Development and validation of nomograms for predicting survival and recurrence in patients with parotid gland cancer

Abstract

Background

To develop nomograms for predicting recurrence risk and long-term survival in patients with parotid gland cancer (PGC).

Methods

A total of 301 consecutive patients with PGC who underwent surgery were enrolled and randomly divided into a training cohort (n=210) and a validation cohort (n=91). Predictive nomograms were constructed based on the independent indicators of overall survival (OS) and disease-free survival (DFS) as determined by multivariate Cox regression analysis. The discrimination and calibration of nomograms were evaluated using C-indices and calibration curves.

Results

Six independent predictors of OS were identified. Incorporating these factors, the nomogram showed good concordance statistics of 0.84 and 0.78 in predicting the 5-year OS in the training and validation cohorts. Five independent predictors of DFS were identified and integrated into the nomogram. The concordance statistics were 0.84 and 0.74 in predicting the 5-year DFS in the training and validation cohorts. The predictive performance of the nomograms outperformed the TNM model. Additionally, the patients were divided into two groups according to the nomogram score, and significant differences in OS and DFS were observed between the high risk and low risk groups. Finally, the role of postoperative treatments was evaluated based on the risk stratification; patients at high risk of disease recurrence showed an improvement in DFS after receiving postoperative treatments.

Conclusions

The nomogram showed good performance in predicting both OS and DFS in patients with PGC. It might be useful for selecting patients for postoperative treatments.

Introduction

Primary parotid gland cancer (PGC) is an uncommon group of malignant tumor accounting for 14–25% of all parotid lesions and 3–6% of all head and neck malignancies (1). However, it primarily affects the major salivary glands, approximately 80% of which involves the parotid gland (2). With its relatively rare incidence and enormous diversity in histopathology types, prospective studies on parotid malignancies have traditionally been difficult to perform on a large scale, leading to a lack of widely accepted consensus regarding the epidemiology, management, and survival outcomes of PGC (3). Given the dearth of prospective evidence regarding PGC, generalizable data on the outcomes and prognosis are largely based on the results of retrospective series. The important prognostic factors for parotid malignancies remain under debate.

Previous studies have identified several prognostic factors and developed prognostic tools that aid in clinical decision-making (4-10). However, many of these studies evaluated salivary gland cancer and did not specially focus on parotid cancer. Other previous studies used the data from public databases, such as SEER, of which many important clinical and tumor factors were missing. Moreover, some previous studies have suggested that PGC patients benefit from postoperative adjuvant treatments, but the role of adjuvant treatments remains controversial.

In this study, using a large sample size, we developed and validated two nomograms for predicting the long-term survival and disease recurrence by combining clinicopathological variables that influenced the prognosis of PGC. In addition, the value of postoperative treatments was evaluated in different risk groups based on the nomogram score.

We presented the following article in accordance with the TRIPOD Reporting Checklist (available at https://dx.doi.org/10.21037/gs-21-250).

Methods

Patient selection

Data of patients with primary PGC were retrospectively collected and analyzed. Patients with primary PGC confirmed through pathological examination, who underwent surgery as initial management with or without postoperative treatments, and with complete clinicopathological, therapeutic and follow-up information were included in the study. Patients with recurrent or metastatic disease at the time of diagnosis, parotid lymphoma or sarcoma, parotid gland metastatic cancer from other primary sites, and secondary malignancy were excluded. The study was conducted in accordance with the Declaration of Helsinki (as revised in 2013). The study was approved by the Ethics Committee of the West China Hospital of Stomatology, Sichuan University (no. WCHSIRB-D-2019-712). Written informed consent was waived because of anonymous and retrospective data analysis.

Between 2000 and 2017, a total of 301 PGC patients diagnosed at West China Hospital of Stomatology, Sichuan University were included into the entire cohort and randomly divided into the training cohort (n=210) and validation cohort (n=91) at a ratio of 7:3 (using sample function in R software) (Figure S1). The patients were followed-up every 3 months in the first 2 years, then every 4–6 months until 5 years, and annually thereafter. The median follow-up period was 61 months (range, 7–201 months).

Predictor identification

The survival endpoints included overall survival (OS) and disease-free survival (DFS). OS was calculated from the date of treatment to the date of death from any cause. DFS was calculated from the date of treatment to the date of disease relapse or death from any cause. Multivariate Cox regression analysis was conducted to identify the independent predictors of OS and DFS in the training cohort. The potential predictors with significance (P<0.05) in the univariate Cox regression analysis were included in the subsequent multivariate analysis (with a forward likelihood ratio method, P<0.05 as the entry and >0.10 as the removal criteria).

