Development and External Validation of Machine Learning-Based Models for Predicting Lung Metastasis in Kidney Cancer: A Large Population-Based Study

Yi, Xinglin; Zhang, Yuhan; Cai, Juan; Hu, Yu; Wen, Kai; Xie, Pan; Yin, Na; Zhou, Xiangdong; Luo, Hu

doi:https://doi.org/10.1155/2023/8001899

International Journal of Clinical Practice

On this page

Abstract Introduction Materials and Methods Results Discussion Conclusions Data Availability Conflicts of Interest Acknowledgments Supplementary Materials References Copyright Related Articles

Research Article | Open Access

Volume 2023 | Article ID 8001899 | https://doi.org/10.1155/2023/8001899

Development and External Validation of Machine Learning-Based Models for Predicting Lung Metastasis in Kidney Cancer: A Large Population-Based Study

Xinglin Yi,¹Yuhan Zhang,²Juan Cai,²Yu Hu,²Kai Wen,²Pan Xie,²Na Yin,²Xiangdong Zhou,¹and Hu Luo¹

Academic Editor: Dawei Cui

Received24 Mar 2023

Revised24 May 2023

Accepted06 Jun 2023

Published20 Jun 2023

Abstract

The accuracy of indices widely used to evaluate lung metastasis (LM) in patients with kidney cancer (KC) is insufficient. Therefore, we aimed at developing a model to estimate the risk of developing LM in KC based on a large population size and machine learning algorithms. Demographic and clinicopathologic variables of patients with KC diagnosed between 2004 and 2017 were retrospectively analyzed. We performed a univariate logistic regression analysis to identify risk factors for LM in patients with KC. Six machine learning (ML) classifiers were established and tuned using the ten-fold cross-validation method. External validation was performed using clinicopathologic information from 492 patients from the Southwest Hospital, Chongqing, China. Algorithm performance was estimated by analyzing the area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, specificity, precision, recall, F1 score, clinical decision analysis (DCA), and clinical utility curve (CUC). A total of 52,714 eligible patients diagnosed with KC were enrolled, of whom 2,618 developed LM. Variables of age, sex, race, T stage, N stage, tumor size, histology, and grade were identified as important for the prediction of LM. The extreme gradient boosting (XGB) algorithm performed better than other models in both the internal validation (AUC: 0.913, sensitivity: 0.873, specificity: 0.809, and F1 score: 0.325) and the external validation (AUC: 0.904, sensitivity: 0.750, specificity: 0.878, and F1 score: 0.364). This study established a predictive model for LM in KC patients based on ML algorithms which showed high accuracy and applicative value. A web-based predictor was built using the XGB model to help clinicians make more rational and personalized decisions.

1. Introduction

Kidney cancer (KC) originates in the kidney and accounts for approximately 2% of all malignancies worldwide [1, 2]. Approximately 350,000 people are newly diagnosed with KC, and 15,000 die from this cancer yearly [3]. According to the 2016 World Health Organization classification of urinary carcinoma, KC incorporates several subtypes, including renal cell carcinoma (90%), transitional cell carcinoma (1%), renal sarcoma (1%), and other kidney tumors [4]. Most patients have a favorable prognosis, and more than half of the patients have an overall survival (OS) of more than ten years.

Although immunotherapy and precision surgery provide patients with KC with a more favorable prognosis, approximately 20% of them have distant metastasis at the time of diagnosis. Once the cancer has spread, the 5-year OS rate severely decreases to approximately 10% [5]. The lungs are the most common site of distant metastasis, accounting for 55% of all metastatic cases of KC [6]. Previous studies demonstrated that although drug treatment is advanced, patients who develop lung metastasis (LM) only have a median survival time of 15 months [7]. Therefore, precise measures to diagnose LM will provide clinicians with more rational decisions. Conventional contrast-enhanced computed tomography (CT) is traditionally used for preoperative diagnosis. However, CT has relatively low sensitivity (62%) and specificity (86%) in predicting LM from KC, leading to the misdiagnosis of many patients with LM resulting in unnecessary surgeries that cannot cure their cancer [8]. Moreover, because the metastatic focus is usually small, many patients do not exhibit respiratory symptoms, resulting in a delayed diagnosis of LM. Although magnetic resonance imaging and biopsy offered high accuracy in detecting LM, the high financial cost and long waiting duration delayed the diagnosis of LM and thus limited its application to all patients with KC [9]. Therefore, a predictive model that conveniently and precisely detects LM in patients with KC is needed, which could help clinicians make more rational treatment decisions, adopt preventive therapy, and improve patient survival. The tumor-node-metastasis (TNM) staging system, UCLA Integrated Staging System (UISS), the tumor stage, size, grade, necrosis (SSIGN) score, and Leibovich score encompass common pathological factors and are frequently used to assess the recurrence and metastatic risk of KC in clinical studies. However, the C-index of these predictive systems was reported to range between 0.723 and 0.80 [10, 11] and was not highly satisfactory.

