Predicting Fracture Risk for Chinese Elderly Osteoporosis Patients by Hybrid Model

doi:10.21203/rs.3.rs-3109583/v1

Download PDF

Research Article

Predicting Fracture Risk for Chinese Elderly Osteoporosis Patients by Hybrid Model

https://doi.org/10.21203/rs.3.rs-3109583/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

Background: In China, there are many elderly people who suffer from osteoporosis. Fractures, the most serious complication of osteoporosis, can greatly affect people’s quality of life and add pressure to the social pension system. Thus, predicting fracture risk in elderly patients with osteoporosis is imperative, which allows for timely treatment and ameliorates the strain on healthcare resources. Machine learning (ML) models based on Electronic Health Records (EHR) can be contracted to predict fracture risk.

Methods: Data of patients with osteoporosis were extracted from the Electronic Health Records (EHR) in Xinhua Hospital. Demographic and clinical features that are readily available from EHR were used to develop prediction models based on 12 independent ML algorithms (Naïve Bayes classifier [BP], logistic regression [LR], decision tree [DT], k-nearest neighbors [KNN], support vector machine [SVM], random forests [RF], extremely randomized trees [ERT], gradient-boosted decision trees [GBDT], adaptive boosting [AdaBoost], categorical boosting [CatBoost], extreme gradient boosting [XGBoost], multilayer perceptron [MLP]) and 3 hybrid ML models (XGBoost + MLP, XGBoost + LR, XGBoost + SVM). A comprehensive importance score was designed to interpret features from several aspects.

Results: A total of 8530 patients with osteoporosis were included for analysis, of which 1090 cases (12.8%) were fracture patients. The hybrid XGBoost and SVM model had the best predictive performance in terms of accuracy and precision (above 90%) among all models. We obtained 20 important features according to the comprehensive importance score, which represents high-risk factors for fractures and was interpreted from a clinical point of view.

Conclusions: The hybrid XGBoost and SVM model can be reliable tools for predicting the risk of fracture in patients with osteoporosis. And the hybrid model had the best predictive performance, which can be used to assist clinicians in identifying high-risk fracture patients and implementing early interventions.

Fracture risk assessment

Machine Learning

Data mining

Fracture prevention

Osteoporosis

China's rapidly aging population is one of the fastest growing in the world, with over 267 million people over the age of 60. They constitute about 18.9% of the nation's total population, a proportion projected to reach 28% by 2040. The increasing aging population presents significant challenges to China’s public health and social development.

One critical health challenge is the rising incidence of fractures, particularly osteoporotic fractures, among the elderly [1]. These fractures pose a considerable burden on the healthcare system [2]. In China, the prevalence of osteoporosis in the elderly (>60 years old) is 36%, with a gender breakdown of 23% in males and 49% in females [3]. Over 2.23 million fractures in China are attributed to osteoporosis [4].

Fractures significantly impact patients' quality of life, posing severe threats and enormous strain on healthcare resources. Despite the high prevalence and considerable harm, fracture diagnosis and treatment rates remain depressingly low in China [5]. The insidious nature of bone density loss, characterized by a protracted course and no overt clinical manifestations in the early stages, makes fracture prevention challenging. In many instances, patients become aware of their decreased bone density only when seeking treatment for other ailments.

Therefore, the ability to predict the fracture risk for elderly people with osteoporosis provides benefits to patients and society. The prediction can be performed based on enormous data already existing in the hospital during patients’ check-ups, called Electronic Health Records (EHR). It normally includes patients’ demographics, progress notes, problems, medications, vital signs, past medical history, and so on [6]. Even though patients might not take the bone examination on purpose, data collected through other examinations might contain some factors that are predictive of fracture. The existing data contain useful information for us to explore. For example, many parameters from blood and urea testing can be monitored and might provide an indication of the risk of bone metabolism. As a result, they might offer predictive information about the risk of fracture for elderly patients with osteoporosis.

The Fracture Risk Assessment (FRAX) tool is a risk-assessment tool developed by the World Health Organization, which calculates the 10-year probability of a major osteoporosis-related fracture [7]. It considers several clinical risk factors (CRFs), including age, sex, body mass index (BMI), and dichotomized risk factors comprising prior fragility fracture, parental history of hip fracture, current tobacco smoking, ever use of long-term oral glucocorticoid use, rheumatoid arthritis, other causes of secondary osteoporosis and high alcohol consumption. Bone mineral density could be added to boost the prediction accuracy. Although FRAX is viewed as one of the most accurate methods for assessing the risk of fractures, there are some limitations to it. First, some important factors related to fractures are not included in FRAX, such as tumble and information included in blood or urine examination indicators, which are available in EHR. Second, FRAX is suitable for people aging from 40 to 90 years. With the increase in population aging, it is necessary to predict the fracture risk of people over 90.

