CHAPTER THREE

MAKING—OR MISSING—THE DIAGNOSIS

It was another one of those bursting-at-the-seams types of days in my clinic. Every scheduled patient showed up, plus a few extra added on. Everybody seemed to have burning concerns that needed immediate attention. One patient had newly diagnosed thyroid disease but the medication was making her feel worse rather than better. Another was having strange twinges in his lower abdomen. One woman had muscle aches in her arms that were now extending to her legs. Another patient was experiencing drilling-type pain in his lower back that spread up to his neck and scalp. One man had a cough that just wouldn’t leave. A woman was concerned because the soles of her feet felt like they were on fire. Another patient said that she simply had no energy and could hardly get through her day.

For each of these presenting symptoms there is a gamut of possible causes—what doctors call the “differential diagnosis”—that range from the prosaically benign to the concerningly urgent to the immediately life-threatening. The name of the game is to come up with a broad differential for each symptom, then prioritize them by likelihood and by severity. Testing for every possible diagnosis is not feasible, so the doctor needs to ask the right questions, listen carefully to the answers, do the right kind of physical exam, and pay attention to the clinical clues.

If you had the luxury of an hour with each patient, you would have the time to diligently sort through each and every possibility. But the reality is that you have just a few minutes to push the majority of diagnoses to the bottom of the list, come up with the most likely few at the top—being careful, of course, to keep in the rare but life-threatening possibilities—and then explain to the patient what you think. You can order labs, X-rays, and the like, but those results won’t come until later. You need to offer the patient, right now, the most likely diagnoses, as well as a plan for how to start treatment and/or further investigation.

It’s a tall order, and an incredibly stressful one. Most textbooks treat diagnosis as a leisurely intellectual process. Medical students are taught to run through each organ system of the body and then consider all the plausible ways that organ system can get pillaged—via trauma, via infection, via metabolic derangement, via cancerous transformation, via toxic exposure, via genetic abnormalities. It’s a thrilling academic exercise, especially if pondered over with a steaming cup of tea and plate of crumpets with jam while the sun eases its languid way across the firmament. Gentle strains of Chopin in the background don’t hurt either.

Unfortunately that is not the way differential diagnosis happens in real life. At best we have a minute or two to navigate the wide chasm of possibilities. For each patient, the presenting symptom could be nothing or could be something. Or could be something horrible. Was the patient with no energy just not getting enough sleep? Or was she anemic, or hypothyroid, or depressed? Could she be suffering from pancreatic cancer? Maybe she was experiencing domestic violence?

Was the patient with muscle aches experiencing a medication side effect or exhibiting the onset of a systemic inflammatory disease? Did the gentleman with abdominal twinges have vascular compromise to his intestines or was he a hypochondriac? Or was he taking some Slavic weight loss concoction he’d purchased on the Internet?

As I raced through my day, I struggled to be thorough while trying to avoid falling too woefully behind schedule. The ulcer gnawing at the pit of every doctor’s stomach is “What if I miss something serious?”

General practitioners such as internists, family doctors, pediatricians, and emergency doctors face the biggest challenges because the diagnostic field is so wide open. We all want to get it right, but we also don’t want to over-order tests that can be harmful or expensive or yield too many false positives or all of the above. We may want to allow some observation time to see if the symptoms self-resolve or progress, but then we worry about missing a serious illness, harming a patient, or getting sued. Some days it feels shocking that we get it right at all.

And we do, on average, get it mostly right. Doctors’ diagnostic accuracy is estimated to be in the range of 90%.1 That, of course, implies a 10% error rate, but on days when it feels like you are being pelted with diagnostic possibilities from every cell of the body and that imminent death is lurking everywhere you turn and you have only minutes to make those decisions, that 90% number is comforting.

On cooler-headed days, though, that 10% error rate is disturbing.

Thirteen years into my clinical practice as a doctor, I took a year-long writing sabbatical. While I was away, a new faculty member took over my patients. When I returned, I sat down with my colleague and got the notable updates on my patients. By all accounts, everything went smoothly during my year away. Then she paused. “There was one thing, though . . .” And from the tone of her voice I could tell it wasn’t something good.

Ms. Romero was a 69-year-old woman who had been in my practice for several years. She was generally well, and our visits were mainly focused on helping her lose weight, exercising her arthritic joints, and adjusting her thyroid and blood pressure medications. Her adult daughter in her home country had suffered a devastating illness, and the stress of that dominated many of our visits.

“There was an anemia,” my colleague said slowly. “And it looked like it was never worked up.” Anemia? Not worked up? I could feel my gut starting to winch painfully. Had I missed an anemia? I fervently hoped it would turn out to be the mild-annoyance type of anemia, as most anemias are.

I was wrong.

“It turned out to be multiple myeloma,” she informed me quietly, and my gut bottomed out to my ankles. I’d missed a cancer.