Machine learning (ML) has emerged as a powerful tool in various fields such as computer vision, security systems, and medicine, where it has shown significant value [12, 13]. An increasing number of studies are demonstrating its potential to improve diagnostics, prognostic predictions, and treatment planning across a range of clinical diseases. In the medicine field, ML algorithms can learn from and make predictions based on data, enabling the creation of personalized, data-driven models that can enhance clinical decision-making and patient care [14]. For instance, Liu et al. employed machine learning models based on a population of 311,408 to predict bone metastasis in patients with ductal carcinoma. They achieved an area under the receiver operating characteristic curve (AUC) of 0.888, a sensitivity of 0.801, and a specificity of 0.837. This high predictive accuracy highlights the potential for ML models to be utilized in determining appropriate treatment strategies for such patients [15]. Similarly, Cheng et al. developed a machine learning-based model using a population of 10,580 to predict the survival of patients with neuroendocrine tumors. They attained an AUC of 0.90, which was significantly greater than that of the American Joint Committee on Cancer (AJCC) seventh staging system. The success of this model demonstrates the utility of ML-based approaches for prognostication and guiding clinical decision-making in oncology [16]. In this study, we attempted to build an exact tool based on ML algorithms by employing a large population of patients with KC from the Surveillance, Epidemiology, and End Results (SEER) database and a real-world hospital.

2. Materials and Methods

2.1. Patients

The patients were extracted from the SEER database (2010–2017), which comprises approximately 30% of the total population in the USA [17]. Patients from the Southwest Hospital in China were also enrolled in the patient cohort. The inclusion criteria were patients with kidney malignancy. Patients who (1) were younger than 18 years old, (2) had unknown T or N stage, (3) had unknown LM, (4) had unknown tumor size, (5) had more than one primary tumor site, and (6) had unknown tumor grade were excluded from the cohort. This retrospective study involving human participants was conducted in accordance with the ethical standards of the institutional and national research committee. Ethical approval was waived by the local Ethics Committee of Southwest Hospital in view of the retrospective nature of the study.

Finally, 52,714 patients were enrolled in this study. We randomly assigned 52,222 patients from the SEER database to the training (70%) and internal test (30%) sets. The training set was used to establish the predictive models, while the latter was used to validate the model’s performance. Subsequently, 492 patients from Southwest Hospital were assigned to the external test cohort, which was used to externally re-validate the models. A detailed patient selection flowchart is shown in Figure 1.

2.2. Feature Selection

We retrospectively selected clinical features using SEERStat software (8.4.0.1) to screen for commonly used variables, including age, sex, T stage, N stage, laterality, tumor size, grade, histology, race, and LM. The T and N stages were determined according to the seventh edition of the AJCC TNM staging system. Histology categories included the following: (8120) transitional cell carcinoma, (8255) adenocarcinoma with mixed subtypes, (8260) papillary adenocarcinoma, (8310) clear cell adenocarcinoma, (8312) renal cell carcinoma, and (8317) chromophobe renal carcinoma and other rare subtypes.

2.3. Model Establishment and Model Performance

First, we used univariate logistic regression to identify features related to LM in the training cohort and then included variables with a value less than 0.05 in the model development process. We measured feature importance using the permutation method in machine learning models, as described in references [14, 18]. Then, utilizing the “Tidymodels” packages, we constructed six machine learning models that incorporated the selected features. These models included logistic regression (LR), extreme gradient boosting (XGB), random forest (RF), support vector machine (SVM), artificial neural network (ANN), and decision tree (DT). These models were developed using the selected variables from the procedures described above and applied to the training cohort. The hyperparameters were optimized using a ten-fold cross-validation and grid search approach, with the specific parameter settings detailed in Supplementary file 1.

