body mass index (BMI) and time spent indoors. Grouping features included calf girth grouping, grip grouping, SARC-F grouping, bone mineral density (BMD), bone density examination, grip, BMD T value, calf girth and SARC-F score. Model_1 included all general features and all grouping features, model_2 included all grouping features. \u003cstrong\u003e(c)\u003c/strong\u003eThe ROC of model_3, model_5 and model_6. Model_3, model_5, and model_6 included all general features and griping grouping, calf girth grouping, and SARC-F grouping respectively. \u003cstrong\u003e(d)\u003c/strong\u003e The ROC of model_4 and model_7. Model_4 and model_7 included all general features and bone density examination and BMD respectively. \u003cstrong\u003e(e)\u003c/strong\u003e The ROC of model_8, model_9, and model_10. The model_8 included all general features, BMD and gripping grouping. Model_9 included all general features, griping grouping, calf girth grouping, SARC-F grouping, and bone density examination. Model_10 included all general features, griping grouping, calf girth grouping, SARC-F grouping, and BMD.\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-6124947/v1/3687cb4c14f90e65b66adfbf.png"},{"id":80756340,"identity":"a7b2ff01-84db-4011-b568-d5e063183304","added_by":"auto","created_at":"2025-04-16 17:55:05","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":231464,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eThe explanation of binary-classification early detection model (model_9) by SHAP. (a) \u003c/strong\u003eThe explanation of risk factors in each comparison in model_9 (features included cooking at home, drinking milk, electronic devices use time, dental implant, dental decay, professional oral cleaning, falls in the past year, life satisfaction, the degree of pain or discomfort, indoor air improvement, drinking, body mass index, time spent indoors, grip grouping, SARC-F grouping, calf girth grouping and bone density examination). SHAP \u0026gt; 0 indicated a higher probability of bone or/and muscle abnormality, while SHAP \u0026lt; 0 indicated a higher chance of normal. A dot indicated a case, and the colors indicated feature values ranging from low to high. About the color distribution of the dots, we can infer the effect of this feature on the classification of bone and muscle conditions.\u003cstrong\u003e (b) \u003c/strong\u003eThe importance of included features in model_9.\u003cstrong\u003e (c)\u003c/strong\u003e The incorporates specific explanations of risk factors in each comparison in model_9 in one case.\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-6124947/v1/73bd996e5a5710185f512272.png"},{"id":80756346,"identity":"a8b6ac26-f15b-4b92-98e3-67ffdd68a0a8","added_by":"auto","created_at":"2025-04-16 17:55:05","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":267985,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eThe performance of multiple-classification early detection models based on XGBoost algorithm. (a)\u003c/strong\u003e The study design for model construction. \u003cstrong\u003e(b)\u003c/strong\u003eThe accuracy rate and area under the curve of model_11 to model_20 both in the training dataset and the testing dataset. Model_11 included all general features and all grouping features, model_12 included all grouping features. Model_13, model_15, and model_16 included all general features and griping grouping, calf girth grouping, and SARC-F grouping respectively. Model_14 and model_17 included all general features and bone density examination and bone mineral density (BMD) respectively. The model_18 included all general features, BMD and gripping grouping. Model_19 included all general features, griping grouping, calf girth grouping, SARC-F grouping, and bone density examination. Model_20 included all general features, griping grouping, calf girth grouping, SARC-F grouping, and BMD. General features included cooking at home, drinking milk, electronic device use time, dental implant, dental decay, professional oral cleaning, falls in the past year, life satisfaction, the degree of pain or discomfort, indoor air improvement, drinking, body mass index (BMI) and time spent indoors. Grouping features included calf girth grouping, grip grouping, SARC-F grouping, BMD, bone density examination, grip, BMD T value, calf girth and SARC-F score. Comparison_① indicated the early detection between normal and abnormal_1.Comparison_② indicated the early detection between normal and abnormal_2. Comparison_③ indicated the early detection between abnormal_1 and abnormal_2. In each comparison, numbers in red indicated the top 3.\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-6124947/v1/741556286fe2c436a77c3b5e.png"},{"id":80757615,"identity":"72fc4492-06cf-4e9c-9e1e-fb6657991ca3","added_by":"auto","created_at":"2025-04-16 18:19:05","extension":"jpg","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":577262,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eThe explanation of multiple-classification early detection model (model_19) by SHAP. (a)\u003c/strong\u003e The explanation of risk factors in each comparison in model_19 (features included cooking at home, drinking milk, electronic devices use time, dental implant, dental decay, professional oral cleaning, falls in the past year, life satisfaction, the degree of pain or discomfort, indoor air improvement, drinking, body mass index, time spent indoors, grip grouping, SARC-F grouping, calf girth grouping and bone density examination). SHAP \u0026gt; 0 indicated a higher probability of bone or/and muscle abnormality, while SHAP \u0026lt; 0 indicated a higher chance or bone and muscle normal. A dot indicated a case, and the colors indicated feature values ranging from low to high. About the color distribution of the dots, we can infer the effect of this feature on the classification of bone and muscle conditions. \u003cstrong\u003e(b) \u003c/strong\u003eThe incorporates specific explanations of risk factors in each comparison in model_19 in one case.\u003c/p\u003e","description":"","filename":"Figure5.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6124947/v1/51cae713da59e0cd9df86cac.jpg"},{"id":81241372,"identity":"98db9cf6-a49e-455e-906f-89a99b26690b","added_by":"auto","created_at":"2025-04-23 22:31:22","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2603670,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6124947/v1/da0c57c8-a313-4be2-b793-48fe80108830.pdf"},{"id":80757365,"identity":"50ca20a3-f0c2-45d6-b8ef-1c3adfc17cd3","added_by":"auto","created_at":"2025-04-16 18:11:05","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":609113,"visible":true,"origin":"","legend":"","description":"","filename":"Supplementarymaterials.