Recently, there are many machine learning methodologies for fracture risk assessment. Atkinson et al. assessed fracture risk using GBM models based on bone images [8]. Vincenzo et al. used several machine learning models to predict fracture risk according to genotypes [9]. Chen et al. developed a hybrid deep learning model to predict the risk of fracture for diabetics [10]. However, there has been no model established based on EHR data for Chinese elderly patients with osteoporosis.

In this study, we aim to predict the fracture risk of elderly people with osteoporosis based on real-world data and discover the factors related to fractures, including the well-known factors as well as those not yet being paid attention to before. Therefore, the study will offer a new clinical direction that helps in the prevention of high-risk groups. The workflow of our study is shown in Figure 1.

Data source

We obtained the EHR of outpatients and inpatients in Xinhua Hospital for the period between July 2012 and October 2017, which consists of diagnosis data, medical records, vital signs, inspection index, medication history, and so on. There are a total of 14418 patients in this dataset.

Data extraction

Senior patients with osteoporosis were selected as our research objective accounting for 8530. Abnormal records are removed from the dataset caused by input errors, such as systolic blood pressure > 300 mmHg, age of menopause <30, height < 30 cm, etc. And we excluded patients with fractures before a diagnosis of osteoporosis or caused by trauma.

From the diagnosis data, we extracted the osteoporosis patients based on the International Classification of Diseases (ICD) -10 code of the disease and represented it in binary form. From medical records data, we obtained patients’ age and gender as well as medical histories, such as smoking history, drinking history, and fracture history. From inspection data, parameters of blood and urine tests were obtained, such as blood Calcium, blood glucose, urine, and so on. If one patient has multiple medical records, the record was selected with the closest time to the diagnosis result. We obtained patients' height, weight, and blood pressure from vital signs data. From doctor’s order and medication data, we obtained the patient's medication status and medication history, such as glucocorticoids, aromatase inhibitors, thyroid hormones, proton pump inhibitors, and other medicines that might affect bone metabolism. From symptom data, we obtained information about arthralgia, weight loss, and puffiness. From bone density data, we collected the bone density of patients' lumbar spine1~4. The flowchart of data extraction is shown in Figure 2.

We started by incorporating all possible features to mine the potential risk factors of the elderly with osteoporosis and then neglected features with too few non-null values. As a result, 146 factors were obtained which can be divided into six categories, including demographic data, medical history, inspection data, illness, medicine, and symptoms. Next, we formulated the task of identifying old osteoporosis patients for their risk of fracture as a binary classification problem by training machine learning models with these features.

Statistical analysis

We conducted descriptive statistical analysis and significance testing on the dataset to ascertain if there were any statistically significant differences between the fracture and non-fracture groups’ indicators. We categorized the characteristics into two groups: categorical and numerical variables. If these characteristics met the criteria for parametric testing, we applied the chi-square test and the t-test, respectively. For those not meeting these criteria, we used non-parametric testing. Further details are presented in Table 1.

Table 1. Baseline demographic and clinical characteristics of the included patients

