Renal cell carcinoma (RCC) roughly comprises 90% of kidney cancer and the incidence of RCC is currently increasing (1). According to a recent report, there were nearly 400,000 new cases and 170,000 kidney cancer-related deaths worldwide in 2018 (2). Clear-cell RCC is the main type of RCC; other subtypes include papillary RCC and chromophobe RCC (3).

Although, with the gradual improvement in imaging detection, the clinical diagnosis and treatment rate of RCC are increasing annually, metastatic RCC is associated with poor prognosis (4, 5). Numerous previous studies have shown that in RCC, lymph node (LN) metastasis is strongly associated with distant metastasis. Clinically, accurate LN status assessment can improve the diagnosis and treatment of RCC (6, 7). Babaian et al. (8) noted that LN dissection (LND) can improve the 5-year survival rate of patients with LN involvement, suggesting that patients with renal cancer undergoing LND may have better survival. LN infiltration is one of the most important predictors of tumor progression and mortality (9). Therefore, it is imperative to investigate the strong predictors of LN metastasis and accurately identify at-risk patients who require LND.

Unclear imaging findings and low intraoperative positive biopsy rates have delayed the diagnosis and identification of early LN metastasis in RCC, thus limiting the therapeutic effect of LND (10). Although computed tomography (CT) and magnetic resonance imaging (MRI) can detect LNs ~1 cm in diameter, enlarged LNs are not necessarily a sign of metastasis. The specificity of using only a single imaging method to diagnose LN metastasis is poor. Published literature suggests that use of a combination of MRI and F-fluoro-2-deoxyglucose positron emission tomography (FDG-PET) may provide the highest accuracy, however, the capacity to perform such imaging is limited by the high cost and the limited availability of the required equipment (6). Therefore, a reliable and accurate predictive tool for the screening of high-risk populations and risk evaluation of LN metastasis in RCC is urgently required. To date, few studies have assessed the predictors of LN metastasis in patients with RCC.

The Surveillance, Epidemiology, and End Results (SEER) database was developed by the US National Cancer Institute and is available for open access analysis. We aimed to build a reliable and accurate tool based on machine learning (ML) algorithms using an extensive number of patients with RCC from the SEER database and the Southwest Hospital in Chongqing, China, for use in screening patients at high risk of developing LN metastasis.

LN metastasis is the main metastatic pathway of RCC. Moreover, several studies have confirmed that patients with RCC with LN metastasis are more likely to develop distant metastasis (11). Once the lesion spreads to a distance, the 5-year survival rate decreases to ~10% (12). Surgical resection is still the primary treatment for RCC owing to its insensitivity to chemotherapy and radiotherapy. Recently, the value of LND in patients with RCC has been a topic of academic debate. Although some studies have not shown a benefit of LND, most studies suggest that LND promotes accurate tumor staging, improves the prognosis, and prolongs OS in patients with RCC (13–15). The 2018 European Association of Urology guidelines for RCC, state that patients with insidious involvement of lymph nodes (cN0/pN1 cM0) who undergo LND have improved prognosis and survival (16, 17). A clinical study revealed that some patients with RCC who underwent isolated LN resection, had good long-term survival (18). However, these recommendations are not supported by strong evidence, and previous studies have not identified which patients derive the greatest benefit from LND owing to the uncertainty of LN metastasis in renal cancer, which often make clinicians face difficulties in selecting the surgical method and scope.

Clinically, the diagnosis of LN metastases using CT or MRI is often difficult. Imaging examination has high specificity but relatively low sensitivity in the identification of LN metastasis in patients with RCC (19). The diameter of a normal LN is usually 1 cm as the upper limit of clinical imaging, but there are often undetected micrometastases (20). Therefore, a convenient tool to accurately predict LN metastasis is urgently required to guide LND. The resection of positive LNs is thought to be a direct therapy method that effectively blocks future LN metastasis.

