One of the toughest parts of oncology care today is deciding when it is safe to prescribe a risky treatment for a patient. Robert Wachter, MD recently wrote a moving article about this struggle in The New York Times, emphasis mine:
“A recent analysis estimated that about 15 percent of patients with advanced cancer might benefit from immunotherapy — and it’s all but impossible to determine which patients will be the lucky ones. Just last week, a study of lung cancer patients demonstrated the overall benefits of combining immunotherapy with traditional chemotherapy. But here, too, the researchers noted that most patients will not respond to the new treatments, and it is not yet possible to predict who will benefit. In some cases, the side effects are terrible — different from those of chemotherapy but often just as dire.”
Is it possible to support physicians with real-world data and technology in determining when a treatment is worth the risk for a specific patient?
Over the last year, our team has been working to apply artificial intelligence technology to this concern for breast cancer patients. More than 250,000 new cases of invasive breast cancer are diagnosed each year, according to the American Cancer Society. While chemotherapy is a life-saving treatment for patients with breast cancer, it also comes with life-threatening neutropenia risk and susceptibility to deadly infections. One in five chemotherapy patients experiences neutropenia, or a reduction in immune function, as a side effect, which causes an increase in mortality by 20 to 30 percent. The development of neutropenia is often idiosyncratic and was previously unpredictable.
We pretrained an AI module on a vast dataset of real-world cancer data to predict chemotherapy-induced neutropenia (CIN) with a receiver operating characteristic area under curve (AUC ROC), in some cases, exceeding 0.9, which exceeds the best published rates by more than 50 percent. That is a major improvement in our ability to predict neutropenia in patients and inform safe oncology care decisions based on evidence. The American Society of Clinical Oncology published the abstract on our research last week in advance of the annual conference in Chicago at the start of June.
The next step with this AI module is to get it into the oncology clinical workflow. This prediction ability can help make clinical trials safer. We also envision physicians being able to use their patients’ specific medical record data to predict their individual neutropenia risk in chemotherapy treatment.
The future of improving patient outcomes is rooted in embracing precision medicine. Precision medicine takes into account the variation in disease, lifestyle, genes, and physical environment to format informed treatment and prevention strategies. Precision medicine allows doctors to identify specific risk factors based on patients’ genetics, identify potential disease mutations for patients with undiagnosed conditions, avoid serious side effects from medication, and ultimately determine the best care for each individual patient. We want physicians and patients to be able to have a conversation about risk together — using precision medicine in a collaborative care relationship that ultimately improves the patient’s health outcomes and quality of life.