Background: Surgery is the standard of care for localized kidney cancer. Diagnostic imaging plays a critical role in disease staging and informs the extent of surgical resection (partial or radical nephrectomy, extended resection). In clinical routine, up to 15% of the tumors initially assessed as T1-T2 on imaging is upgraded to pT3a status post-surgery, implying a higher risk of relapse. The ability to correctly predict pT3a status pre-surgery would inform the surgical approach. An individualized prediction of the risk of clinical T1 or T2 tumors to be upstaged to pT3a is thus of high surgical interest. Methods: UroCCR is a French national network of 37 multidisciplinary teams for kidney cancer management that collects longitudinal data on the routine clinical care of its patients. A retrospective cohort of 4,395 cases of clinical T1-T2 kidney tumors was analyzed to develop a machine learning-based algorithm predictive of post-surgical pT3a upstaging risk at the individual patient level. For each patient, pre-surgical data were collected, including gender, age, symptoms, tumor size, tumor location, RENAL score, ECOG performance status, ASA score, and post-surgical pathological status. Sites were randomly assigned to the training or testing cohort, and their respective patient cases split between cohorts in a 60/40 ratio. Missing values were addressed through imputation performed with a k-nearest neighbor algorithm. Algorithms were trained on a data set of 2,636 patients and hyperparameters were optimized using a Bayes cross-validation (10-fold) approach. The area under the precision-recall curve (prAUC) was used as optimization metric. The performance of the algorithms for pT3a status prediction was then evaluated on the test dataset of 1,759 patients using precision-recall curves. Results: A logistic regression algorithm reached an AUC of 0.77 and a prAUC of 0.41. Higher values of the tumor size or age at surgery, the hilar location and the presence of symptoms at diagnosis were all associated with an increase of the predicted probability of pT3a upstaging. For each patient, Shapley values graphs were generated to display the pT3a upstaging probability and the relative contribution of each feature to the prediction. Three risk groups were defined based on the relative computed probability of pT3a upstaging, which displayed a statistically significant difference in Disease-Free Survival (DFS) (p < 0.0001), suggesting that pre-surgical multimodal data analysis could help predict long-term outcomes. Conclusions: This study suggests that machine learning applied to pre-surgical multimodal data can predict the risk of pT3a upstaging of a localized kidney tumor and inform long-term outcomes at the individual patient level. The results have been validated on an external cohort of 1,759 patients with data from the clinical routine.