The differential diagnosis of anemia is as prodigious as it is varied. Anemia can be caused by deficiencies of nutrients such as iron, B12, or folate. It can be caused by heavy menses, or a bleeding stomach ulcer, or bleeding from anywhere. It can be caused by liver disease and kidney disease. It can also be caused by inflammatory disease, bone marrow disease, and, of course, by cancer. Red blood cells can rupture—hemolyze—and cause anemia. Alcohol, medications, and toxins can cause anemia. HIV and parvovirus B19 can cause anemia. Anemia can be a fellow traveler with a host of chronic diseases. The list is endless.

Ms. Romero didn’t have any symptoms that might suggest some of the more obvious causes. Her other blood tests were basically normal, which ruled out a slew of other causes. She was up to date with her regular cancer-screening tests such as mammogram and colonoscopy, so it wasn’t any of the obvious cancers.

Multiple myeloma is a blood cell cancer that normally trundles along in a protracted, indolent phase before it “presents.” Patients typically come to medical attention because of bone pain and unexplained fractures. The other way it is often diagnosed is when a high calcium level is noted incidentally on a blood test for other reasons.

Ms. Romero never had any of these symptoms—she always felt fine when I saw her—and her calcium was always normal. But it was clear, in retrospect, that she had been quietly developing multiple myeloma during the years that she was under my care.

The family was very upset. Even though the treatment and prognosis weren’t affected by the delay in diagnosis—given the slow biology of the disease—it was still devastating to them. They were angered that this hadn’t been diagnosed earlier.

“I’m sorry to have to give you the bad news,” my colleague said, not unkindly. “But if it were me, I’d want to know.”

It hardly gets more horrible than missing a cancer. Beyond that, I felt I had violated my patient’s trust in the most fundamental way. Ms. Romero had entrusted her health to me for all of these years, assuming that if she came regularly to her appointments I would make sure she would be okay.

Over the years I’ve written a lot about the power of emotions over intellect, but I’d never felt it as acutely as at that moment. The intellectual side of me wanted to immediately tear through Ms. Romero’s chart with a fine-toothed comb to figure out exactly where and how I’d blundered. While I knew this couldn’t change the outcome or undo the mistake, I needed to pinpoint my error. When I was doing my PhD in a biochemistry laboratory, every failed experiment was followed by a rigorous autopsy; it was the only way to do better in the future and it was simply the right thing to do.

And yet. . . .

And yet, I could not bring myself to do it now. My horror at missing this diagnosis, at letting down my patient (not to mention the shame of an error revealed publicly in all its glory) got the better of me. The emotions were so overpowering that I could not manage more than a cursory glance at Ms. Romero’s chart.

Over the years that ensued, I occasionally caught sight of Ms. Romero in the clinic. (It was clear—and frankly appropriate—that she preferred to stay with my colleague rather than return to my practice as all my other patients had after I’d returned from my sabbatical.) Each time I spotted her, I so wanted to approach her—to apologize, to explain, even just to wish her well during this difficult journey. But I was torn. Would I be doing this for her or for me? Was I trying to make amends with honesty or was I just desperate to clear my own conscience?

I tried to imagine it from her perspective: Would my appearance, with however earnest an apology, be welcome or would it be an intrusion? Would I be offering balm for a festering wound or churning up old emotions that had already settled?

I knew from my own research into medical error that what patients typically most want is honest acknowledgment and apology from their doctors. But if Ms. Romero had already made peace over the years with her cancer and its diagnosis, my reappearance might serve only to unsettle the homeostasis. Would it actually help her, or was I being selfish, hoping she’d offer some sort of absolution to assuage my terrible guilt? And if my visit wasn’t well received, I wouldn’t be able to undo it.

It was that latter realization that made me decide not to approach her. I couldn’t countenance the chance of harming her even more. Though maybe I was just making excuses to hide the fact that I was basically chickening out, unable to face my own incompetence straight on.

It was another five full years before I could bring myself to examine her chart in detail. I had just learned that Ms. Romero died—by all accounts peacefully, comfortably, at home with her family. I was heartsick about the whole situation, especially that I might have caused extra pain for Ms. Romero and her family, beyond even the misery of cancer itself.

Now it was time, I decided. I needed to face up to my error, even if belatedly and even if the error hadn’t necessarily changed the outcome of the disease. It was still an error. I’d missed an anemia.

I pored over the chart obsessively, determined to do the analysis I’d been unable to attempt previously. I went at it voraciously, charting her clinical course almost to the day, vowing not to rest until I’d cornered my error and figured out how not to do it again.

In medicine, though, nothing is ever as clear as the textbooks. Diseases—and patients—never seem to follow those flow charts so confidently constructed in the canon of medicine.

Anemia is diagnosed by the hematocrit, which represents the percentage of the blood occupied by red blood cells, and has a fairly wide range of what is considered normal. Depending on the laboratory, normal can range from 39 to 50 for men and 36 to 46 for women, though many women live comfortably in the low-to-mid 30s due to continuing menstrual blood loss.

The hematocrit is measured as part of a CBC (complete blood count), which also includes the platelets that are involved in blood clotting and the white blood cells that are part of the body’s immune system.

