When kidney cancer is diagnosed and has not spread to distant organs (a condition called M0, meaning no distant metastasis), surgeons have several options for removing the tumor. These include partial nephrectomy (PN), which removes only the tumor and a margin of healthy tissue, complete nephrectomy (CN), which removes the entire kidney on the affected side, and radical nephrectomy (RN), a more extensive removal of the kidney along with surrounding fatty tissue and sometimes nearby lymph nodes.
Choosing between these approaches involves balancing cancer control with preservation of kidney function. Removing less kidney tissue better preserves function but raises concerns about complete cancer removal. Removing more tissue ensures wider margins but sacrifices healthy kidney, which can increase long-term risk of chronic kidney disease and associated health problems.
This study used data from over 56,000 patients in the SEER (Surveillance, Epidemiology, and End Results) database to compare overall survival, cancer-specific survival, and other outcomes across the three surgical approaches. The analysis was stratified by tumor grade to account for the influence of cancer aggressiveness on the appropriateness of each approach.
The researchers also developed a deep learning model to predict which surgical approach would give each individual patient the best survival outcome, going beyond population-level comparisons to enable personalized surgical decision support.
For early-stage kidney cancer, partial nephrectomy has become the preferred approach at most centers when technically feasible. Removing only the tumor while leaving the rest of the kidney intact protects overall kidney function, which is critically important for long-term cardiovascular health and quality of life. Chronic kidney disease, which can result from losing too much kidney tissue, itself increases the risk of heart disease and mortality.
However, not all tumors are suitable for partial nephrectomy. Tumors that are very centrally located, close to major blood vessels, or in kidneys with unusual anatomy may require more extensive removal. For tumors with higher grade (more aggressive appearance), some surgeons prefer wider margins to minimize recurrence risk, leading to a preference for complete or radical nephrectomy.
Clinical guidelines generally recommend partial nephrectomy when feasible, but the supporting evidence comes largely from studies comparing outcomes within specific tumor size categories rather than across the full spectrum of M0 disease. This study aimed to provide a more comprehensive comparison across all non-metastatic RCC cases and all tumor grades, using a large enough sample to detect meaningful differences.
The SEER database was chosen because of its size, representativeness, and the availability of both surgical type and survival outcome data for a very large number of patients treated over a decade. The final analytical cohort of 56,534 patients diagnosed with M0 RCC between 2010 and 2019 provided sufficient statistical power for detailed subgroup analyses by grade and surgical approach.
In the overall cohort, patients who underwent partial nephrectomy had significantly better overall survival and cancer-specific survival compared to those who had complete or radical nephrectomy. This finding was robust across multiple statistical analyses including Cox proportional hazards regression and propensity score matching, which adjusts for differences in baseline patient characteristics between surgical groups.
Propensity score matching was a particularly important methodological step in this study. Patients who receive different surgical approaches differ in ways beyond just tumor characteristics, including age, health status, tumor complexity, and surgeon preference. By mathematically matching patients with similar characteristics who received different surgeries, this analysis isolates the effect of surgical choice from the effects of these confounding factors.
The advantage of partial nephrectomy over more extensive surgery was consistent across grade II, III, and IV tumors. For grade I tumors, the analysis found no statistically significant difference in overall survival between partial nephrectomy and the no-surgery group, suggesting that some very low-grade, small tumors may not require immediate surgical intervention. This finding aligns with active surveillance strategies being increasingly used for select low-risk kidney cancers.
Complete nephrectomy and radical nephrectomy showed similar survival outcomes to each other in most grade subgroups, with the key difference between these two approaches being the extent of tissue removal rather than any fundamental difference in cancer control. Both were inferior to partial nephrectomy across most scenarios evaluated.
Beyond population-level comparisons, the researchers trained a deep learning neural network to predict which specific surgical approach would give each individual patient the highest probability of survival at one year. The model was trained on the matched SEER cohort and incorporated multiple patient and tumor characteristics as input features.
The deep learning model achieved an area under the ROC curve (AUC) of 0.844 for predicting 1-year survival, indicating good discriminatory ability. This performance level suggests the model is capturing meaningful patterns in the relationship between patient characteristics, surgical choice, and outcomes beyond what simpler statistical models could detect.
The model outputs a recommended surgical approach for each patient based on their individual clinical profile. In theory, a physician could enter a patient's characteristics into such a model and receive a data-driven recommendation for the surgical strategy most likely to maximize survival. This represents a form of personalized surgical decision support that goes beyond applying generic guideline recommendations.
The researchers emphasize that the model is intended as a decision aid rather than a replacement for clinical judgment. Surgical feasibility, patient preferences, surgeon expertise, and factors not captured in administrative databases all play important roles in real-world treatment decisions. The model provides evidence-based input to an inherently multifactorial decision-making process.
For patients diagnosed with non-metastatic kidney cancer, these results provide strong evidence that partial nephrectomy should be sought out whenever technically feasible. If a urologist recommends complete or radical nephrectomy, it is entirely appropriate for patients to ask whether partial nephrectomy has been evaluated and why it is not being offered. Getting a second opinion at a high-volume kidney cancer center may be worthwhile if partial nephrectomy was not offered.
The grade I finding, that surgery may not improve survival compared to no surgery, has implications for the growing practice of active surveillance in kidney cancer. Small, low-grade tumors can be monitored with regular imaging rather than immediately treated, allowing surgery to be deferred or potentially avoided entirely in patients for whom surgical risks outweigh benefits. This finding adds population-level outcome data supporting that approach.
Patients with higher-grade tumors can also be reassured that partial nephrectomy, when feasible, appears to be equally effective for cancer control while providing the additional benefit of preserved kidney function. The concern that a less extensive surgery might leave residual cancer or increase recurrence risk is not borne out by the population-level data from this study.
Access to partial nephrectomy varies by surgical center and surgeon experience. Hospitals with dedicated robotic surgery programs and high-volume urologists typically have greater capacity to safely perform partial nephrectomy for complex tumors. Patients who are told their tumor is not suitable for partial nephrectomy may benefit from a referral to a center with advanced minimally invasive surgery capabilities.
The study's reliance on registry data introduces inherent limitations. SEER does not capture information about tumor complexity scores, comorbid conditions in detail, patient preferences, or reasons why specific surgical approaches were chosen. These unmeasured factors influence both surgery selection and outcomes, and cannot be fully accounted for even with propensity score matching.
The finding that partial nephrectomy is superior across grades may partly reflect selection bias: surgeons may preferentially perform partial nephrectomy on healthier patients with more favorable tumor anatomy. Even after propensity score matching, residual confounding cannot be fully eliminated from observational data. Randomized controlled trials, while difficult to conduct for surgical comparisons, would provide stronger causal evidence.
The deep learning model, while promising, was developed and validated on the same SEER dataset used for the comparative analysis. External validation on data from different countries, healthcare systems, and time periods will be needed before this model could be considered for clinical deployment.
This large population-level study reinforces and extends the evidence supporting partial nephrectomy as the preferred surgical approach for non-metastatic kidney cancer whenever technically achievable. The survival benefit was consistent across most tumor grades and was robust to multiple analytical approaches including propensity score matching.
The novel finding that grade I kidney cancers showed no survival benefit from surgery versus no surgery provides important population-level support for active surveillance as a management strategy in carefully selected patients, adding quantitative outcome data to what has previously been primarily a theoretically motivated recommendation.
The development of a deep learning surgical decision model demonstrates the potential for AI-assisted personalized treatment planning in surgical oncology. As these models are refined with larger and more diverse datasets and validated prospectively, they may play an increasingly important role in supporting high-quality, individualized cancer care decisions.