Validation exercise: Advancing the Drug Development Process in Metastatic Prostate Cancer through Machine Learning

Validation exercise: Advancing the Drug Development Process in Metastatic Prostate Cancer through Machine Learning

Despite the emergence of a number of new drugs for use in patients with metastatic prostate cancer (mPC), disease which has spread from its origin in the prostate gland to seed new tumors in other sites of the body, the disease remains lethal. Unfortunately, promising new agents can be slow to become approved for use. One of the reasons is that for regulatory agencies, the standard to demonstrate that a new agent is beneficial is measurable prolongation of life, known as average Overall Survival (OS). Determining the average OS for patients enrolled on a clinical trial can take years and investigators have long sought markers to be predictive of OS for a shorter time to regulatory approval. A decline in serum prostate serum antigen (PSA) in response to therapy, while welcomed as evidence of anti-cancer activity, is not directly predictive of OS to be used as a shorter-term marker. However, the availability of techniques in “machine learning,” computational methods which apply statistical models to infer patterns in large sets of data, provides the opportunity to test billions of combinations of PSA characteristics reflective of complex PSA changes over time, and to identify or “learn” patterns associated with survival outcomes. Machine learning can aggregate complex data with a goal of identifying dynamics early in the course of the disease which are predictive of late outcomes. We have assembled data, including over 250,000 PSA measurements from 7,265 mPC cancer patients enrolled on eight clinical trials, in which patients were assigned to a new treatment or to the standard of care. These are completed trials for which the OS outcome is known. If the new treatment provides an improvement in OS it is deemed a “positive” trial, and if the new treatment is not better than the standard of care, it is called a “negative” trial. Computational modeling has been applied to PSA measurements obtained over the course of these trials to ask whether PSA measurements acquired early in a trial, can predict the ultimate outcome of that trial at its standard completion point, typically 5-10 years later. We have developed a computational method to “simulate” the trajectory of a clinical trial for patients with mPC, using PSA data collected within just the first four months of a trial period to then accurately predict OS in completed therapeutic trials. To date we have been able to accurately predict outcomes (both positive and negative) for multiple completed studies, testing different types of mPC treatments, including Androgen Receptor Signaling Inhibitors (agents acting against prostate cancer growth by interfering with androgen receptor signaling), a PARP inhibitor agent (for disease involving DNA repair biology), and chemotherapy. To further evaluate this model's applicability across classes of therapeutic agents for mPC, the VISION study offers a unique opportunity to include the radioligand class of therapy - systemic treatments in which cancer-killing radioactive compounds are targeted to sites of cancer by molecular engineering methods. The VISION study was a groundbreaking study demonstrating that Pluvicto had a positive impact on average OS. Pluvicto is in a distinct drug class from the agents in trials whose data have validated performance of our model so far. VISION study data would allow the simulation method we have developed to be tested more broadly and serve our high-level goal of accelerating the timeline for validation and approval of new therapies for metastatic prostate cancer. In very practical terms, if further validated, the computational approach we have developed could short-circuit the conduct of clinical trials and identify successful new approaches to treat mPC within a few months which otherwise could take up to a decade. Ultimately, we hope that this tool will help fast-track the availability of effective new therapies for all patients with mPC.

[{ "PostingID": 21191, "Title": "NOVARTIS-CAAA617A12301", "Description": "Study of 177Lu-PSMA-617 In Metastatic Castrate-Resistant Prostate Cancer (VISION)" }]