Several evaluators, including AUC, accuracy, sensitivity, specificity, precision, recall, and F1 score, were used to estimate the performance of models in internal and external test cohorts. Decision curve analysis (DCA) and clinical utility curve (CUC) were performed to examine the discriminative and fitting abilities of the models. We then selected the best-performing model to build a web-based online calculator for generalization. In addition, to evaluate the contribution of each variable in prediction, we used an imputation-based method to rank the importance of the selected parameters in the training cohort. Finally, the survival analysis of OS and cancer-specific survival was performed to validate the prognostic value using the Kaplan–Meier method based on the predictive results.

2.4. Statistical Analysis

The age and tumor size variables in this study were measured in a continuous form, and the t-test was used to compare the differences between these two variables. The TNM stage was classified according to the 7th AJCC TNM classification. Other variables were displayed in the categorical form, and the chi-square test was used to compare the differences. A correlation analysis by the Spearman method was performed to describe the relevance among variables and identify highly relevant features to LM. The relevant index categorized three levels: 0–0.4, low; 0.4–0.7, intermediate; and ≥0.7, high. All statistical analyses were performed using R software (version 4.2.1; R Foundation for Statistical Computing).

3. Results

3.1. Baseline Characteristics

In total, 52,714 patients were enrolled in this study. Among them, 2,618 (4.96%) patients with KC were diagnosed with LM. A comparison of characteristics between the LM and non-LM cohorts is summarized in Table 1. Compared with non-LM individuals, individuals with LM were more likely to be elderly (61.3 vs. 59.6), male (71.4% vs. 62.5%), with larger tumor sizes (51.0 mm vs. 34.7 mm), advanced (T3, T4) T stage (71.8% vs. 20.1%) and N stage (33% vs. 2.9%), and higher (III-IV) tumor grade (79.2% vs. 28.4%).

After being randomly divided into training and internal test groups in a 7 : 3 ratio, patients in the training arm (38,335) had characteristics similar to those of the internal (15,667) and external (492) arms (Table 2).

3.2. Univariate and Multivariate Logistic Regression

Based on univariate regression analysis, variables of age, sex, T stage, N stage, tumor size, histology, race, and tumor grade were features with a (Table 3). These variables were used in building six ML algorithms. Multivariate regression analysis showed that older age, male sex, larger tumor size, Asian ethnicity, advanced T and N stage, tumor grade, and histology of renal cell carcinoma were identified as independent factors for LM.

3.3. Correlation Analysis

To recognize the variables relevant to LM and to examine the linear relationship among characteristics, we performed a correlation analysis based on the Spearman method. As shown in Figure 2, no variables exhibited a high linear relationship (index >0.8). In addition, the Spearman relevant analysis showed that characteristics of N stage, T stage, tumor size, and tumor grade were LM’s four most relevant features.

3.4. Model Performance

Receiver operating characteristic (ROC) curves of the internal and external cohorts are shown in Figure 3, indicating that the XGB algorithm exhibited the highest AUC value. Detailed information on the performance is shown in Table 4. In internal and external test cohorts, XGB outperformed the others, with AUC, accuracy, sensitivity, and specificity of 0.913, 0.812, 0.873, and 0.809, respectively, in the internal test and 0.904, 0.872, 0.750, and 0.878, respectively, in the external test. The XGB algorithm demonstrated the third highest F1 scores, following RF and SVM, in the internal test set, while it performed the best in the external test set. Overall, the XGB algorithm outperformed the others in terms of performance.

(a)

(b)

As indicated in Figure 4, DCA curves suggested that XGB had the highest clinical applicability, which meant that clinicians would make a more accurate judgment using the XGB model rather than other ML algorithms. The probability density plot showed that the predictive probability distribution in non-LM patients was extremely high; whereas it was relatively flat in LM patients (Figure 5). A CUC was used to detect the optimal threshold of each predictive cohort. As shown in Figure 5, when the value of the x-axis was >0.05, the XGB model could accurately predict patients with LM.

