NEW YORK – The concept of web-enabled, or "virtual" care, which catapulted into the mainstream during the COVID-19 pandemic, is also becoming ingrained within precision oncology care delivery at many community and rural healthcare institutions in the US through so-called virtual molecular tumor boards.
Community healthcare organizations are streaming in precision oncology experts and specialists through these virtual MTBs, who can remotely review patients' complex genomic test results and advise their local oncologists on appropriate treatments based on the latest published studies and guidelines. But busy oncologists don't always have time to engage with these MTBs, while board experts also need help conducting labor-intensive reviews of molecular test reports and medical data. Amid growing demand for such services, new strategies are needed to make MTBs accessible to more patients.
Many studies have shown that MTB review can increase patients' access to evidence-based precision cancer treatments, giving them the chance for better outcomes.
For example, a case-control study of more than 900 non-small cell lung cancer patients conducted by researchers at the University of Kentucky found that MTB review of molecular test results and other clinical data is an independent, positive predictor of patients' overall survival regardless of whether they lived in urban or rural areas. Similarly, a real-world data analysis of 164 patients with breast and gynecologic cancers who underwent MTB review showed that those who received the full treatment plan recommended by the MTB had better progression-free and overall survival than those who didn't. Patients in complete compliance with the MTB's recommendations had median progression-free survival of nine months and median overall survival of 17.1 months, versus four months and 10.8 months, respectively, for patients who didn't receive MTB-recommended treatment.
Despite the data showing patients benefit from MTB involvement, scaling up access to such expertise has been a challenge. Vida Almario Passero, chief medical officer of VA National TeleOncology, who leads the organization's virtual tumor board, has been focused on putting processes in place so more patients can access this resource. "If it's good for patient care, we think outside the box and try to build the infrastructure for it," she said. "If we always think along the lines of how we've always delivered care, we already know it's not quite enough."
As demand grows for MTB expertise, however, the challenge now is how to best use treating physicians' limited time so they can participate in case reviews with MTBs and the specialists on these boards can see more cases. Early adopters are exploring a myriad of ideas, from engaging coordination teams that help oncologists gather case information for the MTBs, to incorporating artificial intelligence (AI) technologies that can function as assistants.
Finding time for virtual MTBs
In the first two years of the VA's virtual tumor board, it reviewed 233 patient cases from 51 VA facilities in 33 states and Puerto Rico, according to data published on the program earlier this year.
The virtual tumor board, which began in March 2022, includes specialists in hematologic, breast, gynecologic, central nervous system, thoracic, and cutaneous malignancies. The interdisciplinary board involves medical oncologists, surgical oncologists, radiation oncologists, pathologists, radiologists, genetic counselors, clinical pharmacists, advanced practice providers, and nurses. Patients' treating physicians, or a member of their local care team, are encouraged to present their patients' cases and participate in MTB discussions, Passero said.
Upon starting the virtual tumor board, the VA had to spread the word to oncologists at VA medical centers across the country about its availability. Passero said the organization regularly reached out to oncologists to tell them about the virtual tumor board and encourage them to discuss their patients' cases with experts.
However, oncologists at the VA also have packed schedules and may not feel they can carve out time for this additional service. Recognizing this, the VA engaged a coordination team to help treating oncologists gather the information they needed to present their cases to the tumor board. "There is always going to be that time limitation [for physicians]," Passero said, emphasizing the importance of regular communication reminding treating oncologists that the tumor board is meeting. "If there's a tumor board that can be flexible or review cases outside of scheduled patient time, that's a way of getting some buy-in from treating oncologists."
In total, the virtual tumor board at the VA held 113 sessions in the first two years, which attracted 468 unique attendees in 41 states, Washington D.C., and Puerto Rico. The virtual tumor board's growing popularity tracks with the need for telehealth care in the VA since, according to Passero, more than half of veterans (56 percent) who received cancer care through the VA's National TeleOncology service in 2022 lived in rural areas.
A pilot version of the national virtual tumor that Passero led and launched in Pennsylvania in 2018 sought to better understand the needs of rural VA beneficiaries. For example, do patients need to travel long distances to receive care at the VA? "Rural healthcare has a limited amount of oncology specialists," Passero said. "We explored what could we do to maximize the technological infrastructure for these patients."
The VA Pittsburgh Healthcare System built a virtual cancer care network that included a video telehealth platform and a chemotherapy pharmacy and nursing infusion clinic in Altoona, Pennsylvania, a small city in a more rural part of the state. Through the pilot, the VA showed that virtual care was feasible in this rural setting and that it saved patients travel time and cost while maintaining their relationship with their oncologist at the Pittsburgh medical center through virtual appointments.
