789 OSCC patients underwent primary tumor ablation and free flap transplant surgery at Sun Yat-Sen Memorial Hospital between July 2017 and December 2019. Of these, 469 patients were eligible for the study and 320 were excluded based on the following criteria: 1) patients with recurrent OSCC; 2) patients who previously or concurrently diagnosed with other malignant cancers; 3) patients who diagnosed with distant metastasis; 4) patients who previously underwent anticancer therapies (radiotherapy, chemotherapy, or immunotherapy); 5) patients with postoperative infection or with diseases or a history of medication that might have an impact on the complete cell counts (systemic autoimmune disease, hematological disease, long-term steroidal treatment, etc.); 6) patients with incomplete data. A random split-sample method was used to divide 469 patients into primary and validation cohorts. The primary cohort (n=328) was used to extinguish the inflammatory biomarker with the most potent predictive value and to develop a predictive model. We verified the results that obtained from primary cohort in validation cohort. The study was approved by Sun Yat-Sen Memorial Hospital’s Medical Ethics Committee.

The following demographic data were collected: age, sex, BMI, American Society of Anesthesiology (ASA) status and comorbidities. Clinicopathological data, including TNM stage (AJCC, eighth), tumor differentiation, preoperative complete blood cell counts (CBCs) within three days before surgery and CBCs on discharge, were also collected for all patients. Length of stay (LOS) after surgery and operative variables that may have an impact on the prognosis, including duration of surgery, flap type, blood loss and intraoperative blood transfusion, were also recorded.

The counts of lymphocytes, monocytes, neutrophils, and platelets in blood samples 1-3 days before surgery and on the day of discharge were collected. The calculation of systemic inflammatory markers were performed as follows. LMR was calculated as lymphocyte count/monocyte count, NLR neutrophil count/lymphocyte count, and PLR platelet count/lymphocyte count. SIRI was calculated as neutrophil count x monocyte count/lymphocyte count (10⁹/L). The appropriate truncation points of inflammatory indexes were calculated by X-Tile software (Yale University, New Haven). The dynamic change in LMR was obtained as [(LMR on discharge - LMR before surgery)/LMR before surgery]. Systemic inflammatory markers within three days before surgery were used to reflect the patient’s preoperative inflammation state, and systemic inflammatory markers on discharge were used to reflect the patient’s postoperative inflammation state. According to the definition of these inflammatory markers, LMR is proportional to lymphocyte count and inversely proportional to monocyte count, while the other three markers are opposite to it. Therefore, LMR and the other three markers would exhibit reverse trend after the surgery.

The follow-up was carried out by regular vist to oral and maxillofacial clinic and telephone calls. In the first year after discharge, we followed up the patients every 3 months and every 6 months afterwards. The follow-up was conducted until the patient’s death or the end of the study, dated May 31, 2021. Overall survival, the endpoint of interest, was calculated as the interval from operation to death or last follow-up.

The method to establish and validate nomogram was the same as in our previous study (15). Briefly, we constructed a nomogram including the independent variables of OS which was screened by backward stepwise regression analysis. Calibration plots were developed to assess the survival discriminative capability of the nomogram for OSCC patients. For internal validation, the nomogram was fitted repeatedly in 1000 bootstrap samples to calculate the concordance index (C-index).

Quantitative variables following normal distribution were compard by the Student’s t-test otherwise by the Mann-Whitney U test, whereas Categorical variables were compared using χ2 or Fisher’s exact test. The OS curves were plotted graphically using the Kaplan-Meier survival analysis, and the results of subgroups that were divided by the appropriate truncation values of inflammatory parameters were compared. Univariable and multivariable analyses were applied to spot the independent risk parameters for overall survival. Then, a nomogram was generated in the primary cohort based on variables statistically significant in backward stepwise cox regression analysis. The nomogram was developed and validated by rms package in R. The increment in survival estimation capacity of the constructed nomogram compared with TNM system was quantified by AUC and concordance index.

A total of 469 OSCC patients were included in the present research. The demographic data were seen in Table 1. Patients were assigned into primary cohort (n=328) and validation cohort (n=141) at random. In primary cohort, there were 210 males (64%) and 118 females (36%). The pathological grades of most patients were moderately differentiated (47.9%). On the basis of 8^th edition AJCC TNM staging system, predominant patients were in stage III (50%). Half of the tumor were located in oral tongue (51.9%). The validation cohort comprised 93 males (66%) and 48 females (34%). There were 50 well-differentiated cases (35.5%), 75 moderately differentiated cases (53.2%) and 16 poorly differentiated cases (11.3%). Nearly half patients (46.1%) were in stage III. 59.6% of cases occured in the oral tongue. The median LOS in the primary and the validation cohorts was 12 (9, 15) days and 12 (10, 14) days, respectively. No significant difference in the clinicopathological characteristics was found between these two cohorts.

The median OS was 26 (range, 18 to 36 months). 1-year and 3-year OS rates were 85% and 54%, respectively. For preoperative LMR, NLR, PLR and SIRI, the appropriate truncation values were 3.76, 2.37, 164.90 and 1.20, respectively. For postoperative LMR, NLR, PLR and SIRI, the appropriate truncation values were 1.71, 6.58, 195.00 and 4.05, respectively. Consequently, these two cohorts were separated into two sub-groups by the truncation values.