(a)

(b)

(c)

(a)

(b)

3.5. Feature Importance Evaluation

Based on the permutation test, we ranked variable importance for prediction in the three best-performing models (ANN, XGB, and LR). It is not difficult to observe that although the three ranks differ slightly, T stage, N stage, tumor size, and grade still ranked in the top five (Figure 6).

(a)

(b)

(c)

3.6. Calculator Online Establishment

To generalize the predictive model based on the XGB, which performed best among the six algorithms, we built a web-based online predictor, which is available at https://medicalmachinelearning.shinyapps.io/ModelForLungMetastasis/. As shown in Figure 7, as long as the accessible variables are entered into the option box, we can predict the risk of LM in KC. For example, if we select “female” for gender, “60” for age, “Asian” for race, “T1-N0” for stage, “44 millimeters” for tumor size, “8120 (transitional cell carcinoma)” for histology, and “grade I” in the calculator’s input fields, and then press the “Predict” button, the predicted outcome for developing LM will be “No.” This indicates that, based on these inputs, it is less likely for the patient to develop LM.

3.7. Survival Analysis

Based on the predictive results of the XGB model, we performed a survival analysis using the Kaplan–Meier method. As suggested in Figure 8, those who were determined to have LM had a significantly shorter survival time () than those who did not, suggesting a good discriminative ability of the XGB model. Thus, the XGB model can also help clinicians judge the prognosis of patients with KC.

(a)

(b)

4. Discussion

KC is a prevalent urinary cancer with a relatively long survival time in patients without distant metastasis. However, the prognosis was severely impaired in the case of distant metastases, and the 5-year survival rate of those patients was only 12% [19, 20]. As reported in prior studies, the lung was the most common site for distant metastasis, covering approximately 45–50% of all metastatic cases with a poor prognosis of only 7 months [6, 21–23]. Regarding treatment, due to the high resistance to chemoradiotherapy exhibited by this disease, surgical resection was still deemed the most effective treatment for curing KC. However, many patients who underwent surgical treatment were still at risk for LM. Recently, Choueiri et al. demonstrated that adjuvant pembrolizumab after surgical resection significantly prolonged the disease-free survival to 24 months, but adverse events were common (with an incidence of 21.3%) and reduced the OS owing to these adverse effects [24].

In the era of targeted therapy, numerous targeted treatments have been successfully developed, leading to improved clinical outcomes for patients with KC. Multitargeted, small-molecule tyrosine kinase inhibitor (TKI) drugs that act against vascular endothelial growth factor receptors, platelet-derived growth factor receptors, and other kinases are recommended for patients with previously untreated advanced KC. These therapies have contributed to a significant improvement in median progression-free survival (PFS) from 5.5 to 11 months, as well as an increase in median OS from 23 to 26 months [25]. In recent years, immune checkpoint inhibitors (ICIs) have emerged as a crucial therapeutic approach. A meta-analysis has revealed that combining TKI drugs with immunotherapies significantly enhances tumor responses and improves survival outcomes for patients with metastatic KC. This finding suggests a promising future for the treatment of advanced KC [26]. With a personalized therapeutic schedule, the survival of patients with KC may be improved because it could prevent unsuitable patients from adjuvant treatment’s adverse effects. Hence, early attention to those at risk for LM and taking personal preventative measures are important. To the best of our knowledge, risk factors for LM have been examined in several studies [27, 28]. However, there are only a few established predictive models. Lu et al. [20] used the SEER database, in which 10,929 patients were eligible for a nomogram construction, to predict the LM of renal cell carcinoma. Their study demonstrated that clear cell carcinoma pathology was a risk factor for LM compared to other subtypes of KC, which is consistent with our results. They also suggested that parameters such as race, grade, T stage, N stage, surgery, tumor size, and distant metastasis in other sites were independent variables for LM. Similarly, Xu et al. [29] developed machine learning-based models to evaluate the risk of developing lung metastasis in kidney cancer patients using the SEER database. They performed multivariate logistic regression and found that grade, T and N stage, tumor size, and metastasis to other sites, including the bone, brain, and liver, were all risk factors. Ultimately, they established a prediction model with a high AUC. However, Chan’s study also had some limitations. For instance, their study was only conducted using the SEER database, and the model was validated using 10-fold cross-validation without being split into an internal test set. Furthermore, it did not undergo external validation, both of which may limit the generalizability of the model. Besides, these prediction models irrationally included variables of other metastatic sites, which is unsuitable for preoperative evaluation and would dramatically reduce the model’s utility. Molecularly, certain circRNAs such as circ-EGLN3 and SCARB1 [30, 31] were demonstrated to promote effectiveness and predictive value for LM, but these molecular entities are difficult to examine in each person and include a high cost, thus greatly hampering their clinical application value.

