Several studies have shown that percentage of tumor necrosis on histology is strongly correlated with good prognosis.21 Most groups now define a good histologic response as less than 10% viable tumor cells at time of surgery, and a poor response as more than 10%.23 Results of the Pediatric Oncology Group (POG) protocol for localized osteosarcoma (POG 9351), or Children’s Cancer Group (CCG) 7921, found 45% of patients had favorable responses (>90% necrosis) after preoperative chemotherapy.24 However, several clinicians have recently questioned this finding.
Overall, the prognosis for classic osteosarcoma of the extremity remains highly variable, and there has been little improvement over the past 20 years. The prognosis for younger patients, patients with spinal disease, and patients with metastatic disease remains poor. Although some prognostic factors have been identified and shown to predict a good outcome, it seems few patients have these positive factors. In this article, we describe the literature review and meta-analysis we performed to better define recent survival trends for patients with primary osteosarcoma.
Methods
The MEDLINE, PubMed, and Cochrane databases were searched for eligible studies published in English between 2000 and 2011—a decade of recently reported research. We applied the search strategy [“osteosarcoma” OR “osteogenic sarcoma”] AND [“prognosis” OR “treatment” OR “survival”] and selected reports that specifically addressed factors predicting survival in patients with osteosarcoma—reports that were limited to primary osteosarcoma of the pelvis or extremity and provided 5-year overall survival (OS) data. Abstracts of the selected articles were independently reviewed, and the inclusion and exclusion criteria were applied. We excluded basic science studies and those without pediatric patients, those without primary osteosarcoma, those with periosteal or parosteal osteosarcoma, and those that did not report 5-year OS data.
Statistical Analysis
Number or proportion of patients (whichever was reported) with 5-year OS and number or proportion of patients with 90% necrosis were extracted from each study. For each trial, proportion of patients with 5-year OS and 95% confidence intervals (CIs) and proportion of patients achieving 90% necrosis and 95% CIs were determined. We also calculated proportion of patients with 5-year OS and proportion of patients with 90% necrosis with corresponding 95% CIs of studies that included patients with nonmetastatic disease.
We assessed statistical heterogeneity among trials included in the meta-analysis using the Cochran Q test. Inconsistency was quantified with the I2 statistic, which estimates percentage of total across-studies variation caused by heterogeneity rather than chance.25 We considered I2 higher than 50% as indicating substantial heterogeneity. When substantial heterogeneity was not found, the pooled estimate calculated on the basis of the fixed-effects model was reported using the inverse variance method. When substantial heterogeneity was found, the pooled estimate calculated on the basis of a random-effects model was reported using the DerSimonian and Laird26 method, which takes both within- and between-study variations into account.
Publication bias was assessed through funnel plots and with Begg and Egger tests.27,28 Two-tailed P < .05 was considered statistically significant. All statistical analyses were performed with Stata/SE Version 11.0 (StataCorp).
Results
Our literature search yielded 597 articles. We cross-referenced these articles with the MEDLINE, PubMed, and Cochrane search results using the same keywords and discarded the duplicates. The abstracts of these articles were then reviewed in detail. The 40 articles4,6,11,12,14,15,17-19,21,29-58 that met our study inclusion criteria reported on studies that included patients with metastatic and nonmetastatic osteosarcoma. Because of the significant difference in OS of patients with metastatic disease, we also analyzed articles that included only patients with nonmetastatic disease. Sixteen articles6,14,15,29,32-35,39,47,48,51,53,54,55,57 were included in the analysis of patients with nonmetastatic disease.
Figure 1 shows 5-year OS for each of the 40 studies. For studies that compared survival of different groups of patients, the survival of each group is shown separately. For example, Bacci and colleagues39 divided patients into adolescent and preadolescent groups and reported 5-year OS for each. In our analysis, we treated each group independently and reported their 5-year OS separately. For each study, 5-year OS, weight of study, and CI are included. Five-year OS ranged from 19% to 94%. Analysis was performed to determine 5-year OS for all studies based on weight given to each study. The random-effects model used for this analysis (heterogeneity test, Q = 656.23; P < .001; I2 = 93.4%) showed 5-year OS of 63% (95% CI, 60%-66%) for studies that included patients with metastatic and nonmetastatic osteosarcoma.
Figure 2 shows 5-year OS (range, 53%-94%) for each of the 16 studies that included only patients with nonmetastatic disease. The random-effects model used for this analysis (heterogeneity test, Q = 142.08; P < .001; I2 = 89.4%) showed 5-year OS of 71% (95% CI, 67%-76%) for studies that included only patients with nonmetastatic disease.