variables	Fracture(n=1090)	No Fracture(n=7440)	Missing Rate	P value
Demography
Age, years, mean±SD(range)	78.75±9.56 (60-100)	74.7±9.61 (60-109)	0	<0.001
Female, n (%)	853(78.26)	5804(78.01)	0	0.89
Weight, kg, mean±SD(range)	57.88±10.3(35.5-89.5)	58.31±9.61(32-93)	92.83	0.34
Height, cm, mean±SD(range)	162.25±5.7(145-170)	159.98±7.12(140-180)	96.67	0.04
Medical history
Smoke, n (%)	29(2.66)	305(4.1)	51.49	0.04
Vaccined, n(%)	299(27.97)	2734(28.51)	49.57	0.74
Drink, n(%)	13(1.19)	106(1.42)	52.06	0.64
Fracture, n(%)	98(8.99)	144(1.94)	0	<0.001
Inspection
Calcium-blood, mmol/L, mean±SD(range)	2.18±0.29(1.07-3.32)	2.17±0.28(0.43-3.22)	78.47	0.55
C-reactive protein, mg/L, mean±SD(range)	26.09±31.21(1-174)	23.11±24.82(1-141)	84.69	0.53
Na-blood, mmol/L, mean±SD(range)	140.42±3.31(125-148)	140.23±3.36(113-154.37)	71.15	0.25
Apolipoprotein A/B ratio, %, mean±SD(range)	1.63±0.68(0.7-3.55)	1.66±1.21(0.4-33.4)	85.21	0.77
Alkaline phosphatase, U/L, mean±SD(range)	99.45±43.18(27.00-381.00)	86.99±62.99(16.00-1779.33)	79.13	<0.001
α2-globulin, g/L, mean±SD(range)	9.85±1.99(4.20-15.10)	9.79±2.47(5.8-22.6)	93.29	0.23
HDL-C, mmol/L, mean±SD(range)	1.36±0.37(0.64-3.15)	1.47±0.40(0.38-3.88)	84.89	0.01
Total Cholesterol (TC), mmol/L, mean±SD(range)	4.44±0.94(2.75-6.76)	4.8±1.2(2.01-12.55)	84.76	0.01
DXA value of L1, mean±SD(range)	-1.21±1.24(-3.84 - 2.07)	-1.31±1.19(-3.87 - 2.84)	50.33	<0.001
DXA value of Lumbar spine 1, mean±SD(range)	-1.39±1.48(-3.96 - 2.98)	-1.5±1.11(-3.91-2.99)	50.33	0.77
DXA value of Lumbar spine 2, mean±SD(range)	-1.85±1.3(-4 - 2.7)	-1.77±1.21(--4-2.94)	50.36	0.2
Illness
Heart Disease(%)	61(5.6)	564(7.58)	0	0.14
diabetes, n (%)	144(13.47)	2141(22.33)	0	<0.001
Symptoms
Weight loss, n (%)	98(9.17)	982(10.24)	0	0.98
Joint Pain, n(%)	3(0.28)	6(0.08)	0	0.18
Medicines
Acarbose, n (%)	35(3.27)	417(4.35)	0	0.23
Omeprazole, n (%)	126(11.56)	189(2.54)	0	<0.001

Data Processing

The raw data went through a data-wrangling pipeline including missing value handling and unbalanced label processing.

Missing data is common in clinical research, which occurs when the value of the variables of interest is not measured or recorded for all subjects in the sample [11]. There is a high rate of missing data in our dataset so an appropriate data handling criterion needs to be established. If all records with missing values are deleted, the data set will become extremely small; on the other hand, if all the data are considered with missing values imputed, then most of the values would be inferred, and the analysis results would be rendered quite unreliable. XGBoost is applied for the pilot study for its capability of automatically handling missing values. The experiment demonstrated that the model exhibits optimal generalization capabilities when records with a feature missing rate exceeding 55% are deleted. After eliminating records where the missing rate of any feature exceeded 55%, we obtained a total of 1913 samples for subsequent training.

Several methods were used to deal with the missing values, including layered average filling, missMDA [12], and random forest regression [13]. XGBoost was applied to compare the results of the three filling methods, and Area Under Curve (AUC) was used as a measure. Results are shown in Figure 3 that random forest regression filling performs the best out of the three.

The labels in our dataset are extremely unbalanced which can lead to poor predictive performance. One efficient and flexible strategy for solving this problem is to employ sampling techniques before training a classification learning model [14]. We randomly split the dataset into training set and testing set, and then only performed oversampling and undersampling methods to the training set, leaving the testing set unprocessed. In general, a combination of oversampling and undersampling of the majority class yields better classifier performance. Once again, XGBoost was applied here to evaluate several sampling methods. Results are shown in Figure 4 that the combination of the SMOTE and Tomek links outperforms the rest.

Models

The task of identifying fracture risk is formulated as a classification problem by training machine models with features collected from real-world data. The factors from the EHR of patients were represented by a vector x. Each entry of x described a feature, serving as the input to the model. The prediction of the fracture risk is represented by y, which is the output of the model, in the form of binary. The structure of this hybrid model is presented in Figure 5. XGBoost was used for the feature transformation and SVM was used as the classifier to predict fracture risks.

Model evaluation

The complete data set was divided into training set and testing set at a ratio of 4:1. We oversampled the training set by adapting Synthetic Minority Oversampling Technique (SMOTE) to generate the minor class, and then undersample it by Tomek links to avoid overfitting. We repeated the experiment 100 times to get the average metrics score. The parameters are shown in Table 2, and the results of each model are shown in Table 3.