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6124947/v1/20e00833ad06804752bf3e7e.pdf"},{"id":80757366,"identity":"fbb63afb-bb82-4072-88e5-2cd5178c94e2","added_by":"auto","created_at":"2025-04-16 18:11:05","extension":"jpg","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":249660,"visible":true,"origin":"","legend":"","description":"","filename":"Graphicabstract.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6124947/v1/4f53e8bc5b4cd31dcffaf9a8.jpg"}],"financialInterests":"No competing interests reported.","formattedTitle":"Early detection for elderly people with musculoskeletal aging related diseases based on artificial intelligence model","fulltext":[{"header":"INTRODUCTION","content":"\u003cp\u003eWith aging, the prevalence of osteoporosis and sarcopenia is increasing, which is associated with an increased risk of fractures, reduced quality of life, disability, and early death.\u003csup\u003e\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u003c/sup\u003e Osteoporosis, characterized by bone mass reduction and bone microstructure damage, leads to fragile bone and fracture. It also affects muscle condition, causing atrophy and function decline. Sarcopenia, characterized by reduced muscle strength, function, and mass with aging or disease, accelerates osteoporosis and increases the risk of falling. Based on the similar pathophysiology and health influence, osteosarcopenia is proposed to describe the overlap between osteoporosis and sarcopenia. \u003csup\u003e\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u003c/sup\u003e Elderly individuals with osteosarcopenia are more prone to fall, with a higher risk of fracture, disability, and death. Recent data indicates that the prevalence of osteosarcopenia is 18.5% that increases with aging.\u003csup\u003e\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e\u003c/sup\u003e The number of people aged 65 and over will increase to 1.5\u0026nbsp;billion in 2050, a three-fold increase from 2010. In worldwide, the disease burden of musculoskeletal aging-related diseases is heavy.\u003c/p\u003e \u003cp\u003eAging is an inevitable pathophysiological process caused by many factors that lead to a progressive reduction in the ability to resist stress and contribute to a variety of aging-related diseases.\u003csup\u003e\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e\u003c/sup\u003e The prevention and treatment of musculoskeletal aging-related diseases are promising but challenging. One of the challenges is late-diagnosis due to the silent progression and lack of early symptoms. Most patients are diagnosed when they visit a doctor for pain, fracture, or severe physical limitations. Hence, early detection of individuals at high risk for musculoskeletal aging-related diseases, particularly osteosarcopenia, is a significant public health and clinical concern. Previous studies have indicated that female, higher fat mass, low bone mass, early-life tobacco smoking, malnutrition, and multiple factors are the risk factors of osteoporosis or sarcopenia. \u003csup\u003e\u003cspan additionalcitationids=\"CR6\" citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u003c/sup\u003e Additionally, clinicians have access to various tools for assessing osteoporosis and sarcopenia, such as FRAX\u0026copy; and SARC-F. \u003csup\u003e\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e,\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u003c/sup\u003e However, there is no early detection tool validated for osteosarcopenia. It is still necessary to explore the risk factors of musculoskeletal aging-related disease to provide a scientific basis for the early detection of musculoskeletal aging-related disease in elderly.\u003c/p\u003e \u003cp\u003eFactors contributing to disease risk vary among individuals, making heterogeneous data and complex interactions challenging to evaluate using regression methods. Machine learning (ML) methods, however, can process complex data, including both linear and non-linear information, and can identify hidden relationships between variables and outcomes. Generally, ML methods can be classified into supervised and unsupervised algorithms. Supervised ML is suitable for annotated data, whereas unsupervised ML can process datasets that lack class labels. Supervised ML includes random forest, decision trees and, eXtreme Gradient Boosting (XGBoost). Recently, the XGBoost algorithm has been widely used in the risk stratification and early detection of several diseases, including diabetic retinopathy,\u003csup\u003e\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u003c/sup\u003e acquired immune deficiency syndrome\u003csup\u003e\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e\u003c/sup\u003e, endometrial injury,\u003csup\u003e\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u003c/sup\u003e cervical cancer,\u003csup\u003e\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e\u003c/sup\u003e and et al, which indicates its feasibility in the musculoskeletal aging-related diseases. Therefore, this study aims to identify risk factors associated with musculoskeletal aging-related diseases and establish early detection models based on XGBoost algorithm for different musculoskeletal states, including bone abnormality (osteoporosis), muscle abnormality (sarcopenia) or bone-muscle abnormality (osteosarcopenia). Based on the early detection model built in this study, we anticipate to quickly distinguish the high-risk population of musculoskeletal aging-related diseases, so as to diagnosis and treat them as early as possible, which is beneficial to the rational allocation of medical resources and the promotion of elderly health.\u003c/p\u003e"},{"header":"RESULTS","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eBaseline demographics of enrolled subjects\u003c/h2\u003e \u003cp\u003e853 subjects were enrolled in this study, including 419 males (49.12%) and 434 females (50.88%). There was a significant difference in bone-muscle condition between males and females. (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.05) 80% of the subjects were randomly selected for the training dataset, and the remaining 20% for the testing dataset. The baseline characteristic of subjects in the training dataset was similar to it in the testing dataset. More details of enrolled subjects were shown in \u003cb\u003eSupplementary Materials\u003c/b\u003e\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eCombine Logistic regression, LASSO regression and XGBoost to select features\u003c/h3\u003e\n\u003cp\u003eIn the univariate logistic regression analysis, a total of 88 features (70.4%) showed statistically significant differences with a \u003cem\u003eP\u003c/em\u003e-value less than 0.05, including age, education