Alice's Guardian Angels

The story of a missed diagnosis, and how "outside" intervention managed to prevent (a worse) disaster

About a year ago, Alice Tapper, the teenage daughter of Jake Tapper, a prominent television journalist, publicly shared her story of a missed diagnosis of appendicitis. It is harrowing, and the kind of thing I always worry about when I or my loved one get sick, particularly when they have to be seen in an emergency room. You can read and listen to Alice’s own account here.

As a radiologist I have long been concerned with medical errors, and as a diagnostic radiologist, diagnostic errors in particular. The Institute of Medicine raised awareness of this problem in their 2015 report.

It is one thing to recognize that errors occur and another to find ways to prevent them. In this post, I want to think through what happened in Alice's case, and contemplate possible ways in which it could have been avoided.  One caveat: my thoughts are predicated on what few specifics I can glean from news reports, and are missing important details. Her chart was reportedly hundreds of pages long. But the problem began at her admission, and I do have part of her admission note (displayed on the CNN segment about her story).

The Case

Alice, a 14 year old previously healthy girl, had been ill for three or four days with diarrhea, fever and one episode of vomiting just before admission.  She was dehydrated, tachycardic (fast heart rate), and hypotensive (low blood pressure). Her abdomen was soft and nondistended, and diffusely tender, but not focally so. The author of the first clinical note (perhaps a second year pediatric resident?) looked for no less than 5 signs of appendicitis on physical examination. None were seen. The white cell count was elevated.

The differential diagnosis considered was as follows:

1. viral gastroenteritis

2. bacterial gastroenteritis

3. food poisoning

4. appendicitis

Because of the time course and the diarrhea, and the absence of focal tenderness or peritoneal signs, gastroenteritis was considered more likely than appendicitis. Peritoneal signs refers to irritation of the thin membranous lining of the abdominal cavity (the peritoneum), which is exquisitely sensitive to inflammation; when the appendix ruptures it can cause quite marked diffuse pain from this cause. Food poisoning was thought unlikely because the vomiting had resolved and an elevated white cell count would be an uncommon feature.

Next came the supervising ("attending") physician's note. He/she agreed with the trainee that the "overall story [was] most consistent with gastroenteritis". Because the fever had been present for 3 or 4 days, a bacterial cause had to be ruled out; a fever that long would be unusual with viral gastroenteritis, though not impossible. They were focused on bacteria in the colon such as Salmonella, but we should note that bacteria in the appendix could also account for a prolonged fever. Diarrhea and vomiting can also be seen with appendicitis; it was really the absence of focal right lower quadrant tenderness and peritoneal signs that led to appendicitis falling to the bottom of the list.

The supervising physician's note says that Alice's presentation was concerning for sepsis. This indicates that she was quite ill at this point. But apparently because she had improved (with hydration? - the note is obscured here) they elected to admit her and watch her while giving further hydration and waiting for the infection work-up results. Appendicitis remained on their minds, as evidenced by the recommendation that a scan be obtained if the patient developed focal right lower quadrant tenderness.

She was admitted with a diagnosis of gastroenteritis. I do not have documentation of whether appendicitis was again considered by the inpatient intake physician (likely a resident) or the inpatient attending physician. Both these doctors were likely different from the emergency department doctors. At some point, she was transferred to another hospital, so even more doctors got involved.

The precise time course after this is not known to me, but her mother observed that Alice's hands and feet were cold (a sign of vasoconstriction that can be seen with sepsis), her coloration had deteriorated, her abdomen had become more distended and she was just generally sicker. The parents suggested an ultrasound or other test for appendicitis, raised that diagnosis with the clinical team several times, and even conveyed Alice's pediatrician's suggestion of appendicitis. All of this was to no avail. At one point, the parents asked if there was a reason not get an ultrasound; unlike CT scans, ultrasound is entirely risk-free, and nothing in the picture was inconsistent with appendicitis. The team evidently minimized these suggestions because they did not order a further test. The family requested consultation from a specialist (a gastroenterologist) but for some reason this never occurred. It seems this state of affairs persisted for a number of days whereupon she was discharged, then readmitted, all still without a scan.

