Identifying chemotherapy beneficiaries in nasal and paranasal sinus cancers: epidemiological trends and machine learning insights

Abstract

Background

Studies on the epidemiological characteristics, treatment strategies and prognosis of nasal and paranasal sinus cancer are still relatively limited.

Methods

This study analyzed the age-adjusted incidence rates of nasal and paranasal sinus cancer from 1975 to 2020 using SEER database data. We conducted an in-depth examination of patients diagnosed between 2004 and 2015 with SEER*Stat software. A retrospective study from Fujian Provincial Cancer Hospital (2013–2020) provided an external validation set. Multiple imputation methods in R were used to address missing data. Survival analyses were performed using Kaplan–Meier and Cox proportional hazards models. Additionally, ten advanced machine learning models were utilized and evaluated in Python to predict patient survival outcomes.

Results

This study analyzed data from 3,190 patients. The annual percent change (APC) in incidence rates per 100 000 person-years was 0.36 until 2012, subsequently decreasing to − 1.79. Among various predictive models, the gradient boosting classifier demonstrated superior performance with an area under the curve (AUC) of 0.699 and an accuracy rate of 0.708. Chemotherapy did not significantly influence overall mortality risk (HR = 0.93, 95% CI 0.82–1.05, P = 0.27). Chemotherapy showed potential benefits in specific patient subgroups.

Conclusions

This study revealed a declining trend in incidence rates beginning in 2012. The gradient boosting model demonstrated robust performance, playing a crucial role in predicting patient prognosis and the significance of chemotherapy.

Highlights

• 1. The study analyzed 3,190 patients with nasal and paranasal sinus cancer, showing a slight decrease in incidence since 2012.

• 2. Patient demographics and treatment modalities significantly influenced survival rates, with the gradient boosting classifier being the most effective predictive model, emphasizing the necessity of personalized approaches.

• 3. Patients with stage IV cancer, N1 and N2 staging, sphenoid sinus involvement, epithelial tumor diagnosis, and no prior surgery or radiotherapy showed significant benefits from chemotherapy.

Introduction

Nasal cavity and paranasal sinus cancers, a rare but dangerous malignancy, typically arise in the sinus region, including the nasal cavity and paranasal sinuses. Their main histological classifications include squamous cell carcinoma and adenocarcinoma, along with less common subtypes such as neuroendocrine tumors and mucosal melanoma [1]. The incidence of nasal cavity and paranasal sinus cancers is about 5% of all head and neck cancers, and about 1% of all malignancies [2]. Despite a global incidence of only 0.2–1.0 cases per 100,000 individuals per year, this uncommon disease poses a significant health burden [3]. The low incidence of nasal cavity and paranasal sinus cancers, coupled with a scarcity of reported studies, has led to significant controversy and uncertainty in their epidemiology, treatment methods, and prognostic evaluations. Current research indicates inconsistent survival rates for nasal cavity and paranasal sinus cancer prognosis, with overall 5-year survival rates ranging from 20 to 60% [2, 4,5,6]. This variability is likely due to the heterogeneity of tumor subtypes, potential for tumors to remain concealed, varying tolerances to treatment interventions, and complex anatomical factors, all of which complicate the development of standardized prognostic assessment models. Typically, these models focus on single tissue types, such as squamous cell carcinoma and adenocarcinoma of the sinuses, or specific anatomical sites,such as the maxillary sinus or nasal cavity [7, 8]. Moreover, several studies were confined to case reports. Recently there has been a notable lack of systematic and comprehensive clinical prediction models encompassing a wide range of tissue types and anatomical sites. The heterogeneity of these cancers, originating from diverse tissue types and anatomical sources along with limited data, has restricted the use of traditional clinical prediction models. Previous retrospective studies focusing on nasal cavity and paranasal sinus cancers have analyzed multiple prognostic factors, including clinical stage, patient age, treatment modality, and histological subtype of the tumor, emphasizing the multifactorial impact on prognostic outcomes [8,9,10,11].

