https://orcid.org/0000-0002-4924-9785
Correction
1 Apr 2025: Xu L, Wu M, Zhang Y, Kun H, Xu J (2025) Correction: Relationships between multivitamins, blood biochemistry markers, and BMC and BMD based on RF: A cross-sectional and population-based study of NHANES, 2017–2018. PLOS ONE 20(4): e0321774. https://doi.org/10.1371/journal.pone.0321774 View correction
Figures
Abstract
Background
Previous studies have separately suggested a possible association between the vitamin exposure, blood biochemical indicators, and bone density. Our study aimed to investigate the relationship between vitamin exposure serum concentrations, blood biochemical indicator serum concentrations, and BMC and BMD using the NHANES 2017–2018 nutrient survey data. This population-based cross-sectional study aimed to explore these associations.
Methods
In this study, we measured vitamin serum concentrations, serum ion serum concentrations, and serum biochemical indicators in adults participating in the NHANES. Skeletal status was assessed by evaluating BMC and BMD in the whole body. Given the inclusion of multiple variables and diverse data types, we used the RF to fit a multivariable model to estimate the associations between vitamin serum concentrations, blood biochemical indicator serum concentrations, and skeletal status.
Results
Under the dimension reduction and comparison selection of RF model, we identified ALP, CPK, and creatinine serum concentrations as the most important factors associated with BMC and BMD in multiple skeletal sites, and the gender, age, height, weight, and body mass index which were found to be related to BMC and BMD in different skeletal sites. Vitamin D and blood calcium serum concentrations were not the important factors associated with BMC and BMD and the three blood biochemical indexes were more important than the vitamin level for BMC and BMD.
Conclusion
The effect of vitamin serum concentrations and blood calcium serum concentrations on human bone density was not significant. ALP, CPK and creatinine serum concentrations body development indicators were identified as the most important factors related to bone status. The RF model can be used to comprehensively evaluate the effects of vitamin content and blood biochemistry serum concentrations in adults on BMC and BMD.
Citation: Xu L, Wu M, Zhang Y, Kun H, Xu J (2025) Relationships between multivitamins, blood biochemistry markers, and BMC and BMD based on RF: A cross-sectional and population-based study of NHANES, 2017–2018. PLoS ONE 20(1): e0309524. https://doi.org/10.1371/journal.pone.0309524
Editor: Carlos Alberto Antunes Viegas, Universidade de Trás-os-Montes e Alto Douro: Universidade de Tras-os-Montes e Alto Douro, PORTUGAL
Received: April 15, 2024; Accepted: August 13, 2024; Published: January 29, 2025
Copyright: © 2025 Xu et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: The National Health and Nutrition Examination Survey (NHANES) database, which is managed by the National Center for Health Statistics (NCHS) of the Centers for Disease Control and Prevention, provided the data used in this study. The NHANES database is used to assess the health and nutritional status of the US population. The NHANES database can beaccessed at https://www.cdc.gov/nchs/nhanes/index.htm. All relevant data from this database have been provided within the paper and its Supporting Information files.
Funding: The author(s) received no specific funding for this work.
Competing interests: The authors have declared that no competing interests exist.
1. Introduction
The health of the skeletal system is a major public health concern in modern medicine [1]. Bone mass is commonly used to evaluate skeletal status, and bone mineral content (BMC) and bone mineral density (BMD), measured using techniques such as dual-energy X-ray, quantitative ultrasound, computed tomography, or magnetic resonance imaging, are reliable indicators of bone mass [2, 3]. BMC and BMD have long been the focus of research and attention, as increasing bone mass can delay the onset of osteoporosis [4]. Decreased BMC and BMD can lead to unpredictable fractures, pain, disability, significant economic burden, and reduced well-being [5].
BMC and BMD are known to be associated with various factors, including individual growth and nutritional indicators. Previous studies have investigated the individual effects of various factors, including gender, age, height, weight, hormone metabolism serum concentrations, dietary nutrients, and vitamin serum concentrations, as well as blood laboratory indicators, on BMC and BMD [6–11]. However, these studies lack a comprehensive understanding of the combined exposure of multiple vitamins, blood biochemical serum concentrations, and their association with bone health. Most studies have focused on specific bone sites and used BMC or BMD as proxies for overall skeletal health [12, 13], without evaluating the associations between BMC, BMD, and potential exposure factors across various skeletal sites. Additionally, collecting specimens with multiple vitamin exposures and different laboratory indicator serum concentrations presents significant challenges.