level, drinking, weight, dental decay, pain in back and et al. In the LASSO regression analysis, a total of 51 features (40.8%) were included, involving gender, height, BMI, dental decay, smoking, drinking, and et al. Specifically, 14 features were associated with bone or muscle abnormality, such as gender, height, oral ulcer, pain in back and et al. Only the variable \u0026ldquo;the degree of pain or discomfort\u0026rdquo; was associated with at least one abnormality in bone or muscle. Considering univariate logistic regression and LASSO results together, only 42 features (33.6%) were overlapped. Based on the above 42 features, this study built a simple primary screening model based on the XGBoost algorithm. (\u003cb\u003eFigure-1a\u003c/b\u003e) Then, the top 22 features of overall importance were included, involving 13 general features and 9 grouping features. (\u003cb\u003eFigure-1b\u003c/b\u003e) General features included cooking at home, drinking milk, electronic devices use time, dental implant, dental decay, professional oral cleaning, falls in the past year, life satisfaction, the degree of pain or discomfort, indoor air improvement, drinking, BMI, and time spent indoors. Grouping features included calf girth grouping, grip grouping, SARC-F grouping, BMD, bone density examination, grip, BMD T value, calf girth, and SARC-F score. (\u003cb\u003eSupplementary Materials\u003c/b\u003e)\u003c/p\u003e\n\u003ch3\u003eSimple primary screening model based on binary-classification\u003c/h3\u003e\n\u003cp\u003eIn the binary-classification, the abnormal group indicated the subjects with muscle abnormalities, bone abnormalities, or muscle and bone abnormalities. Through the simple early detection model based on binary-classification, subjects classified as the abnormal need the further diagnosis of osteoporosis, sarcopenia, or osteosarcopenia. (\u003cb\u003eFigure-2a\u003c/b\u003e) As \u003cb\u003eSupplementary Materials\u003c/b\u003e shown, ten early detection models were constructed by combining the general features and grouping features as model_1 to model_10. Over the whole, the accuracy rate of model was from 95.90\u0026ndash;100.00%, of which the AUC was all greater than 0.99 based on the training datasets. About the testing dataset, the accuracy rate of model was from 63.74\u0026ndash;92.40%, the AUC of model was from 0.74 to 0.96.\u003c/p\u003e \u003cp\u003eModel_1 included all general features and all grouping features, and the model_2 only included all general features. The accuracy rate and AUC were significantly higher in model_1 than in model_2 both in the training dataset and testing dataset. (\u003cb\u003eFigure-2b\u003c/b\u003e) Model_3 (grip grouping), model_5 (calf girth grouping), and model_6 (SARC-F grouping) were all composed of general features and grouping features related to muscle condition. Based on the training dataset, the performance of model_5 was much better than model_3 and model_6. However, the performance of model_3 was more robust than model_5 and model_6 based on the testing dataset. (\u003cb\u003eFigure-2c\u003c/b\u003e) Regarding the grouping features related to the bone condition, the accuracy rate (96.77%) and AUC (1.00) of model_4 were the same as the model_7 based on the training dataset. In the testing dataset, the accuracy rate and AUC of model_4 (accuracy rate: 70.76%, AUC:0.77) were much higher than model_7 (accuracy rate: 63.74%, AUC:0.74). (\u003cb\u003eFigure-2d\u003c/b\u003e) Taken together, as \u003cb\u003eFigure-2e\u003c/b\u003e shown, the accuracy rate and AUC of model_9 (accuracy rate: 87.13%, AUC:0.92) based on the testing dataset was higher than model_8 (accuracy rate: 79.53%, AUC:0.85) and model_10 (accuracy rate: 83.63%, AUC:0.91). Moreover, the false positive rate and false negative rate of model_9 were 15.11% (13/86) and 10.59% (9/85) respectively.\u003c/p\u003e\n\u003ch3\u003eThe explanation of the optimal binary-classification model\u003c/h3\u003e\n\u003cp\u003eSHAP was used to measure the contributions to the outcome from each feature separately.\u003csup\u003e\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e\u003c/sup\u003e \u003cb\u003eFigure-3a\u003c/b\u003e showed the summary of the XGBoost model explainability with the SHAP in all cases based on the optimal binary-classification model (model_9). SHAP\u0026thinsp;\u0026gt;\u0026thinsp;0 indicated a higher probability of bone-muscle abnormality, while SHAP\u0026thinsp;\u0026lt;\u0026thinsp;0 indicated a higher chance of bone and muscle normal. A dot indicated a case, and the colors indicated feature values ranging from low to high. Some features have distributions between positive and negative SHAP values, such as falls in the past year, life satisfaction, drinking milk, cooking at home, and dental decay. Among them, some features showed a more obvious aggregation in positive or negative SHAP values. For example, the lower feature values of drinking milk and dental decay and the higher feature value of cooking at home, have a positive effect on the model prediction. In model_9, life satisfaction, falls in the past year, BMI, time spent indoors and drinking milk were important factors in the prediction of bone-muscle abnormality. (\u003cb\u003eFigure-3b\u003c/b\u003e) Further, considering that each feature has different values, we explained the one case in detail. As \u003cb\u003eFigure-3c\u003c/b\u003e shown, no falls in the past years, no drinking and professional oral cleaning may be the protective features of bone-muscle abnormality.\u003c/p\u003e\n\u003ch3\u003eSimple primary screening model based on multiple-classification\u003c/h3\u003e\n\u003cp\u003eIn the multiple-classification, there were three groups: normal group (normal in muscle and bone condition), abnormal_1 group (abnormality in muscle or bone condition), and abnormal_2 group (abnormality in muscle and bone condition). (\u003cb\u003eFigure-4a\u003c/b\u003e) As \u003cb\u003eSupplementary Materials\u003c/b\u003e shown, ten models were constructed by combining the general features and grouping features as model_11 to model_20. (\u003cb\u003eFigure-4b\u003c/b\u003e) Over the whole, the accuracy rate of model was from 92.08\u0026ndash;94.72%, of which the AUC was all greater than 0.98 based on the training datasets. About the testing data, the accuracy rate of the model was from 61.40\u0026ndash;85.96%, the AUC of the model was from 0.63 to 0.86. Further in each model, there was three kinds of comparisons including normal vs abnormal_1 (①), normal vs abnormal_2 (②), and abnormal_1 vs abnormal_2 (③). The top 3 models in terms of the AUC of comparisons_① were model_11 (0.92), model_19 (0.85), and model_20 (0.84). Consistently, the top3 models in terms of the AUC of comparisons_② were model_11 (0.97), model_19 (0.93), and model_20 (0.92). About the comparisons_③, the three models with the highest AUC were model_14 (0.74), model_11 (0.70), and model_18 (0.69). Except for model_13, the AUC of comparison_② was the highest among all other models. On the whole, the accuracy rate and AUC of model_19 were 78.95% and 0.80 respectively in testing dataset, of which the performance was better than other models. Moreover, the false positive rate and false negative rate of model_19 were 17.44% (15/86) and 11.76% (10/85) respectively.