July 2019 Edition Vol.11, Issue 7

Can Real-World Data Improve Patient Care?

By Lynne Lederman, PhD

In keeping with the theme of this year’s ASCO Annual Meeting—Caring for Every Patient, Learning from Every Patient—there was a special Clinical Science Symposium entitled, “Using Real-World Data to Advance Research and Care.”

Broader Enrollment Criteria Could Double Eligible Patients with NSCLC

Donald Harvey, PharmD, Winship Cancer Institute, Emory University, presented abstract LBA108, Impact of Broadening Clinical Trial Eligibility Criteria for Advanced Non-Small Cell Lung Cancer (NSCLC) Patients in a Real-World Analysis.1

This study was a retrospective electronic health record (EHR) review using ASCO’s CancerLinQ database records from 2011 to 2018 to determine the effect of broadened clinical trial inclusion criteria on potential clinical trial enrollment of adults with advanced NSCLC.

The broadened inclusion criteria looked at 3 domains: allowing individuals with brain metastases; with another previous or concurrent cancer diagnosis; or with reduced renal function measured by a creatinine clearance as low as 30 mL/min. Traditional criteria exclude brain metastases, prior or concurrent cancer diagnosis, and creatinine clearance <60 mL/min.

Of 10,500 individuals with advanced NSCLC who were identified, 47.7% (5,005 people) would not meet standard trial eligibility criteria. Traditional criteria would exclude 21.2% patients due to brain metastases, 21.5% due to prior concurrent cancer, and 14.4% due to creatinine clearance <60.

Adoption of expanded criteria would exclude only 1.5% (154 people), expanding the potential trial population by 4,851 individuals and including more patients age ≥75 years, thereby nearly doubling the potential NSCLC trial population.

Dr. Harvey concluded, “These data show that in a community oncology population of patients with advanced NSCLCs, expanding eligibility in 3 common criteria can substantially expand those who could enroll in trials, leading to more generalizable information and potentially speeding new therapies to those who need them most.”

ASCO and Friends of Cancer Research have recommended broadening criteria, and more recommendations are expected. The FDA, NCI, and other groups are leading by an example to drive the change to eligibility criteria. For example, the ASCO TAPUR study is using these broader criteria.

Sumithra J. Mandrekar, PhD, Mayo Clinic, discussed this abstract, asking can we capture every patient? She referred to the approval of palbociclib for the treatment of men with metastatic breast cancer—a rare condition that is not well studied in clinical trials—as an example of the use of real-world data from EHRs and post-marketing data to expand the indication of an already-approved therapy.

Dr. Mandrekar made a distinction between “real-world data,” and “real-world evidence.” Real-world data relates to patient health status or delivery of health care routinely collected from EHRs, billing registries, in-home use settings, and the use of mobile devices, etc. Real-world evidence is gathered using data from randomized clinical trials, observational studies, and other trials on medical product usage and potential benefits.

She prefers the term “point-of-care research” to “real-world data.” Advantages of point-of-care research over traditional clinical trials include the efficiency of a single-data system, preservation of data from both protocol-eligible and non-eligible patients, long-term follow-up past-trial completion, and the potential for use in additional studies.

The challenge is that these data are heterogeneous, EHR systems are not interoperable, outcome data are lacking, and data collection adds an additional burden on clinicians. Nevertheless, point-of-care research data have the potential to answer questions about long-term survivors and rare cancers.

Real-World Data for Predictive Survival Model in NSCLC

Nathanael Fillmore, PhD, Associate Director for Machine Learning and Predictive Analytics, VA Boston Cooperative Studies Program Informatics Center, presented a predictive model for survival in NSCLC based on real-world EHR and tumor sequencing data at the department of Veterans Affairs (VA).2 The objective was to create a model that could make accurate individualized survival predictions in newly diagnosed NSCLC.

This pilot program for future models to provide decision support, e.g., concerning therapy for VA clinicians, demonstrated the use of the VA Precision Oncology Data Repository (PODR) for integrative outcomes analysis, and to provide more accurate individualized survival predictions for veterans with NSCLC.

VA PODR includes longitudinal EHR data from the VA Central Cancer Registry and targeted tumor sequencing from the VA precision oncology program. This report included patients (n=356) enrolled in the VA Precision Oncology Program between 2015 and 2017 with a lung cancer primary tumor site and histology consistent with NSCLC.

The predictive model included 48 features, reflecting 18 baseline clinical and demographic characteristics defined from the EHR registry, and the presence or absence in 96 genes in screening panels. A random forest machine learning algorithm was trained to predict 1-year survival and was deemed effective.

The features playing the biggest role include age at diagnosis, tumor size and stage, the number of mutations, and the presence of variations in TP53, KRAS, and STK11. Incorporation of ICD codes for acute and chronic comorbid conditions that occurred in the year prior to diagnosis and associated with visits to the VA hospitals improves performance of the predictive model substantially.

