Can a Novel Scoring System Improve on the Mirels Score in Predicting the Fracture Risk in Patients with Multiple Myeloma?

Stratification of the fracture risk is an important treatment component for patients with multiple myeloma, which is associated with up to an 80% risk of pathologic fracture. The Mirels score, which is commonly used to estimate the fracture risk for patients with osseous lesions, was evaluated in a cohort in which fewer than 15% of lesions were caused by multiple myeloma. The behavior of multiple myeloma lesions often differs from that of lesions caused by metastatic disease, and accurate risk stratification is critical for effective care. To our knowledge, the Mirels score has not been validated specifically for multiple myeloma. Our purpose was: (1) To develop a novel scoring system for the prediction of pathologic fracture in patients with long-bone lesions from multiple myeloma; and (2) to compare the sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and receiver operating characteristic (ROC) area under curve (AUC) between the novel scoring system and the Mirels system. Between 2003 and 2017, 763 patients at one center with the diagnosis of multiple myeloma were reviewed, of whom 174 presented with long-bone disease involvement. Of those, 5% (nine of 174) were missing data or radiographs at a minimum of 1 year and had not reached an endpoint (fracture or surgery) before that time and were therefore excluded. Many patients have more than one lesion; consequently, we used the largest lesion in each patient, resulting in 163 lesions in as many patients. Ten percent (16 of 163) of these patients eventually developed a fracture and 4% (six of 163) underwent prophylactic stabilization (excluded from analysis because of outcome uncertainty). During the study period, prophylactic stabilization was performed at the discretion of the orthopaedic oncologist. Fifty-one percent (83 of 163) of patients were female, and the mean (± SD) age was 60 ± 10 years at radiographic lesion identification. All lesions were characterized before determining whether the patient underwent pathologic fracture. We identified variables associated with pathologic fracture on univariate analysis. Variables independently significant on logistic regression analysis were used to generate scoring algorithms at varying weights and scoring cutoffs for comparison via ROC curves. We then selected a novel score based on ROC performance, and compared the sensitivity, specificity, PPV, and NPV of that scoring system to that of Mirels score. ROC AUCs were