\u003c/p\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003eThe explanation of the optimal multiple-classification model\u003c/h2\u003e \u003cp\u003e \u003cb\u003eFigure-5a\u003c/b\u003e shows the summary of the XGBoost model explainability with the SHAP in all cases based on the optimal multiple-classification model (Model_19). SHAP\u0026thinsp;\u0026gt;\u0026thinsp;0 indicated a higher probability of abnormality, while SHAP\u0026thinsp;\u0026lt;\u0026thinsp;0 indicated a higher chance of normal. A dot indicated a case, colors indicated the feature values ranging from low to high. About the color distribution of the dots, we can infer the effect of this feature on the classification of bone-muscle condition. For example, drinking and indoor air improvement were significantly important for the abnormal_1 group compared to the normal group. Grip grouping, electronic devices use time, falls in the past year, life satisfaction, drinking milk, SARC-F grouping, dental implant, and calf girth grouping were more worthy of attention for the abnormal_2 group compared to the normal group. Then, considering that each feature had different values, we further explained the one case in detail. As \u003cb\u003eFigure-5b\u003c/b\u003e shown, indoor air quality improved (indoor air improvement\u0026thinsp;=\u0026thinsp;1), no dental decay (dental decay\u0026thinsp;=\u0026thinsp;0), and no drinking (drinking\u0026thinsp;=\u0026thinsp;0) may be beneficial to reduce bone and muscle abnormalities.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eThe comparison of XGBoost with other machine learning models\u003c/h3\u003e\n\u003cp\u003eModel_9 and model_19 were the optimal binary-classification model and multiple-classification model respectively. Based on the accuracy rate (model_9: 87.13%; model_19: 78.95%), false positive rate (model_9: 15.11%; model_19: 17.44%), false negative rate (model_9: 10.59%; model_19: 11.76%) and AUC (model_9: 0.92; model_19: 0.80) of the testing dataset, model_9 performed better than model_19. Based on features included in the model_9, further comparison of XGBoost with random forest and decision tree showed that the accuracy rate of all three models was 87.13% (149/171), but the confusion matrix was different. The false negative rate of XGBoost, random forest, and decision tree were 10.59% (9/85), 12.94% (11/85), and 14.12% (12/85) respectively. The false positive rate of XGBoost, random forest, and decision tree were 15.12% (13/86), 12.79% (11/86), and 11.63% (10/86) respectively.\u003c/p\u003e"},{"header":"DISCUSSIONS","content":"\u003cp\u003eTo early detect and concentrate the high-risk population of musculoskeletal aging-related diseases, this study combined logistic regression, LASSO, and XGBoost algorithms to explore the risk factors of musculoskeletal aging-related diseases, then constructed early detection models based on binary-classification and multiple-classification respectively, which provided an important basis for the early diagnosis and early treatment of musculoskeletal aging-related diseases such as osteosarcopenia and sarcopenia.\u003c/p\u003e \u003cp\u003ePrevious studies have indicated the prevalence of osteoporosis among those aged 40 years or older was 5.0% among men and 20.6% among women.\u003csup\u003e\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e\u003c/sup\u003e In clinical practice, the diagnosis and treatment of osteoporosis are relatively well established. However, sarcopenia has only been classified internationally as a disease in the last ten years, which needs more attention. During aging, muscle and bones are intricately connected tissues displaying marked co-change.\u003csup\u003e\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e\u003c/sup\u003e The risk of hip fracture (HR: 2.67), major osteoporotic fracture (HR: 2.04), and death (HR: 1.91) were significantly higher in patients with osteosarcopenia than in those without osteoporosis or sarcopenia.\u003csup\u003e\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u003c/sup\u003e Besides, Yoo et al. indicated the one-year mortality rate of elderly hip fracture patients with osteosarcopenia (15.1%) was significantly higher than that of patients with osteoporosis (5.1%) or sarcopenia (10.3%).\u003csup\u003e17\u003c/sup\u003e It can be seen that elderly patients with osteosarcopenia are more likely to fall and have a significantly increased risk of fracture, disability, and death after a fall. A meta-analysis shows that the overall prevalence of osteosarcopenia is 18.5%.\u003csup\u003e3\u003c/sup\u003e Under limited conditions, osteosarcopenia is the key to the protection of bone-muscle health in the elderly. Of course, osteoporosis and sarcopenia are also important.\u003csup\u003e\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e\u003c/sup\u003e Consequently, this study set up the multiple-classification and binary-classification to suit the different demands of early detection.\u003c/p\u003e \u003cp\u003eIn this study, multiple-classification included normal (bone and muscle condition are both normal), one abnormality (bone or muscle condition is abnormal), and two abnormalities (bone and muscle condition are both abnormal). Based on this kind of classification, the early detection model built in this study was aimed to solve the problem of risk stratification of musculoskeletal aging-related disease, then subjects with different risks should receive differentiated further processing. There was interaction between muscle and bone at the molecular and cellular level. \u003csup\u003e\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e\u003c/sup\u003e Combined with public literature, we speculate that the two abnormalities should be paid more attention than the one abnormality. The AUC of comparison_② (normal vs abnormal_2) is generally better than that of comparison_① (normal vs abnormal_1), further suggesting that the difference gap between normal and bone-muscle abnormal was more obvious. However, the effect of comparison_① (AUC: 0.79) and comparison_② (AUC: 0.77) is similar and are both higher than that of comparison_③(AUC: 0.51) in model_13. In model_14, the effect of comparison_② (AUC:0.84) is better than that of comparison_③ (AUC:0.74) and comparison_① (AUC:0.67). The BMD result is the diagnostic criteria for osteoporosis and osteosarcopenia.