A novel prognostic index for oral squamous cell carcinoma patients with surgically treated

This study aims to develop an applicable prognostic index with conventional factors for predicting outcome of patients with oral squamous cell carcinoma (OSCC). We performed a prospective study in a large cohort of 892 OSCC patients in Fujian, China. All patients were randomly divided into a discovery group and validation group. A prognostic index was developed based on β value of each significant variable obtained from the multivariate Cox regression model. The results from discovery and validation set demonstrated thatthe model-4(included clinical stage, tumor differentiation, ill-fitting denture, oral hygiene and cigarette smoking) was the optimal model. The optimal cutoff points of prognostic index (1.88 and 2.80) were determined by X-tile program which categorized all subjects into low, middle and high risk subsets. Patients in high risk group were at the greatest risk of death compared with those in low risk group (HR: 6.02; 95%CI: 4.33-8.38). Moreover, there was a significant tendency of the worse overall survival with the higher prognostic index (Ptrend <0.001). The discriminatory capacity of prognostic index was 0.661(95%CI: 0.621-0.701). This study developed and validated a prognostic index that is an economical and useful tool for predicting the clinical outcomes of OSCC patients in Southeast China. Future randomized trials with larger cohort are required to confirm our results.


INTRODUCTION
Oral squamous cell carcinomas (OSCC) is the predominant histological type in oral carcinomas and has a rising morbidity and mortality in many countries [1,2]. Despite many advances in the diagnosis and treatment, long-term survival of OSCC only improved slightly over the past few decades, with 5-year survival rate still around 50% [3,4]. It is therefore essential to explore prognostic factors which would be better to predict OSCC patients' prognosis.
Recently, prognostic prediction model with different independent prognostic factors has been proposed and became a promising method for reliable prediction of cancer prognosis. Tertipis et al. [13] utilised HLA class I, CD8+ TILs and clinical characteristics to predict the outcomes of tongue cancer patients. Rietbergen et al. [14] Clinical Research Paper established a prognostic model included HPV infection, comorbidity and nodal stage for oropharyngeal squamous cell carcinoma. Almangush et al. [15] combined tumor budding and depth of invasion into a predictive model for tongue cancer. However, most of previous models contain costly biomarkers and not sufficiently validated with small sample size and short-term follow up.
Therefore, we performed a prospective study with large samples size and long-term follow-up to develop and validate a practical prognostic index (PI) using simple and easily available clinical parameters to predict the outcomes of OSCC patients in Southeast of China.

RESULTS
The demographic characteristics of discovery group and validation group are listed in Table 1. The 5-year overall survival (OS) rate for all OSCC patients was 67.6% (69.6% for discovery group and 66.0% for validation group). There was no significant difference in demographic characteristics between two groups (all P > 0.05).  15), respectively). Additionally, cigarette smoking, illfitting denture and poor oral hygiene also significantly elevated risk of death.
Next, we performed a multivariate Cox regression analysis to develop different models by incorporating potentially significant or important factors obtained from Table 2. As shown in Table 3, the model-4 (including tumor stage, histologic grade, cigarette smoking, illfitting denture, oral hygiene) showed the highest Harrell's c-statistic (0.743), which had significant differences with model-1 (P < 0.001), model-2 (P = 0.036) and model-3  Abbreviations: CT, chemotherapy; RT, radiotherapy; CRT, chemoradiotherapy. a Comorbidity, the simultaneous existence of other medical conditions. To further evaluate the prognostic value of model-4, we developed a composite PI according to five significant factors in model-4 (PI = 0.73×stage II +1.25×stage III +1.52×stage IV +0.46×moderate differentiation +1.28×poor differentiation +0.58×cigarette smoking +0.56×ill-fitting denture+0.59×oral hygiene). The median of PI was 2.30 ranging from 0 to 4.53, and the higher the PI, the poorer the survival. Then, the X-tile program determined 1.88 and 2.80 as the optimal cutoff values with the minimum P value (χ 2 = 85.14, P < 0.001), which divided the cohort into low, middle and high risk subsets (Figure 1a-1b). Moreover, patients in high-risk group had the significantly worst OS than low-risk group (P < 0.001, Figure 1c; Table 4).
Furthermore, we re-examined the four model to validate the above results in an independent set. The model-4 still had the highest Harrell's c-statistic (0.738, 95%CI: 0.695-0.781; data not shown). Moreover, no significant difference of Harrell's c-statistic was observed between discovery group and validation group (P > 0.05). The overall survival rates were significantly different in the different PI groups, with P < 0.001 by log-rank test ( Figure 2).
We then estimated the clinical prediction value of PI in all OSCC patients. As shown in Table 4, patients in high risk group were at the greatest risk of death compared with those in low risk group (HR: 6.02; 95%CI: 4.33-8.38). There was also a significant linear trend in the risk (P trend < 0.001). The 5-year OS rates for three risk subsets were 84.6%, 65.2% and 40.7%, respectively. Kaplan-Meier curve analysis also showed the similar results ( Figure 3). For death prediction, the area under the receiver operating characteristic curve (AUROC) of PI was 0.661(95%CI: 0.621-0.701).

