Fracture risk prediction method and system

Abstract

The invention discloses a fracture risk prediction method and system, and relates to the technical field of health management. According to the prediction method, serious sarcopenia is selected to diagnose and assess the sarcopenia state, the ratio of monocytes to lymphocytes reflects the immune inflammation level, the prediction model is constructed in combination with the femoral neck bone mineral density T value, and the fracture risk influence factors are comprehensively covered; and the risk identification capability on special high-risk groups with sarcopenia, immune dysfunction and the like is effectively improved. And secondly, only three target independent variables are selected, and the three target independent variables are indexes which can be directly obtained or simply calculated in clinical routine examination, so that the data collection difficulty is greatly reduced. Meanwhile, the prediction model is visualized through a column diagram, complex calculation is not needed, a clinician can quickly inquire the fracture risk score through the actual value of the target subject, and the method is suitable for being efficiently applied to conventional outpatient service, hospitalization and other scenes.

Description

Technical Field

[0001] This application relates to the field of health management technology, and in particular to a method and system for predicting fracture risk. Background Technology

[0002] Osteoporotic fractures are the most serious and major complication of osteoporosis. They not only severely affect the long-term prognosis of patients, but also significantly increase the medical burden.

[0003] To identify high-risk groups for fractures at an early stage, various fracture risk assessment tools based on clinical and individual characteristics have been developed. For example, CN202210129046.8 discloses a stepwise regression-based osteoporosis risk prediction system and method for postmenopausal women. The method involves designing a data collection table related to osteoporosis prediction in postmenopausal women to form training and validation datasets; performing univariate analysis on the training dataset, identifying factors with strong correlations and combining them with clinical relevance as candidate factors; using stepwise logistic regression to screen variables and establish a predictive model; calculating the probability of osteoporosis in patients in the established and validation datasets to validate the model; and creating a nomogram of osteoporosis risk in postmenopausal women.

[0004] However, the above application only considers bone mineral density as an influencing factor, resulting in insufficient accuracy in fracture assessment. Summary of the Invention

[0005] Based on this, this application provides a fracture risk prediction method based on comprehensive clinical data; this application also provides a fracture risk prediction system.

[0006] This application provides a method for predicting fracture risk, including: A case sample of hospitalized patients was obtained, and a prediction model was constructed based on the case sample and a binary logistic regression algorithm. The dependent variable of the prediction model was the fracture diagnosis result, and the target independent variables of the prediction model were the femoral neck bone mineral density T-score, the diagnosis of severe sarcopenia, and the ratio of monocytes to lymphocytes. Construct a nodal plot based on the target independent variable and the prediction model; For each of the target independent variables, obtain the actual value of the target subject; Based on the actual values, the fracture risk score of the target subject is obtained from the nomogram.

[0007] Optionally, the prediction model is constructed based on the following steps: An initial model is constructed based on the aforementioned binary logistic regression algorithm; Based on the case sample, the dependent variable and multiple initial independent variables are obtained; Multiple initial independent variables are sequentially input into the initial model to exclude collinearity among the multiple initial independent variables, the target independent variables are selected, and the regression coefficient of each target independent variable is obtained; The prediction model is established based on the dependent variable, the target independent variable, and the regression coefficients.

[0008] Optionally, the initial independent variables include: The study included grip strength, time to 5 sit-ups, 6-meter walking speed, limb skeletal muscle index, total lumbar spine and femoral neck bone mineral density T-score, complete blood count, bone metabolism markers, liver function, kidney function, blood lipids, and fasting blood glucose. The complete blood count included white blood cells, red blood cells, hemoglobin, platelets, neutrophils, lymphocytes, monocytes, the monocyte-to-lymphocyte ratio, and the neutrophil-to-lymphocyte ratio. Bone metabolism markers included parathyroid hormone, osteocalcin, type I procollagen N-terminal propeptide, type I collagen C-terminal peptide, 25-hydroxyvitamin D, serum calcium, and serum phosphorus. Liver function included alanine aminotransferase (ALT), aspartate aminotransferase (AST), gamma-glutamyl transferase (GGT), alkaline phosphatase, and albumin. Kidney function included blood urea nitrogen, serum creatinine, and uric acid. Blood lipids included total cholesterol and triglycerides.

[0009] Optionally, the fracture diagnosis result is obtained based on at least one of the following: thoracic and lumbar spine X-ray (anteroposterior and lateral views), computed tomography (CT) scan, and magnetic resonance imaging (MRI) in the case sample.

