Combining information from multiple bone turnover markers as diagnostic indices for osteoporosis using support vector machines

Context: Osteoporosis (OP) is a progressive systemic bone disease. Dual-energy X-ray absorptiometry (DXA) is routinely employed and is considered the gold standard method for the diagnosis of OP. Objective: We aimed to investigate the potential use of combined information from multiple bone turnover markers (BTMs) as a clinical diagnostic tool for OP. Materials and methods: A total of 9053 Chinese postmenopausal women (2464 primary OP patients and 6589 healthy controls) were recruited. Serum levels of six common BTMs, including BAP, BSP, CTX, OPG, OST and sRANKL were assayed. Models based on support vector machine (SVM) were constructed to explore the efficiency of different combinations of multiple BTMs for OP diagnosis. Results: Increasing the number of BTMs used in generating the models increased the predictive power of the SVM models for determining the disease status of study subjects. The highest kappa coefficient for the model with one BTM (BAP) compared to DXA was 0.7783. The full model incorporating all six BTMs resulted in a high kappa coefficient of 0.9786. Conclusion: Our findings showed that although single BTMs were not sufficient for OP diagnosis, appropriate combinations of multiple BTMs incorporated into the SVM models showed almost perfect agreement with the DXA.