This uncertainty stems from the lack of targeted randomized trials in this patient cohort, highlighting a critical gap in current understanding of treatment strategies for these cancers. Traditional survival analysis methods are not explicitly designed to quantify the proportional contribution of individual factors to outcomes nor to guarantee optimal model parameterization, instead focusing primarily on time-to-event estimations. Machine learning algorithms have become essential tools in diagnosing various tumors and predicting recurrence [12]. Consequently, it is a clear imperative to develop predictive models that comprehensively incorporate prognostic factors. The integration of machine learning, epidemiological methods, and database applications can enhance disease the understanding of diseases and model development. This study aimed to comprehensively explore the epidemiology, clinical characteristics, treatment modalities, and prognosis of nasal cavity and paranasal sinus cancers while establishing a robust predictive model to provide valuable information for optimizing assessment and treatment strategies.

Methods

Data extraction and patient selection

The age-adjusted incidence of nasal cavity and paranasal sinus cancers was thoroughly examined using data from the Surveillance, Epidemiology, and End Results (SEER) database, covering the period 1975–2020, to analyze the prevailing national trends. A specific cohort of patients, diagnosed with nasal cavity and paranasal sinus cancers between 2004 and 2015, was selected for analysis using SEER*Stat (version 8.4.0).

The SEER database collects and provides detailed information on cancer incidence, prevalence, survival, and mortality across diverse populations, making it an invaluable resource for cancer-related research. To investigate trends, risk factors, treatment outcomes, and disparities in cancer epidemiology.

The inclusion criteria were established as follows: the International Classification of Diseases for Oncology, Third Edition codes C30.0 (nasal cavity) to C31.9 (sinus attachment, NOS) were used to identify all documented instances of sinonasal carcinoma within the designated timeframe. Cases were considered eligible if they met the following requirements: they demonstrated a pathological type within the range of codes 8010–8589, had comprehensive follow-up dates, exhibited a survival period exceeding 0 days, were diagnosed with a solitary primary condition, and were confirmed to have carcinoma. The exclusion criteria were as follows: (1) individuals diagnosed posthumously or via death certificates; (2) individuals diagnosed with carcinoma in situ, benign tumors, or borderline tumors; and (3) individuals with incomplete follow-up duration.

This retrospective study was conducted at Fujian Cancer Hospital, focusing on patients diagnosed with nasal cavity and paranasal sinus cancers between 2013 and 2020. These patients constituted the external validation cohort. Figure 1 illustrates the sample selection process from the SEER database and Fujian Cancer Hospital, along with the methodology for constructing the primary dataset and external validation set.

Fig. 1

The flowchart of the study

.Demographic characteristics (age, race, gender) and clinical variables (histological grade, tumor histology, primary site, TNM stage, treatment regimen) were systematically collected. The principal metric of interest in this investigation was overall survival (OS), defined as the duration between the date of diagnosis and either the date of death or the last follow-up.

This research was approved by the Ethics Committee of Fujian Cancer Hospital, China. Due to the retrospective nature of the study and the absence of any interventions on participants, Fujian Cancer Hospital's Ethical Committee waived the need for informed consent. The study was conducted following established guidelines and regulations and was in compliance with the Declaration of Helsinki principles. As all data were anonymized, there was no requirement for individual consent for publication.

Handling missing data and sample cohort partitioning

In order to effectively tackle the problem of missing data, we implemented the multiple imputation methodology using the "mice" package in R (version 4.2.2). This approach played a crucial role in ensuring the dependability and resilience of our dataset. The complete cohort was meticulously divided into a training set and a validation set, adhering strictly to a predefined ratio of 7:3.

In-depth interpretation of age-adjusted incidence rate trends

A comprehensive analysis was conducted to investigate the age-adjusted incidence rates of nasal cavity and paranasal sinus cancers from 1975 to 2020. The SEER database was utilized to obtain incidence rates for nasal cavity and paranasal sinus cancers for trend analysis. A log-linear model was applied using Joinpoint software (version 4.9; National Cancer Institute). The age-adjusted incidence trends were analyzed using the Annual Percent Change (APC) analysis tool provided by the National Cancer Institute's surveillance research program. The APC and average annual percentage change (AAPC) were computed to indicate the direction and magnitude of the observed trend.

