| Background:Vertebral fractures are the most common osteoporotic fracture,causing chronic pain,loss of function,and are associated with increased mortality and heavy burden.Epidemiological data on vertebral fractures in Chinese population are limited and risk factors associated with vertebral fractures have not been adequately identified.A risk assessment system for vertebral fractures with Chinese characteristics has not yet been established and improved.Objectives:The Chinese Osteoporosis Prevalence Study(COPS)was used to investigate the risk factors for vertebral fractures in the Chinese population and to construct a prediction model for the existence of vertebral fractures,in order to provide a scientific and theoretical basis for early screening and diagnosis of vertebral fractures.Methods:The COPS cohort was used as the development cohort,and the Beijing subgroup of the Chinese Vertebral Osteoporosis Study(ChiVOS)was used as the external validation cohort.Data including demographic survey data,physical measurements,bone mineral density,and bone metabolic indexes were first collected,and the 10-year fracture risks were calculated with the use of the FRAX tool.Demographic characteristics were described and significance tests for each factor were performed between the vertebral fracture group and the no vertebral fracture group for overall,males,and females,respectively,and those with significant differences were selected and entered the univariate logistic regression to obtain the odds ratio(OR).Then variables were additionally manually filtrated and entered multivariate logistic stepwise regression to build the model and plot the receiver operating characteristic curves(ROC).The area under the curve(AUC)and Brier scores were calculated in the development cohort,internal bootstrap resampling cohort and external ChiVOS cohort,and calibration plots were drawn to evaluate the discrimination and calibration of the models which were subsequently compared using Delong test and NRI index.The best performing model was selected for visualization and score sheets were produced.Statistical cutoff values were set based on the Youden index.The established prediction model and the risk score system were compared with screening indications from commonly used guidelines,and FRAX in a development cohort.Results:This study included 8422 subjects with a mean age of 57.4 ± 10.1 years.There were 3596 males with a mean age of 57.8± 10.4 years and 4826 females with a mean age of 57.03±9.82 years.Vertebral fractures occurred in 806 individuals with a mean age of 64.6±8.8 years and 48.6%of females.Most variables were significantly different between the overall,male and female vertebral fracture groups and the no vertebral fracture group.After filtrated by univariate logistic regression and multivariate logistic stepwise regression,model 1 was constructed:sex,age,height,weight,urbanity,hunchback,previous fracture history,bone mineral density,and FRAX risk of major fracture within 10 years;model 2:model 1 minus FRAX;and model 3:model 2 minus bone mineral density.the AUC values of the 3 models in the development cohort were 0.775,0.774 and 0.759 respectively,and Brier scores were 0.078,0.078 and 0.080,respectively;validated by internal resampling,the AUCs were 0.772,0.772 and 0.757,and Brier scores were 0.078,0.078 and 0.080,respectively;validated by external cohort,the AUCs were 0.754,0.755,0.740 and the Brier scores were 0.133,0.133 and 0.135.The AUC for model 2 was not significantly different from model 1(p=0.3 57)but significantly higher than model 3(p<0.001)by Delong’s test.The NRI at 5%,10%and 15%cut points obtained by comparison between model 1 and 2 was statistically significant(p1=0.615,p2=0.242,p3=0.704).Model 2 was posted at https://yaooo-9697.shinyapps.io/dynnomapp/and a score sheet was created accordingly.The screening cutoff values for Model 2 and the rating scale were set at 8.6%and 11 points based on the Youden index.Model 2 had an AUC of 0.702,a sensitivity of 75.2%,and a specificity of 65.2%in the development cohort;the rating scale had an AUC of 0.692,a sensitivity of 68.6%,and a specificity of 69.9%.Both had higher AUC and sensitivity than ISCD2019(AUC 0.576,sensitivity 27.7%),vertebral fracture screening indications of Chinese osteoporosis guidelines(AUC 0.639,sensitivity 54.7%),and FRAX major fracture risk(cut-off value 3.25%,AUC 0.646,sensitivity 62.5%).Conclusion:This study constructed a vertebral fracture risk prediction model for the Chinese population,which consists of sex,age,height,weight,urbanity,hunchback,previous fracture history,and bone mineral density.It is superior to existing vertebral fracture screening indications and FRAX with possible clinical implications.This study provides scientific recommendations for anchoring the population for the determination of performing vertebral fracture screening. |
PDF Full Text Request