October 15, 2024
The biopharmaceutical landscape is marked by continuous advancements, breakthrough discoveries, and relentless pursuit of new targeted treatments. But underlying the call for transformative treatments is the vital importance of accurately assessing how new pharmaceutical interventions affect patient health and wellbeing. Despite significant progress in comprehending complex diseases, the current approach to biomarker development remains simplistic, fragmented, and “linear.”
Consider how clinicians usually determine whether to treat a cancer patient with an immune-checkpoint inhibitor — e.g., an antibody to programmed cell-death protein 1 (PD-1) or its associated ligand (PD-L1). The basic assumption is that if a drug targets PD-L1, then measuring PD-L1 levels in a tissue sample will indicate whether a patient will benefit from that drug. The same logic has been applied to cytotoxic T-lymphocyte–associated protein 4 (CTLA-4), lymphocyte-activation gene 3 (LAG-3) proteins, and T-cell immunoreceptors with immunoglobulin and tyrosine-based inhibitory motif domains (TIGITs).
Available literature indicates that such an approach inadequately predicts treatment success (1–6). It also significantly limits drug developers’ and clinicians’ ability to understand patient biology, personalize treatment, monitor responses, identify early indications of relapse, and pinpoint other potential targets for therapy. Achieving genuine precision medicine will require adoption of a comprehensive, multicomponent strategy for biomarker development.
The need for such a transformation is especially evident in assessment of clinical outcomes, which serve as the ultimate measure of success or failure in biopharmaceutical development. The efficacy and safety of novel drugs are scrutinized rigorously through the lens of clinical outcomes, including elements such as treatment response, disease progression, patient survival, and overall clinical benefit. But as the industry grapples with increasingly complex diseases and diverse patient profiles, traditional approaches to evaluating clinical outcomes are proving to be insufficient.
Proliferation of increasingly sophisticated drugs has amplified the urgency for a shift in biomarker development, especially considering that diseases manifest differently across patients. We cannot expect to develop or apply novel drugs effectively based on oversimplistic biomarkers derived with linear thinking. That is especially true for immuno-oncology drugs, which are designed to target different “trails” in a patient’s immune system, affecting numerous biological pathways and cell types that, in turn, cause various reactions in patients. Unlike the one-size-fits-all model that the biopharmaceutical industry is used to, drug developers now require a comprehensive framework for assessing clinical outcomes, one that accounts for the multifaceted nature of patient immune systems, disease pathways, treatment responses, and resistance mechanisms.
This is where composite or multicomponent biomarkers enter the spotlight. Cancer, autoimmune disorders, and many chronic diseases exhibit variable biological pathways and intricate resistance mechanisms. Because one-dimensional biomarkers fail to capture the subtleties of such interactions, multicomponent biomarkers now emerge as a necessity. They could provide researchers and clinicians with holistic, multidimensional insights into a patient’s condition and potential for treatment. However, designing such biomarkers will require an analytical paradigm shift, and machine learning (ML) and artificial intelligence (AI) are coming to play a pivotal role in unraveling the intricate web of patient variability and treatment responses.
The vast amount of biomarker data generated from diverse sources, coupled with heterogeneity in patient demographics (e.g., sex and age), demands the ability to identify and interpret patterns beyond the capabilities of human analysts. Applied correctly, ML and AI offer a solution by leveraging sophisticated mathematical systems to integrate millions of data points and discern classifiers that are critically implicated with clinical outcomes. ML and AI could act as a lens, improving the resolution of what human analysts can discover using traditional methods such as histology, pathology, blood chemistry, and even genomics. By using AI, ML, and biostatistics, we can find patterns and pathways that were previously hidden.
Consider two patients, both 75-year-old males with advanced lung cancer that has metastasized to liver and bone tissues. Both patients have identical medical backgrounds. Samples of their tumors show PD-L1 scores of >50% and are negative for oncogenic mutations. Based on their clinical profiles, their physician starts them on the same guideline-approved regimen of immunotherapy as monotherapy.
Six months pass. One patient is doing well. His disease is shrinking, the metastases have disappeared, he feels great, and his clinical outlook is positive. The second patient remains very sick. He has shown no response to treatment, his primary tumor is growing, more metastases have been found, and he is deteriorating rapidly.
Unfortunately, existing tools failed to provide the required “resolution” to understand differences between these seemingly similar patients. For physicians, diagnostic limitations present daily challenges in delivering the best possible treatment to their patients. For patients, the lack of effective tools translates to a potential waste of time exactly when time is of the essence. Payors find themselves covering costs for expensive treatments that might or might not offer benefits to treated patients. And when pharmaceutical companies are trying to design successful clinical studies, the lack of resolution can lead to 50% efficacy rates with no insights into the reasons behind some patients’ lackluster responses. That rate is simply not high enough to demonstrate treatment efficacy, and drugs that achieve middling outcomes might end up failing during their clinical programs.