Eventually, the father (as I said, a very prominent journalist) took action by searching for and contacting an administrator of the hospital, resulting in an imaging workup that lead to the correct diagnosis of a ruptured appendix. Alice had become septic and had a long painful course of treatment and recovery. Her appendix was finally removed a few months later after the inflammation had died down. It should be noted that precisely when her appendix ruptured is unclear; it is possible that it was early in her course, thought the initial lack or mildness of peritoneal signs argue against this.

Comments on the case

Let me start with my conclusion: what happened here was a failure of the medical team to revise their weighting of the differential diagnosis as she became sicker, compounded by the absence of any mechanism by which the family could gain traction against the thinking of the clinicians. This second point is crucial. There are possible ways to improve both of these factors and I will discuss each in turn. But efforts at improvement will come up against major resistance on the part of the medical establishment as it currently exists.

My purpose here is not to indict a particular physician. A judgment call was made, and as the facts attest, it was wrong. This was not an error by a single physician; it was a collective effort. Between the emergency department visit and the admitting team, likely at least 4 physicians were involved initially, and likely several more over the ensuing course before a correct diagnosis was finally made. Did everyone involved really not know that appendicitis can present without focal right lower quadrant pain? This strains credulity, especially in the face of repeated inquiries about it from the parents. This is the major puzzle: what is it about the system that allowed this group error to happen?

I am confident that those who missed the diagnosis feel remorse, and my purpose here is not to denigrate them. I have made many mistakes myself, a few burned forever in my memory. Such failures are inherent to this difficult task. It is the nature of medical diagnosis. My purpose is to imagine how the system the physicians work in might have been built in such a way to protect them from the inherently imprecise and risky business of diagnosis. I'll say it again: both patients and their physicians need a better system. If a mistake is made along the way, a mechanism is needed to detect and correct it. When it comes to diagnosis, at present there are few such safeguards in place.

The Evolution of the Facts

Medicine is so often presented as scientific. In actuality, though science plays a role, practicing medicine is more akin to piloting a ship through only partly charted seas. What lies below and ahead can only be surmised most of the time, and circumstances carry us all forward relentlessly. The captain must constantly revise his or her course as circumstances warrant. Doing this effectively depends on accurate understanding of changes in the wind and water. The potential for surprises is vast.

Importance of good monitoring

Before the clinician can exercise good judgment, an accurate and timely account of the physical reality of the patient is mandatory. The correct diagnosis may become more clear as time goes on, but the risk of instability also can increase.

While very sophisticated monitoring is routinely used in certain settings (ICU, operating room), the monitoring of many non-ICU patients has barely improved since the 19th century. Once the patient is admitted to the floor, measurements tend to be intermittent. Detection of changes depends largely on the nurses and on the family. I seldom see effective graphical representation of data trends at the bedside. Much seems to hinge on moment to moment readings instead. In monitoring for sepsis, notably lacking is any direct way to follow the intravascular blood volume changes, a key physiologic factor in that condition. Indirect physical signs are used instead. Fluid intake and output measurements are crude and frankly often poorly tracked. Development of tissue edema depends on physical exam, if one is done. In my personal experience, once I was admitted , there were few instances of more than cursory physical exam. Visits from the physician team were confined to a few minutes in the morning, and sometimes a few minutes in the evening.

Further comment on the physical exam is warranted. In her long course, did Alicenot develop peritoneal signs? What was the time course of her edema and blood pressure instability (if it was unstable)? Here is a harsh question: Did they even examine her? Unfortunately the written notes can be unreliable on this point. I recall my own clinic visit to a prominent cardiologist and his fellow for evaluation for heart failure; at no point did either physician lay a hand on me. No auscultation, no lung check, no ankle edema check. I had gotten a cardiac echo a few weeks prior, so I suppose they thought that was sufficient. Nonetheless, their clinic note documented a normal physical examination.

Changeable makeup of team

Even if they did examine her in more than a cursory way, the shifting makeup of the medical team would make serial comparison of physical signs difficult. Often in a hospital setting the clinicians who take care of patients on the floor are not the same ones that admitted the patient; in fact this is more the norm than an exception. The team can vary from day to day, and as I mentioned typically see floor patients for a few minutes in the morning and perhaps evening.