[0010] Optionally, the prediction model is constructed based on the following formula: ; Where p is the probability of a fracture occurring. , 3.296 is the intercept; The T-value is the bone mineral density of the femoral neck. The regression coefficient for the femoral neck bone mineral density T-value; 4.626 is the regression coefficient of the monocyte-to-lymphocyte ratio. The diagnosis of severe sarcopenia was confirmed. The regression coefficient is used to diagnose the severe sarcopenia.

[0011] Optionally, the nomogram is constructed based on the following steps: Determine each of the target independent variables i The regression coefficients and the range of values ​​for each of the target independent variables. ; Calculate the maximum linear effect of each target independent variable within its range of values. and order ; Determine the baseline total score Draw the baseline fractional axis and specify the scaling constant. ; Based on formula Set the number of points on the scale for each retrieval target variable and plot the corresponding target variable axis; Plot the total score axis ; Calculate the adjusted intercept and based on and Map the total score to the fracture risk probability and plot the risk axis; The nomogram is generated by combining the baseline score axis, the target independent variable axis, the total score axis, and the risk axis.

[0012] Optionally, the femoral neck bone mineral density T value is X1, with a value range of [-6.0, 1.0]; the monocyte to lymphocyte ratio is X2, with a value range of [0.1, 1.0]; and the diagnosis of severe sarcopenia is X3, with a value range of [0, 1].

[0013] Optionally, the step of determining the fracture risk score of the target subject in the nomogram based on the actual value includes: Find the actual value of each of the three target independent variable axes, draw a vertical line for each, and convert the actual value into the corresponding current score on the baseline score axis; Add the three current scores together to get the current total score; Find the corresponding current total score on the total score axis, draw a vertical line, and determine the fracture risk score on the risk axis.

[0014] Optionally, it also includes: The predicted probability of fracture risk for the target subject is obtained through the prediction model. The fracture diagnosis result is determined as the risk dependent variable, the fracture risk prediction probability is determined as the risk independent variable, and the risk threshold range and confidence interval are set based on the risk dependent variable and the risk independent variable. At different risk thresholds within the stated risk threshold range, the first clinical net benefit of the prediction model is obtained, wherein the first clinical net benefit is obtained using the following formula: ; Where TP represents the number of true positive individuals, FP represents the number of false positive individuals, and N represents the total number of samples. p t The risk threshold is defined as follows; Obtain the second clinical net benefit of all intervention strategies at each of the stated risk thresholds; Obtain the third clinical net benefit of all non-intervention strategies at each of the stated risk thresholds; Using the risk threshold range as the horizontal axis and the clinical net benefit as the vertical axis, a first net benefit curve of the prediction model, a second net benefit curve of all intervention strategies, and a third net benefit curve of all non-intervention strategies are plotted to form a clinical decision curve. The clinical decision curve is used to determine the clinical applicability of the prediction model within different risk threshold ranges by comparing the positional relationship of the first net benefit curve, the second net benefit curve, and the third net benefit curve.

[0015] This application also provides a fracture risk prediction system for performing the fracture risk prediction method described above. The prediction system includes: The prediction model building module is configured to acquire case samples of hospitalized patients and build a prediction model based on the case samples and a binary logistic regression algorithm; the dependent variable of the prediction model is the fracture diagnosis result, and the target independent variables of the prediction model are the femoral neck bone mineral density T-score, the diagnosis of severe sarcopenia, and the ratio of monocytes to lymphocytes; The nomogram construction module is configured to construct a nomogram based on the target independent variable and the prediction model; The actual value acquisition module is configured to acquire the actual value of the target subject for each of the target independent variables; The fracture risk score acquisition module is configured to acquire the fracture risk score of the target subject in the nomogram based on the actual value.

[0016] Compared with the prior art, the technical solution provided in this application has the following advantages: The aforementioned fracture risk prediction method uses a combination of factors, including a confirmed diagnosis of severe sarcopenia to assess sarcopenia status, a monocyte-to-lymphocyte ratio to reflect the level of immune inflammation, and the femoral neck bone mineral density T-score, to construct a predictive model. This comprehensive approach effectively covers factors influencing fracture risk and significantly improves the ability to identify risks in high-risk populations such as those with sarcopenia or abnormal immune function. Furthermore, this application selects only three target variables, all of which are indicators that can be directly obtained or easily calculated through routine clinical examinations, greatly reducing the difficulty of data collection. Simultaneously, the predictive model is visualized using a nomogram, eliminating the need for complex calculations. Clinicians can quickly query fracture risk scores based on the actual values ​​of the target subjects, making it suitable for efficient application in routine outpatient and inpatient settings.