Nomogram construction and evaluation

A nomogram model was established based on the results of the multivariate analysis. The concordance (C) index was used to measure the discriminative ability of the nomogram, and a calibration plot was drawn to reflect the agreement between the observed outcomes and predicted probabilities (11). A receiver operating characteristic (ROC) curve analysis was conducted to further evaluate the predictive performance of the nomogram for OS and DFS by calculating the area under the ROC curve (AUC). Time-dependent ROC curve analysis was carried out to compare the abilities of the nomogram model and TNM model in predicting OS and DFS.

Risk stratification and postoperative treatment comparison

The sum score of each patient was computed based on the nomogram. The median score was used as the cut-off value for risk stratification. A Kaplan-Meier survival analysis with the log-rank test was performed to assess the significance of the prognostic difference between the low risk and the high risk groups. Subsequently, the clinical outcomes of the low risk and high risk patients who received different postoperative treatments were compared using the same method.

Statistical analysis

Statistical analysis was performed using SPSS (version 25.0, IBM Corporation, Armonk, NY) and R (version 3.6.1, R Foundation for Statistical Computing, Vienna, Austria). All P values in this study were two-sided, and a P value of <0.05 was considered significant.

Results

Demographic and clinicopathological characteristics

The demographic and clinicopathological characteristics of the patients in the training and validation cohorts are summarized in Table 1. In the training cohort, 28.6% of the patients suffered pain, while 19.5% suffered rapid growth of the tumor. Postoperative findings showed that 73.3% of the patients had pT1/T2 stage tumors, while 92.9% had no lymph node metastasis. Of those patients, 62.9% had low- or intermediate-grade tumors, while 37.1% had high-grade tumors. Neck dissection was performed in 8.6% of the patients, while facial nerve (FN) sacrifice was performed in 36.2% of the patients. Positive/close margin was documented in 9.5% of the patients. In terms of treatment modalities in the entire cohort, 85 patients (28.2%) underwent postoperative radiotherapy, 58 (19.3%) underwent postoperative chemoradiotherapy and 31 (10.3%) underwent postoperative chemotherapy.

Table 1. The demographic and clinicopathological characteristics of patients.

Characteristic	Training cohort (n=210)	Validation cohort (n=91)
Age, year
<60	160 (76.2)	69 (75.8)
≥60	50 (23.8)	22 (24.2)
Gender
Female	108 (51.4)	46 (50.5)
Male	102 (48.6)	45 (49.5)
Smoking history
No	181 (86.2)	77 (84.6)
Yes	29 (13.8)	14 (15.4)
Drinking history
No	186 (88.6)	82 (90.1)
Yes	24 (11.4)	9 (9.9)
Rapid growth
No	169 (80.5)	71 (78.0)
Yes	41 (19.5)	20 (22.0)
Pain
No	150 (71.4)	65 (71.4)
Yes	60 (28.6)	26 (28.6)
Facial palsy
No	193 (91.9)	81 (89.0)
Yes	17 (8.1)	10 (11.0)
Grade
Low/intermediate	132 (62.9)	61 (67.0)
High	78 (37.1)	30 (33.0)
pT stage
1/2	154 (73.3)	73 (80.2)
3/4	56 (26.7)	18 (19.8)
pN stage
Negative	195 (92.9)	87 (95.6)
Positive	15 (7.1)	4 (4.4)
Neck dissection
No	192 (91.4)	86 (94.5)
Yes	18 (8.6)	5 (5.5)
FN sacrifice
No	134 (63.8)	65 (71.4)
Yes	76 (36.2)	26 (28.6)
Surgery margin
Negative	190 (90.5)	87 (95.6)
Positive/close	20 (9.5)	4 (4.4)

Values presented as cases (%). FN, facial nerve.

Identification of predictors in the training cohort

The results of the univariate and multivariate Cox regression analyses based on the data from the training cohort are presented in Tables 2 and 3. Predictors with a P value of <0.05, as determined by univariate analysis, were included into multivariate analysis. Finally, multivariate analysis identified age, rapid growth, facial palsy, grade, pT stage, and pN stage as independent predictors of OS. Meanwhile, five indicators, including age, rapid growth, grade, FN sacrifice, and surgery margin, were significantly related to DFS.

Table 2. Univariate and multivariate analyses of OS in the training cohort.