The result is there is often not one person who observes the patient throughout this whole time, so that perceiving trends and changes can be difficult. The importance of continuity of care has long been recognized however. The principal technique depended upon to provide this  has been requiring doctors, often medical residents, to work 24 hour shifts (36 hours at the time of my internship back in the day). This is a reflection of the fact that the patient's full story is mostly conveyed orally from team member to team member, a process with its own name – the "patient hand-off". Nurses follow a similar tradition.

The electronic medical record is a sad and insufficient tool for this kind of communication. This is particularly tragic given its inherent potential. But there is not room here to expand on this observation.

The Evolution of Clinician Thinking (or lack thereof)

Everyone understands, or should understand, that the diagnosis each patient seen in the ED or admitted to the floor carries is provisional, except in those cases where a definitive diagnosis has in fact been made. This is the minority; there is often an element of residual uncertainty. Part of the pattern that clinicians look for unfolds over time, and is not necessarily clear from the very beginning. Changes in the patient must be carefully monitored and the diagnosis revised as warranted.  Note the language of probability used; this must always be the case when talking about the "provisional diagnosis", that is, a working diagnosis that guides our next actions but which we do not necessarily intend to be our final conclusion.

The clinicians here were looking for a pattern of findings to direct their thinking towards or away from the various diagnoses. Alice did not present in a typical fashion for appendicitis, if by typical we mean as having all of the classic signs, as discussed above. But she did have most of the signs, lacking the most specific, which was focal right lower quadrant tenderness. Despite this, appendicitis was always in the differential, and the clinicians knew this.

Alice's presentation was concerning for sepsis, as recognized by the initial attending doctor. Sepsis is extremely serious in its implications. One must take steps to its exclude major sources. The attending suggested an ultrasound of the appendix if Alice developed increased right lower quadrant abdominal symptoms. But development of such tenderness, while it would have clearly pointed to the diagnosis, is not a constant feature of appendicitis. That sign is seen in only a proportion of patients (about 50%), surprisingly insensitive given the emphasis given to it in medical training. In fact, there is known phenomenon where right lower quadrant pain diminishes once the appendix has ruptured, conceivably the case here. This problem of inaccurate signs, symptoms, and tests is rife throughout medicine and is unfortunately simply the nature of the beast. We do not have perfect tests. We have imperfect information at every step.

Role of "classic" signs and symptoms

Jake Tapper pointed to the root of the problem as being that doctors are not aware that appendicitis can present in a nonstandard way. It is certainly true, in my experience, that right lower quadrant pain is over-relied upon as a sign of appendicitis. But I could make similar statements about overreliance on signs and test results of all kinds. Many diseases do not present in a standard way, and doctors are aware of and struggle with this fact all the time. The "standard" or "classic" presentations of disease serve to help the physician create his or her initial ranking of the differential diagnosis, as in this case.

So doctors are constantly revising their ideas about what is going on with patients. The shortcomings in this case were the failure to exclude a somewhat less likely but still completely consistent cause of her symptoms,  a cause which carried serious ramifications, followed by failure to revise the likelihood of ruptured appendix as she got sicker.

Cognitive biases

You can read about cognitive biases in diagnosis. There are so many it's hard to know where to start. After reading about them, I almost despair of ever making a correct diagnosis. Solving this problem is in my opinion never going to come from the direction of correcting human biases, in the individual humans themselves I mean. That's why I believe some kind of innovation is needed to serve as a corrective. I will expand on my idea on ways to do this below.

Misjudgment

As the facts attest, the diagnosis was missed; was this because of an error in judgment?

The initial judgment of the provisional diagnosis was credible even though it turned out to be wrong. When a 14 girl presents with several days of diarrhea with an episode of vomiting, fever and leukocytosis, tachycardia, hypotension, and diffuse abdominal pain, gastroenteritis can be reasonably surmised as the most likely diagnosis. Three days of diarrhea seems a bit much for appendicitis, and there was no focal right lower quadrant pain, a classic sign.  To be clear, appendicitis was considered from the very start; the admitting resident looked for no less than 5 physical signs of it, all absent.

