Nearly two-thirds of US doctors are turning to an AI-powered search engine to make patient decisions, propelling a little-known tech startup to a $15 billion valuation while prompting a debate over the future of patient care.
Miami-based OpenEvidence, founded by serial entrepreneur Daniel Nadler in 2022, said that many of America’s 1 million physicians processed more than 42 million prompts last month. Doctors are using the company’s chatbot to scour medical journals and clinical guidelines to help diagnose and treat patients.
The ads-powered platform has grown largely by word of mouth, penetrating a sector where few tools have seen similar traction. “We’re the first viral consumer internet app in healthcare,” Nadler told the FT in an interview.
The cases of Dr Hadiza Bawa-Garba and RaDonda Vaught
Content note: this article discusses the deaths of patients, including a child. Co-created and edited by Jade Garratt and Bea Poyton.
Dr Hadiza Bawa-Garba is a British paediatrician. RaDonda Vaught is an American nurse. Both made errors in complex, high-pressure systems and both were subsequently prosecuted for it.
Their cases are very different, but they do share a consequence: the implicit message sent to healthcare workers was that honesty about error carries criminal risk. What happened to patients in both cases mattered a great deal – the harm was only too real, and so was the grief of the families and the staff involved, the sometimes-called “second” and “third” victims. But what happened to Bawa-Garba and Vaught afterwards may have put far more patients at risk, by eroding something that safe healthcare depends on – the safety and freedom to speak up and share learning from mistakes.
“Underneath every simple, obvious story about ‘human error,’ there is a deeper, more complex story about the organization.” Sidney Dekker
The Case of Dr Hadiza Bawa-Garba
On 18 February 2011, six-year-old Jack Adcock was admitted to Leicester Royal Infirmary. He had Down’s Syndrome and a pre-existing heart condition, and had come in with vomiting, diarrhoea and breathing difficulties. He died that evening.
Dr Hadiza Bawa-Garba, a paediatric registrar (ST6 – which means she was in year 6 of her post-grad training), led Jack’s care. In the subsequent investigation and legal proceedings, a number of errors in the clinical management of Jack’s case were identified. A chest X-ray indicating infection sat unreviewed for over two hours – Dr Bawa-Garba was not notified it was available and antibiotics were started late as a result. Blood test results were held up for over five hours due to IT failures. Jack’s regular heart medication was discontinued by Dr Bawa-Garba due to his deteriorating condition, but this wasn’t documented, so when his mother visited that evening she administered his usual dose (which was, and still is, standard practice for family members to administer medication to children), unaware it had been stopped. This led to the circulatory shock and subsequent cardiac arrest. During his cardiac arrest, Dr Bawa-Garba briefly confused him with another patient and paused resuscitation before restarting.
These were real errors and the consequences are devastatingly real too. But to understand them only as individual failures would be a misread of the situation.
The ward was short-staffed, and the on-call consultant was off-site. He hadn’t realised he was on call that day, a rota failure that left Dr Bawa-Garba with sole responsibility for the entire unit. The IT system was also malfunctioning and nursing communication was fragmented.
Photo by gorden murah surabaya: https://www.pexels.com/photo/a-hospital-ward-7563452/
Dr Bawa-Garba was working in conditions that systems thinkers would recognise immediately, with performance-shaping factors stacked against her at every turn. In such conditions, the errors she made can be seen not as aberrations but as predictable outcomes of the system she was embedded in. This is local rationality – the idea that people’s actions almost always make sense from inside their situation, given what they knew, the tools they had, and the pressures they were under. She was a trainee doctor managing an entire unit alone because of a rota error, with an IT system that was actively delaying critical results, in conditions that would have stretched a far more experienced clinician. Her actions made sense in context, even where they fell short.
In 2015, Dr Bawa-Garba was convicted of gross negligence manslaughter. She was suspended, then struck off, then reinstated following a crowdfunded legal battle and significant professional outcry, and finally cleared to practise without restrictions in 2021. It had been a decade of professional limbo for a registrar who was, by most reasonable analyses, failed by her system before the system failed her patient.
The outrage across the medical profession was substantial, but one element of the case stood out above the others, and it matters enormously for psychological safety.