Comprehensive model development and evaluation

During the model selection and development phase, we carefully selected ten machine learning algorithms and extensively trained and validated them using Python (version 3.11). These ten machine learning models were specifically chosen to evaluate their effectiveness in addressing the binary classification problem of predicting survival or non-survival after a 3-year period. The chosen models encompassed a wide range of algorithms, including logistic regression, Bernoulli naive Bayes, K-nearest neighbors, support vector classifier, decision tree classifier, random forest classifier, gradient boosting classifier, XGBoost classifier, LightGBM classifier, and CatBoost classifier.

The training and validation process employed tenfold cross-validation and grid search techniques to optimize the hyperparameters of the model. To ensure a comprehensive assessment of the model's performance, multiple metrics including accuracy, precision, recall, F1 score, and area under the curve (AUC) were employed. This approach facilitated a holistic evaluation of the model's predictive capabilities, encompassing various aspects of its performance.

Survival analysis and assessment of chemotherapy effects

For statistical analysis, we used R software (version 4.2.2) and the MSTATA program. To analyze the survival data, the OS rates among various patient groups diagnosed with nasal cavity and paranasal sinus cancers were estimated using the Kaplan–Meier method. Additionally, the log-rank test was used to compare the survival curves. Variables demonstrating significant associations (p < 0.05) in univariate analysis were selected for multivariate modeling. They were then incorporated into multivariate Cox proportional hazards models to precisely determine the impact of each factor on the patients’ overall survival, including the hazard ratios (HRs) and their 95% confidence intervals (CIs). This approach allowed us to identify key variables significantly affecting prognosis. Furthermore, to address discrepancies in the significance of chemotherapy, as indicated by the feature importance in Cox regression, Kaplan–Meier curves, and machine-learning models, we performed a comprehensive 1:1 propensity score matching (PSM) study using calipers of 0.05.

Results

Clinical characteristics

Table 1 summarizes the demographic and clinical characteristics of the SEER cohort (n = 3,190) after multiple imputation. The cohort demonstrated bimodal age distribution, with 43.2% of patients aged < 45 years. Male predominance was observed (61.8%). Tumor laterality distribution showed left-sided (49.2%), right-sided (47.1%), and bilateral (3.7%) involvement. The predominant primary tumor sites were the nasal cavity, accounting for 47.8% of cases, and the maxillary sinus, accounting for 34.2% of cases. Epithelial tumors were the most frequently observed histological type, representing 65% of cases. Treatment modalities included surgery (68.9%), radiotherapy (43.2%), and chemotherapy (37.4%). A comprehensive description of the SEER database was provided prior to the completion of the missing data (Supplement 1). Supplement 1 lists the data from the organization's external validation set, which included 38 patient samples. Significant discrepancies were identified in terms of age distribution, tumor characteristics, and treatment preferences of the patients. In our institutional cohort, the majority of patients belonged to the 45–64 age group, accounting for 63.2% of the sample. Tumor distribution showed reversed site predominance: maxillary sinus (50.0%) > nasal cavity (39.5%). Squamous cell carcinoma was the predominant histological subtype, constituting 81.6% of neoplasms. A discernible inclination towards more advanced malignancy grades was observed, as grade II and III tumors accounted for 15.8% and the 23.7% of instances, respectively. Notably, treatment modalities exhibited substantial heterogeneity, with surgical interventions being employed in only 36.8% of cases. Conversely, radiotherapy was favored by 89.5% of patients, implying a proclivity towards delayed-stage detection and, consequently, limited surgical alternatives.

Table 1 Characteristics of SEER database after data imputation and cleansing

Incidence trends

Figure 2A demonstrates a sustained increase in nasal cavity and paranasal sinus cancer incidence from 1975–2012 (APC = 0.36%). This trend reversed post-2012, showing consecutive decline (APC = − 1.79%). Despite stable rates in the both < 65 and ≥ 65 age groups remaining relatively steady throughout this period, the extensive overall sample size enabled us to observe this trend in the entirety of the data, even in the absence of significant alterations within specific subgroups (Supplement 2). Notably, males showed a transient surge (1975–1978 APC =  + 16.24%), suggesting sex-specific temporal associations. In contrast, the AAPC exhibited no significant upward trend from 1975 to 2012, with a value of only −0.0258 (95% CI − 0.3505, 0.5306) (Supplement 1). Significant age associations were observed in the U.S. sinus cancer incidence (P < 0.001) during 1975–2020. The presented figure demonstrates that there were no statistically significant disparities observed among All Age Deviations, All Cohort Deviations, All Period RR, and All Cohort RR (Supplement 1). Longitudinal analysis revealed age-dependent risk patterns (Fig. 2B), while cross-sectional data showed higher disease prevalence in older populations (Fig. 2C). The lack of statistically significant differences between localized and global trends suggests that the overall incidence pattern does not correlate with age-specific trends.