The NHANES survey conducted between 2017 and 2018 included measurements of vitamin serum concentrations and blood biochemical indicators, providing sample data for assessing bone health in relation to multiple vitamin serum concentrations and biochemical indicators. This study incorporates BMC and BMD from eight skeletal sites (skull, left arm bone, left leg bone, right arm bone, right leg bone, thoracic vertebrae, lumbar vertebrae, and pelvis) as well as whole-body measurements to represent overall skeletal health, offering a more comprehensive approach than previous studies that focused on a single skeletal site. This approach allows for a clearer identification of the specific important underlying factors influencing different skeletal sites. The study also includes individual anthropometric indicators, blood vitamin exposure serum concentrations, and selected blood biochemical indicators as potential influencing factors. Given the positively skewed distribution of BMC and BMD, the variable type for potential factors such as individual growth and development indicators, blood vitamin exposure serum concentrations, and blood biochemical indicator serum concentrations include nominal and continuous variables, with continuous data exhibiting skewed distributions.
Previous studies have used the random forest tree algorithm(RF) for prediction [6], and other research has shown that RF is a suitable ensemble learning algorithm and machine learning method, offering advantages such as independence from variable conditionality [7], high accuracy, sensitivity, and specificity compared to decision trees [7]. Furthermore, RF can also be used for predicting continuous variables and obtain prediction results without significant bias [8]. Therefore, RF is a suitable predictive method for the data in this study, allowing for the evaluation of relevant influencing factors.
This study selects participants from the NHANES survey conducted between 2017 and 2018, which includes assessments of bone density serum concentrations. The objective of this study is to use the vitamin exposure serum concentrations and laboratory examination indicators of the study participants as a basis to apply the RF method and evaluate factors related to BMC and BMD. Furthermore, the study aims to investigate whether specific factors differ in their associations with BMC/BMD across different skeletal sites. Assessing the association between BMC/BMD and the vitamin exposure serum concentrations and blood laboratory indicator serum concentrations will provide a macro-level understanding of the true associations between vitamin exposure, blood indicators, and their impact on BMC and BMD. This study aims to provide new insights into the specific vitamin exposures, blood laboratory indicator serum concentrations, and their combined effects on bone density and bone mineral content.
2. Materials and methods
2.1. Data source
NHANES is a cross-sectional study conducted since 1999, aimed at assessing the health and nutritional status of adults and children in the United States. NHANES involves interviews and physical examinations, focusing on various health and nutrition measurements, and is a major program of the National Center for Health Statistics (NCHS). For more detailed information, please visit the NHANES official website (https://www.cdc.gov/nchs/nhanes/) [9]. Considering that childhood and adolescence are characterized by rapid bone mass accumulation, this study did not include individuals under the age of 18. NHANES screening criteria for bone density: Participants aged 8–59 years were eligible, except for those who were pregnant (self-reported positive pregnancy test and/or self-reported use of radiographic contrast agents in the past 7 days), self-reported weight over 450 pounds or height over 6 feet 5 inches, which were excluded from Dual-energy X-ray absorptiometry (DXA).
Considering that serum serum concentrations of vitamin serum concentrations (i.e., C, D, A, and E) and blood biochemical markers were only fully measured in NHANES 2017–2018, this study utilized data from that cycle, including a total of 9,255 adult participants aged 19–59 years. After excluding incomplete data for BMC and BMD measurements (N = 4,072), incomplete data for vitamin serum concentrations (N = 1,824), and incomplete data for blood biochemical markers, the participant exclusion flowchart, as shown in Fig 1, resulted in a final sample size of 1,471.
