Eliminating Bias and Ethically Integrating AI for Palliative Oncology Care

Ram Prakash Thirugnanasambandam, MBBS, spoke with CancerNetwork^® about considerations for responsibly using artificial intelligence (AI)–based tools in palliative medicine and the treatment of patients with hematologic malignancies. He discussed these strategies in the context of a manuscript he authored and published in the November/December 2025 issue of the journal ONCOLOGY^®.

Thirugnanasambandam, a fellow of Internal Medicine, Hospice and Palliative Medicine, and Geriatric Medicine at the University of Miami, highlighted how clinicians must consider potential data privacy concerns and algorithmic bias associated with AI platforms as part of truly meeting the needs of their patients. Additionally, he noted that AI models must be trained on different datasets encompassing diverse patient populations to reduce the risk of acting on potentially biased information.

Transcript:

One of the main themes that emerged while we were doing this review article is that, yes, AI can be utilized well in palliative care and [hematologic] malignancies, but we also [must] think about the ethical considerations while utilizing it. From a patient perspective, that can be related to data privacy consent from patients as well as bias. For example, one of the studies we looked at in the review article showed AI-based tool healthcare utilization needs in comparison to true needs for a particular patient population, and that may not be reflective of all patients across the country. That can reflect systemic inequities rather than true need of care in these patient populations.

Another standpoint is that while AI can be utilized as a decision-making support, it cannot replace nuanced clinical judgment and empathy in terms of palliative care. It can [conduct] predictive analysis, but it cannot predict the values, goals, and emotional distress that a patient goes through while they go through these difficult diseases. These are ethical considerations that need to take into the bigger picture and have transparency and continuous clinical oversight from a human perspective. We also need to train these AI models on different data—data from diverse patient populations—so that there's no bias when it comes to the utilization of AI.