Limitations include a potential lack of generalizability beyond the veteran population or even the VA PODR data set. Information about specific treatment regimens are not included in the model yet, but exist in the VA PODR data set, so could be included in the future. Mortality prediction is of limited use, and a predictive model to guide treatment would be more useful.

Robert Charles Doebele, MD, PhD, University of Colorado Cancer Center, discussed Dr. Fillmore’s presentation. Big data can identify gaps in current knowledge, identify areas that can be targeted to improve the outcomes of our patients, confirm clinical trial data in ineligible populations, generate evidence that cannot be obtained via clinical trials, and used to detect failure to implement guidelines and administer recommended therapies.

He noted that the VA model looked at the presence or absence of mutations, which would not capture effects of, e.g., oncogenes versus tumor suppressors. Tumor mutations should be linked to specific therapies.

“It’s one thing to characterize treatment therapies, such as chemotherapy, radiation, and surgery, which were captured in the VA database. I think the next step will be linking targeted therapies, since they’re critically important in NSCLC,” Dr. Doebele said.

The inclusion of the use of checkpoint inhibitors and biomarkers, like tumor mutational burden, would probably improve the model.

No Effect of Autoimmune Disease on Outcome After Checkpoint Inhibitor Treatment of NSCLC

Sean Khozin, MD, MPH, Oncology Center of Excellence, FDA, discussed real-world outcomes of patients with advanced NSCLC with or without autoimmune disease (AD) receiving immune checkpoint inhibitors,3 in a retrospective, observational cohort study in patient with advanced NSCLC using ASCO’s CancerLinQ database. Symphony claims data were linked to the CancerLinQ EHR by privacy-preservation tokenization to obtain additional data. Sampling was from January 2011 to November 2018.

Of 2402 patients with NSCLC treated with immune checkpoint inhibitors, 22% (n=531) had AD. PD-1 inhibitors accounted for most of the checkpoint inhibitors used. There was no association between AD status and outcomes, including overall survival, time to treatment discontinuation, time to next treatment, real-world progression-free survival, or incidence of adverse events (AE) other than higher rates of select AE, including endocrine, gastrointestinal, and blood and lymphatic disorders in those in the subgroup with active AD.

Dr. Khozin said, “It’s the real world performance and impact of cancer therapies that counts the most.”

David Kozono, MD, PhD, Dana-Farber Cancer Institute, discussed Dr. Khozin’s presentation. He asked if we are ready to use big data to identify outcomes of uncommon populations. The problem is that only 3% of individuals in the US who are diagnosed with cancer will participate in clinical trials, that number will never increase to 100%, and trials will not duplicate real-world patients.

Patients with both cancer and AD represent a great example of an “uncommon” population, although both cancer and autoimmunity comprise many diseases. Khosin’s study was an “excellent example of the use of real-world data to identify outcomes in an uncommon population.”

There are ethical concerns with database studies like the ones presented here. Can patients be re-identified, particularly when genetic data are included? What are the implications of sharing data with drug companies and insurers? These studies can’t completely replace clinical trials. Big data and clinical trials taken together will enable evidence-based medicine that can be applied to most patients.

Looking Ahead

Audience interest was reflected in a Q&A session that went over the allotted session time. There are many unanswered questions including how to improve data collection. Dr. Fillmore said the original databases were “very messy,” and required time-consuming manual review. The mCODE initiative “may be an opportunity for us to fill in some of those gaps,” he said.

mCODE (Minimal Common Oncology Data Elements) is a collaboration among ASCO, CancerLinQ LLC, the MITRE Corporation, and the Alliance for Clinical Trials in Oncology Foundation (Alliance Foundation) to achieve data interoperability.

mCODE is a set of basic data that would populate all EHRs for patients with cancer. The data elements would be sufficiently clear so that a query related to a particular patient would return a cohort of matched, de-identified patients from a larger database.

The initial set of common cancer data standards and specifications were released at the 2019 ASCO Annual Meeting and are available free at Accessible for download at mCODEinitiative.org.


References

1.     Harvey RD, Rubinstein WS, Ison G, et al. Impact of broadening clinical trial eligibility criteria for advanced non-small cell lung cancer patients: Real-world analysis. J Clin Oncol 37, 2019 (suppl; abstr LBA108).

2.     Fillmore N, Ramos-Cejudo J, Cheng D, et al. A predictive model for survival in non-small cell lung cancer (NSCLC) based on electronic health record (EHR) and tumor sequencing data at the Department of Veterans Affairs (VA). J Clin Oncol 37, 2019 (suppl; abstr 109).

3.     Khozin S, Walker MS, Jun M, et al. Real-world outcomes of patients with advanced non-small cell lung cancer (aNSCLC) and autoimmune disease (AD) receiving immune checkpoint inhibitors (ICIs). J Clin Oncol 37, 2019 (suppl; abstr 110).

Post a Comment

OBR Archives

To view previous issues of OBR green you can visit our archives. The entire library of OBR green articles is searchable.