2.2 Measurement of vitamins
Collection of serum samples is conducted simultaneously with face-to-face interviews. The serum specimens are then processed, stored at -30°C, and transported to the Laboratory Science Department and CDC for analysis. Vitamins A and E are measured using high-performance liquid chromatography (HPLC) and photodiode array detection. Vitamin C is measured using isocratic HPLC, and electrochemical detection is set at 650 mV1. Serum 25-hydroxyvitamin D is measured using an equilibrium radioimmunoassay method. The NHANES Laboratory Manual provides detailed descriptions of the vitamin measurement procedures.
2.3 Measurement of biochemical markers
All methods were measured on the Roche Cobas 6000 (c501 module) analyzer. See Laboratory Method Files (NHANES Laboratory Manual for more detailed information about analyse methodologies, principles, and operating procedures. In addition, contract laboratories randomly perform repeat testing on 2% of all specimens. Based on the above, we collected the following biochemical markers: alanine aminotransferase (ALT)(U/L), albumin, frozen serum (g/dL), alkaline phosphatase (ALP)(U/L), aspartate aminotransferase (AST) (U/L), bicarbonate (mmol/L), blood urea nitrogen (grams/deciliter), chloride (millimole/liter), creatine phosphokinase (CPK) (IU/L), frozen serum creatinine (umol/L), globulin (g/dL), glucose, refrigerated serum (milligrams/deciliter), iron, refrigerated serum (micrograms/deciliter), osmolality (mmol/Kg), phosphorus, potassium, sodium, total bilirubin, total calcium, total cholesterol, total protein, triglycerides, and uric acid serum concentrations.
2.4 Bone density measurement
BMC and BMD: DXA is the most widely accepted method for measuring the body composition, partly due to its speed, ease of use, and low radiation exposure. Whole-body DXA scans were conducted from 2011 to 2018. NHANES DXA examinations provide nationally representative data on the body composition (bone and soft tissue), as well as age, sex, and race/ethnicity, to study the associations between body composition and other health conditions and risk factors such as cardiovascular disease, diabete, hypertension, and physical activity and dietary patterns. DXA scans provide measurements of bone and soft tissue in the whole body, arms, legs, trunk, and head. Measurements of the pelvis, left and right ribs, thoracic vertebrae, and lumbar vertebrae are also obtained. The values for the whole body and regions include: BMC (grams), bone area (cm2), and BMD (grams per square centimeter). It is worth noting that to fully illustrate the relationship between bone status, vitamin serum concentrations, and blood biochemical indices, there may be considerable differences due to different locations of the bones themselves. To avoid the occurrence of such unknown situations, we systematically evaluated the BMC and BMD of the whole body, skull, left arm bone, left leg bone, right arm bone, right leg bone, thoracic vertebrae, lumbar vertebrae, and pelvis.
2.5 Other variables
Sociodemographic factors (age, sex, height, weight) were obtained through face-to-face interviews. Body Mass Index (BMI) data were obtained through physical examinations. BMI is calculated by dividing weights (kg) by the square of height (m2) and classified as non-obese (<30 kg/m2) and obese (≥30 kg/m2).
2.6 Statistical analysis
NHANES Data Analysis Guidelines are retrieved from (https://wwwn.cdc.gov/nchs/nhanes/tutorials/default.aspx). In our study, if the continuous variable followed a normal distribution, it was presented as mean ± standard deviation; otherwise, it was described as median (min, max). We used a RF model to assess the relationships between various vitamins, blood biochemical indices, BMC and BMD [9]. To do this, we first removed several predictor variables that were highly correlated with other variables, and we retained the variables that best reflected the correlations. The variables that were removed were mainly derived variables of the correlated variables (due to the unit conversion). We then fine-tuned the model structure through the model adjustment. We explored using the grid search and ultimately selected the hyperparameters mtry = 5 and n_trees = 150 [9]. Subsequently, we constructed separate models for each body site using the selected hyperparameters. For each model, 70% of the observations were used for training, and 30% were used for testing. The training set was used to build the RF model, including all variables from the included samples as candidate variables, while the validation set was used to validate the model’s performance. In this study, the predicted values and actual values were evaluated using the mean absolute error (MAE) and root mean square error (RMSE) to assess the predictive performance. The coefficient of determination (R-squared) was used to reflect the regression fit of the prediction model. Mean accuracy was used to evaluate the relative importance of variables [10, 11]. Finally, variable importance analysis was conducted, and the variable importance in the RF model was estimated using the average decrease in accuracy calculated by Matlab 2016 [12–14]. This was performed for each different bone block of BMC and BMD models and summarized in the graph, showcasing the top 6 potential factors related to bone status. The statistical analysis of the aforementioned dataset was performed using IBM SPSS Statistics 22, Matlab 2016.