As a technological tool, ML has yielded remarkable results in assisting epidemiologists [32]. Using ML algorithms, Handelman predicted a reduction in diagnostic errors by addressing complex and tedious clinical work [14]. Compared with conventional CT images frequently used in preoperative screening, ML models were good at assigning risk levels for developing LM with high accuracy and convenience. Here, we successfully developed a web-based predictor to predict the risk for LM in patients newly diagnosed with KC, which used easily accessible clinical characteristics and proved to be highly accurate and applicable.

This study used the data of 36,555 patients with KC to establish the ML models. These algorithms have been internally validated in 15,667 patients from the SEER cohort and externally tested in 492 patients from a Chinese cohort. Among the six predictive models, XGB performed the best, with an AUC of 0.913 and 0.904 in the internal and external test cohorts, respectively. DCA and CUC curves showed great discriminative and applicative abilities in the clinic.

Referring to LM risk factors, Thompson and colleagues suggested that a larger tumor size and an advanced T stage were significantly associated with a higher probability of metastasis in renal cell carcinoma [33]. Among 781 KC patients with tumors less than 3 cm, they identified only one patient with a record of distant metastasis. For every 1 cm increase in tumor size, the hazard ratio of metastasis-free survival increased by 0.24, with . Our study also yielded consistent results, indicating that larger tumor size was more strongly associated with LM and identified as an independent factor for LM. Mikami et al. reported that KC with higher grades tended to develop epithelial-mesenchymal transition (EMT), which has been proven to be critical for metastasis [34]. In our study, we observed a higher proportion of high-grade (III-IV) cases in the LM cohort and considered it an independent factor for LM. We also found that positive lymph nodes were more likely to be observed in the metastatic cohort and were one of the most influential characteristics associated with LM in KC patients. Similarly, Dudani et al. discovered that lymph node involvement contributed to distant metastasis and was common in KC, especially in papillary renal cell carcinoma [27]. Blacks, Asians, and African Americans have been reported to have a relatively high mortality risk due to KC [35]. This result may be resulted by higher probability of LM which was identified as a risk factor in this study. Consistently, Vaishampayan et al. found that compared with white patients with KC, black patients had a significantly shorter survival time () [36]. Histology also seemed to suggest distant metastasis of KC. Wang analyzed 36,365 patients with renal cell carcinoma and found that the clear cell subtype had a higher risk of distant metastasis, followed by the papillary and chromophobe subtypes [37]. Rong et al. reported that in multivariate analysis, compared with clear cell carcinoma, sarcomatoid had a relatively high hazard ratio for generating LM. Chromophobe cell carcinoma and collecting duct carcinoma are less likely to develop LM in KC [38]. In this study, we found that the clear cell carcinoma subtype was a risk estimator of LM, while chromophobe and papillary cell carcinomas had a relatively low incidence of LM, similar to previous studies. Moreover, multivariate LR showed that male patients had a tendency to develop LM, which may be correlated with a higher smoking rate in males and requires further investigation.

ML as a black box has a long-term problem of interpretability [39]. To address this problem, we built a web-based free calculator based on the XGB algorithm trained in this study to help clinicians rapidly predict and estimate the probability of LM in patients with KC.

Although the predictive model performed well in estimating LM in patients with KC, this study has several limitations. First, this was a retrospective study, inevitably resulting in a selection bias. Second, the study did not involve common clinical indices, such as marriage and biochemical indices [40]. Finally, the external validation cohort from China only involved Chinese individuals, so more validation arms in other countries are needed to examine the model’s utility further.