Following this experience, Passero and her team turned their attention to implementing the virtual tumor board nationwide. "It's about our virtual team partnering with those sites, a lot of which are rurally located, to make sure patients get their care," she said, adding that the virtual care model ensures that patients in rural communities can "stay where they are" and still access specialists. "We will make sure that we will get them a neuro-oncologist or the lung cancer specialist to … [give] a second opinion or prescribe chemotherapy [or weigh in on] survivorship, palliative, and supportive care."
Looking ahead, Passero's team has ambitions to make the virtual tumor board available to even more patients with hard-to-treat cancers and rare tumors. She hopes to analyze in more depth how the board's recommendations changed clinical decisions. Her team is even working with specialists outside of oncology to build similar virtual case review boards for patients with other diseases.
Helping patients with rare tumors
Identifying precision therapies for patients with rare tumors is one area in which generalist community oncologists may not be that up to date, but where MTB expertise in targetable molecular aberrations driving cancer can be particularly helpful. Within the TCF-001 TRACK (Target Rare Cancer Knowledge) study, the TargetCancer Foundation is using a virtual MTB to identify treatments for patients with seldom-seen tumors.
The TRACK study began in October 2020 with the goal of enrolling 400 patients, who would locally provide blood or tissue samples for next-generation sequencing and have a virtual MTB review the test results. In the trial, after patients are remotely consented, TargetCancer organizes their sample collection through a community clinic or sends a phlebotomist to their homes. The samples are sent to molecular testing lab Foundation Medicine, where tissue samples are sequenced using the FoundationOne CDx comprehensive genomic profiling test and blood samples are assessed using the FoundationOne Liquid CDx assay.
"We've reached patients in at least 41 states, and we've seen over 50 different types of rare cancers," said TargetCancer CEO Jim Palma. "That scope of geographic distribution and cancer type distribution wouldn't be possible in any setting other than a remote setting."
He added that the remote nature of the trial and MTB is beneficial to patients who are "participating fully in the trial from their homes without any need to travel or change their treating physician."
However, figuring out the logistics of the remote study and MTB review are more challenging, Palma acknowledged. Without physical study sites, the research team has to collect all medical records and other information for each patient from their hospital. "Because the study is remote, every patient might be coming from a different institution, so it's almost like starting from scratch every time we have to reach out to a new place," he explained.
The heavy lift on the logistics side getting all the paperwork in order means eligible patients can see delays in getting molecularly profiled and having the results evaluated by the virtual MTB, Palma said. To speed up the information gathering process, the TRACK study has a dedicated team that reaches out to treating physicians, pathology labs, and Foundation Medicine for patient data.
With the help of this team, Palma estimated there's on average about a one-month lag between when a patient is remotely consented for the study and when their test reports are reviewed by the virtual MTB. The wait time was slightly different depending on if patients had blood or tissue testing performed, according to data TRACK researchers shared in June. Liquid biopsy profiling results were available in a median of 9.1 days versus 13.3 days for tissue results. There was also a shorter time between submission of blood samples for testing than with tissue samples, 2.4 weeks versus 4.9 weeks, respectively.
"The time to MTB review really depends on our ability to successfully secure medical records and tissue," Palma said. "It is a limiting factor in a remote study, and the time frame can be expanded if we're not getting a response from a doctor's office or a hospital on medical records or if the pathology lab isn't moving quickly on transferring tissue for testing. Our goal is to do that as quickly as possible because these patients are dealing with very, very difficult-to-treat cancers and oftentimes coming to us at late stages."
Another challenge of running a remote MTB is engaging with the treating physicians. Of the 130 patients included in the results released in June, 128 patient reports were reviewed by the MTB and 87.5 percent of those received MTB recommendations based on comprehensive genomic profiling results.
Palma said the TRACK team provides treating oncologists with a report containing the virtual MTB's treatment recommendations, including the experts' rationale and references to published studies and guidelines underlying those recommendations. "But at that point, it's really up to the patient and their doctor to accept those recommendations, if they want to," he explained. "If we don't have a relationship with that treating physician that can be a roadblock to them engaging with those recommendations, or even accepting them."
Researchers in the TRACK study have not yet reported on the proportion of treating physicians that accept the virtual MTB's recommendations. Patients are followed for one year after enrollment, and Palma expects to report this data after the follow-up period has completed for all enrolled patients.
So far, inviting the treating physician to the MTB meeting appears to be a good way to forge that relationship and build trust between the virtual team and treating oncologist, in Palma's view. However, like the VA virtual tumor board, finding time on a busy community oncologist's calendar has been the biggest hindrance to building this connection. On the other hand, convening regular virtual MTB meetings between a multidisciplinary team from different institutions — including medical and surgical oncologists, genomics experts, molecular pathologists, a basic scientist, a pharmacist, a genetic counselor, regulatory staff, a scribe, an oncology fellow, and a medication acquisition specialist — has been less difficult.
