Look behind the lecture: AI and real-world evidence in cancer care

In this interview, Jeff Elton (Concerto HealthAI, MA, USA) discusses his presentation from ASCO (31 May—4 June, IL, USA) on the impact of AI and real-world evidence on cancer care.

Go to the profile of The Evidence Base
Jun 27, 2019

Please introduce yourself and your institution.

I am the CEO of Concerto HealthAI (MA, USA), a precision oncology company providing real-world evidence (RWE) solutions and technologies for life science companies, health providers and health payers. We believe the future of precision medicine, and precision oncology in particular, is the concept of ‘Precision Evidence’. Concerto HealthAI builds real-world data (RWD) products designed for specific use-cases ranging from regulatory submissions, post-approval studies enabling label modifications, broadened market access and supporting patient-centered solutions. With our artificial intelligence (AI) models, tools, and AI-enabled applications, we hope to enrich and expand the utility of RWD, generate accelerated insights into rare cancers, and generate faster, more precise insights.

What are the challenges of assessing treatment outcomes in cancer patients with autoimmune diseases?

Immune checkpoint inhibitors – a new class of immuno-oncology drugs – have significantly improved survival for many patients with various cancers, including advanced melanoma and advanced non-small-cell lung cancer. Patients with autoimmune diseases, however, have been excluded from clinical trials of these agents because of an assumption that these drugs might significantly worsen autoimmune diseases and overall outcomes in these patients.

"...a recent study used RWD [to show that] patients with autoimmune diseases achieve the same benefits from treatment with immune checkpoint inhibitors as patients without autoimmune diseases."

Our use of RWD has changed that; a recent study used RWD to analyze patients with autoimmune diseases who had been prescribed an immune-oncology agent in a real-world cancer care setting. This allowed researchers to study the validity of the assumption that immune-oncology agents would worsen outcomes for patients with autoimmune diseases and assess the effectiveness and safety of immune checkpoint inhibitors in treating these patients.

Preliminary findings show that patients with autoimmune diseases achieve the same benefits from treatment with immune checkpoint inhibitors as patients without autoimmune diseases; patients with autoimmune diseases can safely benefit from this new class of agents. This work, done collaboratively with the American Society of Clinical Oncology (ASCO) and the FDA, has further illustrated the utility of Concerto HealthAI’s RWD products, analytic solutions and Outcomes Sciences services. 

As patient populations get smaller, and greater numbers of diseases are classed as ‘rare’, why is it important to have validated and precise RWE for use in decision making?

As patient populations become smaller and more diseases are classified as rare, it is increasingly critical that underlying real-world datasets are population-scale and representative enough to generate RWE with a high level of veracity and confidence in the patient populations being studied. This will allow for actionable and relevant insights to be provided.

"...real-world datasets [that] are population-scale and representative enough...will allow for actionable and relevant insights to be provided."

Further, as rarer cancers mean smaller patient populations, patient recruitment into clinical trials becomes an even greater challenge. Rapid-query, large, multi-source-linked and population-scale datasets allow researchers to quickly see the impact of eligibility criteria on patients with a specific tumor type. As a result, clinical studies can be re-designed efficiently to focus study protocols on achieving overall results with the greatest utility for regulatory decisions and clinical practice.

How can machine learning be applied to RWD/RWE?

We are applying AI and machine learning technologies across the clinical pipeline, from preclinical discovery to clinical development, to market access and commercial use cases. For clinical research, Concerto HealthAI applies AI in a few different ways. Unsupervised learning models help identify previously undetected patient cohorts to inform new study concepts focused on the highest responders, as well as those not benefiting from the latest medical innovations.

Semi-supervised learning approaches can further refine patient cohorts and study concepts by identifying the critical clinical characteristics of new sub-populations that can inform a new research design or treatment management approach. Definitive RWD AI algorithms streamline analyses and can be specific to a tumor type or hematological malignancy – they are derived from the collaborative work of our life science partners that include Friends of Cancer Research, FDA, ASCO and others.

"...large, population-scale datasets [can allow] clinical studies to be re-designed efficiently to focus study protocols on achieving overall results with the greatest utility for regulatory decisions and clinical practice."

We are the leading company to apply these approaches to external control arms. Our predictive AI and machine learning models are able to identify patients where their cancers may progress rapidly, negative responders and those for whom treatments are highly durable and effective. Together, this set of industry-leading AI and machine-learning approaches supports the work of clinical researchers and healthcare providers who are establishing new study concepts and/or new patient cohort groupings and will hopefully become a foundational enabler of precision evidence for truly precision oncology.

Why is it important to be able to understand treatment utilization patterns?

Drug approvals are often based on evidence collected from randomized clinical trials (RCT). Treatment outcomes may differ when these drugs are used in real-world practice. This could be due to individual differences between patients, dosage, treatment sequencing or other factors. Use of RWD to evaluate treatment utilization patterns provides clarity on the factors that contribute to best practices and optimal treatment outcomes.

Recent guidance from FDA outlined various new clinical research approaches intended to increase the generalizability of RCT studies to community oncology practitioners – doing this requires understanding the current standards of care and how they might be evolving. Together these approaches and innovations in clinical study designs and intent will advance new innovations and new treatment approaches for patient benefit.

"Treatment outcomes may differ when drugs are used in real-world practice."

What are the challenges of utilizing machine learning and AI in the RWE space?