However, the presentation was also completely consistent with appendicitis, that is, there was nothing that clearly pointed away from it, because right lower quadrant pain is not a constant feature of that disease. It is impractical to test for every possible diagnosis at once. So when does testing for the next most likely diagnoses come into play?

Striking to me was that although there was concern for early sepsis, the second most likely diagnosis was not searched for. As stated in the note by the emergency attending, the clinicians were waiting to see whether the pattern of symptoms changed. There was an error in judgment in thinking that a change would be needed to justify looking for appendicitis, and that the change would be development of right lower quadrant pain. Ultrasound is more sensitive for appendicitis than are physical signs. Once the appendix ruptures, focal tenderness can improve or even resolve because then the inflammation is everywhere in the abdomen.

As time went on and the patient grew sicker, looking for alternative diagnoses became more important. The seriousness of the error increased as the facts evolved and alterative suggestions were ignored. Appendicitis was considered in the top two diagnoses from the start, mentioned prominently in the admission notes, and by the family and the pediatrician repeatedly, but was never looked for. Why?

It is tempting to suppose that the initial error in judgment carried forward as a kind of momentum – "diagnostic momentum" as others have suggested. Like many diagnoses themselves, this is merely a description, not an explanation. Why did this "momentum" occur? The answer has to be something about the way in which the physicians think (biases, etc. as above), but also the structure of power and credibility. The only defensible reason not to get a scan would be that appendicitis was considered extremely unlikely, and no one could have reasonably argued that from a medical perspective. But the suggestion to look for it was coming from a quarter without perceived expertise (in the case of the family) or authority (in the case of the outside pediatrician).

To redress this will require putting in place some kind of robust information and idea feedback to the medical team that they can consider and accept without loss of face.

The Feedback and Its Reception

Mistakes happen. But once a mistake happens, what is the mechanism to correct it? The process of assigning a medical diagnosis to a patient is not inherently redundant; lacking is a robust mechanism to catch errors and correct them.  In my field of radiology for example, I was usually the only person looking at a major diagnostic examination such as a CT scan for cancer staging. If I made a mistake, it was unlikely that anyone would catch it, because there was no mechanism for routine review. This is understandable given the demands on manpower that would entail, but the problem is there nonetheless. To remedy this will require revision in how medical teams are organized and kept accountable. In a team the emphasis is usually on arriving at a consensus opinion. In theory, the team is open to information and suggestions from anyone; in practice, for a trainee to openly question a more senior physician is discouraged by the very structure of medical teams and training. Actively soliciting alternative points of view is not built into the structure of the day. Continued uncertainty inhibits action so is resisted. This is not to say that dissent or criticism will be ignored when it comes, but the reception it receives will depend on the source. If a very senior or respected physician is brought in, you can be sure they will be listened to.

The feedback from "outside"

The same cannot be said of suggestions by the patient's family (or even nurses and the patient's community pediatrician for that matter). As we observe Alice's story, we see that the only direction that a corrective force was coming was from her parents. Of course parents can be wrong as much or probably more than they are right, but they have several undeniable advantages over the clinicians here. They were at the patient's bedside more or less continuously. They knew her baseline appearance, demeanor, and typical response to illnesses like the back of their hands. They had no competing interests or distractions, and because they lacked expertise, there was no risk of loss of face if it turned out they were wrong. They had consulted with Alice's pediatrician in an effort to bolster their questions with some kind of authority. They were Alice's guardian angels (see below), but they were not being heard. In the end, they had to invoke a deus ex machina, so to speak, to change the situation.

In the search for a solution, let me begin by discussing what the patient and family can do right now as the system currently exists, then move on to newer ideas about how the system can be improved.

The Search for a Solution: What the Family Can Do

What can you do if you, as a patient or loved one, find yourself in a similar situation? The first approach is politely and repeatedly quiz the doctors. I suggest you do this in a particular way, a way that forces them to think out loud in front of you. This might slightly annoy them, but don't be too concerned about that; if a physician bristles at this it is a bad sign about their competence. The best doctors are patient in this regard, and even may pause and admit you have a point, then go on to talk about how they could look into that possibility further. But keep in mind you may be wrong yourself. You are not in a position to make diagnoses, but you absolutely can raise them as hypotheses. Ask the doctors explicitly: What else could this be? Why is the clinical picture not consistent with this idea we have? What happens if you are wrong? What tests could you order? What are the pros and cons of doing those?