Table 2

Several models and their parameters
Method	Parameters
LR	L2 regularization strength = 0.4
SVM	Gaussian kernel, C = 1.0, 𝛾=0.008
DT	Max depth = 20
KNN	Number of neighbors = 3
RF	Number of trees = 200
ERT	Number of trees = 400
GBDT	Max number of trees = 500
AdaBoost	Max number of trees = 200
XGBoost	Max number of trees = 300
MLP	Hidden layer size = [50,50,10], Dropout = 0.5

Table 3

Model performance metircs
Models	Accuracy	Precision	Recall	F1
BP	72.43	68.8	75.18	68.71
LR	85.07	79.1	83.45	80.75
DT	83.76	77.55	81.72	79.1
KNN	80.99	74.68	79.49	76.19
SVM	88.64	84.96	82.87	83.77
RF	88.56	84.36	83.55	83.91
ERT	89.37	85.59	84.64	85.05
GBDT	88.12	83.27	84.56	83.82
AdaBoost	86.66	81.24	83.91	82.32
CatBoost	82.92	77.21	81.14	78.3
XGBoost	89.4	85.4	85.19	85.24
MLP	87.49	82.34	84.15	83.11
XGBoost + MLP	90.34	87.75	83.44	85
XGBoost + LR	90.08	87.49	82.86	85
XGBoost + SVM	90.6	89.1	82.79	85.1

And then, we explored the effect of the number of boosted trees on the recall score of the model. We are more interested in recall here because recall measures the ability to predict fracture risk from a population at high risk for fracture. The performance of the hybrid model is consistently better than that of the single XGBoost. As shown in Fig. 6, when the number of trees is up to around 300, there is no significant improvement in the performance.

Explanation of risk factors

We have formulated a strategy to identify critical attributes essential for discerning fracture risk factors. This strategy integrates the significance scores from four crucial perspectives. Weight, the number of times a feature is used to split the data across all trees; Gain, the average gain across all splits the feature is used in; Cover, the average coverage across all splits the feature is used in. The feature importance scores of the above three metrics are shown in Fig. 7. And we also combined SAHP (SHapley Additive exPlanations) [15] values to interpret features. The 20 most important features obtained using SHAP are shown in Fig. 8.

To get the comprehensive feature importance score, we first normalized the values obtained by the four methods. In this way, all values are scaled between 0 and 1.

$${x}^{*}=\frac{x-min\left(x\right)}{max\left(x\right)-min\left(x\right)}$$

Then, these four scores are linearly combined to generate the comprehensive feature importance score.

$$comprehensive feature importance score ={\alpha } \times \text{w}\text{e}\text{i}\text{g}\text{h}\text{t} +{\beta } \times \text{g}\text{a}\text{i}\text{n} +$$

$${\gamma }\times \text{c}\text{o}\text{v}\text{e}\text{r} +{\theta }\times \text{S}\text{H}\text{A}\text{P} \text{v}\text{a}\text{l}\text{u}\text{e}\text{s}$$

Optimal coefficients are determined by evaluating various combinations. For each combination, a composite feature importance score was generated and ranked, isolating the top 20 features. These features were then fed into the model to determine accuracy, precision, recall, and F1-score..

Our model achieved the best results when coefficients for cover and gain were set at 0.3, and weight and SHAP were set at 0.2. The top-ranked features, in order, were Calcium (Ca)-blood, Apolipoprotein A/B ratio-blood, Weight loss, C Reactive Protein (CRP)-blood, Vaccination history, High-density lipoprotein cholesterol (HDL-C)-blood, Alpha2globulin (alpha2-GLO)-blood, Alkaline phosphatase (ALP)-blood, Heart disease, Retinol binding protein (RBP)-blood, Smoking history, Sodium (Na)-blood, Bilirubin (BIL)-blood, Weight, Alkaline granulocyte count-blood, DXA value of Lumbar spine 1, White blood cell esterase-urine, Insulin (2 hours)-blood, Age, and DXA value of L1-L2.

For comparison, all features were input into the model, but only the top 20 features as obtained above are illustrated in Fig. 9. When using only these 20 features, the model was able to achieve a superior classification effect. This suggests that these 20 features are predictive of fractures in elderly patients with osteoporosis, and thus could be considered risk factors for fracture.

This study constructed a hybrid model combining XGBoost and SVM to predict the fracture risk for elderly patients with osteoporosis based on a real-world dataset, and we determined risk factors for fracture by designing a comprehensive feature importance score. All the possible prediction variables were considered as the input to the model in the beginning. The data went through a wrangling pipeline including missing value imputation and imbalanced label handling. Then, we tested several machine learning models, where the hybrid model combing XGBoost and SVM achieved a better result than other benchmarks.

