Prognostic value of inflammation-based scores in patients with osteosarcoma

Systemic inflammation responses have been associated with cancer development and progression. C-reactive protein (CRP), Glasgow prognostic score (GPS), neutrophil-lymphocyte ratio (NLR), platelet-lymphocyte ratio (PLR), lymphocyte-monocyte ratio (LMR), and neutrophil-platelet score (NPS) have been shown to be independent risk factors in various types of malignant tumors. This retrospective analysis of 162 osteosarcoma cases was performed to estimate their predictive value of survival in osteosarcoma. All statistical analyses were performed by SPSS statistical software. Receiver operating characteristic (ROC) analysis was generated to set optimal thresholds; area under the curve (AUC) was used to show the discriminatory abilities of inflammation-based scores; Kaplan-Meier analysis was performed to plot the survival curve; cox regression models were employed to determine the independent prognostic factors. The optimal cut-off points of NLR, PLR, and LMR were 2.57, 123.5 and 4.73, respectively. GPS and NLR had a markedly larger AUC than CRP, PLR and LMR. High levels of CRP, GPS, NLR, PLR, and low level of LMR were significantly associated with adverse prognosis (P < 0.05). Multivariate Cox regression analyses revealed that GPS, NLR, and occurrence of metastasis were top risk factors associated with death of osteosarcoma patients.

reported to be significantly associated with both overall survival and disease-free survival of patients with gastric cancer 15 . In metastatic colorectal cancer, high NLR, PLR, and low LMR were significantly linked to decreased survival time 16 . Elevated GPS, NLR, and PLR were also reported to be associated with poor survival of hepatocellular carcinoma 17 . High NPS level was significantly associated with poor survival in a variety of common cancers 18 . However, little was known about the prognostic role of the inflammation biomarkers in osteosarcoma.
In this retrospective study, we evaluated the clinical significance of pre-treatment inflammation-based scores and determined the independent prognostic factors for patients with osteosarcoma.

Results
Patient characteristics. This study consisted of 162 osteosarcoma patients with complete clinical data ( Fig. 1) including 96 male and 66 female patients. Patients with large tumor may have cancer fever without evidence of infection. Twenty-one patients with cancer fever were excluded in this study, they have already accepted NSAID or anti-cancer treatment outside of our department. No studies have been reported that cancer fever could affect the blood routine test. In this case, three patients with cancer fever were enrolled and they were indispensable among our patient population. Table 1 lists the main features of the analyzed patients. The median age was 18 year-old, and the majority of tumors (89.5%) located in extremities. A total of 93.8% patients had a KPS of ≥ 80, and 80.9% patients received neoadjuvant chemotherapy. 143 patients were diagnosed as stage I-II according to Enneking surgical staging criteria, and the rest were stage III. The number of patients suffered from pathological fracture, local recurrence and distant metastasis were 18, 31 and 78, respectively. All enrolled patients underwent surgery followed by 8-14 cycles of adjuvant chemotherapy.
As shown in Table 4, for histological subtypes, there were 148, 6, 6 and 2 patients diagnosed with conventional, telangiectatic, intramedullary and periosteal osteosarcoma, respectively. No obvious relations were observed between these histological subtypes and either GPS or NLR in other tumors, including gastric or pancreatic cancers 19,20 ; the number of non-conventional osteosarcoma was too small in our current study so that no further statistical analysis could be performed.

Discriminatory ability of inflammation-based scores.
We compared the significance of GPS, CRP, NLR, PLR and LMR to discriminate between survivors and non-survivors using ROC curves. As shown in Table 5, the AUC of GPS was 0.677 (95% CI, 0.592-0.761), with a sensitivity of 40.8% and a specificity of 94.2%. The AUC of NPS was 0.504 (95% CI, 0.415-0.594), with a sensitivity of 6.6% and a specificity of 94.2%. The AUC of CRP was 0.603 (95% CI, 0.514-0.691), with a sensitivity of 26.3% and a specificity of 94.2%. The AUC of NLR was 0.663 (95% CI, 0.578-0.748), with a sensitivity of 50.0% and a specificity of 82.6%. The AUC of PLR was 0.600 (95% CI, 0.513-0.687), with a sensitivity of 72.4% and a specificity of 47.7%. The AUC of LMR was 0.407 (95% CI, 0.320-0.495), with a sensitivity of 30.3% and a specificity of 51.2%. Notably, GPS and NLR showed higher AUC values than CRP, PLR, LMR, and NPS (Table 5 and Fig. 3).