[0017] Furthermore, this application employs a forward stepwise regression analysis method using binary logistic regression to screen variables, accurately identifying core factors significantly impacting fracture risk. The constructed predictive model effectively captures the risk characteristics of elderly individuals and high-risk individuals with no history of fractures. Based on this, a nomogram visually presents the risk score, further improving the accuracy of risk assessment for this specific population and avoiding missed intervention opportunities due to underestimation of risk. This reduces the complexity of clinical decision-making, allowing clinicians to quickly determine a patient's fracture risk level based on the fracture risk score, providing a quantitative basis for intervention formulation. This approach avoids the waste of medical resources and overtreatment associated with comprehensive intervention, while also reducing the risk of fractures due to lack of intervention or misjudgment of low risk, thus achieving precision medicine.

[0018] The fracture risk prediction system provided in this application has all the technical features and beneficial effects of the above-mentioned fracture risk prediction methods, which will not be repeated here. Attached Figure Description

[0019] Figure 1 A flowchart illustrating the fracture risk prediction method provided in this application embodiment; Figure 2 A nomogram of the fracture risk prediction method provided in the embodiments of this application; Figure 3 Clinical decision curves for the fracture risk prediction method provided in the embodiments of this application; Figure 4 A schematic diagram comparing the fracture risk prediction value of the fracture risk prediction method provided in the embodiments of this application with that of other models. Detailed Implementation

[0020] To make the above-mentioned objectives, features, and advantages of this application more apparent and understandable, the specific embodiments of this application are described in detail below with reference to the accompanying drawings. Many specific details are set forth in the following description to provide a thorough understanding of this application. However, this application can be implemented in many other ways different from those described herein, and those skilled in the art can make similar modifications without departing from the spirit of this application. Therefore, this application is not limited to the specific embodiments disclosed below.

[0021] See Figure 1 , Figure 1 This is a flowchart illustrating the fracture risk prediction method provided in an embodiment of this application. The embodiment of this application provides a fracture risk prediction method, including the following steps 101 to 104.

[0022] Step 101: Obtain case samples of hospitalized patients and construct a prediction model based on the case samples and a binary logistic regression algorithm; the dependent variable of the prediction model is the fracture diagnosis result, and the target independent variables of the prediction model are the femoral neck bone mineral density T-value, the diagnosis of severe sarcopenia, and the ratio of monocytes to lymphocytes (MLR).

[0023] Specifically, this application uses a predictive model constructed by selecting severe sarcopenia to assess sarcopenia status and the ratio of monocytes to lymphocytes to reflect the level of immune inflammation, combined with the femoral neck bone mineral density T-score, to achieve comprehensive coverage of factors influencing fracture risk and effectively improve the risk identification ability for special high-risk groups such as those with sarcopenia and abnormal immune function. Secondly, this application selects only three target variables, all of which are indicators that can be directly obtained or easily calculated from routine clinical examinations, significantly reducing the difficulty of data collection. Among these, the diagnosis of severe sarcopenia is based on decreased grip strength, decreased 6-meter stride speed, decreased limb skeletal muscle index, and increased time to sit up in five attempts.

[0024] In some embodiments, the predictive model is constructed based on the following steps: constructing an initial model based on a binary logistic regression algorithm; obtaining a dependent variable and multiple initial independent variables based on case samples; sequentially inputting the multiple initial independent variables into the initial model, excluding collinearity among the multiple initial independent variables, selecting target independent variables, and obtaining the regression coefficient of each target independent variable; and establishing a predictive model based on the dependent variable, target independent variables, and regression coefficients.

[0025] Specifically, this application eliminates collinearity through collinearity diagnosis. Collinearity can distort the estimated regression coefficients of independent variables, and even increase the standard error of the coefficients and make statistical test results unreliable. Therefore, after eliminating collinearity through collinearity diagnosis, the regression coefficients of the target independent variables (femoral neck bone mineral density T-score, diagnosis of severe sarcopenia, MLR) can truly reflect their independent influence on fracture risk, avoiding mutual interference between target independent variables that could lead to bias in the prediction model parameters, thus making the prediction model more reliable. In addition, it can eliminate highly correlated redundant initial independent variables in advance, ensuring that the selected target independent variables are all core factors with independent and significant influence on fracture risk, and avoiding the inclusion of invalid variables in the prediction model.

[0026] In this way, the prediction results can be stably output in different clinical scenarios and different populations, such as the elderly and those without a history of fractures, thereby improving the clinical promotion value and practical application effect of the prediction model.

[0027] In some embodiments, the prediction model is constructed based on the following formula (1): (1); Where p is the probability of a fracture occurring. , 3.296 is the intercept; The T-value is the bone mineral density of the femoral neck. The regression coefficient for the femoral neck bone mineral density T-value; 4.626 is the regression coefficient of the monocyte-to-lymphocyte ratio. The diagnosis was severe sarcopenia. The regression coefficient is used to diagnose severe sarcopenia.