In the realm of medical applications where high dimensionality and limited data are common phenomena, high prediction accuracy and model stability are crucial, and the fusion of XGBoost and SVM into a hybrid model presents several advantages. In simple terms, the hybridization of the two models could enhance accuracy by capitalizing on the different aspects of the data each model excels at capturing. Concretely speaking, the amalgamation could amplify the models' inherent robustness, given XGBoost's aptitude for handling noise and outliers, and SVM's resistance to overfitting when appropriately kernelized and regularized. This can result in more reliable predictions, a highly desirable trait in medical applications. Additionally, the combination model can efficiently manage imbalanced data, a frequent concern in medical scenarios as well, by adjusting the weights on minority classes in XGBoost and leveraging techniques such as SMOTE in SVM. Moreover, despite SVM often being regarded as a "black box", the hybrid model can provide a level of interpretability via the XGBoost component by examining the significance of the features, offering valuable insights into which factors are driving predictions. With XGBoost's capacity for capturing complex non-linear relationships and SVM's proficiency in binary and multiclass classification problems being particularly notable, Both XGBoost and SVM have demonstrated superior performance across diverse tasks[16].

Our comprehensive feature importance score considered four aspects of information (weight, cover, gain, SHAP) to mine the top 20 important features shown in Fig. 9. These features play important roles in this model, which could be viewed as the risk factors for fracture. Some of the features [17] found during this study are well known to be associated with osteoporosis or even fractures, such as smoking history [18], weight loss [19], DXA value of Lumbar spine 1, weight, age, sodium-blood and DXA value of L1-L2 [7].

Moreover, several factors were not well demonstrated in prior studies and appeared to be related to osteoporosis, which should be focused on, such as (i) bone metabolite: Calcium-blood and Alkaline phosphatase (ALP)-blood, (ii) inflammatory response markers: C reactive protein-blood, and (iii) indicators of lipid metabolism: Apolipoprotein A/B ratio-blood and High-density lipoprotein cholesterol (HDL-C)-blood. The ranking of these five factors in our model is 1, 8, 4, 2, and 6 respectively, all of which are very high. The detailed SHAP diagram is shown in Fig. 10.

Our model predicted that calcium (Ca)-blood may be a high-risk factor for fracture. Even though 99% of the body’s calcium is stored in the bones with the remaining 1% in the blood, calcium-blood is much easier to be measured than calcium-bone; therefore, it is a routine laboratory test for patients with osteoporosis. For patients with primary osteoporosis, their blood calcium values are usually within the reference range. As we all know [20] that blood calcium and parathyroid hormone (PTH) levels have reciprocal regulatory effects, such as decreasing concentrations of blood calcium ions, secondary secretion of increased PTH by parathyroid chief cells, stimulation of osteoclast (OC) proliferation, inhibition of osteoblast (OB) activity, and resulting osteoporosis. A recent study [21] in the Pakistani population suggested that patients with osteoporosis have significantly lower serum calcium than controls. Likewise in our model, as shown in Fig. 10 when blood calcium levels decrease (blue), the predicted probability of fracture (SHAP value) increases. Alkaline phosphatase (ALP) is a hydrolase enzyme responsible for removing phosphate groups from many molecules, including nucleotides, proteins, and alkaloids [22]. It belongs to bone formation markers among bone turnover biochemical markers (BTMS) and can react to osteoblast activity and bone formation status. Although not useful in the diagnosis of osteoporosis, ALP plays an important role in the differential diagnosis of several skeletal disorders, determination of bone turnover types, monitoring treatment adherence, and evaluation of drug efficacy [23], and the levels of ALP are usually normal or mildly elevated in patients with primary osteoporosis. A recent study has also shown that the activity of serum total ALP > 129 U/L is used as an indicator for osteoporosis in males [24], consistent with what our model predicts (shown in Fig. 10).

C-reactive protein (CRP) participates in the immune response [25], which is a non-specific and high-sensitivity inflammatory biomarker. The inflammatory processes are involved in a wide variety of physical health problems and systemic chronic inflammation (SCI) often increases with age [26]. Meanwhile, studies [27, 28] also show that older individuals have higher circulating levels of cytokines, chemokines, acute phase proteins, and greater expression of genes involved in inflammation. Moreover, SCI is persistent and ultimately causes collateral damage to tissues and organs over time, such as by inducing osteoporosis. In addition to age, physical inactivity was found to be directly associated with increased anabolic resistance, increased CRP levels, and increased levels of proinflammatory cytokines in healthy individuals [29]. These effects, in turn, promote several inflammation-related pathophysiologic alterations, including osteoporosis [30, 31]. Furthermore, several studies even showed that increased CRP was linked to an increased fracture rate due to osteoporosis [32, 33]. Overall, our model's predictions agree with these views mentioned in Fig. 10.