\u003csup\u003e\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e\u003c/sup\u003e In this study, the only difference between model_13 and model_14 is that model_14 includes whether accept BMD examination, while model_13 includes gripping. Nowadays, the uptake of BMD examination remains low. Besides, there may be recall bias during the questionnaire about the BMD results. Fu et al. found osteoporosis was diagnosed in 86.2% of women and half of men among those willing to uptake BMD examination.\u003csup\u003e\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e\u003c/sup\u003e It suggests that even the acceptance of receiving BMD examination may play a critical role in the primary screening of osteoporosis and osteosarcopenia, which explains the better performance of whether to accept a BMD examination in this study.\u003c/p\u003e \u003cp\u003eHowever, the performance of comparison_③ was not well among all multiple-classification models. And the performance of model based on binary-classification was better than model based on multiple-classification. The more categorical the model, the more complex it is to construct. Gripping and calf dimensions were related to the muscle condition of the upper and lower limbs, respectively. In this study, the AUC of model included griping was higher than that of model that included calf dimension, but both were better than the SARC-F score. Similarly, calf circumference demonstrated greater diagnostic accuracy than SARC-F in Thia older adults.\u003csup\u003e\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e\u003c/sup\u003e However, SARC-F questionnaire has been recommended as a screening tool for sarcopenia. SARC-F has shown reasonable diagnostic accuracy for sarcopenia screening and were demonstrated low sensitivity but high specificity, which may partly explain the poor performance in this study.\u003csup\u003e\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e\u003c/sup\u003e Despite its low sensitivity of SARC-F, it proves to be a useful tool for identifying severe cases in early detection taking advantage of its simplicity. Moreover, the SARC-F questionnaire also can be used to predict the presence of mild cognitive impairment in postmenopausal women.\u003csup\u003e\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e\u003c/sup\u003e Concerning bone condition, whether or not accept BMD examination can still be used to some extent as a substitute for BMD diagnosis for primary screening of musculoskeletal aging-related disease.\u003c/p\u003e \u003cp\u003eBased on the optimal binary-classification XGBoost model, we further compared the performance of XGBoost, random forest and decision tree models. In the training dataset, the accuracy rate of the above three models were the same, but XGBoost model had a lower false negative rate and was better appropriate to early detection demands that decreased the missed diagnosis as much as possible. However, there is still a need to further explore the muscle-related diagnostic tools, including the development of new tools and the adjustment of the cut-off value of the existing tools. In addition, the performance of identifying bone or muscle abnormal (abnormal_1 group) from bone and muscle abnormal (abnormal_2 group) is limited, which indicates more tools that can be used to distinguish between osteoporosis or sarcopenia and osteosarcopenia are still to be explored and developed. To this, further work is required to identify biomarkers, which, in turn, may increase the accuracy rate of diagnosis, risk stratification, and targeted treatments to improve health outcomes.\u003c/p\u003e \u003cp\u003eSpecifically, this study indicated some potential risk factors were interpreted concretively by the SHAP algorithm, providing us with new clues. Most risk factors were modifiable. Addressing these modifiable risk factors can prevent, or at least delay, the onset of musculoskeletal aging-related diseases. Previous studies have indicated that menopause, and current smoking are independent risk factors of osteoporosis in females and males respectively, leisure screen time, and coffee intake played significantly causal roles in sarcopenia, and physical inactivity and poor nutrition are two major risk factors for osteosarcopenia.\u003csup\u003e\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e,\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e,\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e\u003c/sup\u003e In this study, multi-dimensional risk factors were paid attention to, such as indoor air quality and cooking at home. As visualized results show, the improvement of indoor air quality, and cook less frequently at home were potential protective factors against musculoskeletal aging-related diseases. Indoor air pollution from cooking at home has been associated with several diseases, including cognitive decline, cardiovascular diseases, and et al. \u003csup\u003e\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e,\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e\u003c/sup\u003e Besides, solid fuel for cooking the number of solid fuel use potentially facilitates the onset and progression of muscle loss and sarcopenia.\u003csup\u003e\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e\u003c/sup\u003e And chronic exposure to biomass smoke increased the risk of bone resorption and consequent osteoporosis.\u003csup\u003e\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e\u003c/sup\u003e Actually, both infiltration of outdoor air pollution into the indoor space and indoor sources (such as cooking or heating practices and household materials) contribute to unique exposure mixtures influencing indoor air quality.\u003csup\u003e\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e\u003c/sup\u003e Regardless of the source of indoor air pollution, it is recommended to actively improve indoor air quality as suggested by the results of this study. However, more validation studies about the risk factors proposed in this study are still needed in the future.\u003c/p\u003e \u003cp\u003eCombined with statistical analysis and machine learning algorithms, this study provided the potential early detection clues for high-risk populations of musculoskeletal aging-related diseases. Subjects enrolled in this study came from different regions in China and were diagnosed with a variety of fracture types, which has a certain representativeness and lays a population foundation for the generalization of the early detection model. However, there still have some limitations. Firstly, the calf circumference, griping and SARC-F score included in this study are simpler and easier to perform in the early detection process. However, more objective muscle-related indexes based on DXA or bioelectrical impedance analysis were still needed. Secondly, some risk factors are proposed by this study, but the relevant investigation content needs to be in-depth, and then accurate intervention recommendations are put forward. Lastly, the sample size is relatively small. In the future, it is still needed to expand the sample size and include multi-center data as much as possible.