How AI-Powered Analytics and Real-World Evidence Can Improve the Precision of Prostate Cancer Treatment

Analysis of Unstructured EHR Notes Can Help Transform Prostate Cancer Research

Prostate cancer has one of the highest five-year relative survival rates (97.1%), according to aggregate stage data from the SEER (Surveillance, Epidemiology, and End Results) database¹. That’s the good news. The not so good news is that many of the methods used to treat prostate cancer come with risks of long-term side effects including skeletal disease, cardiovascular disease, loss of libido, erectile dysfunction, urinary incontinence, hot flashes, and more – all of which can negatively impact quality of life. 

Worse, in many cases, some of the most aggressive forms of treatment were found to be unnecessary. A study, based on data from the CAPSURE prostate cancer registry, found that 92-98% of patients with the lowest tumor risk scores (and presumably were eligible for a more conservative approach, such as active surveillance) were treated with prostatectomy, radiation, or hormone therapy². However, 10-20% of patients with prostate cancer developed more severe, castration-resistant forms of the disease³ and 44% of patients, ages 45 to 74, developed metastatic prostate cancer⁴. Based on these results, it can be difficult to determine which patients should receive a more conservative course of treatment and which should receive the most aggressive form. Many clinicians and patients choose to err on the side of over-treatment in order to get a definitive result.

Today, thanks to advancements in analytics, using real-world evidence (RWE), researchers are able to pinpoint critical moments and detailed symptoms and sequences of treatment during the patient journey that signal the best course of treatment for each unique patient. 

But, what if urologists, oncologists, primary care clinicians, and life sciences researchers could take a more precise, patient-specific approach to these types of treatment decisions? Today, thanks to advancements in analytics, using real-world evidence (RWE), researchers are able to pinpoint critical moments and detailed symptoms and sequences of treatment during the patient journey that signal the best course of treatment for each unique patient. 

Artificial Intelligence Enables Unprecedented Insight Into the Prostate Cancer Patient Journey

It all starts with being able to mine the details of the patient experience at scale, extracting key information on patterns of patient behavior, unstructured data in clinical notes and other criteria that provide a more detailed, nuanced view of the complete patient journey. Until recently, that had been incredibly difficult to achieve because much of the information required to inform that level of precision was buried in the clinical notes of individual urologists’ electronic health records (EHRs). 

For example, Prostate-Specific Antigen (PSA) levels and Gleason scores are primary measures used to identify disease severity and progression, but these measures are not captured in standard medical claims databases. As a result, attempts to extract meaningful insights from these unstructured datasets required labor-intensive, manual searches that were inefficient. The introduction of artificial intelligence (AI)-powered large language models has made it possible to continually track and analyze this type of unstructured data for hundreds of thousands of patient encounters from EHRs participating in the American Urological Association (AUA) Quality Registry (AQUA) to extract key information. By carefully curating that data to flag keywords and patterns of language consistent with certain clinical cues, it is possible to develop a new data-driven taxonomy for understanding disease progression, treatment and outcomes.

Verana Health has been able to capture approximately 5x more instances of metastasis when expanding from the explicit documentation of the TNM staging system to including more implicit types of evidence.

When applied to assessing disease severity and risk of metastasis in prostate cancer, not every clinician uses the well-known “TNM” staging system⁵, so Verana Health has been able to develop machine learning models that search for keywords in clinical notes. Terms, such as “growing sites of metastasis on scan,”  or “positive bone scan,” can provide critical clues that signal things like metastatic risk. Verana Health has been able to capture approximately 5x more instances of metastasis when expanding from the explicit documentation of the TNM staging system to including more implicit types of evidence⁶. Additionally, phrases such as “demonstrating evidence of androgen resistance” or “becoming hormone refractory,” can indicate development of castration-resistant forms of the disease. This offers a key insight that a therapy may no longer be effective for this patient and may influence a change in treatment. Similarly, by analyzing patterns of diagnosis and patient PSA levels over time, it is also possible to identify patients with localized cancer and evaluate treatment patterns and outcomes,  identifying possibly unnecessary treatments. Gleason score, PSA level, castration resistance and presence of metastasis are key variables that can be curated from unstructured clinical notes to study the prostate cancer journey at scale. 

This data can also be used to help researchers determine if patients are receiving treatments in line with AUA recommendations, based on disease severity. Additionally, they can track patient response to a prescribed therapy. 

RWE has the potential to help support and advance prostate cancer research and subsequent care in several ways, including:

1. Early detection and intervention – Real-world Data (RWD) analysis is helping to identify urinary biomarkers and pathology indicators that enable earlier disease detection and better risk stratification. Disease progression patterns are also informing earlier intervention for those with high-risk forms of the cancer, such as castration-resistant prostate cancer, helping reduce mortality. 
2. Treatment path validation – Data trends can be analyzed to verify that prescribed treatment plans, or active surveillance, are appropriate and effective in managing the disease. For example, tracking new evidence of metastasis throughout a course of treatment could help determine therapeutic efficacy. This data can help inform research that has the potential to modify future standards of care for those with advanced disease, helping to improve patient outcomes. 
3. Clinical trial recruitment – RWE also has the power to deliver insights that can help accelerate and improve prostate cancer trials by identifying urology practices treating patients with specific prostate cancer profiles. This can help advance the application of investigational therapies to patients while reducing the time and resources necessary to secure optimal trial sites.

About Real-World Data from the AQUA Registry

Verana Health has an exclusive partnership with the AUA to curate and analyze data from the AQUA Registry , which includes a 9-year longitudinal database of more than 10 million de-identified patients from 2,200 active clinicians. 

Verana Health takes the structured and unstructured AQUA Registry data a step further to inform research, transforming it—through the VeraQ^Ⓡpopulation health data engine—into high-quality,  research-ready, fit-for-purpose data modules, known as Urology Qdata^Ⓡ.

Qdata Prostate Cancer now includes multiple variables curated from unstructured notes, including PSA level, Gleason score, castration resistance, non-explicit mentions of metastasis and other key indicators of disease severity. 
For more information about Qdata Prostate Cancer and the curated variables available, click here.

1. National Cancer Institute -Cancer Stat Facts: Prostate Cancer ↩︎
2.  Overdiagnosis and Overtreatment of Prostate Cancer – Ian M. Thompson, Jr. MD – American Association of Clinical Oncology Education Book ↩︎
3. Development and prevalence of castration-resistant prostate cancer subtypes – Neoplasia – November 2020 ↩︎
4. Trends in incidence of metastatic prostate cancer in the US – Urology – March 14, 2020 ↩︎
5.  https://www.cancer.org/cancer/types/prostate-cancer/detection-diagnosis-staging/staging.html ↩︎
6. Data on file, Verana Health September 2023. ↩︎