Fig. 2

A The temporal trends and joinpoint analysis of sinus cancer incidence in the SEER database; B the longitudinal age curve; C the cross-sectional age curve

Survival analysis

The study documented patient survival rates as follows: the one-year survival rate was 77% (95% CI 76–79%), the 3-year survival rate was 58% (95% CI 57–60%), and the 5-year survival rate was 49% (95% CI 48–51%). With a median follow-up duration of 49 months, Kaplan–Meier survival curves were used to examine the correlation between demographic factors, clinical treatment modalities, and OS in patients. In an external validation set with a median follow-up of 47.5 months, the one-year survival rate was 70% (95% CI 47–100%), 3-year survival was 70% (95% CI 47–100%), and 5-year survival was 50% (95% CI 27–93%). Separate Kaplan–Meier analyses were carried out stratified by baseline demographic and clinical characteristics, including age, pathological grade, TNM staging, treatment strategies, primary tumor site, and histological type. Survival curves were compared across these categories using log-rank tests, revealing significant differences in survival outcomes between subgroups. The laterality of the primary site (bilateral or unilateral) and ethnicity appeared to be unrelated to prognosis. Figure 3 reveals that non-chemotherapy had better prognosis (HR = 1.54, 95% CI [1.40, 1.69], P = 1.5e-20), radiation therapy was beneficial (HR = 0.75, 95% CI [0.69, 0.83], P = 1.7e-9), and surgery improved outcomes (HR = 2.42, 95% CI [2.21, 2.66], P = 5.0e-84). The worst prognosis was observed for epithelial neoplasms and tumors located in the maxillary sinus. The OS was to be significantly influenced by various factors. These factors included TNM staging, surgery, radiotherapy, chemotherapy, age, ethnicity, pathological grade, primary site, and histological type. In order to mitigate the influence of confounding variables and ascertain the specific impact on OS, a multivariate Cox regression analysis was performed. The outcomes of this analysis confirmed that age played a pivotal role in determining OS. Furthermore, there was a strong correlation between high cancer grading, advanced T, N staging, and M1 stage with poorer survival outcomes. The impact of treatment strategies exhibited variability, with radiation therapy (HR = 0.872) and surgery (HR = 0.658) being significant factors in improving survival. However, the effect of chemotherapy (HR = 0.820) was relatively ambiguous. In terms of histology, basal cell tumors, transitional cell papillomas, and carcinomas displayed relatively better prognoses, whereas the prognosis for unspecified epithelial tumors was the most severe (HR = 1.325). Differences in primary sites also had a significant impact on prognosis, with cancers originating in the frontal sinus having the most pessimistic outlook, while those originating in the nasal cavity had a relatively favorable. These findings highlight the varying impacts of treatment interventions on improving cancer patient prognosis, suggesting the importance of personalized treatment plans (Tables 2, 3).

Fig. 3

Kaplan–Meier curve of OS by subgroup analysis: A T stage; B N stage; C M stage; D age; E grade; F histologic type; G primary site; H gender; I surgery; J radiotherapy; K laterality; L race; M chemotherapy before PSM; N chemotherapy after PSM

Table 2 Univariate Cox proportional hazard model of OS in all patients

Table 3 Multivariate Cox proportional hazard model of OS in all patients

Evaluation, validation of machine learning models

In this machine learning-based study we employed a dataset comprising 3,190 individuals diagnosed with nasal cavity and paranasal sinus cancers to predict survival outcomes over a period of 3 years. The dataset was partitioned into training and validation subsets, with a ratio of 7:3. We investigated ten distinct models, encompassing logistic regression, Bernoulli Naïve Bayes, K-nearest neighbors, support vector classifier, decision tree classifier, random forest classifier, gradient boosting classifier, XGBoost classifier, LightGBM classifier, and CatBoost classifier. In order to comprehensively evaluate model performance, various metrics including accuracy, precision, recall, F1 score, AUC, and confusion matrix were employed (Figs. 4, 5, Supplement 3).