Second opinion

But what if you become convinced your concerns are not being addressed adequately? At the present time, you have limited options. One is to request a second opinion, from someone who is not part of the current team. It is best if this is someone with higher status, more experienced or more specialized. In an acute setting this can be difficult to pull off, and in particular getting someone in to look at an inpatient can be difficult. Community doctors for example, though they may know the patient better than the hospitalists, have been sidelined. This appears to have happened in the Tapper case.

Very unfortunately in the Tapper’s case, consultation from a specialist was requested but never carried out. This was yet another serious error by the medical team.

Families should keep in the mind the possibility of approaching nurses at this point for help. They can be surprisingly effective in getting doctors attention in cases where the nurse is sufficiently concerned. Nurses' insight is a grossly underutilized resource in my opinion. I am thinking of registered nurses here, preferably the most experienced you can find. This does involve them sticking their necks out, but I have found many courageous enough to do this. The fact that I have to invoke courage here tells you something important about the relationship between nurses and doctors.

The option of last resort

The final option is to make a stink – politely as possible so as to be effective as possible. The parents in this story were well educated and from a social class where demanding attention from someone in authority is tolerated. Had they been from a lower rung of society, chances are they would have been less successful at getting attention.

The Search for a Solution: What the Doctors Can Do

How might innovation reduce the kind of errors we see in this case? I'll first mention two steps that could be taken without the need for new technology, followed by a bolder idea – medical guardian angels.

Blue team, red team

This approach was adopted by the military as a training method and later suggested for medicine. The idea is to formalize the opposition. The "blue" team works up the patient in the usual way. Then, the "red" team's responsibility is to ask difficult and challenging questions, such as "What is the worst case scenario? Have we ruled that out? Is the situation time critical requiring immediate action? What else could this be?" And so forth.

Although this blue team/red team approach is sometimes used in teaching, I've never seen it actually used for patient care. It reminds me of another practice which is very common, that of tumor boards. In oncology it is common practice to periodically convene a panel of oncologists, surgeons, radiation therapists and radiologists to discuss cancer patients facing difficult or consequential decisions. The patient's case is presented, then dissected in detail. Opinions are offered, long shot suggestions brought up, disagreements gingerly raised, and a consensus for action formed. Other specialties sometimes have analogous meetings. There is no sense of deliberate challenge or opposition built into this, and I do think it is a better model for medicine than one with echoes of combat. But the goal is the same. In fact, I could see a benefit to deliberately incorporating a bit of the "devil's advocate" role into tumor boards.

The tumor board model is completely impractical for day-to-day care of patients, especially in acute care. But for inpatients, many institutions do convene daily conferences where inpatients are discussed. These "morning report" conferences are formalized in a different way than blue team/red team or tumor board models. The emphasis is on coming to a consensus, and there is often a clear hierarchy for decision making. This will vary depending on the particular individuals involved. How often is the spirit of intense questioning and disagreement, perhaps from caregivers who do not bear direct responsibility for the patient and therefore have an outside perspective, encouraged?

I do think attending physicians could designate someone to play the red team role in rounds. The role could be passed around from day to day. Tough questions could be rewarded and boldness admired. This would be a cultural shift from the current situation where consensus is valued over dissent.

Assistance from software

A number of software-based diagnostic aids have proliferated over recent years, and I can only mention them here. I believe most of these tend to be used by individual physicians, physician assistants and nurses, especially trainees, to jog their memories and help them avoid omissions in their thinking. I am not sure how often, if ever, these are used in the formal setting of hospital rounds or emergency department decision huddles. One of the earliest such tools was Isabel. This one is fascinating, because it was founded not by physicians or other medically trained people, but by a father whose daughter suffered from a missed diagnosis. He wanted to help prevent similar errors from occurring in others. It is striking, and frankly baffling, to me that this (and other) innovations had to come from families instead of from professionals. This says something important, and disturbing, about how innovation in clinical practice methods is adopted. But it is also testimony to the power of love to push for change.

