Balancing AI Mortality Predictions and Patient Autonomy

Palliative care teams are increasingly utilizing longitudinal electronic health record (EHR) artificial intelligence (AI) to identify patients who would benefit from timely goals-of-care conversations. By analyzing health data over time, these predictive models forecast mortality risks and alert clinicians to initiate vital discussions earlier in a patient's disease progression. However, the core challenge lies in integrating these data-driven prompts into clinical workflows without diminishing human empathy or overriding patient autonomy. This article explores how healthcare providers use AI as a supportive screening tool rather than a definitive directive, ensuring that patient values and personal choices remain at the center of end-of-life care.

Understanding Longitudinal EHR Mortality-Prediction AI

Longitudinal EHR mortality-prediction AI works by continuously scanning a patient’s medical history over an extended period. Unlike static models that look at a single hospital admission, longitudinal algorithms analyze trends—such as frequent emergency room visits, declining lab values, weight loss, and changes in medication.

When the AI detects a pattern indicating a high risk of mortality within the next six to twelve months, it flags the patient's chart. This solves a persistent problem in medicine: clinicians often overestimate patient survival and delay difficult conversations until a medical crisis occurs. The AI acts as an objective early warning system.

Triggering Timely Clinician Prompts

Once the AI identifies a high-risk patient, it does not make clinical decisions. Instead, it triggers a nudge—typically via an EHR alert, a dashboard update, or an automated email—to the primary care provider or the palliative care team.

This prompt simply suggests that a goals-of-care conversation may be appropriate. It brings the patient to the team's attention, allowing clinicians to review the chart and decide if an intervention is truly warranted. By moving these discussions upstream to an outpatient or stable inpatient setting, patients are given the time to process information without the pressure of an acute emergency.

AI as a Tool, Not a Mandate

To protect patient autonomy, palliative care teams treat the AI’s output strictly as a screening mechanism. The algorithm calculates statistical probabilities, not certainties, and it knows nothing about a patient’s personal resilience, cultural beliefs, or life goals.

Clinicians exercise "clinical override" before ever approaching the patient. If the medical team feels the AI prompt is premature or inaccurate based on their personal knowledge of the patient, they can dismiss the alert. The technology serves to initiate human review, never to replace clinical judgment or dictate a care pathway.

Preserving Patient Autonomy in Conversations

When a clinician decides to act on an AI prompt, the technology itself usually remains invisible during the consultation. Providers do not tell patients, "The computer says you have a high risk of dying this year." Doing so would cause unnecessary distress and strip the patient of hope and agency.

Instead, the clinician uses the prompt to open a compassionate, patient-led dialogue. They use established communication frameworks to ask the patient about their understanding of their illness, their fears, their priorities, and what a "good day" looks like for them.

Autonomy is preserved through the following practices:

• Seeking Permission: Clinicians ask the patient if they are ready to discuss their future care. Patients retain the absolute right to decline or postpone the conversation.
• Centering Patient Values: The AI does not recommend specific treatments, such as a Do Not Resuscitate (DNR) order. If a high-risk patient expresses a desire for aggressive, life-prolonging treatment, the palliative care team respects and documents that choice.
• Shared Decision-Making: The clinician provides medical context while the patient provides personal context. Together, they build a care plan that aligns medical capabilities with the patient's unique desires.

Guarding Against Algorithmic Bias and Paternalism

A major threat to patient autonomy is the potential for AI to induce medical paternalism—where doctors subtly push patients toward less aggressive care simply because an algorithm predicted a poor outcome. Palliative care teams actively train to avoid this bias.

Furthermore, teams must remain vigilant about data bias. If an EHR algorithm is trained on historical data where minority or underserved populations received less aggressive care, the AI might prematurely flag those populations for palliative withdrawal. By keeping the focus of the conversation on what the patient wants rather than what the algorithm predicts, palliative care teams ensure that technology enhances personalized care rather than automating life-and-death decisions.