A few days after the incident, Bawa-Garba was asked to meet Stephen O’Riordan, the duty consultant at the time of the incident. At the meeting she was asked to reflect on the circumstances, setting out what she should have done differently. She was sent home immediately afterwards and told not to come back until she was asked to. These reflective learning notes are the kind of structured, honest self-appraisal that medical training actively encourages. She reflected at length on the points where, as a trainee, she felt she could have managed Jack’s care better – all the things she would do differently given the chance. Reflective notes aren’t just for the author to learn from: their intended purpose is at least in part to facilitate collective learning – so we can learn from each other’s experiences.
But those reflective notes were obtained by the prosecution, legally, because nothing prevented it – and the prosecution QC, Andrew Thomas, pressed Dr Bawa-Garba on the reflections she made after Jack’s death. “List for us, please, all of the mistakes,” he said. Her own reflections became evidence that she knew she had made errors.
There is something epistemologically dishonest in that. Reflective notes are hindsight documents by definition: written after the outcome is known, in conditions nothing like those on the ward that day. After the event, Bawa-Garba had time and space to process what happened, and the knowledge of how things had ended. She also had the support in making those notes of the consultant who should have been on call that day. The Bawa-Garba on the ward, in the moment, had none of that – she was a junior doctor managing multiple sick children in an understaffed, overstretched and underfunctioning unit, without adequate consultant cover. Treating her reflection after that event as evidence of what she understood during it mistakes the nature of the document (and of causality) entirely.
The very act of learning from the event was weaponised against the person who tried to learn from it.
“Blame is emotionally important, not operationally important.” Todd Conklin
The effect on the profession was significant. The fact that no law protected those notes from prosecution use only sharpens the point: the boundary between learning material and legal scrutiny had never actually existed. If honest reflection can be read by prosecutors and treated as admission of culpability, why would anyone write honestly? A survey found that over 80% of junior doctors changed how they write reflections after this case — they rationally self-censored out of fear.
Reflective practice only works when clinicians believe that candour is safe. Once that belief breaks, so does the practice – a completely understandable outcome. If honesty about work as done risks prosecution, silence becomes the safer choice, and a profession that has learned to stay silent is a much less safe one.
The Case of RaDonda Vaught
On 24 December 2017, 75-year-old Charlene Murphey was admitted to Vanderbilt University Medical Center in Nashville with a subdural haematoma. Two days later, she was recovering well and nearly ready to go home, but needed a PET scan before discharge, something she was a little nervous about. To ease her anxiety ahead of the procedure, she was prescribed Versed (midazolam).
Nurse RaDonda Vaught went to retrieve the drug from their notoriously flaky automated dispensing cabinet. When she typed “VE”, no results appeared (because unknown to her, the system required the generic name of midazolam). Vaught activated the cabinet’s override function. This was not unusual – Vaught herself later explained that nurses had been instructed to use the override function when they couldn’t access medications patients needed, and many experts have said that this is a common occurrence at lots of hospitals. Scanning the results, she selected what she thought was Versed, but was actually vecuronium: a neuromuscular blocking agent that causes full paralysis. She administered it to Charlene Murphey, a patient who had been on the cusp of going home. Charlene lost consciousness, suffered a cardiac arrest, and died.
Vaught immediately disclosed her errors. She told the truth.
The Medical Center terminated her employment in January 2018 after an internal investigation, but in fact recorded Murphey’s cause of death as natural causes and did not report the medication error to the authorities. It took a whistleblower and a federal investigation to bring the full picture to light. In 2022 (four years after the event) Vaught was convicted of criminally negligent homicide and abuse of an impaired adult. Her nursing licence was revoked as a result. Although a Tennessee Bureau of Investigation agent testified that the hospital had a “heavy burden of responsibility” for the error, no further action was brought against the institution.
The inversion here is stark: the nurse who admitted her mistake was prosecuted, whilst the institution that actively concealed its more structural and systemic role to avoid scrutiny got away with it. With echoes of Bawa-Garba, Vaught’s transparency became the mechanism of her prosecution.
“There is almost no human action or decision that cannot be made to look flawed and less sensible in the misleading light of hindsight.” Sidney Dekker
As with Bawa-Garba’s incident, context is essential. The dispensing machines used drug naming conventions that made confusion easy. The override function existed precisely because the system regularly failed staff who needed to access medications quickly: indeed, staff had been taught how to use the override for exactly this reason. As far as Vaught could tell, the only way for Charlene to get the medication she needed was to bypass the system safeguards, which were designed to catch exactly this kind of error.
