Validation of RCC Predicting Model With Emulated-target Trial

Overview

This single-center study utilizes real-world data (2012-2024) from 4700 renal cell carcinoma (RCC) patients at Peking University Third Hospital to: (1) Develop and validate a prognostic prediction model specifically for RCC patients, including those with venous tumor thrombus (VTT); (2) Compare the performance of this new model against existing RCC prediction models in both the overall RCC cohort and the VTT subgroup; (3) Employ an emulated target trial (ETT) methodology to evaluate whether risk-stratified treatment based on the prediction model (grouping patients as high/medium/low risk) improves survival outcomes .

1. Study Design and Technical Approach: This is a single-center, observational study utilizing real-world data (RWD) to achieve three primary objectives within a renal cell carcinoma (RCC) cohort, specifically focusing on patients with venous tumor thrombus (VTT): * Objective 1 (Model Comparison): Compare the performance (discrimination, calibration, stability) of the newly developed prediction model against existing published RCC prediction models within the overall cohort and the VTT subgroup. * Objective 2 (Causal Evaluation via ETT): Employ an Emulated Target Trial (ETT) framework to evaluate the causal impact of risk-stratified treatment guided by the prediction model on survival and health economic outcomes, compared to non-stratified treatment. 2. Study Population and Data Source: * Source: Peking University Third Hospital, Urology Department. * Cohort: A bidirectional cohort comprising: * Retrospective component: \~4000 RCC patients treated between January 2012 and June 2025. * Prospective component: All eligible RCC patients treated between Dec 2025 and Dec 2026 (target total N=4700, including \~446 VTT patients). * Inclusion Criteria: (1) Age 18-80 years; (2) Clinical diagnosis of RCC; (3) Underwent radical/partial nephrectomy. * Exclusion Criteria: (1) Severe missing clinical information; (2) Distant metastasis at baseline; (3) History of other malignancies. * Data Collection: Comprehensive candidate variables include demographics, lab results, imaging, treatment details (surgery, systemic therapy - neoadjuvant/adjuvant), pathology (including VTT level), and follow-up data. 3. Core Methodologies: * Missing Data: Assessed for patterns. Addressed primarily via Multiple Imputation. Sensitivity analyses will evaluate the impact of missing data assumptions. * Outliers: Identified through data review; verified against source; excluded if confirmed erroneous. * Emulated Target Trial (ETT): * Objective: Estimate the effect of adhering to risk-stratified treatment (based on the model's High/Intermediate/Low risk groups) versus non-adherence. * "Treatment" Definition: SHAP analysis will identify key features defining risk groups and inform optimal treatment strategies per group. Adherence is determined by comparing actual treatment received to the predicted risk group's recommended strategy (e.g., type/extent of surgery, use of neoadjuvant/adjuvant systemic therapy/radiotherapy). * Group Assignment: Non-random, based on actual treatment: * Intervention Group (Adherent): Patients whose received treatment aligned with their predicted risk stratum. * Control Group (Non-Adherent): Patients whose received treatment did not align with their predicted risk stratum. * Baseline: Defined as the time of first meeting eligibility criteria (also considered treatment assignment time for ETT). * Causal Inference \& Confounding Control: Uses cloning-censoring-weighting methods to emulate randomization. Directed Acyclic Graphs (DAGs) will identify key confounders for adjustment. Intention-To-Treat (ITT) principle is the primary analysis; Per-Protocol (PP) is secondary. * Follow-up: Starts at baseline. Includes telephone and outpatient visits. Frequency based on EAU RCC guidelines by risk group. Continues until outcome (death, recurrence), withdrawal, or study end. Captures survival, recurrence, QoL, resource utilization/costs. 4. Outcomes: * Primary: Overall Survival (OS). * Secondary: Model performance. 5. Statistical Analysis Plan: * Descriptive Statistics: Mean, SD, median, min, max, Q1, Q3 for continuous variables; counts and percentages for categorical variables. * Model Performance: C-statistic (AUC), calibration plots, Brier score, net reclassification improvement (NRI), integrated discrimination improvement (IDI) for comparing prediction models. * ETT Analysis: Weighted survival analysis (e.g., weighted Kaplan-Meier, weighted Cox regression) for OS. * Subgroup Analyses: (1) VTT presence and level; (2) Type of non-adherence (under-treatment vs. over-treatment); (3) Predicted risk group. * Sensitivity Analyses: For (1) missing data imputation methods, model specifications, definition of adherence, follow-up duration, censoring assumptions. (2) different methods of risk group division based on predicting results for grading and nomogram models. (3) whether participants received neo-adjuvant therapy. 6. Registry Procedures and Quality Assurance: * Source Data \& Management: Primary source data is paper-based (medical records, lab reports, questionnaires, CRFs). Strict procedures for secure storage (fire/water protection), handling, and archival of source documents and certified copies are implemented. * Data Validation Checks: Electronic data entry will incorporate predefined range and consistency checks (e.g., valid dates, plausible lab values, logical relationships between fields) to minimize entry errors. * Source Data Verification (SDV): A subset of records will undergo source data verification comparing the entered registry data against the original source documents (medical charts, CRFs) to assess accuracy and completeness. The extent (%)/method (targeted vs. random) will be detailed in the monitoring plan. * Data Dictionary: A comprehensive data dictionary will define all collected variables, including: variable name, source (e.g., pathology report, lab system), definition, coding (e.g., MedDRA for AEs, ICD-10 for diagnoses), format (numeric, categorical, date), units, allowable values/ranges, and handling rules for missing/special values. * Standard Operating Procedures (SOPs): Detailed SOPs will govern all critical processes: patient identification/screening, informed consent (adhering strictly to described principles), data collection/extraction, source documentation, data entry/management, coding (AEs, medications), SDV, query management, adverse event reporting, statistical analysis, data security/privacy, and audit processes. Training on SOPs is mandatory for all study personnel. * Quality Assurance Plan: Includes protocol/SOP adherence monitoring, regular data quality reviews (completeness, accuracy checks), SDV as above, and readiness for on-site auditing by the sponsor/institution or regulatory bodies. A process for handling protocol deviations is established. * Sample Size Justification: Based on historical data (HR\~0.65 for adjuvant therapy effect in advanced RCC), log-rank calculations determined a simulated RCT would require \~256 events. Considering the ETT design (requiring sufficient sample size for weighting and subgroup analyses) and the center's patient volume/outcome rates, a total sample size of 4700 patients is targeted to ensure adequate power for the primary ETT survival analyses within the study timeframe. * Plan for Missing Data: The primary strategy is Multiple Imputation by Chained Equations (MICE) , assuming data is Missing At Random (MAR). Sensitivity analyses will explore the Missing Not At Random (MNAR) scenario and complete case analysis to assess robustness. Patterns of missingness will be thoroughly investigated and reported.