| Objective:Aging is a complex process of biological system degradation characterized by irreversible accumulation of degenerative changes in the body and increased susceptibility to the disease,eventually leading to the end of life.Although aging is not a disease,it can lower the threshold of aging-related diseases and increase the probability of illness.Osteoporosis?OP?,as one of the common diseases of the elderly,is characterized by the continuous reduction of bone mass,resulting in decreased bone strength and increased susceptibility to fractures.Especially in postmenopausal women and men aged 50 and over.With the increasing aging of the global population,OP will become an important public health problem that all countries in the world need to face.Our team's previous studies have confirmed that there was an interaction between heart and kidney,heart and blood vessels,blood vessels and bones.It is suggested that there is a complex network structure between the organs in the body.However,it did not confirmed whether there was an interaction between kidney and bone in healthy people.Therefore,the purpose of this study was to demonstrate whether there is an association between kidney function parameters and bone mineral density through this cross-sectional studies and the effect of renal function index changes on bone loss was observed in the population.In addition,if the application of artificial neural network model can be used to predict the changes of bone mineral density by screening the indicators from the analysis of a longitudinal study,and its advantages compared with traditionally statistical methods,it provides an important way for early prevention,diagnosis and treatment of osteoporosis.Methods:1.Study population:This was a community-based cross-sectional study in which all subjects were randomly selected from 15 different Shenyang communities between January 2014 and February 2015.Registration requirements for the subjects were as follows:junior middle school education or higher;?35 years;no self-reported history of disease?heart,brain,lung,kidney,liver disease,hypertension,diabetes,hyperthyroidism,tumor,rheumatoid disease,and chronic infection?;no medication use?including estrogen treatment,steroid use,and other drugs?;good health?including normal visual and auditory function,emotionally stable,able to handle family and social relationships,able to adapt to the environment,and capable of learning and committing skills to memory?;capable of caring for themselves and complete daily living activities without difficulty or the need for help;and of sound mind and able to provide self-reported data and sign informed consent.Finally,a total of 550 subjects were selected including 160 men?age 50 to 88?and 390 postmenopausal women?age 47 to89?.In January 2017,550 healthy subjects who were selected in January 2014 made an appointment for a longitudinal follow-up visit to the Geriatric Laboratory of Shengjing Hospital,China Medical University.The study subjects focused on 412 follow-up without osteoporosis in 2014.However,351 subjects were chosen in the longitudinal follow-up,including 92 males and 259 postmenopausal women,which excluded 33subjects without follow-up and 28 subjects with missing data.It was verified that all participants had signed an informed consent.2.Study methods:?1?Data collection:The data recorded included name,age,gender,height,weight,waist circumference,hip circumference,smoking and drinking history,menopausal status,medication history,fractures history,as well as lifestyle?including calcium and vitamin D supplements,milk intake,and physical exercise etc?.Blood pressure was measured on the right arm of the seated participant who had rested for10–15 min in a comfortable temperature environment with a manual stethoscope and a mercury-column sphygmomanometer with an appropriate size cuff.Blood samples were collected between 8:00 and 9:30 AM after all subjects had fasted for at least 10 h.All blood samples were then sent to Shengjing Hospital Inspection Center for analysis?ARCHITECT ci16200 Integrated System?included the following:TP,ALT,AST,TBil,TG,TC,LDL-C,HDL-C,SUA,BUN,SCr,CYSC,FBG,and SCa.The reference range for SUA was 142–420?mmol/L.The reference range of measurements used for SCr was0.51 to 0.95mg/dl and 0.67 to 1.18mg/dl for women and men,respectively.The remaining blood samples were quickly centrifuged,aliquoted and frozen?