Renal cell carcinoma (RCC), or kidney cancer, is often called a silent disease because it typically causes no symptoms until it has grown quite large or spread to other organs. As a result, approximately 30% of patients have metastatic disease at the time of their very first diagnosis, meaning cancer has already spread to distant organs like the lungs, bones, liver, or brain.
For patients with de novo metastatic RCC (meaning metastatic from the start, not relapsed after prior treatment), the prognosis is significantly worse. The 5-year survival rate for metastatic kidney cancer is only around 15%, compared to over 90% for localized disease. Treatment for these patients requires immediate systemic therapies such as immunotherapy combined with targeted drugs rather than surgery alone.
Identifying which patients have metastatic disease at diagnosis is therefore a critical decision point in kidney cancer care. This study developed a simple, clinically practical prediction tool called a nomogram that can estimate a patient's risk of having metastatic disease at diagnosis using only three variables drawn from routine blood tests and CT scan measurements.
The study retrospectively analyzed 461 patients with confirmed RCC treated at a single Chinese university hospital between 2016 and 2025. Patients were randomly assigned in a 7:3 ratio to training (323 patients) and validation (138 patients) cohorts. At initial diagnosis, 87 patients (18.9%) had confirmed distant metastases, with the lung being the most common metastatic site (47%), followed by bone (22.8%) and liver (16.7%).
Starting from 31 candidate clinical variables identified by univariate analysis, the researchers applied three machine learning algorithms simultaneously to select the most important features: LASSO (Least Absolute Shrinkage and Selection Operator) regression, SVM-RFE (Support Vector Machine with Recursive Feature Elimination), and Boruta, a comprehensive feature selection algorithm. Only variables selected by all three methods were retained, ensuring the final predictors were consistently important across different analytical approaches.
This triple-filter approach identified 5 candidate predictors that survived all three selection methods: monocyte count, LMR (lymphocyte-to-monocyte ratio), SIRI (systemic inflammatory response index), AAPR (albumin-to-alkaline phosphatase ratio), and PADUA score. Multivariate logistic regression then confirmed which of these were independently predictive, ultimately selecting 3 final variables for the nomogram.
Multivariate logistic regression identified three independent predictors of having metastatic disease at RCC diagnosis. LMR (lymphocyte-to-monocyte ratio) was protective: each unit increase in LMR reduced the odds of metastasis by 22% (OR = 0.78). A low LMR indicates relatively fewer lymphocytes or more monocytes, reflecting a tumor-promoting immune environment.
AAPR (albumin-to-alkaline phosphatase ratio) was the strongest single predictor (OR = 0.05): patients with very low AAPR had roughly 20 times higher odds of metastasis. Low albumin indicates poor nutritional status and systemic inflammation, while high alkaline phosphatase may reflect oxidative stress and liver involvement, both markers of more advanced disease.
The PADUA score, an anatomical complexity score calculated from CT measurements of the kidney tumor (incorporating tumor size, position, relationship to the collecting system, and other geometric factors), was the only variable that increased metastatic risk: each point higher on the PADUA scale raised the odds of metastasis by 41% (OR = 1.41). This makes intuitive sense: larger, more complex tumors are more likely to have acquired the genetic mutations needed to spread.
The completed nomogram achieved an AUC of 0.771 in training and 0.747 in internal validation, indicating moderate discriminatory ability. While not a perfect predictor, this level of accuracy is clinically meaningful when combined with the tool's key advantage: it uses only three variables that are universally available from routine pre-treatment blood tests and CT imaging, requiring no specialized tests or equipment.
Calibration curves assessed using the Hosmer-Lemeshow test showed excellent agreement between predicted and observed probabilities in both cohorts (P = 0.769 for training, P = 0.603 for validation), meaning the nomogram's numerical probability estimates are reliable and not systematically over- or underestimating risk.
Decision curve analysis showed that the nomogram provided positive net clinical benefit across a wide range of decision thresholds (0% to 63% in training, 0% to 68% in validation). This means that using the nomogram to guide clinical decision-making results in more correct classifications than either treating all patients as metastatic or treating none as metastatic.
The greatest strength of this nomogram is its practicality. All three variables are calculated from data collected during a patient's initial hospital admission blood draw and routine CT scan. No additional tests, no waiting for pathology results, and no specialized equipment are required. A clinician could calculate the score at the bedside within minutes of seeing initial test results.
For patients identified as high-risk by the nomogram, clinicians could initiate a more comprehensive metastatic workup immediately, including bone scans, brain MRI, and thorough chest imaging. For those assessed as low risk, such extensive workups could be reserved for cases where other clinical factors suggest concern. This risk stratification approach optimizes both patient safety and resource use.
The study also found that LMR correlates with immunotherapy effectiveness in metastatic kidney cancer. A high LMR has been associated with better responses to nivolumab immunotherapy in metastatic RCC, suggesting the nomogram could potentially help guide not just detection of metastasis but also initial treatment selection once metastatic disease is confirmed.
This study successfully developed and internally validated a three-variable diagnostic nomogram for predicting metastatic kidney cancer at initial diagnosis. By integrating an immune biomarker (LMR), a nutritional-inflammatory marker (AAPR), and an anatomical complexity score (PADUA), the tool captures three different dimensions of the disease in a simple, clinically accessible format.
The moderate AUC of approximately 0.75 reflects the inherent biological complexity of predicting metastasis: no simple three-variable model will ever perfectly classify all patients. However, the tool's high clinical utility, excellent calibration, and use of universally available variables make it a valuable addition to the clinician's toolkit, particularly in settings with limited access to advanced genomic or molecular testing.
Future steps include external validation in independent multicenter cohorts and prospective testing to confirm the tool performs as expected when applied in real clinical decision-making. Incorporation of additional molecular markers and imaging-based features could further improve its accuracy and help address the current limitation of the relatively small validation cohort.