In the News

Five Predictions For AI And Real-World Data In Oncology

Brigham Hyde

CEO and Founder of Precision Health AI. Partner at the Symphony AI Venture Fund.

As progress in oncology care is advancing faster than ever — especially due to the work of artificial intelligence (AI) and real-world data for clinical decision making — it can be difficult to distinguish what is hype and what will be realized in the near future.

At this year’s American Society of Clinical Oncology (ASCO) conference, the big trend of 2018 was “less is more.” (Full disclosure: Precision Health AI is a co-exclusive licensee of the ASCO CancerlinQ EMR data asset.) Two new clinical best practices made international headlines by recommending areas where people with cancer could safely avoid difficult, risky and costly treatment. The first being that most women in the early stages of breast cancer could skip chemotherapy with guidance from a diagnostic test; the second is that many people with advanced kidney cancer do not need surgery. Both studies were conducted through clinical trials, but I am confident that similarly minimalist findings are on the horizon using AI analysis of existing real-world data.

As the summer turns toward August, I’m thinking ahead to what should be a fascinating fall and winter season on multiple fronts. Here are my five big predictions for what will really change the oncology sector in the next 12 months:

1. The FDA will approve an AI oncology algorithm.

Or, at the very least, it will issue guidance on a clear path for such algorithms to be approved. We have seen increasing action from the FDA around algorithms already this spring. The agency approved the first AI software that can identify a disease (IDx-DR for diabetic retinopathy) without clinician involvement in April, followed shortly thereafter in May with an approval for an AI software that aids doctors in diagnosing wrist fractures (Imagen OsteoDetect).

Mike Miliard at Healthcare IT News reported: “With more machine learning tools expected to get the go-ahead soon, health systems should be partnering with vendors to supply data for better algorithms.”

2. Tumor radiology will use AI to predict a pre-tumor cancer diagnosis undetectable by current human radiologists.

Early and accurate cancer detection is crucial to both improving patient outcomes and reducing the cost of treatment. Last year, Japanese research showed that AI could be used to detect colorectal cancer with 86% accuracy before tumors become malignant and the cancer is much harder to treat. Google has been pursuing this kind of deep learning as well, looking at ways to detect cancer metastases on gigapixel pathology images. It’s just a matter of time before this technology is proven.

3. Real-world data will be used by the FDA to approve a new drug application (NDA).

Or, at least it will be used to remove contraindications or expand a label indication for an existing drug in the U.S. A study published last year in the Journal of the European Academy of Dermatology and Venereology took this a step closer to reality by using real-world treatment data on 317 patients treated for melanoma with the monoclonal antibody ipilimumab. The data showed the treatment was tied to a “40% reduced hazard of dying than those not receiving treatment after ipilimumab.” The real-world data is there and ready for the FDA to put to work.

4. Treatment recommendations and clinical interpretation reports from next-gen sequencing of tumor genomics will include evidence from real-world data, not just randomized control trials.

The first half of 2018 saw important changes to the reimbursement of genomic testing. Specifically, Foundation Medicine’s FoundationOne test received approval from the Centers for Medicare & Medicaid Services (CMS) under its new Advanced Diagnostic Laboratory Test (ADLT) status for reimbursement at $3,500 per covered test. To date, most commercial tests have included some sort of editorialized content from publications or clinical trials instead of real-world data. This new reimbursement makes tumor genomics mainstream and should create a boom in use of this real-world data for clinicogenomics in patient care. Tempus and Foundation Medicine are both well-positioned here because of their access to RWD and the improved delivery systems through which they provide it.

5. Predictive AI models will come to the physician point of care.

Insights without action are just numbers. Getting the power of AI analysis to providers will be the most important trend for the year ahead. Electronic health record (EHR) systems and other clinical cloud vendors are on the brink of including predictive AI data around adverse events, patient outcomes and optimal treatments plans. Currently, the fastest advances in this kind of predictive analytics are in improving EHR and billing documentation, but the clinical application is close and sped along by a shift to value-based care. Real-Time Oncology debuted an oncology treatment calculator at ASCO18 that potentially fits this mold, using 20,000-plus rules to calculate personalized treatment protocols for patients.

Of these five predictions, I am most excited to see the use of real-world data in new drug approvals — meaning that FDA approvals could better incorporate evidence from more diverse populations instead of small clinical trials with homogenous patients. This diversity will help physicians and patients really understand which treatments work and if they are worth it, ultimately improving outcomes and reducing costs. Inversely, I am slightly more skeptical about my final prediction on physician integration at the point of care, if only because EHR still remains the beleaguered gatekeeper to changing and new information in the exam room.

 

Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives

Get in touch to request a demo