In one hospital study with on-site interpreters on the payroll, only 28% of patients with limited English proficiency had full documentation of informed consent for invasive procedures, compared with 53% of English-speaking patients. The interpreters were down the hall. The paperwork still came up short.
That matters to anyone who may one day sign a consent form, or sit with a family member who’s about to. The signature on the page is supposed to represent a real conversation about what the procedure is, what it’s for, what could go wrong, and what the alternatives are.
When that conversation happens through a tablet on a rolling cart, a phone line, or an app translating sentence by sentence, the shape of the conversation changes. So does the question of what a court will later call informed. For families weighing whether something went wrong, a medical malpractice attorney will often look first at what the consent record actually shows, and who was in the room when it was obtained.
The Video Cart at the Bedside
Video remote interpreting is the format most patients now encounter. A screen on a wheeled stand rolls up, a credentialed interpreter appears, and the conversation runs three ways: clinician, patient, interpreter. On paper, it’s close to having a human interpreter in the room. In practice, the setting introduces its own friction.
The camera has to be positioned so the interpreter can see the patient’s face. The audio has to be clean enough to catch a hesitation, not just a yes. If the connection drops in the middle of a risk disclosure, the clinician has a choice: wait, switch to a phone line, or push forward in English and hope.
The last option is where consent stops being informed. The chart may still show that an interpreter was used, but it won’t show that the interpreter was gone when the risk of stroke was explained.
The AI Translation App in a Clinician’s Pocket
The newest layer is machine translation. A clinician opens an app, speaks a sentence, and the phone reads it back in another language. For a quick orienting question, this can be fine. For consent, it’s a different animal.
A systematic review of AI translation in clinical settings reported accuracy of 83 to 97.8% when translating from English, but only 36 to 76% when translating back into English. The direction matters.
Clinicians can push information out reasonably well. Catching what the patient is trying to say in return, including questions and objections, is where the technology thins out. A one-way flow of information is a briefing, not a consent conversation.
Risks of leaning on an app without a qualified interpreter include:
- Idioms that don’t survive translation. A phrase like “we’ll put you to sleep” can land very differently than intended, and the app won’t flag the drift.
- Low-resource languages. Accuracy varies widely by language pair, and the languages with the thinnest training data are often the ones a patient most needs served well.
- No cultural read. A human interpreter notices when a patient defers to a family member, or agrees out of politeness. An app translates the words and moves on.
- No record of the exchange. If the app doesn’t save a transcript, there’s nothing to audit later if the patient says they didn’t understand.
A Bilingual Relative Is Not a Qualified Interpreter
The oldest workaround is still the most common one: a bilingual family member standing at the bedside, translating for a parent or spouse. It’s fast, it’s free, and it’s exactly what the National CLAS Standards tell hospitals not to rely on. Standard 7 is explicit that language assistance should be provided by trained and qualified individuals, and that family and friends, especially minors, shouldn’t be used except in an emergency.
The reasons show up at the bedside in ways that are hard to catch in the moment. A son may soften a cancer diagnosis to protect his mother, and a daughter may skip the sexual history question entirely because asking it of a parent feels extremely difficult.
A 12-year-old pulled in to interpret a surgical risk conversation is being asked to carry a weight no child should carry, and the clinician often has no way to tell what was passed on and what was edited out. When something goes wrong later, the chart typically shows that consent was obtained. It rarely shows who did the interpreting, or whether they were qualified to.
What a Reasonable Consent Conversation Looks Like Now
The standard isn’t a signature. It’s a documented, two-way conversation the patient could reasonably follow, in a language they actually use. Across formats, a few things hold up:
- Qualified interpreter on the record. The interpreter’s name or ID, and the modality used, should appear in the note, not just a checkbox that language services were offered.
- Teach-back, not head nods. The patient explains the procedure and the main risks back in their own words, through the interpreter, before signing.
- Consent form in the right language. A form the patient can read, or a short-form consent with an interpreter’s attestation, beats an English form signed under a verbal translation.
- A pause when the technology fails. If the video drops, the call cuts out, or the app stalls on a key sentence, the right move is to stop and reconnect, not to finish in English.
None of this is unreachable. It’s slower than a nod and a signature, and that’s the point. When a consent conversation later becomes the center of a dispute about what the patient knew, the record of how the language barrier was actually handled is usually where the case turns.