Fig. 4

A Model performance for the 3-year survival. B Accuracy of different models

Fig. 5

The ROC curves for 10 algorithms in internal validation and a gradient boosting model for external validation: A logistic regression; B Bernoulli Naive Bayes; C K-nearest neighbors; D support vector classifier; E decision tree classifier; F random forest classifier; G gradient boosting classifier (internal); H XGBoost classifier; I LightGBM classifier; J CatBoost classifier; K gradient boosting classifier (external)

A thorough examination of the receiver operating characteristic curves and confusion matrices depicted in the graphs was conducted. In accordance with our research objectives, we focused on the predictive performance of the models regarding survival and death. The gradient boosting classifier emerged as the top performer, exhibiting an AUC of 0.699, accuracy of 0.708, precision of 0.727, recall of 0.774, and F1 score of 0.750.

Additional analysis was performed to assess the significance of individual features in the gradient boosting classifier (Supplement 1). The findings demonstrated that the variable "T stage" showed the strongest association, as indicated by its importance score of 0.3094. Conversely, the variable "Gender" showed the weakest association with a score of 0.00171. Furthermore, chemotherapy demonstrated a relatively modest association in predicting the 3-year survival of patients with nasal cavity and paranasal sinus cancers, as evidenced by its importance score of 0.0464.

Detailed clinical and prognostic data were collected from 38 patients with nasal cavity and paranasal sinus cancers at our institution to further validate our gradient boosting machine model. The utilization of this model on an independent external dataset exhibited compelling predictive capability, as evidenced by its discriminatory ability, with an AUC of 0.65, indicating a high level of model validity.

Moreover, the model demonstrated a precision of 0.67, further confirming its reliability and applicability for practical clinical decision-making (Fig. 4).

Investigating the benefits of chemotherapy in nasal cavity and paranasal sinus cancers through PSM

Previous studies showed that chemotherapy was an independent prognostic factor for nasal cavity and paranasal sinus cancer patients [13]. However, our comprehensive analysis employing multivariate Cox regression, Kaplan–Meier survival curve analysis, and machine learning feature importance indicated that chemotherapy may not have a substantial association with patient outcomes. Therefore, this study aimed to examine the association between chemotherapy and the prognosis of patients with nasal cavity and paranasal sinus cancers by comparing those who received chemotherapy with those who did not, while considering their initial characteristics. To account for baseline characteristic variations between these groups, we utilized PSM to account for the observed disparities. Following the application of PSM, no substantial disparities in baseline characteristics were observed, as indicated by the majority of standardized mean differences (SMD) being below 0.1 (Fig. 6).

Fig. 6

Propensity score matching PSM effect evaluated by a love plot

In the dataset adjusted with propensity score matching (PSM), the analysis revealed that chemotherapy showed no statistically significant association with overall mortality risk (HR = 0.93, 95% CI 0.82–1.05, P = 0.27) (Fig. 7). Following PSM analysis of populations with and without chemotherapy, we observed significant differences across various subgroups, including tumor grade, T stage, N stage, patient age, primary site, and treatment modalities. Specifically, these differences were pronounced in tumor grade (P for interaction < 0.001), T stage (P for interaction = 0.013), N stage (P for interaction < 0.001), age (P for interaction = 0.044), primary site (P for interaction = 0.052), and treatment (P for interaction = 0.016).

Fig. 7

Subgroup analysis after PSM

Delving into specific subgroups, we found that the following cohorts benefited significantly from chemotherapy: Grade IV (HR = 0.59, 95% CI (0.44–0.79), p < 0.001), N1 (HR = 0.55, 95% CI (0.34–0.90), P = 0.016), N2 (HR = 0.54, 95% CI (0.37–0.79), P = 0.002), patients with primary site at the Sphenoid Sinus (HR = 0.41, 95% CI (0.20–0.85), P = 0.017), patients with histologic type as epithelial neoplasms (HR = 0.61, 95% CI (0.43–0.88), P = 0.008), and patients who had not undergone surgery or radiotherapy (HR = 0.71, 95% CI (0.59–0.86), p < 0.001). These findings demonstrate significant therapeutic association of chemotherapy.