Characteristic	Univariate analysis			Multivariate analysis
	HR (95% CI)	P value		HR (95% CI)	P value
Age (≥60 vs. <60 years)	3.264 (2.047–5.205)	<0.001		1.903 (1.175–3.082)	0.009
Gender (male vs. female)	1.230 (0.786–1.923)	0.365
Smoking (yes vs. no)	1.191 (0.628–2.260)	0.593
Drinking (yes vs. no)	1.397 (0.718–2.717)	0.325
Rapid growth (yes vs. no)	2.555 (1.565–4.172)	<0.001		1.703 (1.035–2.803)	0.036
Pain (yes vs. no)	2.293 (1.444–3.642)	<0.001		NS
Facial palsy (yes vs. no)	4.444 (2.496–7.915)	<0.001		2.472 (1.366–4.474)	0.003
Grade (high vs. low/intermediate)	4.980 (3.095–8.014)	<0.001		3.121 (1.900–5.124)	<0.001
pT stage (3/4 vs. 1/2)	3.066 (1.950–4.820)	<0.001		2.108 (1.316–3.375)	0.002
pN stage (positive vs. negative)	3.602 (1.972–6.580)	<0.001		3.269 (1.769–6.044)	<0.001
Neck dissection (yes vs. no)	1.969 (1.012–3.830)	0.046		NS
FN sacrifice (yes vs. no)	2.866 (1.820–4.513)	<0.001		NS
Surgery margin (positive/close vs. negative)	3.831 (2.199–6.673)	<0.001		NS

OS, overall survival; HR, hazard ratio; FN, facial nerve; NS, not significant.

Table 3. Univariate and multivariate analyses of DFS in the training cohort.

Characteristic	Univariate analysis			Multivariate analysis
	HR (95% CI)	P value		HR (95% CI)	P value
Age (≥60 vs. <60 years)	2.395 (1.565–3.664)	<0.001		1.809 (1.166–2.808)	0.008
Gender (male vs. female)	1.309 (0.881–1.943)	0.182
Smoking (yes vs. no)	0.932 (0.509–1.706)	0.819
Drinking (yes vs. no)	1.283 (0.700–2.349)	0.420
Rapid growth (yes vs. no)	1.945 (1.238–3.056)	0.004		1.651 (1.032–2.642)	0.037
Pain (yes vs. no)	1.717 (1.129–2.611)	0.011		NS
Facial palsy (yes vs. no)	3.563 (2.032–6.251)	<0.001		NS
Grade (high vs. low/intermediate)	3.142 (2.096–4.710)	<0.001		2.045 (1.322–3.163)	0.001
pT stage (3/4 vs. 1/2)	2.192 (1.455–3.304)	<0.001		NS
pN stage (positive vs. negative)	2.786 (1.577–4.924)	<0.001		NS
Neck dissection (yes vs. no)	1.905 (1.041–3.486)	0.037		NS
FN sacrifice (yes vs. no)	3.217 (2.155–4.801)	<0.001		2.284 (1.478–3.530)	<0.001
Surgery margin (positive/close vs. negative)	3.333 (1.960–5.666)	<0.001		1.853 (1.060–3.240)	0.031

DFS, disease-free survival; HR, hazard ratio; FN, facial nerve; NS, not significant.

Development and validation of a nomogram for predicting OS

The six independent risk factors of OS were integrated into an OS predicting nomogram (Figure 1). The C-index value of the nomogram in the training cohort was 0.797 [95% confidence interval (CI): 0.753–0.841], and the calibration curve showed favorable consistency between the predicted probability of OS and the actual OS (Figure 2A). The AUCs of the nomogram for predicting OS at 3- and 5-year were 0.83 (95% CI: 0.76–0.90) and 0.84 (95% CI: 0.78–0.90) (Figure 2B), which were higher than those of the TNM model (Figure 2C). In the validation cohort, the C-index value of the nomogram was 0.773 (95% CI: 0.679–0.867), and the calibration curve also showed good agreement between the predicted and actual 3- and 5-year OS (Figure S2A). The AUCs of the nomogram were 0.84 (95% CI: 0.67–1.00) and 0.78 (95% CI: 0.65–0.91) in predicting the 3- and 5-year OS (Figure S2B), respectively. The performance of the nomogram was also preferable to that of the TNM model (Figure S2C).

Figure 1.

Prognostic nomograms for predicting the 3- and 5-year overall survival (OS) and disease-free survival (DFS) in patients with primary parotid gland cancer. (A) Prognostic nomogram for OS. (B) Prognostic nomogram for DFS.

Figure 2.

Performance of the prognostic nomograms in the training cohort. Calibration curves of the OS nomogram (A) and the DFS nomogram (D); ROC curves and AUCs of the OS nomogram (B) and the DFS nomogram (E); time-dependent ROC curves compared the OS nomogram and the TNM model (C) as well as compared the DFS nomogram and the TNM model (F).