5. Conclusions

In this study, we used a mainstream powerful machine learning tool to identify the high-risk factors of renal cancer lung metastasis and established a convenient and efficient web tool to help clinical doctors quickly identify those renal cancer patients prone to lung metastasis. This tool will greatly help patients in economically underdeveloped areas or those who are not convenient for puncture biopsy. The power behind this study is based on the large population in the SEER database, and the model has been independently verified by an external team. Future work should focus on increasing the sample size and diversity of ethnicities to validate the machine learning model. Additionally, incorporating more parameters, such as patient symptoms, may help improve performance. These are areas that could be addressed in future research.

Data Availability

The relevant data and code associated with the current manuscript have been uploaded to the following website https://github.com/qq731936287/kun.git.

Conflicts of Interest

The authors declare that there are no conflicts of interest.

Acknowledgments

The authors sincerely thank all authors who took part in this study. This work was supported by the National Natural Science Foundation of China (no. 81873606) and Chongqing Medical Scientific Research Project (no. 2020FYYX012).

Supplementary Materials

Supplementary file 1: the detailed hyperparameters in various machine learning models. (Supplementary Materials)

References

R. E. Gray and G. T. Harris, “Renal cell carcinoma: diagnosis and management,” American Family Physician, vol. 99, no. 3, pp. 179–184, 2019.
View at: Google Scholar
S. Turajlic, C. Swanton, and C. Boshoff, “Kidney cancer: the next decade,” Journal of Experimental Medicine, vol. 215, no. 10, pp. 2477–2479, 2018.
View at: Publisher Site | Google Scholar
J. Ferlay, I. Soerjomataram, R. Dikshit et al., “Cancer incidence and mortality worldwide: sources, methods and major patterns in GLOBOCAN 2012,” International Journal of Cancer, vol. 136, no. 5, pp. 359–386, 2015.
View at: Publisher Site | Google Scholar
S. R. Williamson, B. Delahunt, C. Magi-Galluzzi et al., “The world Health organization 2016 classification of testicular germ cell tumours: a review and update from the international society of urological pathology testis consultation panel,” Histopathology, vol. 70, no. 3, pp. 335–346, 2017.
View at: Publisher Site | Google Scholar
T. J. Mitchell, S. Turajlic, A. Rowan et al., “Timing the landmark events in the evolution of clear cell renal cell cancer: TRACERx renal,” Cell, vol. 173, no. 3, pp. 611–623, 2018.
View at: Publisher Site | Google Scholar
H. Wei, J. Miao, J. Cui et al., “The prognosis and clinicopathological features of different distant metastases patterns in renal cell carcinoma: analysis based on the SEER database,” Scientific Reports, vol. 11, no. 1, Article ID 17822, 2021.
View at: Publisher Site | Google Scholar
G. Hou, X. Li, Y. Zheng et al., “Construction and validation of a novel prognostic nomogram for patients with sarcomatoid renal cell carcinoma: a SEER-based study,” International Journal of Clinical Oncology, vol. 25, no. 7, pp. 1356–1363, 2020.
View at: Publisher Site | Google Scholar
D. C. Jung, H. J. Choi, H. Y. Kim, and K. H. Lee, “Pulmonary metastasis from renal cell carcinoma: characterization using contrast-enhanced CT attenuation value measurements,” Journal of Computer Assisted Tomography, vol. 33, no. 1, pp. 54–57, 2009.
View at: Publisher Site | Google Scholar
P. Alongi, M. Picchio, F. Zattoni et al., “Recurrent renal cell carcinoma: clinical and prognostic value of FDG PET/CT,” European Journal of Nuclear Medicine and Molecular Imaging, vol. 43, no. 3, pp. 464–473, 2016.
View at: Publisher Site | Google Scholar