Discussion
The accuracy of prognostic evaluation is critical for the medical care of patients with osteosarcoma. Adverse predictive factors, including detectable primary metastases, axial or proximal extremity tumor site, large tumor size, and older age, were well established 21 . In addition to these tumor characteristics, systemic inflammation also affected the prognosis of osteosarcoma. Elevated CRP was reported to be significantly associated with poor prognosis [22][23][24][25] . Pretreatment neutrophils count is an independent prognostic factor of extremity osteosarcoma. Moreover, low level of LMR was regarded to be related with adverse OS in osteosarcoma by Liu T, et al. 26 . However, the clinical prognostic value of GPS, PLR, NLR, and NPS in osteosarcoma remained poorly defined. Elevated CRP reflected increased systemic inflammatory response, which involved in tumor development and progression. GPS is a prognostic score comprised of serum CRP and albumin levels. It was first introduced by Forrest et al., who evaluated its prognostic significance in inoperable non-small-cell lung cancer 27 . After that, GPS was found to be a prognostic factor in various other tumors, such as renal clear cell cancer, gastric cancer, gallbladder cancer, and Esophageal squamous cell carcinoma 15,[28][29][30] . Patients with GPS more than zero have elevated CRP or hypoalbuminaemia or even both of them, implying GPS presents not only inflammation status but also nutritional status of cancer patients. Therefore, we speculate that GPS might be a better predictor of the prognosis of cancer than CRP. Patients with high GPS in our study accepted the same therapy with the rest patients. Our results indicated that GPS, rather than CRP, was an independently prognostic factor of osteosarcoma (Table 3). Moreover, GPS also showed a higher sensitivity of 40.8% in comparison with the sensitivity of 26.3% of CRP, while their specificities were the same (Table 5).
Cancer-related inflammation has a role in cancer development and progression 31,32 . Neutrophilia, thrombocytosis, monocytosis and lymphopenia tend to represent a nonspecific response to cancer-related inflammation and are associated with poor survival in cancers [32][33][34][35][36] . Neutrophils interact with tumor cells by producing cytokines and chemokines, which affects tumor cells' proliferation, angiogenesis, and metastases 37 . Tumor-associated macrophages, which arise from blood monocytes, promoted tumor progression and metastases 38 . Lymphocytes play a major role in the immune response by mediating the immunologic destruction of various cancers 39,40 . Platelets were also reported to act as chemoattractants, increasing the migration of ovarian cancer cells 41 . NLR, PLR, LMR, and NPS have been shown to be independent risk factors in various malignant tumors 18,[42][43][44] . These factors when combined may have stronger prognosis value than any single one. Moreover, LMR was suggested as an independent prognostic factor for OS in patients with osteosarcoma by Liu T, et al. 26 . However, among these prognostic scores, only NLR was an independent risk factor in osteosarcoma in our research, in which patients with high NLR accepted the same therapy with patients with the rest patients ( Table 3). The difference might be caused by heterogeneity in patient populations between these two studies. For instance, Liu T, et al. enrolled patients with no prior pre-operative anticancer treatment, but as high as 80.9% patients received neoadjuvant chemotherapy in our study.
Using ROC curve analysis, we have shown that optimal cut-off points of NLR, PLR, and LMR were 2.57, 123.5 and 4.73, respectively. Optimal thresholds of NLR, PLR and LMR in our study were similar to the studies in other malignants [42][43][44] . The AUC of GPS, CRP, NLR, PLR, LMR, and NPS were compared, and GPS and NLR had a markedly larger AUC than CRP, PLR, LMR, and NPS (Table 5 and Fig. 3), which was consistent with the results of multivariate analysis (Table 3).
It is worth noting that our study is the first attempt to evaluate the prognosis significance of these inflammation-based prognostic scores in patients with osteosarcoma. In addition, inflammation-based scores are simple and comprised of components of blood test with low cost. The establishment of predictive model based on inflammation is of great value for the patients in developing countries. However, there are still several potential limitations in the present study. First and most importantly, this is a retrospective, single-institution, and small-sample-size study, providing a lower level of confidence than those randomized controlled trials. Second, the sample size of 162 patients enrolled in our study was not enough to separate into two groups to derive the parameters and confirm their utility, which might cause overestimation of results. Third, the age distribution is atypical in our population, as most of osteosarcoma occurs at pediatric age. Fourth, blood samples were not obtained at same time and with no repeated test, which could introduce irreconcilable bias and negate the utility of the test, because blood parameters are dynamic. Finally, the heterogeneity in the treatments of patients might also affect the results. Our results should be interpreted with cautious according to defections above.
In conclusion, the present study indicated that abnormal pretreatment inflammation-based prognostic scores, such as GPS > 0, CRP > 10 mg/L, NLR > 2.57, PLR > 123.5, and LMR ≤ 4.73, were inversely associated with OS in osteosarcoma. Moreover, occurrence of metastasis, GPS and NLR are robust predictors of osteosarcoma survival. Patients with these two risk factors may need more aggressive chemotherapy and closely follow-up to improve clinical outcomes, according to our observations. Due to the limitations of retrospective studies, further prospective studies are still warranted.