Currently, precision medicine is summarized into four concepts: predictive, personalized, preventive, and participatory. Integrating big data with ML algorithms is becoming a clinical necessity. Recent large-scale studies have selected this novel and efficient means to investigate the clinical issues related to the early diagnosis and treatment of cancer metastasis; however, few studies have attempted to evaluate the risk of LN metastasis in patients with confirmed RCC using ML methods (21, 22). For example, Pin et al. (23) reported that patient age at surgery, largest tumor diameter, presence of preoperative symptoms, cT stage, cN stage, and serum biomarkers were associated with LN metastases in patients with RCC. Similarly, Wenle et al. (24) showed that a nomogram-based prediction method based on quantifying the risk of LN invasion in patients with RCC may have practical clinical application. Compared with traditional statistical methods, ML can construct an optimal mathematical model and constantly adjust the model parameters to effectively prevent overfitting. Therefore, in this study, six ML algorithms were developed and validated to predict LN metastasis in patients with RCC. Among them, the XGB model showed outstanding performance in both the internal and external validations, with an AUC of 0.930 and 0.958 in the internal and external tests, respectively. The DCA curve and CUC also showed good applicability.

Our analysis confirmed that the M stage, tumor size, T stage, and tumor grade were the four most important determinants in forecasting the results, which is consistent with the results of previous studies (10, 25, 26). Umberto et al. (10). found that the M stage at diagnosis and tumor size were the most important independent predictors of LN metastases. In addition, patients with RCC with a tumor diameter > 7 cm (cT2a or higher) have been shown to have a significantly increased risk of LN progression (25). Several researchers have found that LN metastases in patients with RCC are dependent on many factors, but especially on tumor size and tumor grade. Pantuck et al. (26) found that only 6% of tumors with Fuhrman grades 1–2 had LN involvement, and 26% of tumors with high Fuhrman grade had LN involvement. In patients with RCC, LN metastasis is more common in men than in women. Smoking and alcohol consumption are well-established risk factors in RCC (27) and are associated with a relatively high incidence of LN metastasis in men with RCC. A novel finding of this study is that patients with renal cancer whose tumor site was on the left side had a higher risk of LN metastases. A retrospective study using two different national databases showed that patients with left-sided RCC were more likely to have a higher tumor grade, LN positivity, and distant metastasis than those with right-sided disease (28). The spread patterns of the lymphatic system may have contributed to these differences. The effect of the side of the tumor on prognosis warrants further in-depth study.

This study has several strengths. Based on a large sample size and internal and external validations, our study ensured credibility and authenticity. In addition, the survival outcomes of the N1 group were worse than those of the N0 group, which is consistent with the results of previous studies. Several studies have shown that 5-year survival rates are often <40% for patients with LN involvement (18, 29). This further indirectly verified the reliability of the prediction results of this study. Nevertheless, this study also has some limitations. First, the use of retrospective data may have led to data bias. Second, this study did not include specific biochemical indicators. Although this avoids the effects of differences in levels of testing at different institutions, some specific biochemical parameters that were omitted warrant further investigation. In addition, owing to the inevitable differences in the level of diagnosis and treatment in different countries or regions, the external validation cohort only included Chinese patients, and so the generalizability to other countries is unclear. However, our study represents an important step forward in developing a model to predict the risk of LN metastasis in patients with confirmed RCC.

In conclusion, our study showed that integrating ML algorithms and clinical data can effectively predict LN metastasis in patients with RCC. Subsequently, a freely available online calculator (see text footnote 1) based on the XGB model was built to quantify the risk of LN metastasis in patients with RCC conveniently and accurately.

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

The studies involving human participants were reviewed and approved by Ethics Committee of the First Affiliated Hospital of PLA Army Medical University. The patients/participants provided their written informed consent to participate in this study.

YZ and XY: conceptualization. KP: methodology and project administration. XY: software. YZ: validation, writing—original draft preparation, and writing—review and editing. PX: formal analysis. ZT: investigation. NY: resources. LZ: data curation. QD: visualization. HL: supervision. All authors contributed to the article and approved the submitted version.

The work was funded by the National Natural Science Foundation of China (authorization number: 81873606).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.