2.7 Random forest trees
RF is an ensemble learning method based on the Bagging approach, which combines multiple decision trees for classification and regression tasks [15]. In traditional decision trees, the selection of features for splitting is based on selecting the optimal feature from the feature set at the current node. However, in RF, at each node of the base decision tree, a random subset containing a particular feature k is selected from the feature set, and the optimal feature for splitting is chosen from this subset. For the input dataset, where represents the feature vector and represents the corresponding label. Let the randomly sampled dataset form the training dataset for each decision tree, and the prediction result for each decision tree, where is the number of decision trees. The prediction formula is shown in Eq (1-1) [16].
For randomly selected decision trees, feature importance screening, namely feature contribution assessment, Gini index or out-of-bag (OOB) error rate is commonly used as an evaluation metric. In this study, the Gini index is employed as the assessment criterion for the importance of factors BMC and BMD. The variable importance scores are denoted as VIM, and the Gini index is represented as GI. Assuming there are m features denoted as, X1,X2,…,Xn, with K categories, and the proportion of category k in node m is pmk. The formula for calculating the Gini index is shown in Eq (1-2).
The importance of a feature (Xj) at node m represents the computation of the Gini index variation before and after the branching of that node (GIl and GIr):
(1–3)
For each tree in the ensemble, a recursive process involves random sampling of both samples and variables. If a feature (Xj) exists in a certain node of a particular tree i among the T decision trees in the forest, the importance of that feature XC is determined as follows:
(1–4)
(1–5)
The obtained importance scores are normalized to derive the final results [17].
The final results include the top six selections as shown in our Table 3 and Fig 2, and the complete set of results can be found in S2 Table.
3. Results
3.1 Sample characteristics
This study included a total of 1,471 participants (see Table 1), with males and females accounting for 48.74% and 51.26% of the sample, respectively. The average age of the participants was 33.04 ± 14.64 years, with a height of 166.293 ± 9.49 cm, weight of 75.86 ± 20.65 kg, and BMI of 27.30 ± 6.68. Descriptive statistics were employed to assess the differences in BMC and BMD among different body parts. All Abbreviation of Charactors are included in Tables 1 and 2.
3.2 Variable selection evaluation
In the RF model, all input variables carry certain weights, and by comparing the magnitude of these weights, the relative importance of each influencing factor on the outcomes can be assessed. As the study involved a considerable number of variables, a detailed table of weights can be found in S2 Table. To better illustrate the important weight variables related to different bone structures, we selected the top 6 variables with high importance using the RF model as independent variables. Table 3 provides a detailed listing of the top 6 variables for the importance of all input indicators regarding BMC and BMD for various bone regions, including the skull, left arm bone, left leg bone, right arm bone, right leg bone, thoracic vertebrae, lumbar vertebrae, pelvic bone, and the entire body. The most important variables, their importance rankings, and the sensitivity and specificity of the variable sets are displayed in S1 Table. Furthermore, weight graphs were created for the top 6 influencing factors for different bone structures, as shown in Fig 2. The RF model was evaluated for the average absolute error (MAE) and root mean square error (RMSE) between predicted and actual serum concentrations. The predictive performance was assessed using the determination coefficient R² to reflect the regression fit of the prediction model. The partly specific evaluation results are shown in Fig 3, and all model evaluation values are included in S2 Table and S1, S2 Figs.
3.3 Variable selection comparison
3.3.1 Skull.
For instance, the top six major associated factors with BMC in the skull are: age, ALP, height, weight, LBXSTR (triglycerides, refrigerated serum (mg/dL)), LBXSPH (phosphorus (mg/dL)); the top six major associated factors with BMD in the skull are: age, ALP, sex, weight, height, creatinine phosphokinase (CPK) (IU/L)).