The association between lipid and bone metabolism has become an increasing focus of interest in recent years [34]. Our model also predicts two lipid metabolism-related indicators: Apolipoprotein A/B (apoA/B) ratio-blood and High-density lipoprotein cholesterol (HDL-C)-blood. A study by Dennison et al [35], investigating the correlation between bone mineral density (BMD) and lipid profiles, observed that total spine BMD was inversely correlated with levels of apoA but positively associated with levels of apoB in males and females, which agrees with our model results. However, regarding HDL-C, our model's trends don’t appear to match exactly what has been reported. On the one hand, Ackert-Bicknell et al. proved that there is sufficient evidence to conclude that bone metabolism and HDL-C are genetically linked, and HDL-C can interact directly with osteoblasts and osteoclasts [36]. Yamaguchi et al [37] investigated the correlation between plasma lipid levels and BMD and found that low levels of HDL-C were associated with an increased risk of vertebral fracture (similar to those predicted by our model). Nevertheless, another study reported the opposite result [35]. On the other hand, Hsu et al [38] conducted a study that aimed to analyze the association between plasma lipid profile and BMD, bone mineral content (BMC), and osteoporotic fractures in 7137 Chinese males, and 4585 premenopausal and 2248 postmenopausal females. No significant correlation between whole-body BMC and levels of HDL-C was detected. Similarly, another study showed no association between HDL-C and BMD [39]. Several factors may be responsible for these discrepancies. Firstly, the differences in non-modifiable characteristics of the subjects, including age, sex, and medication history may have introduced bias. Secondly, the differences in modifiable characteristics, including cigarette and alcohol consumption, or physical activity among the study subjects may also have led to bias. Apart from these reasons, the use of different research methodologies may have affected the results. Therefore, the relationship between lipid profile and bone metabolism warrants further investigation.

To the best of our knowledge, some of the remaining factors predicted by our model, such as heart disease [17], Retinol binding protein (RBP) -blood [40], and Bilirubin (BIL)-blood [41, 42], also seem to have some relationship with fractures and future work is suggested to focus more on these indicators.

Several limitations of our work need to be mentioned here. Firstly, the retrospective and observational nature of our study may lead to inevitable bias. Secondly, given that the raw data is imbalanced, we have performed oversampling and undersampling of the training set, which might lead to deviation from the true value. Thirdly, the patient’s data came from a single center in China. Therefore, further research with large samples and multiple centers is necessary to validate our model’s performance.

In conclusion, the hybrid model incorporating XGBoost and SVM has demonstrated utility as a predictive tool for assessing fracture risk in elderly patients. Furthermore, our study has identified several risk factors for fracture, providing valuable new insights for elderly patients with osteoporosis in clinical practice.

Machine learning

EHR

Electronic Health Records

CRFs

Clinical risk factors

BMI

Body mass index

ICD-10

International Classification of Diseases and Tenth Revision

XGBoost

Extreme Gradient Boosting

SVM

Support vector machine

CRP

C-reactive protein

ALP

Alkaline phosphatase

HDL-C

High-density lipoprotein cholesterol

Total Cholesterol

DXA

Dual energy X-ray absorptiometry

Random forests

AUC

Area Under Curve

SMOTE

Synthetic Minority Oversampling Technique

Logistic regression

Decision Tree

KNN

k-Nearest neighbors

ERT

extremely randomized trees

GBDT

Gradient-boosted decision trees

AdaBoost

Adaptive Boosting

MLP

Multilayer Perceptron

Naïve Bayes classifier

CatBoost

Categorical Boosting

SHAP

Shapley Additive exPlanations

RBP

Retinol binding protein

BIL

Bilirubin

PTH

parathyroid hormone

osteoclast

osteoblast

BTMS

bone turnover biochemical markers

SCI

systemic chronic inflammation

BMD

bone mineral density

BMC

bone mineral content.

Acknowledgments

This work was funded by the National Key R&D Program of China (2020YFC2005502), the National Natural Science Foundation of China (82072142), and the Science and Technology Commission of Shanghai Municipality (Project No. 19401900500).

Author contributions

The experiments are designed by Xf G and Aj X, and performed by Mh L, Zc M, and Jw R. Data analysis is conducted by XW, and Mh L, XW, Xd X, and Aj X drafted the manuscript.

Availability of data and materials

The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.

Ethics approval and consent to participate

All procedures followed complied with the ethical standards of the responsible committee on human experimentation (institutional and national) and the Helsinki Declaration of 1975, as revised in 2000(5). Informed consent was obtained from all patients included in the study.

Consent for publication

Not applicable.

Competing interests

Menghan Liu, Xin Wei, Xiaodong Xing, Zicheng Ma, Jiwu Ren, Xiaofeng Gao, Ajing Xu declared that they have no conflict of interest.

