Establishment And Validation Of Machine Learning Model For Predicting The Risk Of Reduced Bone Mineral Density

ObjectiveTo establish and verify the risk model of bone mineral density reducti on predicted by simple and easily available clinical laboratory indicators,a nd improve the clinical attention to Reduced bone density.MethodsThe clinical data of patients who underwent dual-energy X-ray bone densitometry together with routine blood,urine and biochemical indexes such as liver and kidney function and electrolyte glucose in Chifeng Municipal Hospital from August 2018 to January 2022 were collected as the training data for machine learning;the clinical data of patients who also underwent the above examination items in Chifeng Municipal Hospital from February 2022 to June2022 were collected as the machine learning data for this study Validation data.The data sets were divided into normal and reduced bone density groups according to the results of dual-energy X-ray bone density examination.The t-test,LASSO regression,and logistic regression were used to screen statistically significant clinical laboratory indicators for model construction,to establish a predictive BMD hypoplasia line graph,to visualize the model,and to quantify the contribution of each indicator to the prediction of BMD hypoplasia,and to verify the predictive efficiency of the line graph by ROC curve,correction curve,DCA curve,and CIC curve.Results1.Training data were collected from 3050 patients: 1525 in the group with reduced bone mineral density and 1525 in the group with normal bone mineral density;validation data were collected from 497 patients: 261 in the group with reduced bone mineral density and 236 in the group with normal bone mineral density;a total of 59 clinical and laboratory indicators were included.2.41 characteristics were screened into LASSO regression by t-test,and11 characteristics were further screened into logistic regression,and the final logistic multi-factor regression analysis obtained 6 indicators associated with reduced bone mineral density: age(OR: 1.0157;P<0.0001),mean red blood cell volume(OR: 1.0087;P< 0.0001),blood alkaline phosphatase(OR: 1.0025;P<0.0001)were positively associated with reduced bone mineral density,and eosinophil ratio(OR: 0.9846;P<0.0001),blood uric acid(OR: 0.9997;P<0.0001)and male(OR: 0.8771;P<0.0001)were protective factors.3.The derived 6 indices were used for the construction of the bone density abnormality model,and the model was visualized by drawing a column line diagram.The column line graphs can show the contribution of each index to the reduced bone mineral density separately.Using Point 10 as an example,the corresponding increased values of the risk of reduced bone mineral density were similar for women,for every 8 years increase in age,for every 8%decrease in eosinophils,for every 13 fl increase in mean red blood cell volume,for every 40 IU/L increase in alkaline phosphatase and for every 400 μmol/L decrease in uric acid.4.The model was internally validated on the training data,and the AUC of the predictive analysis ROC was 0.848,and the sensitivity and specificity were 0.796 and 0.765,respectively;the model was externally validated on the validation data,and the AUC of the predictive analysis ROC was 0.806,and its sensitivity and specificity were 0.686 and 0.793,respectively;the predicted probability of the calibration curve was similar to the actual probability,suggesting that The DCA curve and CIC curve both reflect that the prediction model has good clinical application value.Conclusions1.The 6 indexes of age,sex,mean volume of red blood cells,blood alkaline phosphatase,eosinophil ratio,and blood uric acid were important predictors of reduced bone mineral density.2.The above 6 indexes constructed a model of bone mineral density abnormalities with reliable prediction accuracy and high differentiation.3.The column line graphs drawn by the above 6 indicators can quantify the risk of reduced bone mineral density and its related predictive contribution of the subjects.