3.3.2 Four limbs bone.
For the left arm, the top six major associated factors with BMC are: height, weight, sex, LBDSCRSI (frozen serum creatinine (umol/L)), BMI, age; the top six major associated factors with BMD in the left arm are: sex, weight, height, LBDSCRSI, ALP, CPK; For the left leg, the top six major associated factors with BMC are: height, weight, sex, BMI, LBDSCRSI, CPK; The top six major associated factors with BMD in the left leg are: height, sex, weight, ALP, LBDSCRSI, BMI; For the right arm, the top six major associated factors with BMC are: height, weight, sex, LBDSCRSI, BMI, age; The top six major associated factors with BMD in the right arm are: sex, weight, height, LBDSCRSI, age, BMI; For the right leg, the top six major associated factors with BMC are: height, weight, BMI, sex, LBDSCRSI, CPK; The top six major associated factors with BMD in the right leg are: height, weight, sex, BMI, CPK, LBDSCRSI.
3.3.3 Spine and pelvis.
For the thoracic vertebrae, the top six major associated factors with BMC are: weight, height, ALP, BMI, age, LBXVIE (alpha-tocopherol (micro/dL)); the top six major associated factors with BMD in the thoracic vertebrae are: ALP, height, age, weight, sex, LBXSUA (uric acid (mg/dL)); For the lumbar vertebrae, the top six major associated factors with BMC are: height, ALP, age, weight, LBDSCRSI, BMI; The top six major associated factors with BMD in the lumbar vertebrae are: ALP, height, age, weight, sex, BMI; For the pelvic bone, the top six major associated factors with BMC are: LBDSCRSI, height, age, weight, sex, BMI; The top six major associated factors with BMD in the pelvic bone are: weight, BMI, height, age, LBDSCRSI, LBXSUA.
4. Discussion
This study primarily explores the associations between various vitamin exposure serum concentrations and blood biochemical indicators and DXA-derived BMC and BMD measurements in U.S. adults included in NHANES during 2017–2018. The goal is to derive the most relevant variables associated with BMC and BMD through the RF model. Considering the advantages of RF ensemble learning in mitigating overfitting and noise resistance due to the introduction of two sources of randomness (independent sampling during the construction of each decision tree and random sampling with replacement for rows and columns (attributes) of the training set [18], allowing duplicate samples), we directly include all potential influencing factors related to bone density in the machine model. Additionally, as our selected outcome variables, BMC and BMD, are continuous skewed variables, and the vitamin serum concentrations, electrolyte serum concentrations, and most laboratory biochemical indicators are also continuous skewed distributions, we opted for the adaptability of RF, which can handle both continuous and discrete variables without the need for normalization. The RF model has the advantage of automatically outputting the importance of variables, achieving a good dimensionality reduction effect. Through the aforementioned RF analysis, we found that the impact of blood vitamin serum concentrations on BMC and BMD may have been overestimated. In the multivariate RF model, we observed that blood biochemical indicators and general indicators of growth and development have greater influence weights than vitamin serum concentrations.
4.1. Impact of blood vitamin serum concentrations on BMC/BMD
Assessing the BMC and BMD for various bone regions, including the skull, left arm bone, left leg bone, right arm bone, right leg bone, thoracic vertebrae, lumbar vertebrae, pelvic bone, and the entire body, we did not find any vitamins to be important variables for BMC and BMD. In other words, in the jointly established model considering blood biochemical indicators, individual indicators, and blood vitamin serum concentrations, vitamin serum concentrations were not the optimal correlated variables for BMC/BMD. Previous studies have reported positive effects of vitamin B and vitamin D on BMD, while vitamin C may have a negative correlation with bone density [19–21]. Although such effects were considered, the RF model, which evaluates the contribution of all input variables to the model, indicated that the importance of vitamin serum concentrations is not significant under the combined influence of the body itself, the vitamin exposure, and blood biochemical indicator serum concentrations. This may explain why most vitamins have been reported to be related to bone BMC and BMD, but interventions through vitamin supplementation to regulate bone density serum concentrations are often challenging to achieve effective results [22]. One of our aims is to discover the relationship between vitamin D and various bone densities through NHANES 2017–2018 nutrient survey data and bone density survey data. However, we did not find that vitamin D are important influencing factors for any bone BMC or BMD.