Collaboration GCKD. Global, regional, and national burden of chronic kidney disease, 1990–2017: a systematic analysis for the Global Burden of Disease Study 2017. Lancet. 2020;395:709–33.
Lin X, Xiong D, Peng YQ, Sheng ZF, Wu XY, Wu XP, Wu F, Yuan LQ, Liao EY. Epidemiology and management of osteoporosis in the People's Republic of China: current perspectives. Clin Interv Aging. 2015;10:1017–33.
Chen P, Li Z, Hu Y. Prevalence of osteoporosis in China: a meta-analysis and systematic review. BMC Public Health. 2016;16:1039.
Zeng Q, Li N, Wang Q, Feng J, Sun D, Zhang Q, Huang J, Wen Q, Hu R, Wang L, et al. The Prevalence of Osteoporosis in China, a Nationwide, Multicenter DXA Survey. J Bone Miner Res. 2019;34:1789–97.
Yu F, Xia W. The epidemiology of osteoporosis, associated fragility fractures, and management gap in China. Arch Osteoporos. 2019;14:32.
Menachemi N, Collum TH. Benefits and drawbacks of electronic health record systems. Risk Manag Healthc Policy. 2011;4:47–55.
Kanis JA, Hans D, Cooper C, Baim S, Bilezikian JP, Binkley N, Cauley JA, Compston JE, Dawson-Hughes B, Fuleihan E-H. Interpretation and use of FRAX in clinical practice. Osteoporos Int. 2011;22:2395–411.
Atkinson EJ, Therneau TM, Melton LJ 3rd, Camp JJ, Achenbach SJ, Amin S, Khosta S. Assessing fracture risk using gradient boosting machine (GBM) models. J Bone Miner Res. 2012;27:1397–404.
Forgetta V, Keller-Baruch J, Forest M, Durand A, Bhatnagar S, Kemp J, Morris JA, Kanis JA, Kiel DP, McCloskey EV. Machine learning to predict osteoporotic fracture risk from genotypes. BioRxiv 2018:413716.
Chen Y, Yang T, Gao X, Xu A. Hybrid deep learning model for risk prediction of fracture in patients with diabetes and osteoporosis. Front Med. 2022;16:496–506.
Austin PC, White IR, Lee DS, van Buuren S. Missing data in clinical research: a tutorial on multiple imputation. Can J Cardiol. 2021;37:1322–31.
Josse J, Husson F. missMDA: a package for handling missing values in multivariate data analysis. J Stat Softw. 2016;70:1–31.
Hong S, Lynn HS. Accuracy of random-forest-based imputation of missing data in the presence of non-normality, non-linearity, and interaction. BMC Med Res Methodol. 2020;20:1–12.
Liu B, Blekas K, Tsoumakas G. Multi-label sampling based on local label imbalance. Pattern Recogn. 2022;122:108294.
Lundberg SM, Lee S-I. A unified approach to interpreting model predictions. Adv Neural Inf Process Syst 2017, 30.
Chang W, Liu Y, Wu X, Xiao Y, Zhou S, Cao W. A New Hybrid XGBSVM Model: Application for Hypertensive Heart Disease. IEEE Access. 2019;7:175248–58.
LeBoff M, Greenspan S, Insogna K, Lewiecki E, Saag K, Singer A, Siris E. The clinician’s guide to prevention and treatment of osteoporosis. Osteoporos Int. 2022;33:2049–102.
Yoon V, Maalouf N, Sakhaee K. The effects of smoking on bone metabolism. Osteoporos Int. 2012;23:2081–92.
Van Loan MD, Johnson HL, Barbieri TF. Effect of weight loss on bone mineral content and bone mineral density in obese women. Am J Clin Nutr. 1998;67:734–8.
Chiodini I, Bolland MJ. Calcium supplementation in osteoporosis: useful or harmful? Eur J Endocrinol. 2018;178:D13–D25.
Shahida S, Rehman S, Ilyas N, Khan MI, Hameed U, Hafeez M, Iqbal S, Elboughdiri N, Ghernaout D, Salih AA. Determination of blood calcium and lead concentrations in osteoporotic and osteopenic patients in Pakistan. ACS omega. 2021;6:28373–8.
Kuo T-R, Chen C-H. Bone biomarker for the clinical assessment of osteoporosis: recent developments and future perspectives. Biomark Res. 2017;5:1–9.
Kyd P, De Vooght K, Kerkhoff F, Thomas E, Fairney A. Clinical usefulness of bone alkaline phosphatase in osteoporosis. Ann Clin Biochem. 1998;35:717–25.
Fink HA, Litwack-Harrison S, Taylor BC, Bauer DC, Orwoll ES, Lee CG, Barrett-Connor E, Schousboe JT, Kado DM, Garimella PS. Clinical utility of routine laboratory testing to identify possible secondary causes in older men with osteoporosis: the Osteoporotic Fractures in Men (MrOS) Study. Osteoporos Int. 2016;27:331–8.