Kaplan-Meier curve for overall survival demonstrated a survival benefit for lower preoperative NLR (P=0.022, Figure 1A). No significant difference of OS rate was observed between higher and lower preoperative PLR groups (P=0.101, Figure 1B). A survival benefit was observed in higher preoperative LMR group (P<0.001, Figure 1C) and lower SIRI group (P=0.005, Figure 1D). Specifically, in the primary cohort, OSCC patients who in the subgroup of preoperative LMR > 3.76 enjoyed a longer survival period than those in the subgroup of preoperative LMR < 3.76 (30 months vs 21 months, P < 0.001) (Figure 1C). Similarly, OSCC patients who had a postoperative LMR > 1.71 underwent significantly longer survival time than those in low postoperative LMR subgroup (30 months vs 23 months, P < 0.001) (Figure S1C). The same tendency was found in the comparison of OS between two subgroups of NLR and SIRI (P < 0.05), but not seen in comparison of OS between the subgroups of PLR (P>0.05).

Furthermore, the predictability of these inflammatory indexes was compared by AUC. The AUC for preoperative LMR, NLR, PLR and SIRI was 0.589, 0.557, 0.534 and 0.538, respectively, for 1-year survival (Figure 1E) and 0.659, 0.574, 0.582 and 0.622, respectively, for 3-year survival (Figure 1F). Compared with that of preoperative NLR, PLR and SIRI, the AUC of preoperative LMR for 1- and 3-year OS of OSCC patients was larger (P<0.001). For postoperative LMR, NLR, PLR and SIRI, the AUC was 0.593, 0.544, 0.533 and 0.558, respectively, for 1-year survival (Figure S1E) and 0.668, 0.586, 0.520 and 0.629, respectively, for 3-year survival (Figure S1F). The AUC of the postoperative LMR was also drastically larger than that of the other postoperative indexes (P<0.001). Taken together, these data suggested that LMR possesses a superior prognostic significance for OSCC patients than the other three indexes.

We then assessed whether LMR had any potential linkage with clinicopathological characteristics. In primary cohort, the preoperative LMR was significantly associated with TNM stage (P<0.05). However, postoperative LMR had no statistical relationship with TNM staging (P>0.05, Table S1).

Subsequently, the 1- and 3-year OS in accordance with the truncation values of preoperative LMR and dynamic change of LMR was analyzed in Table 2. The result showed that the patients in subgroup of low preoperative LMR bear significantly lower 1- and 3-year survival rates than those in subgroup of high preoperative LMR (79.3% and 32.8% vs 88.9% and 66.7%, P<0.05). Likewise, OS in high and low dynamic LMR change subgroup was also statistically different. The 1- and 3-year OS were 74.4% and 37.0% in the low dynamic LMR change subgroup, 86.7% and 58.5% in high dynamic LMR change subgroup.

Univariate and multivariate analysis screened out potentially independent predictive parameters. All statistically significant variables found in the univariate analysis were incorporated as covariates in multivariate analyses. In primary cohort, the univariate analysis showed that age, hypertension, stroke, histological grade, TNM stage, preoperative albumin, preoperative LMR, postoperative LMR, dynamic LMR change and intraoperative RBC transfusion were significant prognostic factors. In multivariate analyses, dynamic LMR change remained an independent predictive variable of OS (OR: 2.492, 95% CI: 1.246–4.981, p = 0.010, Table S2). Hypertension, histological grade, TNM stage, preoperative LMR and intraoperative blood transfusion were also identified to be independent predictive variable of OS.

Backward stepwise cox regression identified hypertension, histological grade, TNM stage, preoperative albumin, preoperative LMR and dynamic LMR change as independent predictors for survival estimation. These identified parameters were incorporated to construct a predictive nomogram (Figure 2). The corresponding points for each parameters could be got by drawing a vertical line from the parameter axis to the points axis. After summing up all points, we can obtained the 1-year or 3-year survival probability by locating the position in the bottom axis which was right below the position of the total points in points axis.

Subsequently, survival estimation capacity of developed nomogram was compared with that of TNM system. In primary cohort, the concordance index indicated that the developed nomogram holds a better prognostic ability for overall survival than TNM system (0.776, 95% CI: 0.729-0.823 vs 0.651 95% CI: 0.602-0.700, Table 3). With regard to ROC curve, nomogram presented larger area under ROC curve than TNM system for predicting 1- and 3-year overall survival (0.820, 95% CI: 0.765-0.875 vs 0.702, 95% CI: 0.645-0.760 (1-year overall survival); 0.801, 95% CI 0.728-0.874 vs 0.607, 95% CI 0.528-0.686 (3-year overall survival), Figure 3). In validation cohort, the concordance index of developed nomogram was 0.800 (95% CI: 0.733–0.867), which was dramatically higher than that of TNM system (0.707, 95% CI: 0.640–0.774, Table 3). Likewise, AUC for developed nomogram was also bigger than that of TNM system (0.833, 95% CI: 0.752-0.914 vs 0.731, 95% CI: 0.645-0.817(1-year OS); 0.805, 95% CI: 0.702-0.908 vs 0.719, 95% CI: 0.613-0.825 (3-year OS), Figure 3). These results indicated that the nomogram integrating the clinicopathological variables and inflammatory biomarkers was a superior method for the OS prediction of OSCC patients. Lastly, the calibration curve of nomogram was presented graphically to determine the matching degree of actual observation rates and nomogram-predicted rates. As shown in Figure 4, an excellent match was observed, indicating the reliable predicting ability of the established nomogram for overall survival of OSCC patients.