We constructed the predictive model of the integrated and developed LMR-SSIGN-MAPS score in patients with localized ccRCC. A higher LMR-SSIGN-MAPS score was significantly associated with poor DFS. The finding identified that the LMR-SSIGN-MAPS which consisted of preoperative biomarkers of inflammation, PRAT image-based scoring system, and pathology features demonstrated its strengths of simplicity and high predictive power.
Monocytes can differentiate into macrophages and dendritic cells. Furthermore, tumor-associated macrophages (TAMs) are the most dominant subpopulation of myeloid cells[19], which have been correlated with a poor prognosis[20]. Myeloid-derived suppressor cells (MDSCs) with tumor-promoting properties such as elimination of adapted antitumor immune responses and promotion of metastasis are a population of heterogeneous bone marrow cells[21].The levels of monocytes can, at a certain level, represent the tumor burden of patients with cancer[5].T cells are the predominant immune cell populations in the tumor microenvironment (TME) on ccRCC, with an average of 51% in the samples[22].Comparatively low levels of lymphocytes may weaken tumor automonitoring and defense, leading to decreased antitumor efficacy[23].Multiple studies have confirmed that LMR is a superior prognostic predictor for some tumors[5].Therefore, patients with RCC with high LMR tend to experience advantageous urological outcomes. Our study proved that LMR was a significant independent risk variable (P = 0.009), and low level of LMR (< 2.80) is associated with decreased DFS compared with high LMR (> 2.80) in patients with localized ccRCC (P = 0.008).
In this study, although the Fuhrman grade and size were obtained by the LASSO Cox regression screening from 44 variables, (i) the SSIGN was constructed on the basis of clinic stage, size, grade, and necrosis which includes Fuhrman grade and size11; (ii) the regression coefficient of SSIGN was significantly higher than both Fuhrman grade and size; (iii) the SSIGN was reconfirmed as a significant risk variable by multivariate Cox regression analysis (P < 0.001); and (iv) with 20 years of follow-up, the SSIGN continued to be a helpful prognostic tool, and the model kept its high predictive power for contemporary RN and PN patients[24].Therefore, the SSIGN score was chosen for constructing a prognostic model, and high SSIGN scores were associated with decreased DFS in patients with localized ccRCC (P < 0.001). However, SSIGN only included oncological features, and in our prediction model, SSIGN alone had a Harrell's C-index of 0.782, whereas in combination with LMR and MAPS, Harrell's C-index improved significantly to 0.854 in the training cohort.
Cancer cells can reprogram the metabolism of neighboring noncancerous cells to provide extra energy substrates and metabolites for accelerated neoplasm growth[25].Wei et al. reported that ccRCC cells produce a parathyroid hormone-related protein that facilitates PRAT browning through a protein kinase A activation, leading to the release of excessive lactate to promote ccRCC growth, infiltration, and metastasis[8]. Adipose tissue secretes various adipokines, in particular vascular endothelial growth factor and interleukin 6, which are responsible for the induction of angiogenesis and inflammation and correlated with a more invasive pathological profile and poor prognosis of RCC[26].The MAPS was a score system calculated by measuring the thickness of the posterior renal fat (representing visceral obesity possibly) and the extent of perirenal fat stranding (representing organ inflammation possibly) and correlated with a progression-free survival in localized RCC[10].Consistent with this report, our study proved that MAPS was a significant risk variable (P < 0.001), and high MAPS was correlated with a decrease in DFS compared with lower MAPS in patients with localized ccRCC (P < 0.001).
A growing body of predictive algorithms and nomograms were usually used for predicting the outcomes of RCC after surgery[27, 28].In our study,the LMR-SSIGN-MAPS model was constructed for predicting DFS in patients with localized ccRCC.Since the Harrell’s C-index is more applicable to censored data[29], the Harrell’s C-index is used to assess the model’s predictive power. In training and validation cohorts, the LMR-SSIGN-MAPS model (Harrell’s C-index, 0.854, 0.848) has higher accuracy compared with SSIGN score (Harrell’s C-index, 0.782, 0.772). Based on the LMR-SSIGN-MAPS model, the K-M survival analysis showed that higher LMR-SSIGN-MAPS scores were significantly correlated with poorer DFS in the overall cohort (P < 0.0001). Therefore, the integrated prognostic model LMR-SSIGN-MAPS was in agreement with the view that multiple markers integration can be used to provide higher accuracy and predictive efficacy[30].
However, certain limitations with this model warrant further discussion. Firstly, the enrollment of patients exclusively from a single center and the dependency of model formulation and verification on retrospective datasets predispose to a selection bias. Secondly, due to the slow update of the patients’ image storage and retrieval system, less image data were available for early patients, and fewer figures were obtained as the time lengthened. Thirdly, the routine blood tests did not contain C-reactive protein before 2013 in our center, so it was not possible to incorporate all potential correlates like C-reactive protein and the modified Glasgow prognostic score for LASSO Cox regression analysis. In addition, due to hematological indicator labs were more variable data and changed daily, the LMR-SSIGN-MAPS model (Harrell’s C-index, 0.854, 0.848) has slightly higher accuracy compared with SSIGN-MAPS model (Harrell’s C-index, 0.847, 0.839) in training and validation cohorts.Finally, since the validations were conducted internally, we were unable to rule out the possibility that the choice of variables and thresholds led to overfitting of the model, requiring an external validation in the future.