Furman D, Campisi J, Verdin E, Carrera-Bastos P, Targ S, Franceschi C, Ferrucci L, Gilroy DW, Fasano A, Miller GW. Chronic inflammation in the etiology of disease across the life span. Nat Med. 2019;25:1822–32.
Franceschi C, Garagnani P, Vitale G, Capri M, Salvioli S. Inflammaging and ‘Garb-aging’. Trends in Endocrinology & Metabolism. 2017;28:199–212.
Furman D, Chang J, Lartigue L, Bolen CR, Haddad F, Gaudilliere B, Ganio EA, Fragiadakis GK, Spitzer MH, Douchet I. Expression of specific inflammasome gene modules stratifies older individuals into two extreme clinical and immunological states. Nat Med. 2017;23:174–84.
Ferrucci L, Fabbri E. Inflammageing: chronic inflammation in ageing, cardiovascular disease, and frailty. Nat Reviews Cardiol. 2018;15:505–22.
Fedewa MV, Hathaway ED, Ward-Ritacco CL. Effect of exercise training on C reactive protein: a systematic review and meta-analysis of randomised and non-randomised controlled trials. Br J Sports Med. 2017;51:670–6.
Redlich K, Smolen JS. Inflammatory bone loss: pathogenesis and therapeutic intervention. Nat Rev Drug Discovery. 2012;11:234–50.
Straub RH, Cutolo M, Pacifici R. Evolutionary medicine and bone loss in chronic inflammatory diseases—a theory of inflammation-related osteopenia. Seminars in arthritis and rheumatism. Elsevier; 2015: 220–8.
Schett G, Kiechl S, Weger S, Pederiva A, Mayr A, Petrangeli M, Oberhollenzer F, Lorenzini R, Redlich K, Axmann R. High-sensitivity C-reactive protein and risk of nontraumatic fractures in the Bruneck study. Arch Intern Med. 2006;166:2495–501.
Eriksson AL, Movérare-Skrtic S, Ljunggren Ö, Karlsson M, Mellström D, Ohlsson C. High‐sensitivity CRP is an independent risk factor for all fractures and vertebral fractures in elderly men: the MrOS Sweden study. J Bone Miner Res. 2014;29:418–23.
Tian L, Yu X. Lipid metabolism disorders and bone dysfunction-interrelated and mutually regulated. Mol Med Rep. 2015;12:783–94.
Dennison EM, Syddall HE, Aihie Sayer A, Martin HJ, Cooper C. Lipid profile, obesity and bone mineral density: the Hertfordshire Cohort Study. J Association Physicians. 2007;100:297–303.
Ackert-Bicknell CL. HDL cholesterol and bone mineral density: is there a genetic link? Bone. 2012;50:525–33.
Yamaguchi T, Sugimoto T, Yano S, Yamauchi M, Sowa H, Chen Q, Chihara K. Plasma lipids and osteoporosis in postmenopausal women. Endocr J. 2002;49:211–7.
Hsu Y-H, Venners SA, Terwedow HA, Feng Y, Niu T, Li Z, Laird N, Brain JD, Cummings SR, Bouxsein ML. Relation of body composition, fat mass, and serum lipids to osteoporotic fractures and bone mineral density in Chinese men and women. Am J Clin Nutr. 2006;83:146–54.
Jia L, Cheng M. Correlation analysis between risk factors, BMD and serum osteocalcin, CatheK, PINP, β-crosslaps, TRAP, lipid metabolism and BMI in 128 patients with postmenopausal osteoporotic fractures. Eur Rev Med Pharmacol Sci. 2022;26:7955–9.
Huang N, Zhou J, Wang W, Wang Q, Tang Y, Sun Y, Wang D, Chen S. Retinol-binding protein 4 is positively associated with bone mineral density in patients with type 2 diabetes and osteopenia or osteoporosis. Clin Endocrinol. 2018;88:659–64.
Zhao J, Zhang M, Quan Z, Deng L, Li Y, He B. Systematic influence of circulating bilirubin levels on osteoporosis. Front Endocrinol. 2021;12:719920.
Jurado S, Parés A, Peris P, Combalia A, Monegal A, Guañabens N. Bilirubin increases viability and decreases osteoclast apoptosis contributing to osteoporosis in advanced liver diseases. Bone. 2022;162:116483.

Download PDF

Version 1

posted

You are reading this latest preprint version

Predicting Fracture Risk for Chinese Elderly Osteoporosis Patients by Hybrid Model

Status:

Version 1

Abstract

Figures

Background

Methods

Results

Model evaluation

Explanation of risk factors

Discussion

Conclusions

Abbreviations

Declarations

References

Status:

Version 1