[0028] Specifically 3.296 is the intercept, meaning that when all objective variables are 0, the log-odds of the event is... A value of 3.296 corresponds to a fracture probability of 0.036. In other words, under baseline conditions (where X1, X2, and X3 are all 0), the risk of fracture is approximately 3.6%.

[0029] The T-value is the bone mineral density of the femoral neck. This represents the regression coefficient for the femoral neck bone mineral density (T-score). That is, for every 1 unit increase in the femoral neck T-score, the logarithmic odds of the event decreases by 0.741, with an odds ratio of 0.477. In other words, for every 1 unit increase in the femoral neck T-score, the risk of fracture decreases by 52.3%.

[0030] The ratio of monocytes to lymphocytes is given, and 4.626 is the regression coefficient for this ratio. That is, for every 1 unit increase in the monocyte-to-lymphocyte ratio, the logarithmic odds of the event increase by 4.626, with an odds ratio of 102.113. In other words, for every 1 unit increase in the monocyte-to-lymphocyte ratio, the risk of fracture increases by 101.113 times.

[0031] The diagnosis was severe sarcopenia. This represents the regression coefficient for diagnosing severe sarcopenia. Specifically, compared to not having severe sarcopenia, the logarithmic odds of the event increase by 0.353 when severe sarcopenia occurs, with an odds ratio of 1.423. In other words, the risk of fracture is 1.423 times higher when severe sarcopenia occurs compared to when it does not.

[0032] In some embodiments, the initial independent variables include: grip strength, time to 5 sit-ups, 6-meter walking speed, skeletal muscle index (ASMI) of the extremities, total lumbar spine and femoral neck bone mineral density T-scores, complete blood count, bone metabolism markers, liver function, kidney function, blood lipids, and fasting blood glucose; complete blood count includes white blood cells, red blood cells, hemoglobin, platelets, neutrophils, lymphocytes, monocytes, monocyte-to-lymphocyte ratio (MLR), and neutrophil-to-lymphocyte ratio; bone metabolism markers include parathyroid hormone, osteocalcin, type I procollagen N-terminal propeptide, type I collagen C-terminal peptide, 25-hydroxyvitamin D, serum calcium, and serum phosphorus; liver function includes alanine aminotransferase (ALT), aspartate aminotransferase (AST), gamma-glutamyl transferase (GGT), alkaline phosphatase, and albumin; kidney function includes blood urea nitrogen, serum creatinine, and uric acid; and blood lipids include total cholesterol and triglycerides.

[0033] Specifically, the initial independent variables encompass a multi-dimensional set of indicators, including muscle mass, strength and function (grip strength, 5 sit-ups time, 6-meter stride speed, ASMI), bone mineral density (T-scores of total lumbar spine and femoral neck bone mineral density), bone metabolism markers (parathyroid hormone, osteocalcin, etc.), immune inflammation (leukocytes, lymphocytes, MLR, etc.), liver and kidney function (alanine aminotransferase, serum creatinine, etc.), and glucose and lipid metabolism (fasting blood glucose, total cholesterol, etc.). This comprehensive approach captures physiological, metabolic, and functional risk factors associated with fractures. This rich pool of initial independent variables provides ample screening material for forward stepwise regression and collinearity diagnosis in binary logistic regression. It allows for the precise identification of core variables (such as femoral neck bone mineral density T-score, diagnosis of severe sarcopenia, MLR) that have an independent and significant impact on fracture risk, eliminating redundant and irrelevant variables. This avoids predictive bias caused by insufficient variable dimensions, ensuring that the final predictive model is both concise and focused on key risk points, balancing predictive accuracy and clinical applicability.

[0034] In some embodiments, a fracture diagnosis is obtained based on at least one of thoracic and lumbar spine X-rays (anteroposterior and lateral views), computed tomography (CT), and magnetic resonance imaging (MRI) from a case sample.

[0035] Specifically, thoracic and lumbar spine X-rays (anteroposterior and lateral views) can quickly observe vertebral body morphology, sequence, and obvious fracture lines. CT scans can clearly show vertebral bone details, the degree of fracture comminution, and spinal canal compression. MRI has higher sensitivity in identifying occult fractures, such as fresh compression fractures and bone marrow edema, as well as soft tissue injuries. Appropriate imaging examinations can be selected based on the specific circumstances of the case, or multimodal imaging can be used for complementary verification to effectively avoid the limitations of a single imaging technique and ensure the accuracy and reliability of fracture diagnoses.