But the dispensing cabinet was only part of it. There was no second-nurse verification required for medications pulled via override. Vecuronium – a paralytic agent with no legitimate emergency use case that would require rapid override access – should never have been available via that route at all. Vanderbilt removed it from the override list after Charlene’s death, a relatively simple safeguard that had always been available, but hadn’t been made. There was also no medicine barcode scanner in the radiology department where the drug was administered, which meant a standard check that would have caught the error at the bedside couldn’t happen.
These are not factors that excuse what happened – Charlene Murphey died, and that matters very much – but they do help explain it, and there is a meaningful difference between explanation and absolution. If we address any one of those systemic factors, the outcome might have been different. If instead we focus only on the individual – blame, shame, retrain or dismiss – none of those conditions change. The next nurse walks into the same ward, faces the same broken system, and the same or very similar thing happens again.
The Observation Effect
Psychological safety is not primarily about our own direct experience. Most nurses will never be prosecuted for a clinical error. But when we observe it happen to someone else – someone who made a mistake, admitted it, tried to learn from it, and was then criminalised for it – it affects how we calculate the risks of speaking up and being open about our decisions and mistakes. We don’t need it to happen to us personally for the lesson to land. Vicarious experience is more than sufficient to alter our behaviour, even if those cases are very rare: we may observe 99 successful speaking up attempts, but if we observe just one where the consequences were severe, that’s the one that looms large.
The Bawa-Garba and Vaught cases sent a powerful signal, at large scale, across entire professions: that honesty about error carries serious risk, and that doing the right thing after a mistake offers no protection. Bawa-Garba’s case also showed that our reflective learning notes can be turned against us.
The result is all too predictable: clinicians become more guarded, errors go unreported, near-misses go undiscussed and reflective practice becomes performative rather than genuine. The very data that organisations need to improve patient safety is systematically suppressed.
The Learn Not Blame campaign, which emerged in the UK partly in response to Bawa-Garba, and the widespread commentary from nurses and nursing bodies following the Vaught conviction, both speak to exactly this: that criminalising error doesn’t make healthcare safer: it produces a culture of silence, which ultimately results in far greater risk to the patients the system is meant to serve.
Learning, or Punishing
RaDonda Vaught went to work on 26 December 2017 to help patients like Charlene Murphey. She went to work to do a good job. So did Hadiza Bawa-Garba on 18th February 2011, the day Jack Adcock was brought into her ward. Neither arrived intending to cause harm, but both were working in systems propped up by workarounds that were well known and under-addressed. When things went wrong, both did what we say we want people to do: they were honest about it.
A reasonable objection at this point: surely individuals must be “accountable”? Surely we can’t simply excuse errors because the system was imperfect?
Of course. And that’s not what anyone serious in this space is arguing. The point is not that we as individuals bear zero responsibility. It is that individual accountability and systemic learning are not the same thing; and treating them as equivalent, by prosecuting individuals as the primary response to systemic failure, produces neither justice nor, more importantly, improvement.
What happened to both Bawa-Garba and Vaught after the incidents speaks volumes about what it can cost a person to tell the truth, and about the gap between the cultures that organisations often say they want and the ones that get built. A Just Culture distinguishes between human error, at-risk behaviour, and recklessness. The vast majority of clinical errors fall into the first category: things that happen to competent, well-intentioned people working in imperfect systems. And somewhere right now, after one of those errors, a nurse is deciding whether or not to file an incident report, and a trainee doctor is wondering whether to write honestly in their reflective log. They are making that calculation in the shadow of, admittedly, rare cases like these, in systems that are intended to enable safe candour, but on occasion, dramatically fail to do so.
Vaught and Bawa-Garba weren’t reckless people. They were people doing their best in imperfect systems, who then did the right thing afterwards. If we want people to speak up, we have to make speaking up actually safe. That means in practice, not just in principle, and it means we need to take care of the next person who tells the truth.
If you’d like to explore this with your team, our Learning Lab has a simulation of that day at Leicester Royal Infirmary, where you make the decisions the doctors faced, with only what they knew at the time. As in the real events, a child dies, so please bear that in mind before running it.
Note: This piece was researched in depth from multiple sources (many included below), but some factual errors may remain. We always welcome corrections and clarifications.