Deep Neural Networks Outperform the CAPRA Score in Predicting Biochemical Recurrence After Prostatectomy

Abstract

BackgroundUse of predictive models for the prediction of biochemical recurrence (BCR) is gaining attention for prostate cancer (PCa). Specifically, BCR occurs in approximately 20–40% of patients five years after radical prostatectomy (RP) and the ability to predict BCR may help clinicians to make better treatment decisions. We aim to investigate the accuracy of CAPRA score compared to others models in predicting the 3-year BCR of PCa patients.Material and MethodsA total of 5043 men who underwent RP were analyzed retrospectively. The accuracy of CAPRA score, Cox regression analysis, logistic regression, K-nearest neighbor (KNN), random forest (RF) and a densely connected feed-forward neural network (DNN) classifier were compared in terms of 3-year BCR predictive value. The area under the receiver operating characteristic curve was mainly used to assess the performance of the predictive models in predicting the 3 years BCR of PCa patients. Pre-operative data such as PSA level, Gleason grade, and T stage were included in the multivariate analysis. To measure potential improvements to the model performance due to additional data, each model was trained once more with an additional set of post-operative surgical data from definitive pathology.ResultsUsing the CAPRA score variables, DNN predictive model showed the highest AUC value of 0.7 comparing to the CAPRA score, logistic regression, KNN, RF, and cox regression with 0.63, 0.63, 0.55, 0.64, and 0.64, respectively. After including the post-operative variables to the model, the AUC values based on KNN, RF, and cox regression and DNN were improved to 0.77, 0.74, 0.75, and 0.84, respectively.ConclusionsOur results showed that the DNN has the potential to predict the 3-year BCR and outperformed the CAPRA score and other predictive models.

A Recursive Partitioning Analysis Demonstrating Risk Subsets for 8-Year Biochemical Relapse After Margin-Positive Radical Prostatectomy Without Adjuvant Hormone or Radiation Therapy

“…The Cancer of the Prostate Risk Assessment Postsurgical (CAPRA-S) score, which incorporates six tumor characteristics to determine a score from zero to twelve, has demonstrated high accuracy rates towards predicting biochemical recurrence [39]. Other approaches including computational modeling with Deep Neural Networks and genomic assays such as Decipher have also shown promise [16,40]. Finally, liquid biopsies assessing minimal residual disease by detecting either circulating tumor DNA or cells are also being investigated [41,42].…”

“… 39 Other approaches including computational modeling with Deep Neural Networks and genomic assays such as Decipher have also shown promise. 16 , 40 Finally, liquid biopsies assessing minimal residual disease by detecting either circulating tumor DNA or cells are also being investigated. 41 , 42 However, their roles in specific subsets of patients with localized prostate cancer after prostatectomy is currently unclear, and we provide evidence for a discerning RPA using regularly available clinical and pathologic variables.…”

“…With the increased sensitivity of PSA detection, the definition of biochemical failure is controversial, as reflected by the various thresholds for PSA failure used in prospective trials. 14 , 15 , 16 Because a lower PSA at the time of salvage intervention has been correlated with improved bRFS, a first rising PSA > 0.1 ng/mL (followed by confirmation with a second PSA) or initiation of salvage intervention for rising PSA was chosen as the definition of biochemical failure rather than 0.2 ng/mL. 17 , 18 bRFS was measured from date of prostatectomy to date of failure.…”

Single-Center Experience of Focal Thermo-Ablative Therapy After Pelvic Radiotherapy for In-Field Prostate Cancer Oligo-Recurrence

“…Associating the PSA-DT, biopsy Gleason score and interval from primary therapy to biochemical failure could further stratify patients according to recent recommendations, notably to select patients benefiting from early ADT initiation after non-metastatic PCa relapse (49,50). In this matter, advances in PCa genomics or even radiomics and artificial intelligence could also generate new hopes with regard to personalized medicine (51)(52)(53). Tumor mutational profiles, such as driver mutations in TP53 or alterations in other tumor suppressor genes, could be associated with disparate outcomes among oligo-metastatic PCa, possibly identifying in the near future patients with aggressive features who may benefit from intensified treatment (54)(55)(56).…”

Post-Operative Radiotherapy in Prostate Cancer: Is It Time for a Belt and Braces Approach?

“…Simultaneously, artificial intelligence-guided tools are rising, taking into account large volumes of heterogeneous data to produce effective prediction models. For instance, using multicentric data from 5,043 patients, a deep learning model was able to predict 3-year BCR with an area under the curve of 0.70 (0.84 when adding post-operative data), surpassing the more conventional CAPRA-S recurrence risk score (AUC 0.63) ( 63 ). This approach could allow closer monitoring and early treatment of high-risk patients as pre-defined by the model.…”