4.2 Impact of blood biochemical indicator serum concentrations on BMC and BMD
Our study found that ALP, CPK and creatinine serum concentrations were identified as the most important factors related to bone health, apart from individual indicators. ALP was found to be an important correlated variable for BMC and BMD for all bone regions except for the left and right leg bones and the pelvic region. The substantial correlation between ALP and BMC/BMD was confirmed in our study, ALP plays a crucial role in bone formation and mineralization and is a commonly used bone turnover marker representing osteoblast activity. CPK is an enzyme involved in the metabolism of competence, mainly found in skeletal muscle, While skeletal muscle affects the bone health, thus, ALP, CPK is the most important factors related BMC and BMD. The influential mechanism of creatinine serum concentrations may be because it is related to the metabolic function, while the disorder of human metabolic function may induce the imbalance of osteoblast and osteoclast function, thus affecting the bone health, but which requires further research.
A previous study showed that an increase in serum ALP is associated with the loss of BMD and more teoporosis in females [23]. Consistent with a previous correlation study between serum T-ALP and lumbar BMD in young individuals, indicating a negative correlation. In clinical practice, bone alkaline phosphatase (APK) is closely associated with BMD, and further exploration is needed regard the relationship between ALT and BMD [24, 25].
CPK and LBDSCRSI serum concentrations were also found to be significant for the limbs (left and right upper and lower limbs), possibly due to their strong correlation with the movement level of the limbs compared to other bone regions. Regarding other blood biochemical indicators, a previous multivariate regression analysis examined the relationship between the protein intake and BMC and BMD, suggesting a harmful association between protein intake and bone health. However, our evaluation of blood albumin serum concentrations did not reveal a significant correlation with bone density, possibly due to differences in the population characteristics of our study sample. One of our aims is to discover the relationship between blood calcium serum concentrations and various bone densities through NHANES 2017–2018 nutrient survey data and bone density survey data. However, we did not find that blood calcium serum concentrations are important influencing factors for any bone BMC or BMD.
4.3. Impact of general growth and development indicators on BMC and BMD
Previous studies has suggested that obesity may lead to skeletal damage [26, 27], while others have reported a positive promoting effect of fat tissue resulting from the weight gain on the bone structure. The contradictory results might not solely be due to different definitions of obesity and overweight but could also be attributed to traditional statistical methods being unable to identify the nonlinear relationship between BMI and BMD, as well as BMC. In contrast, our study utilized the RF model, taking advantage of its ability to analyze variables without conditional restrictions. Through machine learning, we effectively identified the associations between various potential influencing factors and the outcomes of BMC and BMD. Our findings indicate a significant correlation between weight, BMI serum concentrations, and BMC/BMD.
4.4. Strengths and limitations
Our study possesses several strengths. The use of the RF model systematically evaluated the impact weights of blood vitamin serum concentrations and blood biochemical indicators on BMC and BMD across eight different bone regions and overall bone health. This approach avoids the limitations of previous research that focused solely on vitamin serum concentrations or individual blood indicators and considered the influence on bone density in a single region. Moreover, by representing vitamin exposure serum concentrations through clear vitamin concentrations and capturing specific electrolyte and liver and kidney function serum concentrations through blood biochemical indicators, our study employed clinical outcomes such as BMC and BMD in eight different bone regions and the entire body. This approach provides a more comprehensive reflection of the body’s real serum concentrations and conditions.
However, our study also has some potential limitations. The use of cross-sectional survey data limits our ability to make causal inferences, necessitating evidence accumulation through large-scale prospective cohort studies and experimental research. Due to a lack of suitable data for bone metabolism(i.e. osteocalcin, cross-linking telopeptide of type I collagen) in the blood biochemical indicators, these variables were not included in the analysis. Additionally, our study focused on a population of U.S. participants, which may limit the generalizability of the results to other ethnic groups.
