Research On The Relationship Of Blood Biochemical Indicators With Bone Health And Establishment Of A Screening Model For Osteoporosis

Backgroud and Purpose:With the aging of the world’s population,osteoporosis,as a public health problem,has gradually become more noteworthy.Osteoporosis is due to the decrease in the amount of bone tissue and the degradation of bone microstructure,which leads to a decrease in bone strength.These diseases often increase the risk of fractures.Bone Mineral Density（BMD）is currently the gold standard for diagnosing osteoporosis before treatment.Several epidemiological studies have shown that as BMD decreases,the risk of fracture increases.Bone health is affected by many factors.The currently recognized risk factors for osteoporosis include history of fractures,low body mass index（BMI）,drug treatments that affect bone metabolism,diseases that affect bone metabolism,long-term smoking and excessive drinking,and excessive caffeine intake.In addition,some epidemiological studies have shown that osteoporosis and BMD are related to some human biochemical indicators,including blood sugar,blood lipids,liver function,kidney function,sex hormones and other indicators.However,these studies often have some problems.Samples and indicator ranges are limited.And observational research faces the shortcomings of inability to estimate causality.As a chronic disease,osteoporosis could cause fractures,which are more serious consequences.Screening for osteoporosis and early warning of bone loss are very necessary.Extensive examination of BMD is not the current development of osteoporosis screening tools,which are mostly limited to high-risk groups.They are screened based on recognized risk factors such as age,BMI,and menopausal status.And most of these studies are not based on Chinese people.There is still much room for improvement in the screening effect of these tools.Based on the above background,we conducted a three-part study.The first part established a multi-center bone health database based on ordinary Chinese adults,and used this to study the correlation between various blood biochemical indicators and bone health.The second part uses Mendelian Randomization（MR）method to search the effect of circulating sex hormone-binding globulin（SHBG）concentration on BMD,and observe the causal relationship between the two.The third part is based on the multi-center bone health database,using the extreme gradient boosting（XGBoost）algorithm to establish a model for screening osteoporosis with blood biochemical indicators for normal and healthy people in China（Biochemical Indicators Screening Osteoporosis Based on XGBooost,BISOBX）.Aimed to help screen population which without obvious risk factors for osteoporosis.Methods:Part 1:Collecting the health examination data of the general community population in 5 medical centers,and select the data records with BMD examination.By this way,we established a multi-center bone health database as our research population after excluding the factors that significantly affect bone metabolism.We Selected 12 biochemical indicators as targets:Fasting Plasma Glucose（FPG）,serum uric acid（UA）,serum calcium（CA）,serum alkaline phosphatase（ALP）,alanine aminotransferase（ALT）,aspartate aminotransferase（AST）,triglycerides（TG）,total cholesterol（TC）,high density lipoprotein cholesterol（HDL-c）,total bilirubin（TBIL）,albumin（ALB）,and estimated glomerular filtration rate（e-GFR）calculated based on serum creatinine（Scr）.Based on BMD and osteoporosis（or osteopenia）as the outcome,with baseline characteristics such as age,BMI,and blood pressure as confounding factors,to analyze the correlation between various biochemical indicators and bone health.According to the difference of age and gender,subgroup analysis was carried out to analyze the relationship between various biochemical indicators and bone health in the whole population,men,women,men over 50,and postmenopausal women.Part 2:We used Mendelian randomization method,with circulating SHBG concentration as the exposure and BMD as the outcomes,with the help of instrumental variable single nucleotide polymorphism（SNP）to study the causal relationship between exposure and outcome.The aggregated SNP data for the association between exposure and outcome comes from several large-scale meta-studies of genome-wide association studies（GWAS）,most of which are based on Europeans.There were 9 outcomes,including BMD in four different skeletal parts which included forearm（FA）,femoral neck（FN）,lumbar spine（LS）and heel（HL）.Other five outcomes were toatal body BMD in different age groups,including under 15 years