Which treatment strategy best fits the biology of each individual patient
For patients living with multiple myeloma, the abundance of available treatments has long posed a quiet paradox: more options do not automatically mean better choices. Researchers at the University of Miami's Sylvester Comprehensive Cancer Center have developed an AI model called GigaTIME that reads immune signals hidden within routine bone marrow biopsy slides, offering a way to match each patient's biological reality to the therapy most likely to help them. The work, presented at the 2026 ASCO annual meeting, suggests that the immune environment within a tumor may be as telling as its genetic profile—and that artificial intelligence may soon help spare patients from treatments they do not need while guiding them toward those they do.
- Multiple myeloma patients face a growing menu of powerful therapies, but without a reliable way to match treatment to biology, some endure toxic regimens that may offer them little benefit.
- GigaTIME, trained on bone marrow biopsy slides, detects levels of CD16—a marker tied to the immune cells that daratumumab activates—revealing immune landscapes invisible to standard clinical staging.
- The contrast in outcomes was stark: low-CD16 patients who received daratumumab stayed event-free at 86.8% after 18 months, while those who did not collapsed to just 28.6%, underscoring how much biology-blind treatment decisions can cost.
- High-CD16 patients showed comparable outcomes regardless of whether they underwent stem cell transplant, raising the possibility that one of oncology's most intensive interventions could become selectively, rather than routinely, deployed.
- The tool is not yet in clinics—prospective validation, larger datasets, and direct comparison with measured biomarkers lie ahead—but researchers describe the signals as strong enough to mark the beginning of a new era in treatment personalization.
Multiple myeloma patients today have access to treatments that didn't exist a decade ago—powerful immunotherapies, expanded stem cell transplant programs, and complex drug combinations. The challenge has never been the options themselves, but knowing which patient truly needs which approach, and which might safely be spared the most grueling interventions.
Researchers at Sylvester Comprehensive Cancer Center believe the answer may already be sitting in a pathology archive. Using an AI model called GigaTIME, they analyzed routine bone marrow biopsy slides from 212 newly diagnosed patients and found that immune signals embedded in those images could predict treatment response with remarkable precision. The key marker was CD16, associated with natural killer cells—the very immune cells that daratumumab, a widely used monoclonal antibody, depends on to attack myeloma.
The findings were difficult to ignore. Among patients with low AI-predicted CD16 levels, those who received daratumumab alongside standard therapy remained event-free at 86.8% after 18 months. Those who received standard therapy alone fared far worse, with only 28.6% avoiding disease progression at the same point. Meanwhile, patients with high CD16 levels showed similar outcomes whether or not they underwent stem cell transplantation—suggesting that one of oncology's most intensive procedures might be reserved for those whose biology actually calls for it.
Research scientist Arjun Raj Rajanna, presenting the work at the 2026 ASCO annual meeting, described the conceptual shift: rather than asking which drug works best on average, the AI asks which strategy fits this particular patient's immune biology. Team leader C. Ola Landgren was careful to note that stem cell transplants are not obsolete—but that the decision to pursue one may increasingly be guided by biology rather than blanket protocol.
The tool remains in research, with prospective validation and expansion to more diverse datasets still ahead. But the signals are strong enough that Landgren sees GigaTIME as a potential turning point—a moment when artificial intelligence moves from streamlining clinical workflows to genuinely illuminating the biology that determines who lives longer, and who suffers less.
Multiple myeloma patients have never had more treatment options. Doctors can now reach for powerful immunotherapies, expanded access to stem cell transplants, and drug combinations that didn't exist a decade ago. The problem is deciding which patient needs which treatment—and which ones might safely skip the most intensive, most toxic approaches altogether.
Researchers at Sylvester Comprehensive Cancer Center, part of the University of Miami Miller School of Medicine, believe they've found a way to answer that question by looking at something doctors already have: routine bone marrow biopsy slides. Using an artificial intelligence model called GigaTIME, they discovered that immune signals hidden in those slides could predict which patients would respond best to specific therapies, potentially sparing some from unnecessary toxicity while directing others toward the treatments most likely to help them.
