Pathologic fracture a poor prognostic factor in osteosarcoma: Misleading conclusions from meta-analyses?

Abstract Aim Recently published meta-analyses have concluded that pathologic fracture is a negative prognostic factor in osteosarcoma. But several confounding variables were not accounted for in the index studies, thereby compromising internal validity. Methods A multivariable survival analysis of a retrospective cohort of 131 patients with conventional, high-grade osteosarcoma of the extremity long bones treated with neoadjuvant chemotherapy and surgical resection was performed. Results There were no significant differences in clinicopathologic variables between the 21 patients who suffered pathologic fracture and the 110 patients who did not in standard bivariable statistical tests. Hazard ratios for decreased overall and disease-free survival of patients with pathologic fracture failed to reach statistical significance in univariable Cox proportional hazard regression. Furthermore, pathologic fracture did not significantly affect patient outcome (hazard ratio for overall survival, 1.15 [95% CI 0.56–2.38], P = 0.71 or disease-free survival, 1.01 [95% CI 0.53–1.91], P = 0.98) after controlling for confounding factors not accounted for in prior meta-analyses, such as tumor size, chemotherapy response, and proximal tumor location. Conclusions Pathologic fracture is not a significant prognostic factor for extremity osteosarcoma after controlling for other established prognostic factors. Although a useful statistical method, meta-analysis can generate false conclusions if important confounding factors are ignored. Analysis of individual patient data, which would require collaboration among different groups, would circumvent this limitation of meta-analysis.