\u003c/p\u003e \u003cp\u003eIn the elderly, musculoskeletal diseases are common and significantly increase the risk of falls, fractures, and disability, which are also associated with multiple comorbidities. This study mainly explored the risk factors for musculoskeletal aging-related diseases and built the early detection model of high-risk populations. Through early detection models, we can simply and quickly identify high-risk populations of musculoskeletal aging-related diseases, and then carry out diagnostic examinations, so as to promote early diagnosis and early treatment, further reducing its disease burden. Moreover, compared to osteoporosis, efficacious biomarkers for sarcopenia and osteosarcopenia are still currently lacking. It is anticipated that those in the field of aging will figure out answers to these questions and in turn, promote the development of precision medicine for elderly health.\u003c/p\u003e "},{"header":"METHODS","content":"\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003cdiv id=\"Sec12\" class=\"Section3\"\u003e \u003ch2\u003eSubject enrollment\u003c/h2\u003e \u003cp\u003eAll subjects were hospitalized patients with fractures at Beijing Jishuitan Hospital affiliated with Capital Medical University in China from January 2024 to December 2024, aged 50 years and above. The study strictly adhered to the Declaration of Helsinki, and all subjects were required to sign informed consent before participating in the study. This study has been approved by the Ethics Committee of Beijing Jishuitan Hospital affiliated to Capital Medical University. (No. K2024-273-00)\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003eFeature collections\u003c/h2\u003e \u003cp\u003eIn this study, a musculoskeletal aging-related disease questionnaire was designed by the project team to collect the related epidemiological factors. This questionnaire comprised 125 features divided into two sections (general features and grouping features). General features included 12 parts as follows: baseline, cognitive function, covid-19 infection, diet habit, disease history, family history, oral health, physical activity, EQ-5D survey, lifestyle, mental health and pharmacohistory. Grouping features included grip strength, calf girth, SARC-F score, bone mineral density (BMD) T value, BMD Z value, grip grouping, calf girth grouping, SARC-F grouping, and bone density examination. On the day of hospitalization, this questionnaire was administered to the enrolled subjects by trained staff using a tablet computer. All subjects were capable of completing the questionnaire independently.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003eBone and muscle condition\u003c/h2\u003e \u003cp\u003eThe classification of subjects enrolled in this study was mainly based on the condition of bone and muscle. The bone condition was assessed based on the BMD results from dual energy x-ray absorptiometry (DXA) examination. If the subject was diagnosed as osteoporosis (T value \u0026le; -2.5) or osteopenia (-2.5\u0026thinsp;\u0026lt;\u0026thinsp;T value \u0026le; -1.0), it would be classified as the abnormal of bone condition. The muscle condition was assessed based on calf girth, griping and SARC-F score.\u003csup\u003e\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e\u003c/sup\u003e Abnormal calf girth was defined as less than 34 cm for men and 33 cm for women. Abnormal grip strength was defined as less than 28 kg for men and 18 kg for women. A SARC-F score of 4 points or higher was considered abnormal. If subjects exhibited at least one abnormality in calf girth, grip strength, or SARC-F score, they were classified as abnormal muscle condition.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec15\" class=\"Section2\"\u003e \u003ch2\u003eSubject grouping methods\u003c/h2\u003e \u003cp\u003eAbnormalities in muscle or bone are both common among the elderly. In particular, the co-existence of muscle and bone abnormalities deserves more attention. Based on the assessment of bone and muscle condition, there were two grouping methods (binary-classification and multiple-classification) in this study. (1) Binary classification. Subjects enrolled in this study were classified as the normal group (subjects with normal bone condition and normal muscle condition) and the abnormal group (subjects with abnormal bone condition and/or abnormal muscle condition). (2) Multiple classification. Subjects enrolled in this study were classified as the normal group (subjects with normal bone condition and normal muscle condition), abnormal_1 group (subjects with abnormal bone condition or abnormal muscle condition), and abnormal_2 group (subjects with abnormal bone condition and abnormal muscle condition).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec16\" class=\"Section2\"\u003e \u003ch2\u003eStatistical analyses\u003c/h2\u003e \u003cp\u003eCategorical variables were evaluated using the chi-square (X\u003csup\u003e\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u003c/sup\u003e) or Fisher\u0026rsquo;s exact test. According to the type of feature, mice package (V.3.17.0) was used for data interpolation primary, specifically including \u0026ldquo;polr\u0026rdquo;, \u0026ldquo;polyreg\u0026rdquo;, \u0026lsquo;logreg\u0026rsquo;, and \u0026lsquo;pmm\u0026rsquo; methods. Univariate logistic regression analysis (rms package: V.6.7-1) was performed to determine independent risk factors associated with musculoskeletal aging-related diseases. Next, least absolute shrinkage and selection operator regression (LASSO) was also performed to select potential risk factors associated with musculoskeletal aging-related diseases. (glmnet package: V.4.1-8) The above statistical analyses were performed in R Studio (2023.09.0\u0026thinsp;+\u0026thinsp;463). Statistical significance was defined as a two-tailed \u003cem\u003eP\u003c/em\u003e-value\u0026thinsp;\u0026lt;\u0026thinsp;0.05.