Development and validation of a nomogram for predicting DFS

Five variables identified as independent indicators of DFS in the training cohort were used to construct a nomogram for predicting DFS (Figure 1). The C-index value of the nomogram in the training cohort was 0.727 (95% CI: 0.677–0.776), and the calibration plot showed good agreement between the predicted probability of DFS and the actual DFS (Figure 2D). The AUCs of the nomogram score in predicting DFS at 3- and 5-year were 0.77 (95% CI: 0.70–0.84) and 0.84 (95% CI: 0.78–0.90) (Figure 2E), which were higher than those of the TNM model (Figure 2F). Similar results were observed in the validation cohort. The C-index value of the nomogram was 0.698 (95% CI: 0.627–0.769), and the calibration plot also showed good agreement between the predicted and actual 3- and 5-year DFS (Figure S2D). The AUCs of the nomogram score were 0.75 (95% CI: 0.64–0.86) and 0.74 (95% CI: 0.63–0.85) in predicting the 3- and 5-year DFS (Figure S2E). The nomogram in the validation cohort also showed a better performance compared with the TNM model (Figure S2F).

Risk stratification and postoperative treatment comparison

By applying the median value of the nomogram score calculated in the training cohort, patients from the three cohorts (training, validation, and entire cohort) were stratified into two separate groups: high risk and low risk. In each cohort, the low risk group had significantly better OS and DFS than the high risk group (all P<0.01, Figure 3).

Figure 3.

Kaplan-Meier curves for risk stratification. Kaplan-Meier curves of OS for risk stratification in the training cohort (A), the validation cohort (B) and the entire cohort (C); Kaplan-Meier curves of DFS for risk stratification in the training cohort (D), the validation cohort (E) and the entire cohort (F).

Based on risk stratification, we compared the benefits of different postoperative treatments in the entire cohort (Figure 4). Postoperative treatments did not prolong the survival of patients in the high risk and low risk groups. By contrast, postoperative chemoradiotherapy decreased the OS compared with surgery alone. In the high risk group, patients who received three postoperative treatments had higher DFS than those who did not. Similar results were found in the low risk group, except for postoperative chemoradiotherapy.

Figure 4.

Kaplan-Meier curves of postoperative treatments for risk stratification. Kaplan-Meier curves of postoperative treatments in the high risk group (A) and the low risk group (B) in terms of OS; Kaplan-Meier curves of postoperative treatments in the high risk group (C) and the low risk group (D) in terms of DFS.

Discussion

Due to its remarkable heterogeneity and rarity, the risk stratification and prediction of oncological outcomes of PGC remains an unsolved challenge. Therefore, the development of comprehensive and applicable predicted tools is warranted to enable clinicians make effective clinical decisions. In the present study, we developed and validated two nomogram models that showed greater ability in predicting OS and DFS in patients with PGC after surgery. We also compared the different postoperative treatments based on risk stratification using the nomogram scores. These visualized nomogram models enable clinicians to estimate both the risk of disease recurrence and long-term survival in patients with PGC, which is also helpful in selecting patients for postoperative therapies.

PGC has been discussed in previous studies as a type of salivary gland cancer (6,9,12-15). However, patients with PGC showed distinct clinical outcomes compared with those with non-parotid cancers (12). The present study specifically focused on PGC and included a larger sample size than most published studies except for those using the public database. The entire cohort was obtained from one of the largest stomatology hospitals in China, which possessed good generalizability and representativeness of patients with PGC. Based on the clinicopathological features of the population, the present study constructed two nomograms for predicting OS and DFS, and both showed greater discrimination and consistency in the training and validation cohorts. Furthermore, the addition of clinical signs and postoperative pathological features made the nomogram’s performance more precise and preferable. Meanwhile, the identified predictive factors can be readily obtained for PGC patients who underwent surgery, making the nomogram feasible in clinical practice.

For patients with PGC, six independent risk factors, including age, rapid growth, facial palsy, grade, pT stage, and pN stage, were identified through univariate and multivariate analyses. A nomogram predicting OS was established based on the above predictors. The C-index values calculated in the training cohort (0.797) and validation cohort (0.773) suggested that the nomogram performed superior discrimination for OS prediction. Meanwhile, the calibration curve showed good consistency between the nomogram predictions and the actual observations. It also showed better performance than the TNM model, which is currently the most commonly used model for predicting the clinical prognosis of PGC.