M. Sun, G. De Velasco, P. K. Brastianos et al., “The development of brain metastases in patients with renal cell carcinoma: epidemiologic trends, survival, and clinical risk factors using a population-based cohort,” European Urology Focus, vol. 5, no. 3, pp. 474–481, 2019.
View at: Publisher Site | Google Scholar
J. Lin, M. Yin, L. Liu et al., “The development of a prediction model based on random survival forest for the postoperative prognosis of pancreatic cancer: a SEER-based study,” Cancers, vol. 14, no. 19, Article ID 4667, 2022.
View at: Publisher Site | Google Scholar
R. Hu, Z. Tang, E. Wu, Q. Mo, R. Yang, and J. Li, “RDC-SAL: refine distance compensating with quantum scale-aware learning for crowd counting and localization,” Applied Intelligence, vol. 52, no. 12, pp. 14336–14348, 2022.
View at: Publisher Site | Google Scholar
R. Hu, Q. Mo, Y. Xie et al., “AVMSN: an audio-visual two stream crowd counting framework under low-quality conditions,” IEEE Access, vol. 9, pp. 80500–80510, 2021.
View at: Publisher Site | Google Scholar
G. S. Handelman, H. K. Kok, R. V. Chandra, A. H. Razavi, M. J. Lee, and H. Asadi, “eDoctor: machine learning and the future of medicine,” Journal of Internal Medicine, vol. 284, no. 6, pp. 603–619, 2018.
View at: Publisher Site | Google Scholar
W.-C. Liu, M.-X. Li, S.-N. Wu et al., “Using machine learning methods to predict bone metastases in breast infiltrating ductal carcinoma patients,” Frontiers in Public Health, vol. 10, Article ID 922510, 2022.
View at: Publisher Site | Google Scholar
X. Cheng, J. Li, T. Xu, K. Li, and J. Li, “Predicting survival of patients with rectal neuroendocrine tumors using machine learning: a SEER-based population study,” Front Surg, vol. 8, Article ID 745220, 2021.
View at: Publisher Site | Google Scholar
S. Leem, I. Huh, and T. Park, “Enhanced permutation tests via multiple pruning,” Frontiers in Genetics, vol. 11, p. 509, 2020.
View at: Publisher Site | Google Scholar
M. Adibi, P. A. Kenney, A. Z. Thomas et al., “Prediction of pulmonary metastasis in renal cell carcinoma patients with indeterminate pulmonary nodules,” European Urology, vol. 69, no. 2, pp. 352–360, 2016.
View at: Publisher Site | Google Scholar
R. L. Siegel, K. D. Miller, and A. Jemal, “Cancer statistics,” CA: A Cancer Journal for Clinicians, vol. 67, no. 1, pp. 7–30, 2017.
View at: Google Scholar
Z. Lu, C. Yang, W. He, J. Zhou, and R. Xiang, “Nomogram to predict risk and prognosis of synchronous lung metastasis in renal cell carcinoma: a large cohort analysis,” Medicine, vol. 101, no. 27, Article ID e29764, 2022.
View at: Publisher Site | Google Scholar
S. A. Padala, A. Barsouk, K. C. Thandra et al., “Epidemiology of renal cell carcinoma,” World Journal of Oncology, vol. 11, no. 3, pp. 79–87, 2020.
View at: Publisher Site | Google Scholar
T. Chandrasekar, Z. Klaassen, H. Goldberg, G. S. Kulkarni, R. J. Hamilton, and N. E. Fleshner, “Metastatic renal cell carcinoma: patterns and predictors of metastases-A contemporary population-based series,” Urologic Oncology: Seminars and Original Investigations, vol. 35, no. 11, 2017.
View at: Google Scholar
S. Lin, Y. Zheng, Z. Qin et al., “Surgical intervention in renal cell carcinoma patients with lung and bronchus metastasis is associated with longer survival time: a population-based analysis,” Annals of Translational Medicine, vol. 7, no. 14, p. 323, 2019.
View at: Publisher Site | Google Scholar
T. K. Choueiri, P. Tomczak, S. H. Park et al., “Adjuvant pembrolizumab after nephrectomy in renal-cell carcinoma,” New England Journal of Medicine, vol. 385, no. 8, pp. 683–694, 2021.
View at: Publisher Site | Google Scholar
P. C. Barata and B. I. Rini, “Treatment of renal cell carcinoma: current status and future directions,” CA: A Cancer Journal for Clinicians, vol. 67, no. 6, pp. 507–524, 2017.
View at: Publisher Site | Google Scholar
F. Quhal, K. Mori, A. Bruchbacher et al., “First-line immunotherapy-based combinations for metastatic renal cell carcinoma: a systematic review and network meta-analysis,” European Urology Oncology, vol. 4, no. 5, pp. 755–765, 2021.