-80??.The estimated glomerular filtration rate?eGFR?was calculated using the Modified Chronic Kidney Disease Epidemiology Collaboration?CKD-EPI-ASIA?equation.The CKD-EPI-ASIA equations are as follows:Male:SCr?0.9mg/dl,eGFR-CKD-EPI-ASIA=141×?SCr/0.9?-0.411×?0.993?age×1.057;SCr>0.9mg/dl,eGFR-CKD-EPI-ASIA=141×?SCr/0.9?-1.209×?0.993?age×1.057;Female:SCr?0.7mg/dl,eGFR-CKD-EPI-ASIA=144×?SCr/0.7?-0.329×?0.993?age×1.049;SCr>0.7mg/dl,eGFR-CKD-EPI-ASIA=144×?SCr/0.7?-1.209×?0.993?age×1.049?In addition,frozen serum of 550 subjects in storage?-80??were examined in January2017.Enzyme linked immunosorbent assay?ELISA?was used to detect the specific indicators?including P1NP,CTX1,DHVD3,TESTO,IGF-1,GDF-11,FGF-23 and BMP4?related to bone metabolism in serum.The content of each index in human serum was quantitatively determined by the double antibody sandwich ELISA method.Lumbar spine BMD was examined using dual-energy x-ray absorptiometry?Discovery-Wi S/N88155?at the Bone Density Testing Laboratory of Shengjing Hospital of China Medical University.The LBMD of the participants included data from the first,second,third,and fourth lumbar vertebrae.BMD values were defined as grams per square centimeter.The subjects were then classified according to the criteria of the World Health Organization?WHO?as normal BMD group?T-score?1.0?,osteopenia group?2.5<T-score<1.0?,and osteoporosis group?T-score?2.5?.The T-Scores of all the subjects were based on the average of the data from the four vertebrae.?2?Statistical analysis and establishment of BP neural network model.All statistical analyses were made by SPSS version 17.0?Chicago,USA?,The Kolmogorow-Smirnov test was used for normal distribution analyses,For continuous variables,the mean and standard deviation?SD?were calculated.The one-way analysis of variance?ANOVA?was used to examine the comparative analysis between different groups.According to the variance heterogeneity of the test,the least squares difference?LSD?or Dunnett's T3method was used for post hoc comparisons.For the categorical variables expressed as number?percentile??n[%]?,and the chi-square test was used to analyze comparisons between different groups.The correlation between renal function indexes and other variables was analyzed by Pearson correlation analysis.Multiple regression analysis was adopted using the lumbar spine BMD as the dependent variable,The SUA and eGFR was used as the independent variable.Meanwhile,the subjects were divided into quartiles?1–4?of SUA and eGFR,with quartile 1 showing the lowest levels of SUA and eGFR.Binary logistic regression was performed using LBMD decline?those classified as having osteopenia and osteoporosis?as the dependent variable and quartiles of SUA and eGFR as the independent variable to analyze associations between them.Tests for a linear trend across quartiles were performed by modeling the median value of each quartile as a continuous variable.Two models were applied as follows:a model adjusting for age and BMI,and another model adjusting for age,BMI,waist cir,Hip cir,blood pressure,cigarette smoking,alcohol consumption,milk intake,calcium and vitamin D supplements,physical exercise,fracture history,TP,TBil,ALT,AST,TC,TG,HDL-C,LDL-C,FBG,and SCa.A value of P<0.05 was considered as statistical significance.The non-conditional single factor logistic regression was used to screen the indicators of evaluation.The multivariate logistic regression model was constructed to predict OP by selecting various indicators.The neural network model is constructed by using Matlab7.1software program.The examinations and inspection indicators of the subjects were standardized,and the effect of the factor dimension on the data statistics was eliminated.In this study,three layers of BP neural network were used to investigate the grouping of data samples;using the 4:1 random sampling method,the study subjects were classified into two groups:263 training samples and 88 sets of test samples.The implicit layer uses the formula:l=?m+n?1/2+a?m is the number of input nodes and n is the number of output nodes?,0<a<10.The activation function of the hidden layer is tansig,in the form of y=1/[arctan?x?