5. Conclusion
Using the RF model and data from NHANES 2017–2018 for adults aged 19–59 years in the United States, we explored the relationship between exposure to multiple vitamins, body serum concentrations of biochemistry and BMC and BMD. Under RF dimension reduction and comparison selection, the effect of vitamin serum concentrations on human bone density was not significant. ALP, CPK and creatinine serum concentrations were identified as the most important factors related to bone status, and body development indicators were also important variables related to BMC and BMD, but vitamin D and blood calcium serum concentrations may not affect the serum concentrations of BMC and BMD. We suggest using statistical methods such as RF to comprehensively evaluate the effects of vitamin content and blood biochemistry serum concentrations in adults on BMC and BMD. This open up a new direction for further exploring the evidence between multiple vitamin exposure, blood indicator exposure, and BMC/BMD.
Supporting information
S1 Table. Summary of the model evaluation table.
https://doi.org/10.1371/journal.pone.0309524.s001
(DOCX)
S2 Table. All weighting of different variables on skeletal conditions at different sites.
https://doi.org/10.1371/journal.pone.0309524.s002
(XLSX)
S2 Fig. Evaluation plot of the fitting results.
https://doi.org/10.1371/journal.pone.0309524.s004
(PNG)
References
- 1. Fessele KL. Bone Health: Introduction. Seminars in oncology nursing. Apr 2022;38(2):151272. pmid:35461738
- 2. Fraga MM, de Sousa FP, Szejnfeld VL, de Moura Castro CH, de Medeiros Pinheiro M, Terreri MT. Trabecular bone score (TBS) and bone mineral density (BMD) analysis by dual X-ray absorptiometry (DXA) in healthy Brazilian children and adolescents: normative data. Archives of osteoporosis. Jun 15 2023;18(1):82. pmid:37318639
- 3. Pedrosa M, Ferreira MT, LA EBdC, MP MM, Curate F. The association of osteochemometrics and bone mineral density in humans. American journal of physical anthropology. Nov 2021;176(3):434–444. pmid:33852736
- 4. Muñoz M, Robinson K, Shibli-Rahhal A. Bone Health and Osteoporosis Prevention and Treatment. Clinical obstetrics and gynecology. Dec 2020;63(4):770–787. pmid:33017332
- 5. Pai SN, Jeyaraman N, Jeyaraman M, Shyam A. Osteoporosis ‐ An Imminent Ethical and Legal Debacle? Journal of orthopaedic case reports. Sep 2023;13(9):1–3. pmid:37753139
- 6. Wang J, Shi L. Prediction of medical expenditures of diagnosed diabetics and the assessment of its related factors using a random forest model, MEPS 2000–2015. International journal for quality in health care: journal of the International Society for Quality in Health Care. Apr 27 2020;32(2):99–112. pmid:32159759
- 7. Esmaily H, Tayefi M, Doosti H, Ghayour-Mobarhan M, Nezami H, Amirabadizadeh A. A Comparison between Decision Tree and Random Forest in Determining the Risk Factors Associated with Type 2 Diabetes. Journal of research in health sciences. Apr 24 2018;18(2):e00412. pmid:29784893
- 8. Ellis K, Kerr J, Godbole S, Lanckriet G, Wing D, Marshall S. A random forest classifier for the prediction of energy expenditure and type of physical activity from wrist and hip accelerometers. Physiological measurement. Nov 2014;35(11):2191–203. pmid:25340969
- 9. Doubleday K, Zhou H, Fu H, Zhou J. An Algorithm for Generating Individualized Treatment Decision Trees and Random Forests. Journal of computational and graphical statistics: a joint publication of American Statistical Association, Institute of Mathematical Statistics, Interface Foundation of North America. 2018;27(4):849–860. pmid:32523325
- 10. Chicco D, Warrens MJ, Jurman G. The coefficient of determination R-squared is more informative than SMAPE, MAE, MAPE, MSE and RMSE in regression analysis evaluation. PeerJ Computer science. 2021;7:e623. pmid:34307865