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec17\" class=\"Section2\"\u003e \u003ch2\u003eMachine learning models\u003c/h2\u003e \u003cp\u003eSubjects were randomly assigned to an 80% training dataset (n\u0026thinsp;=\u0026thinsp;682) and a 20% testing dataset (n\u0026thinsp;=\u0026thinsp;171). Combining the results of univariate logistic regression and LASSO regression, then this study built a XGBoost model (xgboost package: V.1.7.8.1) to further select the potential risk factors of musculoskeletal aging-related diseases based on the importance of each feature. A total of 22 characteristic features (13 general features and 9 grouping features) was included to construct the early detection models based on XGBoost algorithm. (\u003cb\u003eFigure-1b\u003c/b\u003e) And there were ten models by combining general features (n\u0026thinsp;=\u0026thinsp;13) and grouping features (n\u0026thinsp;=\u0026thinsp;9) based on binary-classification (model_1 to model_10) and multiple-classification (model_11 to model_20) respectively. (\u003cb\u003eFigure-1a\u003c/b\u003e) \u0026ldquo;Multi:softmax\u0026rdquo; and \u0026ldquo;binary:logistic\u0026rdquo; were the main objective parameters of multi-classification models and binary-classification models respectively. The eta value, num_class, max_dep, and nrounds were 0.1, 3, 4, and 200 respectively. The performances of the XGBoost models were assessed by the area under the curve (AUC) and accuracy rate (%). (pROC package: V.1.18.5) Then, shapley additive explanations (SHAP) algorithm (shapviz package: V.0.9.7) was used to interpret and extract SHAP values for each feature based on the optimal multiple-classification model and the optimal binary-classification model respectively. Random forest (randomForest package: V.4.7\u0026ndash;1.1) and decision tree (rpart package: V.4.1.24) algorithms were used to compare with the optimal XGBoost model, and the accuracy rate (%), false negative rate (%), and false positive rate (%) were the evaluation index. The above statistical analyses were performed in R Studio (2023.09.0\u0026thinsp;+\u0026thinsp;463). Statistical significance was defined as a two-tailed \u003cem\u003eP\u003c/em\u003e-value\u0026thinsp;\u0026lt;\u0026thinsp;0.05.\u003c/p\u003e \u003c/div\u003e "},{"header":"Declarations","content":"\u003cdiv id=\"Sec18\" class=\"Section2\"\u003e \u003ch2\u003eDATA AVAILABILITY\u003c/h2\u003e \u003cp\u003eData not presented in the manuscript will be provided upon reasonable request to the corresponding authors. The data are not publicly available due to privacy or ethical restrictions.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec19\" class=\"Section2\"\u003e \u003ch2\u003eCODE AVAILABILITY\u003c/h2\u003e \u003cp\u003eThe underlying code for this study [and training/testing datasets] is not publicly available but may be made available to qualified researchers on reasonable request from the corresponding author.\u003c/p\u003e \u003c/div\u003e\n\u003ch2\u003eCONFLICTS OF INTEREST\u003c/h2\u003e \u003cp\u003eThe authors declare no competing interests.\u003c/p\u003e \u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eM.L, M.G, Y.Z and X.J conceived this study. M.L, S.L and C.C initiated the reported analysis. M.L, S.L, C.C and K.C performed figure 1-5. M.L and C.C drafted this manuscript, and all authors contributed to data interpretation and critical revision of this manuscript. All authors have read and approved the manuscript.\u003c/p\u003e\u003ch2\u003eACKNOWLEDGEMENTS\u003c/h2\u003e \u003cp\u003eThis study was supported by the Beijing Municipal Public Welfare Development and Reform Pilot Project for Medical Research Institutes (NO. JYY2023-8, JYY2023-11), the Project supported by Beijing Jishuitan Research Funding (NO. KYYC202301), the Natural science research Beijing Jishutian Hospital (No. ZR-202402), and the National Key Research and Development project (2024YFC3044700).\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eFaidra Laskou et al., Associations of osteoporosis and sarcopenia with frailty and multimorbidity among participants of the Hertfordshire Cohort Study. J Cachexia Sarcopenia Muscle. 13, 220\u0026ndash;229 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBinkley N, B. B, Beyond FRAX: it's time to consider \"sarco- osteopenia\". J Clin Densitom 12, 413\u0026ndash;416 (2009).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eShanping Chen et al., Global epidemiological features and impact of osteosarcopenia: A comprehensive meta-analysis and systematic review. J Cachexia Sarcopenia Muscle 15, 8\u0026ndash;20 (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJun Guo et al., Aging and aging-related diseases: from molecular mechanisms to interventions and treatments. Signal Transduct Target Ther. 7, 391 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eD. Di et al., Early-life tobacco smoke elevating later-life osteoporosis risk: Mediated by telomere length and interplayed with genetic predisposition. J Adv Res. S2090-1232, 00083\u0026ndash;00083 (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eB. Kirk, J. Zanker, G. Duque, Osteosarcopenia: epidemiology, diagnosis, and treatment\u0026mdash;facts and numbers. J Cachexia Sarcopenia Muscle 11, 609\u0026ndash;618 (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTine Kolenda Paulin, Linnea Malmgren, Fiona E McGuigan, K. E. Akesson, Osteosarcopenia: Prevalence and 10-Year Fracture and Mortality Risk \u0026ndash; A Longitudinal, Population-Based Study of 75-Year-Old Women. Calcif Tissue Int 114, 315\u0026ndash;325 (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJ A Kanis, O Johnell, A Oden, H Johansson, E. McCloskey, FRAX and the assessment of fracture probability in men and women from the UK. Osteoporos Int. 19, 385\u0026ndash;397 (2008).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJuergen Bauer et al., Sarcopenia: A Time for Action. An SCWD Position Paper. J Cachexia Sarcopenia Muscle. 10, 956\u0026ndash;961 (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDariusz Tarasewicz et al., Development and Validation of a Diabetic Retinopathy Risk Stratification Algorithm. Diabetes Care 46, 1068\u0026ndash;1075 (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDandan Tang et al., Study on the prediction performance of AIDS monthly incidence in Xinjiang based on time series and deep learning models. BMC Public Health 25, 780 (2025).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBohan Li, Hui Chen, Xiaona Lin, H. Duan, Multimodal learning system integrating electronic medical records and hysteroscopic images for reproductive outcome prediction and risk stratification of endometrial injury: a multicenter diagnostic study. Int J Surg 110, 3237\u0026ndash;3248 (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBinhua Dong et al., Development, validation, and clinical application of a machine learning model for risk stratification and management of cervical cancer screening based on full-genotyping hrHPV test (SMART-HPV): a modelling study. Lancet Reg Health West Pac 55, 101480 (2025).