Such a situation is a losing battle for most if not all stakeholders. However, by designing a biomarker that enables identification of patient differentiators, we can optimize how clinicians choose first-line therapies and select patients for clinical trials. Both of those factors could improve success rates for late-stage clinical trials and create an ecosystem-wide win–win situation.
The convergence of AI, system biology, and bioinformatics signals a new era in diagnostic tools. Such technologies could help clinicians to differentiate seemingly similar patients and choose optimal treatments. Pharmaceutical companies could leverage such solutions to rescue groundbreaking drugs from clinical-trial failure: Identifying which patients are most likely to benefit from a drug could help in demonstrating effective treatment. Maybe that is true for only a subgroup of patients, but with such findings, companies still could save valuable drugs that otherwise might have been abandoned.
A study by King’s College London and the London School of Economics in the United Kingdom found that the majority of cancer drugs approved by the European Medicines Agency (EMA) did not improve patient survival or quality of life during clinical trials (7). Even therapies that did improve survival showed limited benefits, with patients surviving for an extra 5.8 months maximum (7). Given the high failure rate of oncology drugs in phase 3 trials (>85%) (8), improved matching of drugs and patients could enhance clinical outcomes and expand the market with effective treatments.
AI’s role extends beyond predicting drug efficacy; it also can forecast toxicity potential. Approved drugs have undergone rigorous safety testing and are relatively well tolerated, providing clear clinical benefit beyond their toxicity. However, we know that individual patients have unique genetic predispositions, exhibit variable physiological responses, and present other factors that influence susceptibility to adverse events. In some cases, adverse events can be debilitating and even fatal.
With its capacity to analyze vast data sets and discern intricate patterns, AI introduces a tailorable approach to understanding potential drug toxicity, marking another significant step toward personalized medicine. By integrating AI-driven toxicity predictions into treatment-selection processes, healthcare providers can make well-informed choices when prescribing medications. That would help providers to align treatment with a patient’s unique biological makeup, reducing the likelihood of adverse events. By anticipating risks, clinicians also can make timely adjustments in treatment plans to enhance patient safety. That is true precision medicine.
AI has emerged as a transformative tool. It can guide drug developers and clinicians away from simplistic thinking about biomarkers and therapy selection toward a more comprehensive and beneficial understanding of patient variability and treatment response. AI and ML hold immense promise in healthcare settings. In fact, health and biomedical applications probably represent the most important uses that we can find for these technologies. Advancements in AI and ML could help not only to enhance clinical outcomes, but also to save valuable biopharmaceuticals from premature abandonment, predict the likelihood of toxicity events, and save many lives.
1 Shen X, Zhao B. Efficacy of PD-1 or PD-L1 Inhibitors and PD-L1 Expression Status in Cancer: Meta-Analysis. BMJ 362, 2018: k3529; https://doi.org/10.1136/bmj.k3529.
2 Bravaccini S, Bronte G, Ulivi P. TMB in NSCLC: A Broken Dream? Int. J. Mol. Sci. 22, 2021: 6536; https://doi.org/10.3390/ijms22126536.
3 Havel JJ, Chowell D, Chan TA. The Evolving Landscape of Biomarkers for Checkpoint Inhibitor Immunotherapy. Nat. Rev. Cancer 19(3) 2019: 133–150; https://doi.org/10.1038/s41568-019-0116-x.
4 Indini A, Rijavec E, Grossi F. Circulating Biomarkers of Response and Toxicity of Immunotherapy in Advanced Non-Small Cell Lung Cancer (NSCLC): A Comprehensive Review. Cancers (Basel) 13(8) 2021: 1794; https://doi.org/10.3390/cancers13081794.
5 Gibney GT, Weiner LM, Atkins MB. Predictive Biomarkers for Checkpoint Inhibitor-Based Immunotherapy. The Lancet Oncol. 17(12) 2016: e542–e551; https://doi.org/10.1016/S1470-2045(16)30406-5.
6 Jørgensen JT. Companion Diagnostic Assays for PD-1/PD-L1 Checkpoint Inhibitors in NSCLC. Exp. Rev. Mol. Diagn. 16(2) 2016: 131–133; https://doi.org/10.1586/14737159.2016.1117389.
7 Davis C, et al. Availability of Evidence of Benefits on Overall Survival and Quality of Life of Cancer Drugs Approved by European Medicines Agency: Retrospective Cohort Study of Drug Approvals, 2009–13; BMJ 359, 2017: j4530; https://doi.org/
8 Wong CH, Siah KW, Lo AW. Estimation of Clinical Trial Success Rates and Related Parameters. Biostatistics 20(2) 2019: 273–286; https://doi.org/10.1093/biostatistics/kxx069.
Ofer Sharon, MD, is a physician and chief executive officer of OncoHost Ltd., Hamelacha Street 17, Floor 1, Binyamina-Giv’at Ada, Israel; https://oncohost.com. Please direct email inquiries to marketing and communications manager Lior Alperovich; [email protected].
You May Also Like