old,15-30 years old,30-45 years old,45-60 years old and over 60 years old.Main body analysis adopts fixed-effect inverse variance weighted（IVW）method,sensitivity analysis adopts weighted median（Weighted median）analysis and MR-Egger analysis,as well as IVW analysis after removing a small part of SNPs.The intercept item of MR-Egger analysis was used to test the directional pleiotropy,and the MR-PRESSO（Mendelian Randomization Pleiotropy Residual Sum and Outlier）method was used to test the horizontal pleiotropy.Part 3:Based on the excluded population in the multi-center bone health database,we randomly divide it into training set and test set at a ratio of 80%and 20%.With osteoporosis as the response variable,in terms of biochemical indicators and the baseline characteristics are used to screen predictor variables that have predictive effects on osteoporosis.Three different types of machine learning algorithms including XGBoost are used to establish screening models.The other two algorithms are Support Vector Machine（SVM）and simple artificial neural network（ANN）,which were compared the screening effect to the model BISOBX based on the XGBoost algorithm.The training set data was used to train the model.And the back judgment and internal testing of the model are carried out in the training set and the test set respectively.The screening effect evaluation method includes the classifier confusion matrix and its derived related indicators,as well as the receiver operating characteristic（ROC）curve,the precision-recall（PR）curve and the corresponding area under curves（AUC）.,Osteoporosis Self-assessment Tool for Asian（OSTA）,which is a traditional screening tool for Asian postmenopausal women,was compared the screening effect with BISOBX in suitable populations.Results:Part 1:Partial correlation coefficient（PCC）of each biochemical index and baseline characteristics with BMD shows that the age is the most important negatively correlated variable in the whole population（PCC=-0.228）,female（PCC=-0.504）and postmenopausal female（PCC=-0.477）,while its relative importance is very weak in men（PCC=-0.031）.ALP is the relatively most important biochemical indicator variable in the whole population and all subgroups（PCC is-0.223,-0.130,-0.245,-0.151,-0.237,respectively）.As a positively correlated variable,FPG has a certain relative importance in the whole population and all subgroups.But in the whole population and men,the relative importance of e-GFR（PCC is 0.064 and 0.060,respectively）exceeds that of FPG（PCC is 0.055 and 0.043,respectively）.In the overall multivariate logistic regression analysis,the Odd Ratio（OR）and 95%confidence interval（CI）of FPG for osteopenia and osteoporosis were 0.926（0.885,0.969）and 0.839（0.757,0.929）;UA’s OR（95%CI）for osteopenia and osteoporosis were 0.921（0.876,0.968）and 0.601（0.533,0.679）,respectively;the ORs（95%CI）of e-GFR for osteopenia and osteoporosis were 0.705（0.671,0.741）and 0.759（0.682,0.845）;the ORs（95%CI）of TC for osteopenia and osteoporosis were 1.115（1.061,1.172）and 1.297（1.168,1.440）;the OR（95%CI）of ALP for osteopenia and osteoporosis were 1.522（1.452,1.596）and 1.789（1.644,1.946）.In the multivariate logistic regression analysis of males and females,UA and ALP are both significant independent factors related to osteopenia and osteoporosis.For men,HDL-c has a significant positive correlation with osteopenia（OR=1.133,95%CI:1.064-1.206）and osteoporosis（OR=1.218,95%CI:1.053-1.410）;ALB has a significant negative correlation with osteopenia（OR=0.869,95%CI:0.816-0.925）and osteoporosis（OR=0.805,95%CI:0.686-0.945）;e-GFR has a significant negative correlation with osteopenia（OR=0.588,95%CI:0.552-0.625）and osteoporosis（OR=0.508,95%CI:0.429-0.602）.For women,FPG has a significant negative correlation with osteopenia（OR=0.916,95%CI:0.848-0.991）and osteoporosis（OR=0.790,95%CI:0.680-0.919）.In addition,FPG showed a significant negative correlation factor of osteopenia in men（OR=0.932,95%CI:0.880-0.988）,while TG and TC showed a significant positive correlation factor of osteopenia,OR（95%CI）were 1.136（1.068,1.208）and 1.068（1.004,1.137）,respectively.For women,TC is a positive factor for osteoporosis（OR=1.279,95%CI:1.092-1.497）,and CA is a positive factor for osteopenia（OR=1.148,95%CI:1.043-1.264）.Part 2:In the IVW analysis,SHBG level is negatively correlated with FA BMD（effect=-0.26,P=0.022）and HL BMD(effect=-0.09,P=3.19×10^-9)with causality.However,there is no significant causality between FN BMD and LS BMD.In participants aged 45 to 60（effect=-0.16,P=0.047）and participants older than 60 years（effect=-0.19,P=0.006）,there was a causal relationship between SHBG