The recent burgeoning of artificial intelligence, or more precisely large language models using deep learning, hold great promise for expansion of these tools. They must be tempered by human judgment, not only because of the stakes, but also because these models flat out make things up. I asked ChatGPT-4 for a differential diagnosis in Alice's case; it listed appendicitis as the second most likely diagnosis at 20% probability, with gastroenteritis first at 45%. It also listed several other less likely conditions. When then given updated information about continuing sepsis, it listed ruptured appendix as the most likely diagnosis (see the post entitled “Appendix: A Conversation With Chat GPT-4”).  If this analysis had been done by someone with sufficient expertise, based on everything available including the family's input, and then presented to the clinicians, I suspect an ultrasound would have been ordered. Clinicians frequently employ tests in this way, to resolve dissonance.

Guardian angels

Here is the point where I veer off from the conventional narrative. So far I’ve discussed steps the family might take, and tools the medical team might use to improve care. Now I want to move into the future to an entirely new concept. Explaining this fully would requires a book-length treatment , but let me introduce the idea.

It is important for me to emphasize that this idea is entirely unusual and not at all on the radar of most physicians. It will be opposed strongly at first, as most truly new ideas are.

I use the metaphor, and I think a powerful one, of a guardian angel. Angels have been invoked throughout history not only by Christianity but by Judaism and Islam, and similar creatures by other traditions. To be clear, I do not regard guardian angels as actual beings. But, like Santa Claus, we have little problem describing in considerable detail just what their behavior and mission would consist of if they did in fact exist. Give us a scenario and we can tell you what they would likely do. They watch over us, they see everything, they hold our wellbeing as their highest value. I believe that humans can become each other's medical guardian angels given the right tools.

Who fulfills this role in real life right now when we get sick? For the lucky among us, it is our family, and in the case of children most often their parents. Alice's parents were the guardian angels in this story, and without their intervention, her outcome could have been dire.

Not everyone will be capable of being an effective "guardian angel" in a situation like this. For that reason, an effective technology will necessarily require a human interface with training to observe and enter the appropriate information into a system bolstered by artificial intelligence. This role would be a new profession. I suspect this plan would be most effective if that person was viewed as entirely independent of the medical team, but respected by it. In practice, both the team and the guardian must answer to each other.

The blue team/red team idea presented earlier is a first step in this direction, capable of being implemented immediately. I am proposing here that it be formalized, assigned to an independent professional, given powerful technological support (e.g. AI etc.), and move away from a strictly "devil's advocate" stance towards, well, a guardian angel stance.

This will require not only creation of a new profession (I suspect many RNs could be recruited into this role) but implementation of ways to pay for it. But in the end, I think physicians will see this innovation as powerful protection – a guardian angel for them as well as their patients. Tell physicians that you are developing a technology to replace them, and you will encounter intense resistance. Prove to them that you can offer protection against making mistakes, and they will embrace your offer so fast it will make your head spin.

My belief is that as time goes on, medical institutions, pharmaceutical companies, insurance companies and other powerful players in health care will increasingly rely upon artificial intelligence agents to interact with and make decisions about patients. Patients must have their own AI to interact back, for their own protection. Their guardian angel must look out for their welfare exclusively. I stress the importance of keeping the independent nature of this entity.

Conclusion

It is not that the doctors didn't think of the possibility of appendicitis in Alice; it was that they had too much confidence in the course they had set, and dismissed suggestions that something treacherous might have been overlooked. A guardian angel was needed, and one appeared in the person of Alice's parents. This is the power of love – adversity can sometimes elevate humans in this way.

But of course they were lucky as well for a number of reasons; a way to systematize and strengthen this force is needed.  Professional medical guardian angels is my proposed answer. Implementation requires innovation in monitoring technology, automated data entry, medical literature retrieval, artificial intelligence tools and more. At first these guardians may be entirely human, but perhaps eventually they can be largely automated.

It is a massive task, the work of centuries. We better get started.