- 11. Robeson SM, Willmott CJ. Decomposition of the mean absolute error (MAE) into systematic and unsystematic components. PloS one. 2023;18(2):e0279774. pmid:36800326
- 12. Janitza S, Strobl C, Boulesteix AL. An AUC-based permutation variable importance measure for random forests. BMC bioinformatics. Apr 5 2013;14:119. pmid:23560875
- 13. Leidy NK, Malley KG, Steenrod AW, et al. Insight into Best Variables for COPD Case Identification: A Random Forests Analysis. Chronic obstructive pulmonary diseases (Miami, Fla). 2016;3(1):406–418. pmid:26835508
- 14. Rocher-Ros G, Stanley EH, Loken LC, et al. Global methane emissions from rivers and streams. Nature. Sep 2023;621(7979):530–535. pmid:37587344
- 15. Kutluer N, Solmaz OA, Yamacli V, Eristi B, Eristi H. Classification of breast tumors by using a novel approach based on deep learning methods and feature selection. Breast cancer research and treatment. Jul 2023;200(2):183–192. pmid:37210703
- 16. Xu N, Zhang Z, Shen Y, et al. Compare the performance of multiple binary classification models in microbial high-throughput sequencing datasets. The Science of the total environment. Sep 1 2022;837:155807. pmid:35537509
- 17. Chen CC, Schwender H, Keith J, Nunkesser R, Mengersen K, Macrossan P. Methods for identifying SNP interactions: a review on variations of Logic Regression, Random Forest and Bayesian logistic regression. IEEE/ACM transactions on computational biology and bioinformatics. Nov-Dec 2011;8(6):1580–91. pmid:21383421
- 18. Schwarz DF, König IR, Ziegler A. On safari to Random Jungle: a fast implementation of Random Forests for high-dimensional data. Bioinformatics (Oxford, England). Jul 15 2010;26(14):1752–8. pmid:20505004
- 19. Sadat-Ali M, Al-Turki HA. Vitamin D supplements and bone mineral density. Lancet (London, England). Apr 12 2014;383(9925):1293. pmid:24725573
- 20. Clements M, Heffernan M, Ward M, et al. A 2-Year Randomized Controlled Trial With Low-Dose B-Vitamin Supplementation Shows Benefits on Bone Mineral Density in Adults With Lower B12 Status. Journal of bone and mineral research: the official journal of the American Society for Bone and Mineral Research. Dec 2022;37(12):2443–2455. pmid:36128889
- 21. Lan KM, Wang LK, Lin YT, et al. Suboptimal Plasma Vitamin C Is Associated with Lower Bone Mineral Density in Young and Early Middle-Aged Men: A Retrospective Cross-Sectional Study. Nutrients. Aug 29 2022;14(17) pmid:36079812
- 22. Middelkoop K, Micklesfield LK, Walker N, et al. Influence of vitamin D supplementation on bone mineral content, bone turnover markers, and fracture risk in South African schoolchildren: multicenter double-blind randomized placebo-controlled trial (ViDiKids). J BONE MINER RES. 2024; 39 (3): 211–221. pmid:38477739
- 23. Cherif R, Mahjoub F, Sahli H, et al. Clinical and body composition predictors of bone turnover and mineral content in obese postmenopausal women. CLIN RHEUMATOL. 2018; 38 (3): 739–747. pmid:30341704
- 24. Salamat MR, Momeni S, Rastegari AA. Relation between Biochemical Parameters and Bone Density in Postmenopausal Women with Osteoporosis. Adv Biomed Res. 2023; 12 162. pmid:37564448
- 25. Fu Y, Wang G, Liu J, et al. Stimulant use and bone health in US children and adolescents: analysis of the NHANES data. EUR J PEDIATR. 2022; 181 (4): 1633–1642. pmid:35091797
- 26. Robinson D, Bailey A, Hale A, et al. Bone Density, Biochemistry and Life-Style METHOD INFORM MED. 2018; 32 (03): 233–236.
- 27. Warodomwichit D, Sritara C, Thakkinstian A, et al. Causal inference of the effect of adiposity on bone mineral density in adults. Clinical endocrinology. May 2013;78(5):694–9. pmid:23045999