The work centers on a simple but powerful observation. Two patients with identical genetic risk profiles and the same clinical stage can have completely different immune environments in their bone marrow—and those differences matter enormously when it comes to how they'll respond to treatment. Daratumumab, a monoclonal antibody that teaches the immune system to attack myeloma cells, depends entirely on having a robust immune response. Stem cell transplantation can extend the time before cancer returns, but it carries serious side effects and temporarily weakens immunity, raising infection risk. The question becomes: which patients actually need it?
The research team, led by C. Ola Landgren, analyzed bone marrow biopsies from 212 newly diagnosed multiple myeloma patients in the HealthTree Foundation registry. They used GigaTIME to estimate levels of CD16, a biomarker associated with natural killer cells—the immune cells that daratumumab activates. Then they tracked how these patients fared on standard therapy with bortezomib, lenalidomide, and dexamethasone, either with or without daratumumab added to the regimen.
The results were striking. Patients with low AI-predicted CD16 levels who received the standard three-drug combination without daratumumab saw their disease progress quickly, needing a new treatment within months. But those same low-CD16 patients who received daratumumab added to their regimen had dramatically different outcomes: 86.8 percent remained event-free at 18 months, compared to just 28.6 percent of those treated without it. For patients with high CD16 levels, outcomes were comparable whether or not they received a stem cell transplant—suggesting that transplant decisions could become more selective, biology-driven rather than one-size-fits-all.
Arjun Raj Rajanna, the Sylvester research scientist presenting these findings at the 2026 American Society of Clinical Oncology annual meeting, framed the shift plainly: instead of asking which drug combination works best overall, the AI approach asks which treatment strategy fits the biology of each individual patient. Longer time on initial therapy translates directly into longer disease control, better quality of life, fewer treatment-related toxicities, and less disruption to daily living.
Landgren emphasized that the findings don't suggest stem cell transplants are obsolete. Rather, they support an emerging concept: transplant decisions may become increasingly personalized, driven by biology rather than protocol. Understanding immune biology at diagnosis may prove just as important as understanding the tumor's genetic makeup.
The tool remains in the research phase. The team plans to validate these findings prospectively, compare AI-predicted CD16 levels with directly measured immune biomarkers, and expand the model to larger and more diverse patient datasets while incorporating additional immune markers. But the signals are strong enough that Landgren sees this as potentially the beginning of a new era—one where artificial intelligence moves beyond automating workflows to become a genuine tool for biologic discovery and clinical decision support.
Citas Notables
Instead of asking which drug combination is best overall, we are using AI to ask which treatment strategy best fits the biology of each individual patient.— Arjun Raj Rajanna, Sylvester research scientist
Understanding immune biology at diagnosis may be just as important as understanding the tumor's genetic makeup.— C. Ola Landgren, director of the Sylvester Myeloma Institute
La Conversación del Hearth Otra perspectiva de la historia
Why does the immune environment matter so much more than we thought?
Because daratumumab and other immunotherapies don't work in a vacuum. They're asking the patient's own immune cells to do the killing. If those cells aren't there or aren't activated, the drug can't do its job. Two patients with identical tumors can have completely different immune neighborhoods in their bone marrow.
And the AI is reading those neighborhoods from slides that already exist?
Exactly. Doctors already take bone marrow biopsies. The AI extracts immune signals from those same slides—things human eyes might miss or take hours to quantify. It's not adding a new test; it's reading what's already there more completely.
The 86.8 percent versus 28.6 percent difference seems enormous. Is that real?
It's real in this cohort of 212 patients. But that's why they're calling it a research tool. Those numbers need to hold up in prospective studies with new patients before doctors can rely on it in the clinic. The pattern is compelling, but it's not yet proven.
What happens to patients who get the wrong treatment based on this?
That's the validation question. Right now, this is identifying patterns in retrospective data. Before it guides real treatment decisions, they need to show it works prospectively—that predicting CD16 levels actually changes outcomes when doctors use it to choose therapy upfront.
Does this mean some patients will avoid stem cell transplants?
Potentially, yes. The data suggests high-CD16 patients do equally well with or without transplant. That matters because transplant is brutal—it wipes out your immune system temporarily, increases infection risk, and disrupts your life for months. If you don't need it, avoiding it is a win.
What's the next hurdle?
Expanding to larger, more diverse patient populations and validating the findings in a prospective trial. They also want to measure CD16 directly and compare it to what the AI predicts. And they're working to add more immune markers to the model, not just CD16.