AI is not yet part of the front-line of treatment decisions for patients, nor is it replacing expert clinicians, epidemiologists or regulators in interpreting study results. That said, AI and machine learning approaches are increasingly part of the field of RWD. Natural language processing approaches augment human clinical expertise in making unstructured data available for machine analysis. Under Software as a Medical Device (SaMD), FDA has allowed software device approvals based on retrospective RWD with machine learning algorithms. AI and machine learning have well established places in supporting complex analyses in very large data sets or extracting more ‘signal’ from smaller and more complex data. If we use the progress in digital pathology and digital radiology, we should expect to see AI and machine learning models and analyses that streamline and speed up solutions, inform new hypotheses that extend beyond the published literature and increasingly augment decision-making for greater precision and confidence. 

"...AI and machine learning models and analyses [should] streamline and speed up solutions, inform new hypotheses that extend beyond the published literature and increasingly augment decision-making for greater precision and confidence."

How does this research fit into the FDA’s vision of increased application of RWE in submissions?

The FDA is encouraging the use of RWD and RWE in regulatory applications to bring patients and physicians important answers and treatments sooner. Concerto HealthAI shares this vision and has developed solutions in line with guidance provided by the FDA. Concerto HealthAI has focused on developing deeply abstracted datasets that follow the work of ASCO, Friends of Cancer Research and made recent submissions with RWD which the FDA has indicated is a suitable source of RWE for regulatory submission. In this process Concerto HealthAI often links data from different sources to provide a richer view of the patient treatment journey and the highest confidence of interpretation.

Our methods and approaches are always submitted for independent and peer-based review and assessments to assure the broadest confidence and utility. As oncology and other research focuses on smaller populations and finding a higher likelihood of beneficial response it becomes all the more important that real-world data has utility and veracity for regulatory, market access and treatment decisions. As Concerto HealthAI’s solutions make this kind of research possible, it will bring more patient benefit through innovative and faster research and treatments.

What do you think is the future of computer-based technologies in the RWE space?

The field of RWE is evolving rapidly. Concerto HealthAI sees that technologies for analyses and engineered RWD products need to evolve in lockstep. The nature of data fields and data sources integrated together need to be supported by the analytic environment in which these data are placed, for example, if electronic medical records, claims, genomic and patient-reported outcomes data are integrated together.

"...AI and machine-learning approaches...will hopefully become a foundational enabler of precision evidence for truly precision oncology."

Similarly, the models and the tools used to find patterns and glean insights need to know the types of data they can “consume.” Finally, if regulatory submission is the intent, there is a need to work within large datasets, but then reduce these to the selected cohort of interest with an approach to data provenance and integrity being established as part of that submission.

This requires good clinical practices being documented and in place and control from end-to-end for data products development and analyses conducted. Concerto HealthAI has adopted a few principles and approaches that we believe are unique and necessary for the field. First, data sources should be arrived at collaboratively, keeping the patient and need for new innovations at the forefront. Second, researchers and those providing services to patients need an array of new solutions and innovations – therefore any technologies need to use the latest open standards and microservices architectures to assure that the solutions from multiple companies can play together as a system for patient benefit. We are entering a period where precision evidence generation and utility will be accelerating – this will come through the data-technologies interplay and a very open and collaborative environment across company solutions.

"...data sources should be...keeping the patient and need for new innovations at the forefront."

What were your highlights of ASCO?

1. Clinical trial eligibility research abstract presentation

This presentation was a significant highlight for us at ASCO 2019 as it utilized our RWD for a powerful analysis of the impact of inclusion/exclusion criteria on clinical trial recruitment and design and led the authors to “urge all clinical trial sponsors to adopt these criteria.” This abstract, “Impact of broadening clinical trial eligibility criteria for advanced non-small cell lung cancer patients: Real-world analysis,” was also a part of ASCO 2019’s Press Program, which highlights the top studies presented that year.

Researchers conducted a RWD analysis of patients with advanced non-small cell lung cancer that demonstrate broadening clinical trial eligibility criteria would increase by about 46% the number of patients who could participate, and potentially bring significant benefit to patients.

2. Autoimmune research abstracts

Concerto HealthAI RWD was used to study patients with evidence of autoimmune diseases who are frequently excluded from clinical trials. A series of four research abstracts, through a collaboration between ASCO and the FDA, found that such patients who received immuno-oncology treatments in the real world had similar outcomes to their counterparts without preexisting autoimmune disorders.

The main study, “Real-world outcomes of patients with advanced non-small cell lung cancer (aNSCLC) and autoimmune disease (AD) receiving immune checkpoint inhibitors (ICIs),” was presented at an oral session by lead author Sean Khozin, FDA.

3. AI model for predicting survival

Our validated, peer-reviewed AI model was presented at ASCO 2019 during a poster session. This model predicts survival of lung cancer patients 3-12 months from their last clinical visit to a high degree of accuracy. Results from this AI model were significantly better than a baseline Cox-PH model and compared very favorably to other survival models in the literature created using AI and comparable machine learning techniques.

Concerto HealthAI’s model is useful for assessing patient risk and treatment options, along with evaluating cost and quality of care or determining trial eligibility. Derived from 55,000 patient electronic medical records, this AI model will allow clinical researchers to gain deeper insights into key variables impacting patient survival and greatly improve how they design studies.

Go to the profile of The Evidence Base

The Evidence Base

Community, Future Medicine

The Evidence Base is a community site covering the latest news, opinion and insight into the collection and application of real-world data to real-world problems.
56 Contributions
0 Following

No comments yet.