I read a lot of studies in this field. Most of them confirm what I already believe. This one stopped me.
A paper published in May 2026 by Mazziotti, Tolento and Ondusko at Oregon Health & Science University, "Language access in the neonatal intensive care unit: inequities, legality, practice, and call to action", surveyed 189 NICU staff across all ten AAP districts. It found that 40% of respondents disagreed that language discordance leads to worse quality of care in their unit. Fifty-nine percent disagreed that it affects outcomes.
I sat with that for a while. Not because I think those clinicians are wrong about their patients. But because I spend a lot of my time in conversations with healthcare leaders who take language access seriously, who cite the evidence on LEP outcomes, who are building programs and policies. And still, across a national sample, the majority of NICU staff are not connecting the language in which care is delivered to what happens to the babies and families in their unit.
If that number surprises you, it surprised me too. And when something like that surprises me, I stop trying to explain it away. I start asking what we, institutions, management, vendors and regulators, got wrong.
1. The Problem Is Not That Clinicians Don't Care
NICU nurses and neonatologists are not indifferent to their patients. The NICU is one of the most emotionally demanding settings in medicine. The staff in that survey are working long shifts with medically fragile infants and families in crisis. The question is not whether they care. The question is whether the language access solutions that have reached them were designed in a way that made the connection between language and outcomes visible and felt.
On Care Culture Talks, Nadav Shimoni, MD, early-stage investor at A-Squared Ventures and one of the sharpest thinkers I've interviewed on healthcare AI, said something that stuck with me. He described "some naivety of innovators trying to innovate from outside without fully appreciating the difficulty of the day-to-day clinicians are experiencing and the complexity of what we call clinical workflow." He added that even free is not really free: "the burden of integration and change management is actually, in time and perhaps even bothering the staff type of resources. It costs the health system quite a lot."
That is a precise description of how language access tools have often been introduced into clinical settings.
With good intentions.
With policy backing.
And without enough humility about what it actually costs a nurse mid-shift to route through a new system.
2. What We Got Wrong on Adoption
When a tool is hard to reach, it gets skipped. When it gets skipped enough times, the behavior calcifies. The clinician develops a workaround, a bilingual colleague, a family member, a simplified English explanation that seems to work. The interpreter call that did not happen does not produce an immediate visible failure. It produces a discharge conversation that was 70% understood, a medication schedule that was partially clear, a follow-up that did not happen three weeks later, readmission.
None of that links back to Tuesday's shift in the NICU. So the belief that language access does not significantly affect outcomes in this unit is not cynicism. It is the honest perception of someone who has not seen the downstream data, because the downstream data is not surfaced in a way that reaches them.
This is where institutions and regulators share responsibility alongside vendors. The compliance frameworks around language access, Title VI, Section 1557, state-level provisions that 81% of respondents in this survey were unaware of, are built around documentation and legal obligation. They were not built to make the clinical impact of language access legible to a night-shift nurse. That is a design failure, not a malice failure.
The 2026 NEJM Catalyst study on AI for language access in surgical care found something relevant here: patients did not view AI and professional interpretation as competing. They valued each for different moments in the encounter. What that study described, without using the word, is workflow fit. The right tool, available at the right moment, without adding friction to a shift that is already stretched.
When that fit exists, the interpreter gets called. When the interpreter gets called consistently, the outcomes data starts to connect. When the outcomes data connects, the perception of NICU staff starts to shift. That chain has to be built deliberately. It has not been built deliberately enough.
3. What Humility Looks Like in Practice
Nadav also said something I think about when we talk to health systems about adoption. He described how solutions need to arrive with a clear and objective financial ROI, frictionless integration and design that was done together with the staff, not handed to them. "From the beginning it won't happen," he said, "but I think we can get there in many places. But it requires a lot of dedicated effort."
That is the honest posture. Not: here is the tool, here is the policy, here is the training module. But: we need to understand your shift. We need to know where the friction lives. We need to design the entry point so that using it is easier than not using it.
For language access specifically, that means on-demand access that lives in the workflow the staff already have. Sam Frenkel, MD, an emergency medicine physician who described his experience with real-time interpretation in a frontline ER story on our blog, put it simply: "Instant is key." Not convenient. Not nice to have. Key. The NICU is not the ER, but the logic is the same: when reaching the tool costs more than skipping it, it gets skipped. It means the difference between a quick check-in with an AI interpreter and a scheduled video call for a consent conversation being made clear and easy to navigate. It means outcome data being fed back to clinical leadership in a form that reaches the unit, not just the compliance office.
For language access specifically, that means on-demand access that lives in the workflow the staff already have.
The NACHC Accelerator 2026 cohort, which is building this kind of infrastructure for the 52 million Americans served by community health centers, is one model for what deliberate adoption design looks like at scale. The NICU setting has its own specific emotional and operational texture. It will need its own version.
4. What This Study Is Really Asking Us to Do
The authors frame their findings as a call to action. I read it as a mirror.
Fifty-nine percent of NICU staff do not yet believe language barriers significantly affect outcomes in their unit. That is not a fact about those clinicians. It is a fact about the ecosystem they are working in, the tools they were given, the data they can and cannot see, the culture around interpreter use that has or has not been built in their institution.
If we want that number to change, the answer is not more posters in the hallway or a revised onboarding module. It is showing up at the bedside with a solution that was designed with enough humility to actually fit there. And then staying long enough to close the loop between the language access decision and the outcome that followed.
That work is harder than writing a policy. It is also the only thing that changes what people believe.
.svg){kind=small}
%201.png)