Methods
Patients. We retrospectively reviewed the electronic medical records for 495 patients with osteosarcoma who were accepted by our department from January 2006 to December 2013. The inclusion criteria for primary studies were as follows: (i) All patients diagnosed with histologically confirmed osteosarcoma. (ii) All patients received no anti-cancer treatment before. Patients with either of the following diagnoses were excluded from the final analysis, including patients who already have neutrophilia, high CRP, or high procalcitonin, or if they have clinical evidence of infection or any other inflammatory conditions, or patients who were treated with anti-cancer therapy or non-steroid anti-inflammatory drug (NSAID) previously, or those whose clinical data were incomplete. Antibiotics were not prescribed for patients enrolled. NSAID is an efficient approach for cancer fever treatment, which is also wildly used in patients with osteosarcoma [45][46][47] . NSAID was reported to affect blood test 48,49 , then patients with record of NSAID treatment before blood test were excluded. The study was approved by the medical ethics committee of Shanghai Jiaotong University Affiliated Sixth People's Hospital.    Table 3. Univariate and multivariate Cox proportional hazard regression analyses of overall survival. GPS = Glasgow prognostic score; CRP = the C-reactive protein; NLR = neutrophil-lymphocyte ratio; PLR = platelet-lymphocyte ratio; LMR = lymphocyte-monocyte ratio; NPS = neutrophil platelet score; HR = hazard ratios; CI = confidence interval.
Scientific RepoRts | 6:39862 | DOI: 10.1038/srep39862 The GPS was constructed as following 27 : patients with both elevated CRP(> 10 mg/L) 22 and hypoalbuminaemia (< 35 g/L) were allocated a score of 2; patients with neither of the biochemical abnormalities were allocated a score of 0; and others with either of the biochemical abnormalities were allocated a score of 1. The NPS was calculated as follows 18 : patients with a neutrophil count ≤ 7.5 × 10 9 /L and platelets ≤ 400 × 10 9 /L were scored 0, patients with neutrophils > 7.5 × 10 9 /L or platelets > 400 × 10 9 /L were scored 1 and patients with both neutrophils > 7.5 × 10 9 /L and platelets > 400 × 10 9 /L were scored 2. NLR was defined as a simple ratio of the absolute neutrophil count over lymphocyte count. PLR was defined as a ratio of the platelet count over the absolute lymphocyte count. LMR was defined as a ratio of the absolute lymphocyte count over the monocyte count.   Table 5. Sensitivity, specificity, positive likelihood ratio, and negative likelihood ratio for inflammationbased scores. GPS = Glasgow prognostic score; CRP = the C-reactive protein; NLR = neutrophil-lymphocyte ratio; PLR = platelet-lymphocyte ratio; LMR = lymphocyte-monocyte ratio; NPS = neutrophil platelet score; ROC = eceiver operating characteristic analysis; AUC = the area under the curve. * GPS and NPS are categorical variables. ** The cut-off value of CRP was determined by reference rather than ROC curve. Patient follow-up. All 162 patients were followed after the completion of adjuvant chemotherapy. The intervals for follow-up were every 3 months for the first 4 years, then every 6 months until December 10, 2015. The routine follow-up examinations included physical examination, laboratory tests, chest CT and radiographs of the operated limb. Bone scans were performed every 6 months. The study was approved by the medical ethics committee of Shanghai Jiaotong University Affiliated Sixth People's Hospital.
Statistical analysis. All statistical analyses were performed by SPSS statistical software (Version 17.0, IBM Corp.). Overall survival (OS) was defined as the time from the date of diagnosis to the date of the last follow-up or death from any cause. The optimal cut-off values of NLR, PLR and LMR were determined by receiver operating characteristic (ROC) analysis, using OS as end-point. Kaplan-Meier analysis was performed to plot the survival curve. To determine the independent prognostic factors, univariate analysis and multivariate analysis, expressed as hazard ratios (HR) and 95% confidence, were performed by Cox Regression Model. Area under the curve (AUC) of ROC was calculated and compared to evaluate the discriminatory ability of the inflammation-based prognostic scores. P values were two-sided, and P < 0.05 was considered statistically significant.