Novel AI platform could help identify patients most likely to benefit from clinical trials

Researchers have demonstrated that a new artificial intelligence (AI)-based platform can help doctors and patients assess the benefits of specific treatments being tested in clinical trials. According to a recent study published by Orcutt et al. natural medicine. AI platforms can help make informed treatment decisions, understand the expected benefits of new treatments, and plan future care.

background

Clinical trials of potential new treatments can be limited because less than 10% of cancer patients participate in clinical trials. As a result, clinical trials do not represent all patients with that type of cancer. Even if clinical trials show that a new treatment strategy produces better outcomes than standard treatments, there are many patients for whom the new treatment will not be effective.

Research methods and results

In the study, researchers developed TrialTranslator, a machine learning framework designed to translate clinical trial results to real-world populations. By emulating 11 groundbreaking cancer clinical trials using real-world data, we are able to reproduce the results of real-world clinical trials, making it easier to understand which individual patient groups are affected by clinical trials. It is now possible to identify which groups respond well to treatment and which groups do not.

Researchers used Flatiron Health's national database of electronic medical records to determine what is considered standard of care for four of the most prevalent advanced solid malignancies in the United States: advanced non-small cell lung cancer and metastatic lung cancer. We emulated 11 landmark randomized controlled trials investigating anti-cancer drug regimens. Breast cancer, metastatic prostate cancer, metastatic colorectal cancer.

They found that patients with low- and intermediate-risk phenotypes (machine learning-based traits used to assess a patient's underlying prognosis) had similar survival outcomes to those observed in randomized controlled trials. revealed to have a duration- and treatment-related survival benefit. . In contrast, patients with high-risk phenotypes showed shorter survival and lower treatment-related survival benefit compared with patients in randomized controlled trials.

“With this framework and our open source calculator, we are able to [physicians] “This is to determine whether the results of the Phase III clinical trial can be applied to individual cancer patients,” the co-senior study authors explained. Ravi B. Parikh, MD, MPP, FACPHe is an associate professor in the Department of Hematology and Medical Oncology in the School of Medicine and medical director of the Data and Technology Applications Shared Resource at Emory University's Winship Cancer Institute. “[T]”His research provides a platform for analyzing the real-world generalizability of other randomized trials, including trials with negative results,” he continued.

conclusion

Their findings showed that machine learning can identify groups of real-world patients to whom the results of randomized controlled trials are difficult to generalize.

“[R]”The prognosis of real-world patients is likely to be more heterogeneous than that of participants in randomized controlled trials,” the study authors noted. “[The study] “This suggests that patient prognosis, rather than eligibility criteria, is a better predictor of survival and treatment response,” the researchers emphasized.

The researchers recommend that prospective clinical trials “should consider more sophisticated methods of assessing patient prognosis at the time of entry, rather than relying solely on strict eligibility criteria.”

Additionally, they echoed the recommendation by ASCO and Friends of Cancer Research that “efforts should be made to improve the representation of high-risk subgroups in randomized controlled trials, given the effects of treatment on these patients.'' Quoted. [patients] may be different from other participants,” the study authors emphasized.

“This AI platform [physicians] And it is up to the patient to decide whether the results of a clinical trial are applicable to the individual patient,” Dr. Parikh emphasized. “Furthermore, this study could help researchers identify subgroups where new treatments may not work, potentially facilitating new clinical trials for these high-risk groups.”

“Our work demonstrates the huge potential of leveraging AI.[machine learning] Harnessing the power of rich yet complex real-world data to advance precision medicine at its best,” the co-senior study authors emphasized. Dr. Chi Longprofessor of biostatistics and computer and information science, founding director of the University of Pennsylvania Cancer Data Science Center, and associate director of quantitative data science at the University of Pennsylvania Abramson Cancer Center.

“With proper oversight and evidence, pathology, radiology, or electronic medical record information could soon be analyzed to predict whether a patient will respond to a particular treatment or to diagnose cancer earlier. There will be a growing trend of AI-based biomarkers that can help diagnose cancer or diagnose cancer earlier, resulting in better patient prognosis.'' Dr. Parikh concluded.

Disclosure: Research for this study was supported by a grant from the National Institutes of Health. For full research author disclosure, please visit: nature.com.