Despite the study's challenges, Palma believes remote MTBs suit rare cancer treatment very well because many community oncologists tend to be generalists and may not be familiar with the treatment or clinical trial options for very rare tumors they might see once every few years.
"Rather than a patient having to seek out that expertise, [the TRACK trial] is bringing it to them," he said. "In rare cancers, it's absolutely critical that you're treating [patients] with real awareness of the landscape of what's available for that rare cancer including precision medicines and biomarker-defined therapies."
AI assistants for MTBs
On the other side of this coin, experts in these tumor boards are also under increasing pressure to provide MTB services to more institutions and clinical trials. One startup hoping to help MTBs extend their reach using AI is Allagi.ai, led by Alex Gavryushkin, associate dean for research at the University of Canterbury in New Zealand, who is focused on biological data science.
A rate-limiting factor for MTBs, in Gavryushkin's experience, is the labor involved. "It's an extremely laborious process to analyze all the data, to go through the literature, look at what medications are available, if multiple regimens are available, and how to combine them," he said.
AI, however, could be used to take a first pass at identifying potential treatments for patients based on their molecular profile and other data and reduce the human labor needed to review each case. "The vision here [at Allagi.ai] is that a lot of what [MTB experts] do can be automated such that the cases can be processed far quicker and done remotely," Gavryushkin said.
In the example of the challenge for the TRACK study, collecting and processing disparate patient data, Gavryushkin noted that AI language models can process patient records alongside literature. The algorithm can then present treatment options for human experts to consider and decide if they want to make those recommendations to patients. This type of tool could also help MTB experts review more cases in less time, and conversely, make it easier for treating oncologists to send their patients' data to an MTB for review.
Allagi.ai is working on developing such an AI assistant for molecular tumor boards to support clinical decision-making. The startup, which is in stealth mode, is developing algorithms that can process unstructured clinical and molecular information from patient records to extract relevant data, run relevant bioinformatics analyses on the data, link the results of the analyses to clinical and biomedical knowledge, including information on drug trials, and generate an actionable interactive report for human MTB to review and approve. In January, Allagi.ai was among the winners of the MIT Solve Cure Xchange Challenge "Health AI for Good," which provides funding and incubator support for startups.
This type of AI support is already popping up in some health systems. Another group in Australia explored the effect of incorporating AI within a decision support system for an MTB convened by the Australian Molecular Screening and Therapeutics program. The AI-driven decision support system matched patients to treatments and trials by mining a precision oncology guideline database developed by two Australian cancer institutes and a clinical trials registry.
Data researchers collected from this MTB showed that patients' participation rate in matched clinical trials increased slightly after the AI decision support system was implemented. Prior to the implementation of the AI system, the clinical trial participation rate based on MTB recommendations was 10 percent, and 12 months after, it rose to 12.2 percent.
AI is good at processing and making recommendations for a typical patient case, Gavryushkin noted, but where these machine learning models falter is with so-called "edge cases" that don't have a lot of data supporting them for the AI to be trained on. "If there is a very rare case, where the symptoms might look very similar to something else, and if you just tick the boxes and don't look at the bigger picture, you simply overlook that rare case," he said. "That's where you need the human expertise. It's hard for AI to pick up those rare events."
The future adoption of AI in this setting will depend on whether the algorithms have been validated on large, diverse datasets, so they can help patients with common or rare cancers. But patients' views on data privacy can make it harder for data scientists to access the necessary datasets.
Many AI algorithms are being trained on publicly available, de-identified clinical and genomic data, but some patients do not want their data included in the training sets or want to scrub their data from AI algorithms. "We always enforce that if a patient does not want their data to be used going forward, there should be a technically sound solution to removing that from the data," Gavryushkin said, but that's not always an easy task.
He noted that once algorithms are trained on a dataset, it is difficult to determine whether a specific data point is stored somewhere in the neural network. "The only way currently to 100 percent guarantee that a patient's data was removed and not memorized anywhere in the AI is to retrain everything from scratch, which is expensive," he said. Short of that, Gavryushkin speculated that if an algorithm was trained in a way where it maintains each data point in connection to the final network of information, it could allow just one patient's data to be removed without running the training process again.
The challenge with AI in healthcare at present is also the error rate. In other settings, like AI-based language translation, if the result has a 1 percent error rate, it's not a serious problem, Gavryushkin said. But in healthcare, such an error rate could cost a patient's life. That's why he predicts that AI will be best harnessed as a supplement to human decision-making in medicine.
"The vision is that highly capable AI assistants will be sitting on molecular tumor boards and interacting with experts," he said. "Every patient in the world who has cancer should be able to access those services."