DISCUSSION
In this large cohort prospective study, we developed and validated a novel prognostic index based on a superior   prognostic model for OSCC. The model consisted of five significant factors: tumour stage, histologic grade, cigarette smoking, ill-fitting denture and oral hygiene. Moreover, there was a significant tendency of the worse OS with the higher PI value. Among the significant prognostic factors in the present study, high clinical stage and poor tumor differentiation showed strong associations with risks of mortality among OSCC patients, which are traditional prognostic predictors and have been well-documented in many previous studies [16]. Cigarette smoking was found to be an independent predictor of survival in OSCC, which is consistent with other studies [11]. The carcinogens contained in tobacco smoke may affect the prognosis of OSCC by changing the oral microenvironment and inducing recurrent inflammatory responses [17,18]. Additionally, ill-fitting dentures could give rise to chronic mucosal irritation, and subsequently cause trauma and recurrent oral inflammation, which results in release of inflammatory mediators and growth factors that may be associated with a high rate of recurrence and a poor response to radiotherapy [19,20]. The possible mechanism of poor oral hygiene on the worse outcome of OSCC may be that poor oral hygiene may induce the imbalance of oral flora and lead to postoperative inflammation easily, which may activate the specific chemokine that can regulate cell proliferation, survival and metastasis [20,21].
Although several previous studies have created prognostic prediction models with different discriminatory ability for the prognosis of oral cancer [13][14][15]22], the limitations of small sample size and short-term follow-up could not be avoided. Moreover, some of the parameters incorporated are not routine clinical measurements, which may limit the use of these models in clinical practice. In this study, we incorporated five conventional prognostic parameters to develop an easy applied prognostic prediction model, which is a credible tool to evaluate outcome as these variables are inexpensive and can be simple to calculate by physicians. The PI value with high discriminatory ability would better assess clinical prognosis and facilitate the development of individual treatment.
Nevertheless, there are several limitations in our study. First, we only included clinical features and lifestyle factors in the model, and did not consider serum markers and other novel biomarkers, which may limit the discriminatory capacity of PI within a certain range. However, taking more factors into account would increase the additional measurement cost, and make the model cannot gain widespread acceptance. Hopefully, other routine clinical measurements will also be considered to improve our model in future studies. Second, the variables were measured only at the start of the study but not at follow-up. It therefore needs further studies to get a further validation.
In conclusion, we developed and validated a prognostic index determined by a superior prognostic model for predicting the clinical outcomes of OSCC patients in Southeast China. The PI may be an economical, widely available and useful tool to plan therapeutic strategies and guide the schedule of individualized treatment. Future randomized trials with larger cohort are required to confirm our results.

Study design and population
All study subjects were consecutively recruited between December 2003 and December 2015 at The First Affiliated Hospital of Fujian Medical University (Fujian, China). As described previously [23], subjects were included if they met the following conditions: (1)

Data collection
Information on demographics and lifestyle habits were obtained by trained interviewers through face-toface interview with a structured questionnaire. Clinical characteristics (clinical stage, treatment types, tumor differentiation, histological types, comorbidity, etc.) were collected from medical records. Smokers were defined as those who had smoked at least 100 cigarettes during their lifetime. Oral hygiene status at diagnosis was evaluated by physicians through oral inspection. Ill-fitting denture refers to the denture with sharp or rough surfaces, or having overextended flanges, or lacking of stability and retention.

Prospective follow-up
Follow-up data were obtained by telephone interview and medical records of readmission. Telephone interview was conducted every six months until the patient died or the final follow-up dated June 31, 2016. 892 OSCC patients were followed up for 3,124 person years. The median follow-up time was 70.7 months for all patients. During the follow-up period, a total of 271(30.38%) patients died (262 deaths from OSCC), 113(12.67%) were lost to follow up and 508(56.95%) were still alive. Those who were still alive or who died from other causes or who were lost to follow up were considered to be censored data. The primary endpoint was OS, which was calculated from the date of diagnosis to the date of death from any cause or the last follow up.

Statistical analysis
Two-stage analyses were performed through randomly divided 892 patients into a discovery set (n = 446) and a validation set (n = 446). Univariate Cox regression model was utilized to estimate the associations between variables and OS. Four different multivariate Cox regression models were used to assess independent factors in OSCC prognosis. The superior prognostic model was evaluated and validated using Harrell's c-statistic and AIC. Then, a PI for OSCC was developed based on β value of each significant variable obtained from the superior model. The optimal cutoff points PI were identified using the X-tile program [24]. Trend test for PI was performed by entering PI subgroup as continuous variable in the regression model, and P value was obtained from Wald chi-square test. AUROC was calculated to evaluate the clinical prediction value and discriminatory capacity of PI. Survival curves were generated by the Kaplan-Meier method, and compared by the log-rank test. Statistical significance was defined when P < 0.05. All statistical analyses were conducted with R software (version 3.1.1).