+1].The output function of the output layer is logsig,in the form of y=1/[1+exp?-x?].The number of training sessions for the BP network is initially set up 5000 times,the target error is set up 0.01,and the learning rate is set to 0.1.The training of the network ends because of the minimization of the mean square error of the training group.Eventually,the prediction effect of BP neural network model and logistic regression model on osteoporosis was compared by ROC curve.Results:In this cross-sectional study,we found that SUA and eGFR decreased with BMD decline in postmenopausal women,and the comparison between groups was statistically significant?P<0.05?.SUA and eGFR were significantly positive correlation with LBMD?R=0.120,p<0.05;r=0.159,p<0.01?.After adjustment for many potential confounding variables including age,BMI,waist cir,hip cir,smoking and alcohol history,milk intake,calcium and vitamin D supplements,physical exercise,fracture history,SBP,DBP,TP,ALT,AST,TBil,TG,TC,HDL-C,LDL-C,FGB,the SCa,SUA and eGFR were still closely associated with LBMD,P values were 0.002 and 0.020 respectively.When binary logistic regression analyses using LBMD decline?including osteopenia and osteoporosis?as the dependent variable and quartiles of eGFR and SUA groups as the independent variables to analyze,compared with the lowest quartile 1 group of SUA and eGFR levels,the odds ratios of SUA quartile 2 group?3 group and 4 group were 0.38?95%CI 0.18-0.82,P<0.05?,0.33?95%CI 0.15-0.72,P<0.01?and 0.30?95%CI0.14-0.67,P<0.01?,respectively,with P=0.004 for the trend in SUA;the odds ratios for subjects with LBMD decline in eGFR quartile 3 group was 2.62?95%CI 1.21-5.67,P<0.05?,and that of quartile 4 group was 4.08?95%CI 1.65-10.09,P<0.01?,respectively,with P=0.002 for the trend in eGFR.However,eGFR and SUA were not associated with LMBD in men aged 50 and over.In the longitudinal study,the results of single factor logistic regression analysis:age?p<0.05?,height?p<0.05?,weight?p<0.05?,BMI?p<0.05?,BSA?p<0.05?,hip cir?p<0.05?,calcium and vitamin D supplements?p<0.05?,TC?p<0.05?,SUA?p<0.05?,eGFR?p<0.05?,P1NP?p<0.05?,CTX1?p<0.05?,and DHVD3?p<0.05?.The results of multivariate logistic regression analysis:a total of 5factors entered the model,including age?p<0.01?,BSA?p<0.01?,eGFR?p<0.05?,P1NP?p<0.05?and CTX1?p<0.05?,the accuracy of the whole training set was 73.79%,the sensitivity was 85.45%,and the specificity was 55.80%.BP neural network model for predicting OP:we established three layer BP neural network model,the indicates as input variables were screened by single factor analysis,the hidden layer is 7,output layer is non-OP or OP,The training of the network ends because of the minimization of the mean square error of the training group,the accuracy rate was 88.64%,the specificity was82.86%,the sensitivity was 92.45%.The results of ROC curve comparison:the area under the ROC curve of the multivariate logistic regression model was 0.706,the 95%confidence interval was?0.648,0.648?,the area under the curve of BP neural network model was 0.895,95%confidence interval was?0.857,0.857?.Conclusions:This cross-sectional study showed that the normal physiological range of SUA and eGFR is linearly associated with LBMD,and the two indexes of renal function were independent of LBMD decline?osteopenia and osteoporosis?in postmenopausal women.Therefore,it is speculated that higher SUA and eGFR levels in the physiological range may play a protective role in bone loss.In this longitudinal study,BP neural network method was adopted to predict osteoporosis by selecting each indicators and the accuracy is better.Compared with the traditional logistic regression model,the predicting accuracy,sensitivity and specificity of BP neural network model were relatively high.It provides a new means of exploration for us to study osteoporosis.The early prevention,diagnosis and treatment for disease also provide the help.It is helpful to the selection of biomarkers,and to achieve the goal of multiple and big data mining. |
PDF Full Text Request