# CT-based variables of perirenal fat are risk factors for assessing pathological T-stage in patients with clear cell renal cell carcinoma

## TL;DR

This study shows that CT-based features of perirenal fat can help predict the severity of kidney cancer before surgery, improving treatment decisions.

## Contribution

The study introduces a predictive model using perirenal fat characteristics and preoperative biomarkers to assess T-stage in ccRCC patients.

## Key findings

- Perinephric fat stranding, RENAL Nephrometry Score, and preoperative platelet counts are independent predictors of high T-stage in ccRCC.
- The predictive model outperforms radiologists in predicting high T-stage with an AUC of 0.867.
- The model shows good discrimination and calibration, and is clinically useful according to decision curve analysis.

## Abstract

Advances in the treatment of clear cell renal cell carcinoma (ccRCC) toward risk-based therapy intensity modulation have necessitated patient-specific assessments of preoperative T-stage.

A derivation cohort of 218 ccRCC patients with known pathological results and preoperative biomarker data was used to develop and validate a predictive model for preoperative T-stage. Perirenal fat characteristics were determined by the measurement or evaluation of preoperative CT images. Multivariate logistic regression was used to identify the predictive factors and to develop a predictive model, which was then internally validated using cross-validation and bootstrapping. Model discrimination was assessed using calibration plots. Finally, nomograms were both plotted to visualize model efficiency.

A total of 218 patients (152 males and 66 females) with pathological tumor T-stage (151 low T-stage and 67 high T-stage) were enrolled in this study. Perinephric fat stranding (PFS), RENAL Nephrometry Score (RNS), and preoperative platelet (PLT) counts were independent predictors of a high ccRCC T-stage. The performance of the predictive model built on these variables notably surpasses that of radiologists (AUC: 0.867 vs 0.680; delong test, p < 0.001). Internal validation with a K-fold cross-validation (K = 10) and a bootstrap method showed good discrimination. The model also demonstrated proper calibration. According to the decision curve analysis, the model was found to be clinically useful (risk threshold probability: 6%–85%).

PFS, RNS, and preoperative PLT provided excellent predictions of a high pathological T-stage for ccRCC patients. This predictive model can serve as a reference to aid clinicians and surgeons in clinical decision-making.