[0036] Step 102: Construct a nodal plot based on the target independent variable and the prediction model.

[0037] Please see Figure 2 , Figure 2This is a nomogram illustrating the fracture risk prediction method provided in this application. In this application, during clinical diagnosis, the actual value of the target independent variable can be input into the nomogram to obtain a fracture risk score and directly read the corresponding fracture risk probability. Thus, the prediction model is visualized through the nomogram, eliminating the need for complex calculations. Clinicians can quickly query the fracture risk score using the actual value of the target subject, making it suitable for efficient application in routine outpatient and inpatient settings.

[0038] In some embodiments, nomograms are constructed based on the following steps: determining each target independent variable. i regression coefficient and the range of values ​​for each target independent variable. ; Calculate the maximum linear effect of each target independent variable within its range of values. and order Determine the baseline total score (e.g., 100 points), plot the baseline fractional axis and specify the scaling constant. Based on formula Set the number of points on the target variable for each retrieval query and plot the corresponding target variable axis; plot the total score axis. ; Calculate the adjusted intercept and based on and The total score is mapped to the probability of fracture risk, and a risk axis is plotted. A nodal plot is generated based on the combination of the baseline score axis, the target independent variable axis, the total score axis, and the risk axis.

[0039] Specifically, the binary logistic regression algorithm, also known as the binary logistic regression model, has non-linear characteristics. A nomogram is a visualization based on the logistic regression equation. When the regression coefficients of each target variable are converted into scores, it essentially maps the contributions of different target variables to logit(p) into linear scores. The total score of the entire nomogram is linearly related to logit(p). These values ​​are obtained by mapping the probability intervals output by the predictive model to score intervals. When logit(p) changes rapidly, the corresponding probability change is small; conversely, the opposite is true.

[0040] In some embodiments, the femoral neck bone mineral density T value is X1, with a value range of [-6.0, 1.0]; the ratio of monocytes to lymphocytes is X2, with a value range of [0.1, 1.0]; and the diagnosis of severe sarcopenia is X3, with a value range of [0, 1].

[0041] Thus, the maximum effect of each variable calculated in this application is as follows: ; ; It can be determined that X1 is the maximum effect, that is, the score of femoral neck bone mineral density T-value corresponds to the baseline total score.

[0042] Step 103: Obtain the actual value of the target subject for each target independent variable.

[0043] Specifically, the actual values ​​of each target independent variable are obtained for the target subjects, ensuring that the prediction process is based on the subject's current physiological and examination status, thus guaranteeing the authenticity and reliability of the fracture risk score and probability calculation results from the outset. The actual values ​​of the target independent variables for different subjects naturally exhibit individual differences, directly determining the uniqueness of their fracture risk. Obtaining the actual values ​​accurately captures these individual differences, making the prediction no longer a one-size-fits-all group assessment, but a personalized judgment for the target subject, making the risk assessment results more closely aligned with their actual health condition.

[0044] In addition, the target independent variables are all derived from routine clinical examinations or simple calculations. Obtaining their actual values ​​does not require additional special or complex tests. It can be completed by relying on routine outpatient and inpatient diagnosis and treatment processes, making it convenient to operate without additional medical costs and significantly reducing the operational burden on clinical staff.

[0045] Step 104: Based on the actual values, obtain the fracture risk score of the target subject in the nomogram.

[0046] Specifically, this application obtains risk scores from nomograms based on the actual values ​​of target independent variables of the target subjects, which can convert the impact of each target independent variable on fracture risk into quantifiable scores, avoid the bias of subjective assessment, and greatly improve the accuracy of fracture risk assessment.

[0047] Nom plots replace the complex formula calculations of the Logistic model by visualizing the correspondence between axes. Clinicians no longer need to manually substitute the formula to calculate logit(p). They only need to locate the corresponding target independent variable axis based on the actual value and quickly read the score through the vertical line. The operation is intuitive and simple, and even non-statistical professionals can easily master it, significantly improving the efficiency of clinical assessment.

[0048] In some embodiments, the step of determining the fracture risk score of a target subject in a nomogram based on actual values ​​includes: finding the actual values ​​on the three target independent variable axes and drawing a vertical line for each; converting the actual values ​​into corresponding current scores on the baseline score axis; adding the three current scores to obtain the current total score; finding the corresponding current total score on the total score axis and drawing a vertical line; and determining the fracture risk score on the risk axis.