To date, several similar models have been used to predict individualized survival in patients with PGC. Ali et al. reported that 301 patients underwent surgery for salivary gland cancer. They identified five indicators including age, clinical T4 stage, grade, perineural invasion, and tumor size to generate a nomogram for predicting the OS (6). The nomogram had a preferable C-index value of 0.809. However, patients with submandibular and sublingual gland cancers were also included in this study. Another study conducted by Mannelli et al. used six predictors, including age, histological grade, perineural invasion, pathological lymph node status, resection margins, and distant metastasis, to create a nomogram. But this nomogram was only verified using the internal validation technique of bootstrapping (4). Consistent with previously reported nomograms, we also identified that age, grade, tumor stage, and lymph node stage were predictive factors of OS. In addition, two invasive clinical signs, rapid growth and facial palsy, were incorporated into the model. FN dysfunction can be observed in up to 25% of patients with PGC and can predict worse disease outcomes (16). Terakedis et al. showed a 17% incidence of facial palsy in their cohort and demonstrated that clinical FN palsy was a risk factor for survival (2). Rapid tumor progression has also been proven to be significantly related to worse OS in patients with salivary duct carcinoma in the univariate analysis (17).

In the current study, a nomogram model for the assessment of the risk of disease recurrence after surgery was developed based on five variables, including age, rapid growth, grade, FN sacrifice, and surgery margin. The current nomogram performed well in predicting DFS, as supported by a C statistic of 0.727 in the training cohort. Its good performance was also confirmed in the validation cohort, with a C statistic of 0.698. Compared with the TNM model, the present nomogram showed greater discriminatory ability in DFS, suggesting its reliability for evaluating recurrence in patients with PGC.

Furthermore, some previous studies have constructed similar recurrence-predicted nomograms. Vander Poorten et al. (18) first presented a prognostic model for predicting the relapse risk of patients with parotid carcinoma after surgery. In this model, age, clinical T stage, clinical N stage, pain, skin invasion, FN dysfunction, perineural growth, and positive surgical margins were identified as predictors for tumor recurrence and incorporated into the pre- and postoperative nomograms. The model’s performance was subsequently validated in a series of internal and external cohorts (5,19,20). In the present study, except for the predictors identified in other studies, we also revealed that rapid growth was significantly associated with DFS. Similar results were also reported in another study (17).

Postoperative risk prediction of clinical outcomes has the potential to select eligible candidates for adjuvant treatments. Surgery remains the cornerstone of managing PGC, which could be generally used to cure low-stage, low-grade tumors (21), but is not sufficient for patients with high-risk features (22,23). Postoperative radiotherapy improved locoregional control and survival for selected patients with poor prognostic features (24-26). Besides, systemic treatment has traditionally been used for metastatic or recurrent disease, while postoperative chemotherapy has rarely been studied (27). Postoperative chemoradiotherapy, a more intensive adjuvant therapy, did not improve the survival rates in previous studies (28,29).

This study also attempted to discuss the benefits of postoperative therapies in different risk strata. First, according to the established nomogram score, patients were divided into a high risk group and a low risk group. The low risk group exhibited significantly better OS and DFS than the high risk group in the training, validation, and entire cohorts. Survival analysis revealed that adjuvant treatments could significantly increase the DFS in patients with high risk of disease recurrence. Patients categorized in the low risk group who underwent postoperative radiotherapy or chemotherapy also had a higher DFS than those who underwent surgery alone. However, when comparing postoperative treatments with surgery alone, no OS benefits were observed, regardless of whether the patients were at high risk or at low risk. Unfortunately, surgery followed by chemoradiotherapy aggravated the oncological outcomes, which could be explained by the fact that chemoradiotherapy is mainly used to treat patients with more adverse factors. Overall, with the improvement in disease control, patients categorized in the group at high risk of recurrence could be considered ideal candidates for postoperative treatments, and postoperative radiotherapy or chemotherapy was also optional for patients at low risk of recurrence.

The present study has several limitations. First, the current nomogram model was derived based on a population from a single institution. Although the analysis of the validation cohort showed that the nomogram also performed well, the results need to be verified in the external cohort. Second, this was a retrospective study in which selection bias was unavoidable. However, bias was minimized by including a large cohort of consecutive patients. In addition, a longer follow-up time is required, especially for patients with some slow-growing pathological types.

Conclusions

In conclusion, the current nomogram models showed good performance in the prediction of disease recurrence and long-term survival in patients with PGC who underwent surgery. It could provide reliable information about the oncological outcomes of these patient populations. The application of the nomogram models in risk stratification could facilitate clinicians in selecting suitable candidates for postoperative treatments.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki (as revised in 2013). The study was approved by the Ethics Committee of the West China Hospital of Stomatology, Sichuan University (No. WCHSIRB-D-2019-712) and individual consent for this retrospective analysis was waived.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.