Recurrence and metastasis are the main reasons for poor prognosis of osteosarcoma. It is of great significance to accurately predict the recurrence risk of patients and carry out appropriate monitoring and intervention for patients with high recurrence risk. In recent years, increasing researchers began to pay attention to the role of immune microenvironment in tumor development. A recent meta-analysis included more than 120 literature, systematically summarizing the effects of various immune cells such as B cells, NK cells, macrophages and all T cell subsets on the clinical outcomes of tumor patients. The results showed that cytotoxic T cells and memory T cells were beneficial for survival in different tumor types, while the prognosis effect of other immune cells such as B cell, NK cell, macrophages and some helper T cell subsets are associated with tumor type and stage [
35]. However, the comprehensive analysis of 22 TIIC types in osteosarcoma is still unsatisfactory. Zhang C et al. [
5] used ESTIMATE algorithm to obtain an immune score for each osteosarcoma case from TCGA database, and osteosarcoma cases were divided into the high or low score groups. Difference analysis showed that M0 macrophages and naive B cells were lower in high immune score group than in low immune score group, while M1 macrophages, resting dendritic cells, and M2 macrophages were higher. Then, a risks core model based on different immune related genes was established to predict the prognosis of osteosarcoma. Weifeng Hong et al. [
36] used the similar method as Zhang C to investigate the immune infiltration of osteosarcoma from Target database. The result indicated that T cells CD8, T cells CD4 memory activated, M1 macrophages and M2 macrophages were higher in high immune score group than in low immune score group, while Plasma cells, T cells CD4 naive (0.90%), T cells CD4 memory resting, M0 macrophages and Mast cells resting were lower. Deng C et al. [
37] used CIBERSORTX algorithm explored the changes of 22 immune cell types infiltration of osteosarcoma from Target database after neoadjuvant chemotherapy. The result indicated that CD8+ T cells, CD3+ T cells, PD-L1+ immune cells and Ki67 + CD8+ T cells in osteosarcoma increased after neoadjuvant chemotherapy. In our research, CIBERSORTX algorithm was used to analyze gene expression in osteosarcoma and obtain the proportion of 22 TIIC types. The results revealed that macrophages account for a large proportion of 22 immune cell types, which were consistent with Morrison C’s study [
38]. Then Kaplan-Meier curves were drawn to investigate the prognostic value of 22 TIIC types. The results indicate that naive B cells and Monocytes are associated with poor prognosis of osteosarcoma. As an important component of humoral immunity, B lymphocytes can exert a two-way role in tumor. B lymphocytes can positively regulate tumor immune process by producing anti-tumor antibodies, secreting various cytokines, and acting as antigen-presenting cells, and negatively regulate tumor immune process by inhibiting the proliferation of immune-activated T cells [
39]. In addition, studies have shown that elimination of B lymphocytes not only helps to inhibit tumor progression and recurrence, but also significantly increases the sensitivity of patients to chemotherapy [
40]. As the precursor of macrophages, monocytes can be recruited into tumor tissues and polarized into M1 macrophages or M2 macrophages in different tumor environments [
41]. Sottnik JL [
42] and Tuohy JL [
43] all revealed that monocytes are negatively associated with disease free survival of canine osteosarcoma. Cersosimo F. et al. reviewed that M2 macrophages was associated with OS metastasis and poor patient prognosis [
44], which partly support our analysis results.
At present, several immune risk score models used to quantify the immune status and indicate the prognosis of patients have been proposed in colorectal cancer and breast cancer respectively. These immune risk score models was proposed based on the counts of two lymphocyte populations of tumor centers and invasion margins [
45,
46]. The immune risk score model can be used to identify patients with recurrent risk or benefit from corresponding adjuvant chemotherapy. However, there have no studies involve the construction of immune risk score model in osteosarcoma. In our study, we constructed an immune risk score model based on eight immune cell types selected by forward stepwise approach from 22 TIIC types. Different from the traditional immunohistochemical method, the immune cells we used to construct the immune risk score model was screened by CIBERSORTX algorithm. It has more advantages than traditional methods, such as simple operation, accurate results and more immune cell markers. As expecting, our immune risk score model showed a well prognostic value. The patients in the high risk group had a poor prognosis may relate to stronger cell proliferation and repair ability. In conclusion, the immune risk score model has a good clinical application value and worth popularizing. In addition, considering the intuition of the immune risk score model in clinical application, we constructed a nomogram model based on the eight selected immune cell types to present the prediction results simply and clearly. Nomogram is a model that quantifies the probability of an event individually and accurately by integrating multiple predictors based on multi-factor analysis. So far, there is no perfect nomogram model, and the nomogram model in this study can only serve as a reference for clinicians.
Some limitations of this research should be discussed. Firstly, without complete clinical information, in addition to survival and follow-up data, other important clinical information such as age, gender, staging, metastasis and chemotherapy are not available, so that we cannot explore the independence of the prognostic value of the immune risk score model. Besides, it is also impossible to compare the immune risk score model with traditional GTM stage. Secondly, the heterogeneity of the location of immune infiltration was not taken into account in this research. Thirdly, samples were downloaded from different datasets to enlarge the sample size, which may influent the repeatability of outcome. Finally, limited prognostic osteosarcoma data can be searched in the GEO database, so there may be some bias in the results.