Discussion

The objective of our epidemiological investigation was to conduct a comprehensive analysis of a substantial dataset obtained from the SEER database, with the purpose of examining the patterns in the occurrence of nasal cavity and paranasal sinus cancers between the years 1975 and 2020. To accomplish this, we employed age-adjusted incidence rates, connected-point regression analysis, and APC modeling techniques to elucidate the intricate interplay between age, period, and cohort within the framework of social transformations. The findings of particular studies indicated that the prevalence of nasal cavity and paranasal sinus cancers exhibited a relatively consistent trend throughout the duration of the investigation, spanning from 1973 to 2004 [2, 14]. A separate investigation revealed a statistically significant decline in the overall occurrence of primary non-squamous non-small cell carcinoma from 1973 to 2009, with an APC of − 21.5 and a 95%CI ranging from − 22.0 to − 21.0 ⁴. A notable finding was the declining incidence trend since 2012 (APC = −1.79), potentially reflecting improved environmental/occupational safety measures. This trend, which has not been previously addressed in scholarly literature, may suggest a shifting disease burden; one plausible hypothesis is that improved occupational safety measures and air quality regulations may have reduced exposure to established carcinogens including formaldehyde, wood dust, and industrial particulates [15,16,17,18,19].

The APC modeling analyses revealed a rise in incidence rates with advancing age, indicating the potential accumulation of long-term exposure risks and a heightened susceptibility to disease. Upon conducting joinpoint analyses considering age and race, no significant alteration in the trend was observed, implying that these declining trends may be constrained by specific population attributes. This discovery underscores the necessity for future investigations to undertake comprehensive analyses on more stratified subgroups, in order to comprehensively comprehend the individual or collective contributions of these factors to nasal cavity and paranasal sinus cancer incidence.

This study represents one of the most extensive investigations in the field thus far, employing sophisticated machine learning techniques to examine the clinical attributes and future outlook of individuals afflicted with nasal cavity and paranasal sinus cancer [7, 11]. Through the development of a prognostic model based on a gradient augmentation approach, this study successfully achieved precise predictions of 3-year survival rates. Furthermore, the validity of this predictive model was confirmed by applying it to an external validation set. The investigation also identified the five most significant factors associated with prognosis, namely tumor size, surgical intervention, age, primary tumor location, and lymph node metastasis, that showed significant associations with prognosis. Furthermore, multifactorial regression analysis revealed significant prognostic associations, including tumor grade, distant metastasis, and histological type. Surgical intervention remains a pivotal component of contemporary nasal cavity and paranasal sinus cancer management, encompassing both traditional open surgery and more recent endoscopic techniques. In cases where surgical resection is incomplete or residual tumors remain adjunctive therapies such as chemotherapy and radiotherapy are employed as conventional treatment alternatives. Despite the potential benefits of chemotherapy in terms of increasing organ retention, improving local control, and decreasing metastasis rates, which may show indirect associations with enhanced survival, there is a lack of evidence regarding its direct association with survival.

One study found that in patients with locally advanced squamous cell carcinoma of the paranasal sinuses treated with neoadjuvant chemotherapy, the 2-year OS rate was 61.4% (89% of whom were T4 patients), along with an orbital preservation rate of 81.5%. This suggests that neoadjuvant chemotherapy achieves relatively good survival and organ preservation outcomes when treating this type of disease [13]. Other studies have also provided support for the efficacy of multimodality therapy, encompassing adjuvant radiotherapy, adjuvant radiochemotherapy, or neoadjuvant therapy, in showing substantial associations with improved patient outcomes [7, 20]. Nevertheless, it is important to note that nasal patients may experience divergent treatment outcomes. A particular study discovered that radical radiotherapy and post-surgical adjuvant radiotherapy exhibited no significant disparities when compared to radical radiotherapy alone in nasal patients. [5, 21,22,23,24,25,26,27]