[0049] Specifically, by locating the actual values ​​on the three target independent variable axes and converting them into current scores, the independent contribution of each target variable to fracture risk can be clearly quantified. The total score is then obtained by summing the scores. The entire process relies on the visual axis structure of the nomogram, allowing scoring to be completed without needing to master complex statistical formulas or model principles. This enables clinicians and nurses to quickly learn and efficiently complete assessments, solving the problems of complex calculations and difficulty in clinical application of traditional models, making it suitable for rapid application in daily clinical scenarios.

[0050] Please see Figure 3 , Figure 3 Clinical decision curves for the fracture risk prediction method provided in the embodiments of this application.

[0051] In some embodiments, the fracture risk prediction method further includes: obtaining the predicted fracture risk probability of the target subject through a prediction model; determining the fracture diagnosis result as the dependent risk variable, determining the predicted fracture risk probability as the independent risk variable, and setting a risk threshold range (0.14~0.61) and confidence interval based on the dependent and independent risk variables; obtaining the first clinical net benefit of the prediction model at different risk thresholds within the risk threshold range; obtaining the second clinical net benefit of all intervention strategies at each risk threshold; obtaining the third clinical net benefit of all non-intervention strategies at each risk threshold; plotting the first net benefit curve of the prediction model, the second net benefit curve of all intervention strategies, and the third net benefit curve of all non-intervention strategies on the risk threshold range as the horizontal axis and the clinical net benefit as the vertical axis, respectively, to form a clinical decision curve; the clinical decision curve is used to determine the clinical applicability of the prediction model within different risk threshold ranges by comparing the positional relationship of the first net benefit curve, the second net benefit curve, and the third net benefit curve.

[0052] The first clinical net benefit is obtained by the following formula (2): (2); Where TP represents the number of true positive individuals, FP represents the number of false positive individuals, and N represents the total number of samples. p t This is the risk threshold.

[0053] Please refer to Tables 1 to 4 for details. Tables 1 to 3 contain a single objective variable, while Table 4 contains multiple objective variables.

[0054] In univariate analysis, severe sarcopenia was significantly associated with fracture risk (P=0.007). However, after including femoral neck bone mineral density (T-score) and MLR in the multivariate model, the independent effect of this variable was no longer significant (P=0.383). This may be because there is a strong correlation between severe sarcopenia and bone mineral density, with bone mineral density mediating the impact of severe sarcopenia on fracture risk to some extent. Nevertheless, given the important physiological role and clinical interpretability of severe sarcopenia in fracture occurrence, it was included in the final prediction model to improve the clinical applicability of risk assessment.

[0055] Table 1

[0056] Table 2

[0057] Table 3

[0058] Table 4

[0059] Specifically, this application employs a fracture risk prediction model based on logistic regression to obtain the predicted fracture risk probability for target subjects. Subsequently, using the actual fracture diagnosis as the dependent risk variable and the predicted probability as the independent risk variable, a risk threshold range (e.g., 0.05–0.90) and confidence interval are set, and the clinical net benefit of the model under different risk thresholds is calculated. The net benefit comprehensively considers the benefit of correct predictions (true positives) and the over-intervention caused by incorrect predictions (false positives), reflecting the comprehensive clinical value of the prediction model under different risk decision-making conditions.

[0060] Furthermore, using the risk threshold as the x-axis and net benefit as the y-axis, net benefit curves for the predictive model, all intervention strategies, and all non-intervention strategies are plotted to form a clinical decision curve. By comparing the relative positions of the various net benefit curves, the clinical applicability of the model within each risk threshold interval can be determined. If the first net benefit curve is higher than the second and third net benefit curves, it indicates that the predictive model provides a higher clinical net benefit within that risk threshold interval, i.e., it has higher clinical decision-making value.

[0061] Thus, through the above steps, this application can quantitatively assess the net clinical benefit of the prediction model at different fracture risk thresholds, enabling an intuitive and visual assessment of the clinical application value of the prediction model. This method can provide a basis for individualized intervention decisions and improve the clinical translational value and practicality of fracture risk prediction models.

[0062] The predictive efficacy of the fracture risk prediction method in this application and the FRAX scoring tool is compared.

[0063] The specific calculation process is as follows: Using the receiver operating characteristic (ROC) curve, the area under the curve (AUC) and the 95% confidence interval (CI) of the corresponding model can be obtained. In the ROC curve, each threshold corresponds to a set of sensitivity and specificity. The Youden index of all thresholds is calculated, and the highest Youden index corresponds to the strongest overall performance in terms of sensitivity and specificity.

