Author Correction: Prostate cancer therapy personalization via multi-modal deep learning on randomized phase III clinical trials

Andre Esteva, Jean Feng, Douwe van der Wal, Shih-Cheng Huang, Jeffry P. Simko, Sandy Devries, Emmalyn Chen, Edward M. Schaeffer, Todd M. Morgan, Yilun Sun, Amirata Ghorbani, Nikhil Naik, Dhruv Nathawani, Richard Socher, Jeff M. Michalski, Mack Roach, Thomas M. Pisansky, Jedidiah M. Monson, Farah Naz, James Wallace, Michelle J. Ferguson, Jean-Paul Bahary, James Zou, Matthew P. Lungren, Serena Yeung, Ashley E. Ross, Michael J. Kucharczyk, Luis Souhami, Leslie Ballas, Christopher A. Peters, Sandy Liu, Alexander G. Balogh, Pamela D. Randolph-Jackson, David L. Schwartz, Michael R. Girvigian, Naoyuki G. Saito, Adam Raben, Rachel A. Rabinovitch, Khalil Katato, Howard M. Sandler, Phuoc T. Tran, Daniel E. Spratt, Stephanie Pugh, Felix Y. Feng, Osama MohamadNpj Digit. Medicine 2023 · DOI

Introduction

Personalizing therapy in oncology remains a significant challenge, particularly in prostate cancer where treatment intensity must be carefully balanced against potential side effects. Traditionally, clinicians rely on established guidelines, such as those from the National Comprehensive Cancer Network (NCCN), to categorize risk and guide therapy. However, these methods often lack the precision required for truly individualized care.

This correction addresses a critical data curation issue in a previously published study that utilized Multimodal Artificial Intelligence (MMAI) to improve prognostic predictions. The original research aimed to demonstrate that AI could outperform standard clinical guidelines by analyzing pretreatment biopsy samples. However, it was later discovered that some post-treatment tissue had been inadvertently included in the validation dataset. This correction provides a re-validation of the MMAI model on a "clean" subset of 931 patients, ensuring that the model's predictive power is evaluated in a strictly pretreatment clinical scenario.

Methods and Data Curation

The study employs a Multimodal Artificial Intelligence (MMAI) framework. This approach integrates deep learning analysis of digital pathology scans—specifically, the morphology of prostate tissue slides—with standard clinical variables. By combining these disparate data types, the model aims to capture complex biological signals of disease progression that are often invisible to the human eye or poorly captured by traditional grading systems.

The primary focus of this correction is a sensitivity analysis conducted on a refined validation cohort. The original validation set consisted of 1,109 patients. After a rigorous audit, the authors filtered out cases that did not have strictly pretreatment slides available, resulting in a corrected validation cohort of 931 patients. This step was essential to eliminate "data leakage," where information from the future (post-treatment state) might artificially inflate the model's performance. The model was developed and validated using data from randomized phase III clinical trials, providing a high level of evidence for its potential clinical utility.

Key Results

The re-evaluation on the corrected cohort confirmed that the MMAI model significantly outperforms NCCN guidelines across all primary clinical endpoints. The model demonstrated a consistent and statistically significant "lift" in the Area Under the Curve (AUC) for predicting long-term outcomes, including distant metastasis and overall survival.

As shown in the performance summary below, the MMAI model achieved an AUC of 0.83 for 5-year distant metastasis, compared to 0.72 for the NCCN guidelines. The improvement (Delta) remained substantial across all metrics, ranging from 0.06 to 0.11. Notably, the comparative test $p$-values were less than 0.001 for nearly all outcomes, reinforcing the statistical robustness of the AI's predictive advantage.

Table 1: Performance comparison between MMAI and NCCN guidelines across various clinical outcomes in the corrected validation cohort (n=931).

Discussion and Clinical Implications

The most significant takeaway from this correction is the robustness of the MMAI model. Despite the removal of post-treatment samples and the resulting reduction in cohort size, the model's performance remained stable and superior to standard care guidelines. This suggests that the AI is capturing fundamental biological signals of disease aggressiveness that are present at the time of initial diagnosis.

This correction also serves as a vital case study in the importance of rigorous data auditing and transparency in medical AI. Ensuring that "pretreatment" models are validated on strictly pretreatment data is essential for building clinical trust and meeting regulatory requirements.

While these results are promising, the authors acknowledge the need for further external validation. Testing the model on diverse, real-world data from community hospitals—rather than controlled clinical trials—is the next necessary step. Ultimately, the goal is to integrate these MMAI scores into the standard clinical workflow, providing oncologists with a powerful tool to better personalize treatment for patients with prostate cancer.

Summary updated

BibTeX
@article{esteva2023author,
  author = {Andre Esteva and Jean Feng and Douwe van der Wal and Shih-Cheng Huang and Jeffry P. Simko and Sandy Devries and Emmalyn Chen and Edward M. Schaeffer and Todd M. Morgan and Yilun Sun and Amirata Ghorbani and Nikhil Naik and Dhruv Nathawani and Richard Socher and Jeff M. Michalski and Mack Roach and Thomas M. Pisansky and Jedidiah M. Monson and Farah Naz and James Wallace and Michelle J. Ferguson and Jean-Paul Bahary and James Zou and Matthew P. Lungren and Serena Yeung and Ashley E. Ross and Michael J. Kucharczyk and Luis Souhami and Leslie Ballas and Christopher A. Peters and Sandy Liu and Alexander G. Balogh and Pamela D. Randolph-Jackson and David L. Schwartz and Michael R. Girvigian and Naoyuki G. Saito and Adam Raben and Rachel A. Rabinovitch and Khalil Katato and Howard M. Sandler and Phuoc T. Tran and Daniel E. Spratt and Stephanie Pugh and Felix Y. Feng and Osama Mohamad},
  title = {{Author Correction: Prostate cancer therapy personalization via multi-modal deep learning on randomized phase III clinical trials}},
  journal = {Npj Digit. Medicine},
  year = {2023},
  doi = {10.1038/s41746-023-00769-z},
}
Figure