View at: Publisher Site | Google Scholar
S. Dudani, G. de Velasco, J. C. Wells et al., “Evaluation of clear cell, papillary, and chromophobe renal cell carcinoma metastasis sites and association with survival,” JAMA Network Open, vol. 4, no. 1, Article ID e2021869, 2021.
View at: Publisher Site | Google Scholar
Y. Zhao, G. Liu, Q. Sun, G. Zhai, G. Wu, and Z. C. Li, “Validation of CT radiomics for prediction of distant metastasis after surgical resection in patients with clear cell renal cell carcinoma: exploring the underlying signaling pathways,” European Radiology, vol. 31, no. 7, pp. 5032–5040, 2021.
View at: Publisher Site | Google Scholar
C. Xu, Q. Zhou, W. Liu et al., “Dynamic predictive models with visualized machine learning for assessing the risk of lung metastasis in kidney cancer patients,” JAMA Oncology, vol. 2022, Article ID 5798602, 2022.
View at: Google Scholar
L. Lin and J. Cai, “Circular RNA circ-EGLN3 promotes renal cell carcinoma proliferation and aggressiveness via miR-1299-mediated IRF7 activation,” Journal of Cellular Biochemistry, vol. 121, no. 11, pp. 4377–4385, 2020.
View at: Publisher Site | Google Scholar
J. Sun, S. Pan, H. Cui, and H. Li, “CircRNA SCARB1 promotes renal cell carcinoma progression via Mir- 510-5p/SDC3 axis,” Current Cancer Drug Targets, vol. 20, no. 6, pp. 461–470, 2020.
View at: Publisher Site | Google Scholar
Y. Zhang, X. Yi, Z. Tang et al., “Using machine learning to predict lymph node metastasis in patients with renal cell carcinoma: a population-based study,” Frontiers in Public Health, vol. 11, Article ID 1104931, 2023.
View at: Publisher Site | Google Scholar
R. H. Thompson, J. R. Hill, Y. Babayev et al., “Metastatic renal cell carcinoma risk according to tumor size,” The Journal of Urology, vol. 182, no. 1, pp. 41–45, 2009.
View at: Publisher Site | Google Scholar
S. Mikami, M. Oya, R. Mizuno, T. Kosaka, and Y. Okada, “Invasion and metastasis of renal cell carcinoma,” Medical Molecular Morphology, vol. 47, no. 2, pp. 63–67, 2013.
View at: Publisher Site | Google Scholar
S. Luzzago, C. Palumbo, G. Rosiello et al., “Racial and ethnic differences in survival in contemporary metastatic renal cell carcinoma patients, according to alternative treatment modalities,” Cancer Causes and Control, vol. 31, no. 3, pp. 263–272, 2020.
View at: Publisher Site | Google Scholar
U. N. Vaishampayan, H. Do, M. Hussain, and K. Schwartz, “Racial disparity in incidence patterns and outcome of kidney cancer,” Urology, vol. 62, no. 6, pp. 1012–1017, 2003.
View at: Publisher Site | Google Scholar
J. Wang, C. Zhanghuang, X. Tan et al., “Development and validation of a nomogram to predict distant metastasis in elderly patients with renal cell carcinoma,” Frontiers in Public Health, vol. 9, Article ID 831940, 2021.
View at: Publisher Site | Google Scholar
A. Pecoraro, C. Palumbo, S. Knipper et al., “Histologic subtype, tumor grade, tumor size, and race can accurately predict the probability of synchronous metastases in T2 renal cell carcinoma,” Clinical Genitourinary Cancer, vol. 18, no. 5, pp. 610–618, 2020.
View at: Publisher Site | Google Scholar
A. J. Banegas-Luna, J. Peña-García, A. Iftene et al., “Towards the interpretability of machine learning predictions for medical applications targeting personalised therapies: a cancer case survey,” International Journal of Molecular Sciences, vol. 22, no. 9, Article ID 4394, 2021.
View at: Publisher Site | Google Scholar
D. Luo, Q. Liu, W. Yu et al., “Prognostic value of distant metastasis sites and surgery in stage IV colorectal cancer: a population-based study,” International Journal of Colorectal Disease, vol. 33, no. 9, pp. 1241–1249, 2018.
View at: Publisher Site | Google Scholar

Copyright

Copyright © 2023 Xinglin Yi et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

PDF Download Citation

Download other formats

Order printed copies