levels and body BMD.With the growth of age,the causal relationship between SHBG levels and total body BMD gradually becomes significant.There is no evidence that there is a causal relationship between SHBG levels and total body BMD in participants under 45 years old.In the analysis of the weighted median method,SHBG level is negatively correlated with FA BMD（effect=-0.40,P=0.005）and HL BMD(effect=-0.09,P=8.93×10^-6)with causality.However,in the analysis of total body BMD at five age groups,SHBG levels have no significant causal relationship with the systemic BMD of participants over 45years old.In the IVW analysis after removing 2 SNPs,the SHBG level has a significant negative effect on the BMD at FA（effect=-0.25,P=0.034）and HL(effect=-0.07,P=1.28×10^-5).MR-Egger analysis did not show the existence of directional pleiotropy,MR-PRESSO analysis showed that there was horizontal pleiotropy in the analysis of the two outcomes（HL BMD and 45-60 years old total body BMD）.The results after removing outlier SNPs showed that here is a causal relationship between the SHBG level and HL BMD（effect size=-0.08,P=0.001）.Part 3:The two variables of TBIL and blood pressure were removed from the variable screening,and the remaining biochemical indicators and baseline characteristics showed predictive effects on osteoporosis in the whole population or male（female）.BISOBX has shown a strong screening effect in both the return judgment of training set and the prediction for test set.Its ROC curve performance is excellent in the training set（AUC=0.996,95%CI:0.995-0.997,P<0.001）and the test set（AUC=0.996,95%CI:0.995-0.997,P<0.001）.Due to the imbalance of the sample caused by the prevalence of osteoporosis,the PR curve is not as good as the ROC curve,and its AUC is 0.89 and0.729,respectively.BISOBX has a higher area under the curve than models based on SVM and ANN algorithms,and has a higher sensitivity（training set:99.77%;test set:97.62%）and specificity（training set:93.97%;test set:93.81%）at a cutoff of 0.5.In the osteoporosis screening for postmenopausal women,BISOBX（AUC=0.965,95%CI:0.955-0.974,P<0.001）was rather better than OSTA（AUC=0.755,95%CI:0.723-0.787,P<0.001）.The most important variable in the BISOBX screening model is age,which accounts for 23.06%of importance,followed by e-GFR,ALP,UA,BMI,and HDL-c,with importance accounting for 11.69%,10.03%,9.18%,6.40%and 5.13%respectively.The importance of other variables accounted for less than 5%.Conclusion:Part 1:Within the normal range,higher levels of serum or blood sugar are related to higher BMD,and are related to lower risk of bone loss and osteoporosis.High levels of serum calcium may be related to low bone mass for women.The increase in serum alkaline phosphatase may be related to high levels of bone metabolism,bone turnover or osteoporosis.High levels of triglyceride and total cholesterol affect the bone health of men and women to some extent.High levels of HDL cholesterol and low levels of albumin may be related to low bone mass in men.In people with normal liver function,there is no evidence that there is a significant association between serum total bilirubin levels and osteoporosis.The glomerular filtration rate is significantly positively correlated with bone mass.Part 2:There may be a correlation between circulating SHBG concentration and BMD,which has a negative effect on BMD to a certain extent.SHBG levels have a significant negative causality relationship with FA BMD,and they are robust in sensitivity analysis.SHBG level and HL BMD have a significant negative causal relationship,and it is still significant after removing the level of pleiotropy.There may be a negative causal relationship between SHBG levels and total body BMD in people above 45 years old,which needs further confirmation.There is little evidence that SHBG levels are causally related to FN,LS,and total body BMD in participants under 45 years old.The causality between SNBG levels and BMD in men or wemen need further research.Part 3:BISOBX had an excellent effect in screening for osteoporosis in our research population（the general community population with approximate health in China）.The screening based on blood biochemical indicators maked BISOBX applicable to a wider population of people,not only limited in people with high-risk factors of bone loss.BISOBX had better performance than the screening model established by SVM and ANN algorithms.Compared with the traditional osteoporosis screening model OSTA,BISOBX could screen osteoporosis patients in the general population more accurately,and has made some fillings in the field of osteoporosis screening.