[0064] Please see the results. Figure 4 , Figure 4 A schematic diagram comparing the fracture risk prediction value of the fracture risk prediction method provided in this application embodiment with that of other models. The results showed that the prediction model constructed based on the femoral neck bone mineral density T-score, a confirmed diagnosis of severe sarcopenia, and the ratio of monocytes to lymphocytes (…). Figure 4 A) The area under the curve (AUC) was 0.747, the 95% CI was 0.677–0.816, P < 0.001, the sensitivity was 77.1%, the specificity was 68.4%, and the maximum Yoden index was 0.455.

[0065] Major osteoporotic fracture risk prediction model based on FRAX score ( Figure 4 B) The AUC was 0.687, the 95% CI was 0.611–0.762, P < 0.001, the sensitivity was 52.4%, the specificity was 81.9%, and the maximum Yoden index was 0.343.

[0066] Hip fracture risk prediction model based on FRAX score ( Figure 4 C) The AUC was 0.718, the 95% CI was 0.648–0.789, P < 0.001, the sensitivity was 57.1%, the specificity was 77.6%, and the maximum Yoden index was 0.347.

[0067] This indicates that the fracture risk prediction method proposed in this application is superior to the FRAX scoring tool in predicting fracture risk.

[0068] In summary, this application employs forward stepwise regression analysis with binary logistic regression to screen variables, accurately identifying core factors significantly impacting fracture risk. The constructed predictive model effectively captures risk characteristics in elderly individuals and high-risk individuals with no history of fractures. Furthermore, combining this with a nomogram to visually present risk scores further enhances the accuracy of risk assessment for this specific population, preventing missed intervention opportunities due to underestimation of risk. This reduces the complexity of clinical decision-making, allowing clinicians to quickly determine a patient's fracture risk level based on the fracture risk score, providing a quantitative basis for intervention formulation. This approach avoids the waste of medical resources and overtreatment associated with comprehensive intervention, while also reducing the risk of fractures due to inaction or misjudgment of low risk, thus achieving precision medicine.

[0069] Accordingly, this application provides a fracture risk prediction system for performing the above-mentioned fracture risk prediction method. The prediction system includes: a prediction model construction module, a nomogram construction module, an actual value acquisition module, and a fracture risk score acquisition module. The prediction model construction module is configured to acquire case samples of hospitalized patients and construct a prediction model based on the case samples and a binary logistic regression algorithm. The dependent variable of the prediction model is the fracture diagnosis result, and the target independent variables of the prediction model are the femoral neck bone mineral density T-score, the diagnosis of severe sarcopenia, and the ratio of monocytes to lymphocytes. The nomogram construction module is configured to construct a nomogram based on the target independent variables and the prediction model. The actual value acquisition module is configured to acquire the actual value of the target subject for each target independent variable. The fracture risk score acquisition module is configured to acquire the fracture risk score of the target subject in the nomogram based on the actual values.

[0070] The fracture risk prediction system provided in this application has all the technical features and beneficial effects of the above-mentioned fracture risk prediction methods, which will not be repeated here.

[0071] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0072] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A method for predicting fracture risk, characterized in that, include: A case sample of hospitalized patients was obtained, and a prediction model was constructed based on the case sample and a binary logistic regression algorithm. The dependent variable of the prediction model was the fracture diagnosis result, and the target independent variables of the prediction model were the femoral neck bone mineral density T-score, the diagnosis of severe sarcopenia, and the ratio of monocytes to lymphocytes. Construct a nodal plot based on the target independent variable and the prediction model; For each of the target independent variables, obtain the actual value of the target subject; Based on the actual values, the fracture risk score of the target subject is obtained from the nomogram.

2. The fracture risk prediction method according to claim 1, characterized in that, The prediction model is constructed based on the following steps: An initial model is constructed based on the aforementioned binary logistic regression algorithm; Based on the case sample, the dependent variable and multiple initial independent variables are obtained; Multiple initial independent variables are sequentially input into the initial model to exclude collinearity among the multiple initial independent variables, the target independent variables are selected, and the regression coefficient of each target independent variable is obtained; The prediction model is established based on the dependent variable, the target independent variable, and the regression coefficients.

3. The fracture risk prediction method according to claim 2, characterized in that, The initial independent variables include: The study included grip strength, time to 5 sit-ups, 6-meter walking speed, limb skeletal muscle index, total lumbar spine and femoral neck bone mineral density T-score, complete blood count, bone metabolism markers, liver function, kidney function, blood lipids, and fasting blood glucose. The complete blood count included white blood cells, red blood cells, hemoglobin, platelets, neutrophils, lymphocytes, monocytes, the monocyte-to-lymphocyte ratio, and the neutrophil-to-lymphocyte ratio. Bone metabolism markers included parathyroid hormone, osteocalcin, type I procollagen N-terminal propeptide, type I collagen C-terminal peptide, 25-hydroxyvitamin D, serum calcium, and serum phosphorus. Liver function included alanine aminotransferase (ALT), aspartate aminotransferase (AST), gamma-glutamyl transferase (GGT), alkaline phosphatase, and albumin. Kidney function included blood urea nitrogen, serum creatinine, and uric acid. Blood lipids included total cholesterol and triglycerides.

