By Deborah Borfitz 

June 14, 2022 | Wider availability and uptake of mammography screening globally among women under age 60 closely align with a threefold increase in their participation in breast cancer clinical trials over the past two decades, according to a new global analysis of 2,511,046 breast cancer patients from 4,674 clinical trials by Phesi, a data-driven provider of clinical development analytics products and solutions powered by artificial intelligence (AI). The numbers jumped most dramatically, from 39% to 87%, between the 2005-2009 and 2010-2014 time periods. 

Those demographics should be prompting sponsor companies to take a fresh look at how they are designing studies in terms of who should and could be included, according to Gen Li, Ph.D., CEO and founder of Phesi. More can be done to improve clinical trial diversity while being mindful of the duty to protect subjects from harm. “We should not be deterred unnecessarily [from expanding eligibility] if we don’t have the facts to support it.” 

Likewise, careful consideration needs to be given to the physicians engaged as investigators, he continues. In Poland, for instance, population screening tests have been introduced only recently relative to the U.S. and participation rates remain low. This translates into later detection and worse survival odds, which would inevitably impact treatment outcomes of these patients participating in clinical trials. 

But given the overall pattern of earlier detection globally, trials of therapies for more aggressive types of breast cancer in younger women can be expected since there is more opportunity to intervene, Li says. This younger demographic includes women still in their childbearing years whose exclusion isn’t necessarily warranted based on the balance of risks to benefits. 

The treatments under investigation may be personalized not only to patients’ specific cancer type, subtype, and stage but their genes and genomic profile of their tumor. Outcomes also tend to be highly treatment-specific, with HER2-positive breast cancer on the one hand being highly curable when aggressively treated with targeted therapies and triple-negative breast cancer having comparatively few treatment options and a generally worse prognosis. 

Breast cancer was the most studied disease area globally in 2021, the Phesi analysis reveals. The heterogeneity of breast cancer in terms of biology and natural history dictates that those clinical trials be of situation-specific design, Li says. Across the cancer spectrum, the emphasis in the pharmaceutical industry is shifting to situations involving more aggressive disease subtypes and innovative forms of treatment such as chimeric antigen receptor (CAR) T-cell therapy. 

Phesi just announced it has used AI to create a digital twin for a serious side effect, cytokine release syndrome (CRS), for patients following CAR-T therapy to support development efforts in this space. The digital twin is an alternative to gathering data from individuals in a randomized, double blind, placebo-controlled trial since too few patients are available for reaching conclusions. It can function as a synthetic control arm to accelerate the evaluation of treatments for CRS, Li says.  

A digital twin can likewise be constructed for metastatic breast cancer, Phesi reports in its global data analysis. The approach uses the pattern of distribution of this subset of patients relative to all others, based on their level of functioning (i.e., ECOG performance status) and other parameters aligned with a particular clinical trial design to synthesize a patent profile to “mimic patient characteristics at baseline in a real clinical trial.”  

Costly Miscalculations  

Phesi is “very passionately trying to propel this industry to know better about our own patients, before they design trials and design programs for treating them,” says Li. An otherwise promising drug candidate could fail. 

A consistent theme seen by Phesi is that “the clinical development industry by and large has not been very careful in designing their trials [in terms of protecting] the reproductive system of not just younger women but men as well,” Li says. Although strategies exist to protect fertility (e.g., drugs that temporarily shut down the ovaries to protect them during chemotherapy), they tend to be overlooked early on, prompting a lot of costly protocol amendments down the road. 

A single protocol amendment can cost study sponsors more than half a million dollars, on average, for a phase 3 clinical trial, according to 2018 research by the Tufts Center for the Study of Drug Development. “It’s an issue that’s important to patients but also to pharmaceutical companies to more effectively and cost-effectively bring innovative medicines to patients,” Li notes. 

The motivation to participate in an intervention trial can also differ significantly between people based on their disease subtype, including HER2-positive (more common in white women) and triple negative breast cancer (more common in black women), he says. Although HER2-positive cases are six times as prevalent as the triple-negative variety, these women are no easier to enroll than the triple-negatives in part because alternative treatments are more readily available. 

Studies are also more often using comprehensive biomarker combinations requiring the participation of patients diagnosed with triple-negative disease, as noted in the published Phesi analysis. 

Li says he has observed that “linear thinking” is typical in the clinical trials industry when the real world is, conversely, dynamic and requires thinking outside the lines. “You have to understand the relationship between different cuts of the data.” 

Phesi was an exhibitor at the recent Bio-IT World Conference & Expo, where industry excitement about artificial intelligence and big data was palpable. But Li says his interest is not in the technology itself but AI as a tool for solving some of the practical problems plaguing clinical trials. 

The truth is that answers sometimes arise from a very simple analysis, he concludes. But other times, a seemingly straightforward question simply exposes the multitude of other variables that have yet to be considered.