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJunyi Zhang et al., Insights into geospatial heterogeneity of landslide susceptibility based on the SHAP-XGBoost model. J Environ Manage 332, 117357 (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLinhong Wang et al., Prevalence of Osteoporosis and Fracture in China: The China Osteoporosis Prevalence Study. JAMA Netw Open 4, e2121106 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eK. M. Kim et al., Longitudinal Changes in Muscle Mass and Strength, and Bone Mass in Older Adults- Gender-Specific Associations Between Muscle and Bone Losses. J Gerontol A Biol Sci Med Sci 73, 1062\u0026ndash;1069 (2018).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJun Il Yoo, Hyunho Kim, Yong Chan Ha, Hyuck Bin Kwon, K. H. Koo, Osteosarcopenia in Patients with Hip Fracture Is Related with High Mortality. J Korean Med Sci 33, e27 (2018).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eOscar Rosas-Carrasco, Betty Manrique-Espinoza, J. C. L\u0026oacute;pez-Alvarenga., Beatriz Mena-Montes, I. Oma\u0026ntilde;a-Guzm\u0026aacute;n, Osteosarcopenia predicts greater risk of functional disability than sarcopenia: a longitudinal analysis of FraDySMex cohort study. The Journal of nutrition, health and aging 28, 100368 (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBen Kirk, Sarah Miller, Jesse Zanker, G. Duque, A clinical guide to the pathophysiology, diagnosis and treatment of osteosarcopenia. Maturitas 140, 27\u0026ndash;33 (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFan Yu, W. Xia, The epidemiology of osteoporosis, associated fragility fractures, and management gap in China. Archives of Osteoporosis 14, 32 (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eS-H Fu et al., Screening of Fracture Risk and Osteoporosis Among Older Long-term Care Residents: A Prospective Study. J Nutr Health Aging. 27, 1255\u0026ndash;1261 (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eEkasame Vanitcharoenkul et al., Evaluating SARC-F, SARC-CalF, and calf circumference as diagnostic tools for sarcopenia in Thai older adults: results from a nationwide study. BMC Geriatr 24, 1043 (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJia-Yu Guo et al., The application of Chinese version of SARC-F and SARC-CalF in sarcopenia screening against five definitions: a diagnostic test accuracy study. BMC Geriatr 24, 883 (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMar\u0026iacute;a S Vallejo et al., Risk of sarcopenia: A red flag for cognitive decline in postmenopause? Maturitas 194, 108193 (2025).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eXianxian Yang et al., Prevalence and risk factors associated with osteoporosis among residents aged above 20 years old in Chongqing, China. Arch Osteoporos 16, 57 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMingchong Liu et al., Causal Roles of Lifestyle, Psychosocial Characteristics, and Sleep Status in Sarcopenia: A Mendelian Randomization Study. J Gerontol A Biol Sci Med Sci 79, glad191 (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTingting Xu et al., Association between solid cooking fuel and cognitive decline: Three nationwide cohort studies in middle-aged and older population. Environ Int 173, 107803 (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHsiao-Chi Chuang et al., Long-term indoor air conditioner filtration and cardiovascular health: A randomized crossover intervention study. Environ Int 106, 91\u0026ndash;96 (2017).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhigang Hu, Yufeng Tian, Xinyu Song, Fanjun Zeng, A. Yang, Associations between indoor air pollution for cooking and heating with muscle and sarcopenia in Chinese older population. J Cachexia Sarcopenia Muscle 14, 2029\u0026ndash;2043 (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHirak Saha, Bidisha Mukherjee, Banani Bindhani, M. R. Ray, Changes in RANKL and osteoprotegerin expression after chronic exposure to indoor air pollution as a result of cooking with biomass fuel. J Appl Toxicol 36, 969\u0026ndash;976 (2016).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJared Radbel, Meghan E Rebuli, Howard Kipen, E. Brigham, Indoor air pollution and airway health. J Allergy Clin Immunol 154, 835\u0026ndash;846 (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiang-Kung Chen et al., Asian Working Group for Sarcopenia: 2019 Consensus Update on Sarcopenia Diagnosis and Treatment. J Am Med Dir Assoc 21, 300\u0026ndash;307.e302 (2020).\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Osteoporosis, Sarcopenia, Osteosarcopenia, Risk factor, Machine learning","lastPublishedDoi":"10.21203/rs.3.rs-6124947/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6124947/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eLate-diagnosis is one of the main bottlenecks in musculoskeletal aging-related diseases prevention, and it is urgent to build early detection model. Twenty-two features were included to build early detection models based on binary and multiple classification respectively by XGBoost. In testing, the accuracy rate (63.74%~92.40%) and AUC (0.74\u0026thinsp;~\u0026thinsp;0.96) of binary-classification models were higher than the accuracy rate (61.40% ~85.96%) and AUC (0.63\u0026thinsp;~\u0026thinsp;0.86) of multiple-classification models. The optimal binary-classification model had an accuracy rate of 87.13% and an AUC of 0.92 in testing, including cooking, drinking milk, electronic devices use time, dental implant, dental decay, professional oral cleaning, falls in the past year, life satisfaction, the degree of pain or discomfort, indoor air improvement, drinking, body mass index, time spent indoors, grip grouping, SARC-F grouping, calf girth grouping and bone density examination. In elderly, musculoskeletal aging-related diseases can be early detected by model based on epidemiological factors.\u003c/p\u003e","manuscriptTitle":"Early detection for elderly people with musculoskeletal aging related diseases based on artificial intelligence model","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-04-16 17:55:00","doi":"10.21203/rs.3.rs-6124947/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"9811d14b-189c-45d6-9648-29b06c9dfab9","owner":[],"postedDate":"April 16th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":47170062,"name":"Health sciences/Biomarkers"},{"id":47170063,"name":"Health sciences/Risk factors"}],"tags":[],"updatedAt":"2025-04-23T22:23:14+00:00","versionOfRecord":[],"versionCreatedAt":"2025-04-16 17:55:00","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-6124947","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6124947","identity":"rs-6124947","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}