4. The fracture risk prediction method according to claim 1, characterized in that, The fracture diagnosis result is obtained based on at least one of the following: thoracic and lumbar spine X-ray (anteroposterior and lateral views), computed tomography (CT) scan, and magnetic resonance imaging (MRI) from the case sample.

5. The fracture risk prediction method according to claim 1, characterized in that, The prediction model is constructed based on the following formula: ； Where p is the probability of a fracture occurring. , 3.296 is the intercept; The T-value is the bone mineral density of the femoral neck. The regression coefficient for the femoral neck bone mineral density T-value; 4.626 is the regression coefficient of the monocyte-to-lymphocyte ratio. The diagnosis of severe sarcopenia was confirmed. The regression coefficient is used to diagnose the severe sarcopenia.

6. The fracture risk prediction method according to claim 2, characterized in that, The column chart is constructed based on the following steps: Determine each of the target independent variables i The regression coefficients and the range of values ​​for each of the target independent variables. ; Calculate the maximum linear effect of each target independent variable within its range of values. and order ; Determine the baseline total score Draw the baseline fractional axis and specify the scaling constant. ; Based on formula Set the number of points on the scale for each retrieval target variable and plot the corresponding target variable axis; Plot the total score axis ; Calculate the adjusted intercept and based on and Map the total score to the fracture risk probability and plot the risk axis; The nomogram is generated by combining the baseline score axis, the target independent variable axis, the total score axis, and the risk axis.

7. The fracture risk prediction method according to claim 6, characterized in that, The femoral neck bone mineral density T-value is X1, with a value range of [-6.0, 1.0]; the monocyte to lymphocyte ratio is X2, with a value range of [0.1, 1.0]; and the diagnosis of severe sarcopenia is X3, with a value range of [0, 1].

8. The fracture risk prediction method according to claim 6, characterized in that, Based on the actual values, the step of determining the fracture risk score of the target subject in the nomogram includes: Find the actual value of each of the three target independent variable axes, draw a vertical line for each, and convert the actual value into the corresponding current score on the baseline score axis; Add the three current scores together to get the current total score; Find the corresponding current total score on the total score axis, draw a vertical line, and determine the fracture risk score on the risk axis.

9. The fracture risk prediction method according to claim 1, characterized in that, Also includes: The predicted probability of fracture risk for the target subject is obtained through the prediction model. The fracture diagnosis result is determined as the risk dependent variable, the fracture risk prediction probability is determined as the risk independent variable, and the risk threshold range and confidence interval are set based on the risk dependent variable and the risk independent variable. At different risk thresholds within the stated risk threshold range, the first clinical net benefit of the prediction model is obtained, wherein the first clinical net benefit is obtained using the following formula: ； Where TP represents the number of true positive individuals, FP represents the number of false positive individuals, and N represents the total number of samples. p t The risk threshold is defined as follows; Obtain the second clinical net benefit of all intervention strategies at each of the stated risk thresholds; Obtain the third clinical net benefit of all non-intervention strategies at each of the stated risk thresholds; Using the risk threshold range as the horizontal axis and the clinical net benefit as the vertical axis, a first net benefit curve of the prediction model, a second net benefit curve of all intervention strategies, and a third net benefit curve of all non-intervention strategies are plotted to form a clinical decision curve. The clinical decision curve is used to determine the clinical applicability of the prediction model within different risk threshold ranges by comparing the positional relationship of the first net benefit curve, the second net benefit curve, and the third net benefit curve.

10. A fracture risk prediction system for performing the fracture risk prediction method as described in any one of claims 1 to 9, characterized in that, The prediction system includes: The prediction model building module is configured to acquire case samples of hospitalized patients and build a prediction model based on the case samples and a binary logistic regression algorithm; the dependent variable of the prediction model is the fracture diagnosis result, and the target independent variables of the prediction model are the femoral neck bone mineral density T-score, the diagnosis of severe sarcopenia, and the ratio of monocytes to lymphocytes; The nomogram construction module is configured to construct a nomogram based on the target independent variable and the prediction model; The actual value acquisition module is configured to acquire the actual value of the target subject for each of the target independent variables; The fracture risk score acquisition module is configured to acquire the fracture risk score of the target subject in the nomogram based on the actual value.

Images