The therapeutic efficacy of various chemotherapy regimens remains controversial. Some studies suggest that appropriate chemotherapy regimens can enhance the survival rates of patients. Others have emphasized the significance of these treatments in protecting vital organs. Meanwhile, there is a viewpoint that these treatments may not significantly impact the overall survival period. A study highlighted that synchronous chemotherapy is not linked to local regional recurrence or OS in squamous cell carcinoma of the nasal cavity [28]. Hanna et al. conducted a study involving 46 participants with paranasal sinus tumors undergoing induction chemotherapy, finding partial remission in 67%, progressive disease in 24%, and stable disease in 9% [26]. Another study suggested that an integrated treatment approach markedly improves the OS rate in patients with sinonasal malignancies [13]. Previous studies have taken analysis for the people of anatomic subsites of certain sinuses, yet no research has thoroughly examined the prognostic implications of the anatomical locations of nasal and paranasal sinuses in relation to chemotherapy.

In the present study, the machine learning model determined that chemotherapy exhibited a relatively low level of feature importance. Furthermore, the results from the multivariate regression analysis and K-M survival curve analysis did not demonstrate a significant enhancement in prognosis associated with chemotherapy.

Employing a sophisticated propensity score matching technique, the analysis, depicted in a comprehensive forest plot, revealed that certain subgroups—especially patients with Grade IV cancer, N1 and N2 classifications, sphenoid sinus involvement, a diagnosis of epithelial neoplasms, and those without previous surgical or radiotherapy treatments—might experience benefits from chemotherapy interventions. Our study offers a novel insight, indicating that patients with sphenoid sinus involvement might benefit from chemotherapy. This study represents the initial attempt to aggregate patients diagnosed with diverse subtypes of nasal cavity and paranasal sinus cancers, aiming to examine the correlation between their heterogeneity and the selection of chemotherapy approach.

This study presents a comprehensive examination of the occurrence and prognostic determinants of nasal cavity and paranasal sinus cancers over a period of 45 years. It represents the first extensive analysis of the various subtypes of this disease, shedding light on the intricate association between pathological heterogeneity and the selection of chemotherapy strategies. Furthermore, we devised an innovative prognostic prediction model for nasal cavity and paranasal sinus cancers by employing state-of-the-art machine learning techniques. The objective of this model is to equip physicians with a user-friendly clinical tool that can be utilized to personalize patient management and establish follow-up strategies.

However, it is important to acknowledge the limitations of our study. The relatively limited sample size is a consequence of the constraints imposed by the scope of data available in the SEER database, a common challenge encountered in studies pertaining to rare diseases. The absence of critical information regarding biomarkers, HPV and EBV viral infections, specific chemotherapy regimens, radiotherapy doses, and targeted therapies limited our model's ability to include these variables. In the SEER database, there was a lack of detailed data concerning the sequencing of chemotherapy and treatment in the neoadjuvant, concurrent, and adjuvant settings, as well as specific systemic therapy considerations for various tumor histologies. This limitation potentially affected the precision of our prognostic assessments [29, 30].

Additionally, the exclusion of patients with sinus cancer from targeted and immunotherapy trials has contributed to gaps in our understanding of these areas.

In order to enhance the generalizability and accuracy of the models, it is imperative for future studies to undergo prospective validation in a broader range of populations. Additionally, it is crucial to investigate the responsiveness of distinct subtypes of sinus cancer to chemotherapy, as well as the potential synergistic associations of chemotherapy when combined with surgical and radiotherapy treatment modalities. Building upon the insights derived from the present model, forthcoming research endeavors should delve deeper into the biological mechanisms that influence prognosis.

Currently, the understanding of nasal cavity and paranasal sinus cancer epidemiology is limited, particularly regarding the impact of occupational and environmental factors. Subsequent investigations should adopt a broader approach by incorporating potential risk factors, such as genetics and lifestyle, and examining their associations with established risk factors to attain a more comprehensive understanding of the etiology of nasal cavity and paranasal sinus cancers and formulate effective prevention strategies. Integration of these multidimensional factors could provide deeper insights for disease management across the prevention–diagnosis–treatment continuum.

Conclusions

In summary, this study on nasal cavity and paranasal sinus cancers revealed a declining trend in incidence rates beginning in 2012. The gradient boosting model demonstrated robust performance, playing a crucial role in predicting patient prognosis and the significance of chemotherapy. Chemotherapy has potential benefits in specific patient subgroups, emphasizing the necessity of personalized treatment approaches in oncology.