Dr. Kyle Lavin told me early in our Care Culture Talks conversation that he thinks of himself as an interpreter. A palliative care psychiatrist, he translates an overwhelming medical system into choices a seriously ill patient can make. That is the heart of palliative care language access. In palliative care and in mental health in oncology, language is the clinical instrument: pain, fear, goals and new diagnoses all reach the clinician through words. When patient and clinician do not share a language, that instrument loses precision when it matters most. This post explains what is at stake and what responsible AI medical interpreting has to get right.
Why are words the clinical tool in palliative care and psycho-oncology?
Palliative care and psychiatry diagnose, plan and support patients almost entirely through conversation. No scan shows a patient's goals of care.
Dr. Lavin has practiced for almost 20 years and co-founded Cerula Care four years ago to bring behavioral health care to people living with cancer. He explained that palliative care is a one-year fellowship clinicians can enter from 11 different specialties. Most come from family medicine or internal medicine. He came from psychiatry, so he sees every palliative patient through a psychiatrist's lens.
He described the three areas a palliative team focuses on:
- Pain and symptom management, so quality of life stays as high as possible whatever the illness is doing.
- Goals of care and advance care planning: understanding who the patient is, what matters most to them and whether they are fully hearing what the medical system is telling them.
- Support for a patient and a family going through something most people face only once.
Two of those three are conversations, and the third depends on them. For a palliative psychiatrist, Dr. Lavin said, words are the main tool, both for gathering the information that leads to a diagnosis and for building the treatment plan. That is why reaching patients who do not speak English as their primary language is something he has lived personally.
The mental health burden in this population is large. The National Cancer Institute's clinician summary on depression estimates that depression affects up to about a quarter of people with cancer, well above the rate in the general population, and reports that only around 5% of patients see a mental health professional. Every one of those missed or delayed conversations is harder to have across a language barrier.
What happens when palliative care conversations cross a language barrier?
Goals of care discussions break down quietly when language access breaks down, and the research documents it.
Dr. Lavin put it plainly: navigating a serious illness is hard even when a patient has every resource, including language, financial support and health literacy. Take language away and it becomes much harder. In his own practice he sees patients who speak Spanish, Russian and many other languages.
The evidence matches his experience. A systematic review in the Journal of Pain and Symptom Management examined ten studies of interpreter use in palliative care for cancer patients with limited English proficiency. When professional interpreters were not part of the care, patients and families came away with a poorer grasp of diagnosis and prognosis during goals of care conversations, and patients had worse control of pain and anxiety at the end of life.
Those are the decisions Dr. Lavin described: whether to continue cancer-directed therapy, whether to accept life support during an acute decline, when to shift the focus toward comfort. If a preference gets lost between two languages, the result can be care the patient never wanted.
He works in two settings today. In the hospital, his team relies on in-person interpreters, which works well when one is available, with some wait. In the outpatient clinic, video interpreters on a rolling iPad cart offer more languages and shorter waits. Both models ask clinicians to speak in short sentences and pause often, which interrupts the natural flow a goals of care conversation depends on.
Why do psychiatric emergencies and behavioral health interpreting raise the bar?
In behavioral health, a misheard phrase can become a misdiagnosis, and a misdiagnosis can follow a patient for the rest of their life.
Dr. Lavin was direct about the stakes in psychiatry. A new diagnosis of a thought disorder or schizophrenia changes how a patient is treated, how their family responds and how the health system sees them from then on. “It takes one time of doing that incorrectly to basically unravel a patient’s life and their support,” he said.
Psychiatric emergencies compress that risk into minutes. Clinicians assess mood, thought content, safety and decision-making capacity largely through what the patient says and how they say it. Tone, hesitation and word choice are clinical data. Federal guidance recognizes this. In its letter on the Section 1557 language access provisions, the HHS Office for Civil Rights explains that a qualified interpreter must convey meaning accurately and impartially while keeping the tone, sentiment and emotional level of the original statement. In behavioral health interpreting, that legal standard is also the clinical standard.
This is why Dr. Lavin wants clinicians closely involved in AI-supported care for the next few years. He would like to see an evidence base showing that errors are rare and that the benefits clearly outweigh the risks before the culture of care changes. He also cautioned against vendors moving faster than clinicians are ready to go, because that creates friction instead of progress.
[VISUAL: Where language carries the diagnosis in serious illness]
Why does AI interpreting matter for mental health in oncology at scale?
The language gap in community oncology is too wide for one bilingual hire to close.
Cerula Care exists because supportive care for people with cancer has mostly lived inside large academic medical centers, while community oncology practices and smaller regional health systems rarely offer it. Dr. Lavin shared findings from an abstract his team was preparing for ASCO: Medicaid and BIPOC patients reach Cerula Care at higher rates than they typically access behavioral health services, arrive with higher depression and anxiety scores and respond to treatment better. In his words, that is the population the model was built to reach.
Language is the next step in that equity mission. Today, Cerula Care serves Spanish-speaking patients through a bilingual care manager. One of its main clients, a large community oncology group in Maryland, needs coverage in more than 20 languages, including Cantonese. Dr. Lavin sees AI-powered interpreting as the only realistic way to close that gap at scale.
The national numbers point the same way. The GSA Translation and Interpretation Services SIN 541930 Ordering Guide (December 2025) counts 29.6 million people with limited English proficiency in the United States. For community practices, staffing in-person coverage for long-tail languages at every hour is out of reach. AI-first medical interpreting with 295+ language access options in the same interface is built for that coverage problem, on a flat monthly subscription that replaces per-minute billing and at no cost to the patient.
Dr. Lavin also pointed to format. He appreciated seeing the conversation appear on screen in real time, because clinicians do not have to stop and wait after every short sentence, and the exchange feels closer to a natural conversation for patients who can read along. In a palliative family meeting, that rhythm matters.
What does responsible AI medical interpreting look like in these settings?
Dr. Lavin's view combines patient choice, close clinician involvement and patience with adoption. Each principle applies well beyond oncology.
Ask the patient first
He described a stepwise consent model. At the start, patients are asked whether they are comfortable with an AI agent, and they can leave the AI interaction at any point if they become uncomfortable. He added that some behavioral health studies suggest certain patients find it easier to raise sensitive or shameful topics with an AI agent, because of the distance it creates. Choice protects both groups of patients.
Treat preference and clinical risk as separate questions
When I suggested that the choice between AI and a person is often about sensitivity more than risk, he agreed that preference and permission matter, and noted that clinical risk is still part of the picture. Health systems need clear rules for both.
Keep clinicians close to the highest-stakes moments
New psychiatric diagnoses, capacity assessments and decisions about life support are the encounters where he wants clinicians most involved while the evidence matures.
Move at the pace of the care team
Physicians tend to be risk averse. Studying outcomes together and treating vendors as partners leads to adoption that lasts, while pressure from outside leads to resistance.
Build with clinicians who have practiced
His advice to technology companies entering healthcare was to keep the patient experience at the center and recruit practicing clinicians to shape the product from the ground up, even if that slows things down. He also challenged physicians to step outside traditional academic paths and join digital health teams.
For health system leaders, these principles lead to concrete steps: identify the palliative care, psycho-oncology and psychiatric emergency encounters that carry the highest stakes, record how each patient prefers to receive language support and make sure that support is available around the clock, including nights and weekends.
Dr. Lavin ended our conversation by admitting that these technologies can be scary, and that the opportunity to improve access is still huge if we move responsibly. I left convinced that palliative care and psychiatry are where AI medical interpreting has to prove itself first, because in these rooms the words are the treatment. You can hear the full episode and the rest of the series on Care Culture Talks.
.svg){kind=small}