Despite the common practice of ordering “executive profiles”—broad arrays of lab tests that are done to “check everything”—there is, in fact, no reason for a CBC to be ordered for an asymptomatic adult patient. There is no clinical guideline that recommends routine CBCs because the risk of false positive values (abnormal results when in fact there is no disease) is so high. A CBC should be ordered only when there is a clinical reason to do so, such as a bleeding issue (to see if the hematocrit has dropped and to check for sufficient platelets for clotting), an infection (to check for an abnormal white count), or a new symptom of fatigue (to look for anemia). This stands in contrast to tests for things such as cholesterol, where there are particular age recommendations for routine screening and robust clinical data to back up the utility in asymptomatic patients.

There wasn’t any reason for me to have ordered a CBC based on Ms. Romero’s other conditions, so the few CBCs that were in her chart before I left for sabbatical were likely ordered by other people for other reasons (if she had come to the ER, for example, with a pneumonia).

There were one or two CBCs in the computer from nearly a decade earlier, and her hematocrit had been 37 or 38. Then, about five years before I left, she was admitted to the hospital for abdominal pain. At that time, a CBC was done—likely looking for an elevated white count that might suggest an infection. The hematocrit was 35 and the next day it was 31. An acute drop like that can be from active bleeding—but there was no evidence of that in the chart. In a hospital setting a drop in the hematocrit can also be an artificial drop after aggressive IV hydration. (Because the hematocrit is a percentage, it can appear to drop if you suddenly pump in a lot of fluid. The number of red blood cells—what you are actually concerned about—is still the same.)

The abdominal pain turned out to be nothing serious and Ms. Romero was discharged. For some reason, she did not get an appointment with me and was seen in the walk-in clinic the following month. The CBC was repeated, and now the hematocrit was back up to 35, basically the same as when she had been admitted to the hospital. That doctor may have attributed the drop to 31 to hydration, and now it was back to baseline. It was likely assumed she “lived” in the 35 range and that the 31 was a spurious value.

I can’t say for sure, though, because the doctor’s thought process was not documented in the chart. At my next visit with Ms. Romero I’d probably seen the results in passing and may have made the same assumption. But I can’t confirm that, because I, too, hadn’t written anything in my note from that day about the hematocrit. I probably hadn’t seen any need to write out my thoughts on a presumably normal lab value.

Separate from the issue of how we doctors mentally viewed Ms. Romero’s lab results, there’s also the physical manner in which we viewed those results. In our EMR at that time, the first result in the set would come up, and the rest would plug through—one by one—as you hit the return key. In this manner you’d see the CBC, then the metabolic profile, then the liver function tests, then the thyroid test, etc. However, for each individual lab test, you could also select a Trend function, in which you could compare this lab test to similar ones in the past.

Trending is extremely helpful, for example, in following the blood sugars in someone who is diabetic, allowing you to observe the trend over time. But if a test is normal, there generally isn’t a reason to take the extra steps to examine the trend. Even in EMRs that automatically compare current results to prior results, the default setting is usually just to the past few rounds of blood tests. To do the trend over a long period of time, you have to make a conscious decision.

When I did the trend on Ms. Romero’s CBC now, I could see a gradual drift downward over the decade, but it obviously hadn’t caught my eye in real time. Perhaps I’d only compared her CBC to the one just prior, and I’d never done the trend all the way back. Maybe I hadn’t noticed the drift because I wasn’t actively “following” her blood count over the years, as I might have been, say, for a woman with large uterine fibroids who was experiencing bleeding.

I could see in the chart that shortly before I left for my sabbatical, Ms. Romero had gone to the walk-in clinic for symptoms of dizziness. A CBC at that time showed a hematocrit of 30. The doctor who’d evaluated her noted that this was below her baseline of 35. But her dizziness resolved at a follow-up visit—it was attributed to a concomitant viral syndrome—and the hematocrit of 30 must have fallen off that doctor’s radar.

When Ms. Romero saw me a few weeks later, the focus of the visit was on her ill daughter. In my note from that day, I had commented about the resolution of her dizziness. But there was nothing written about the hematocrit of 30. Perhaps I simply hadn’t seen the CBC. Or maybe I had, but didn’t do a trend to compare to the earlier values (though 30 is low enough that it should have caught my attention by itself). Or maybe she’d been weeping in my office about her daughter—as had happened before, and we’d spent the entire time processing her grief. Or maybe I was rushing and hadn’t been as thorough as I should have been. I can’t say.

Reconstructing my thinking about why one particular strand of hay did or did not stand out in my mind compared to all the other strands of hay in the vast haystack is impossible, especially for a medical encounter years ago amid hundreds, even thousands, of other medical visits I’d done. (Though this is precisely what is expected in malpractice cases.) I wished fervently that I could reconstruct my mind-set, but it would be intellectually dishonest to even intimate that I could. In reality, I have no idea what I thought at the time.

When my colleague took over Ms. Romero’s case a few months later, she was seeing a “new” patient and so evaluated everything from scratch. The low hematocrit of 30 jumped out at her immediately and she repeated the test. It was now 23. The hematocrit had plummeted in the few months’ gap of our handover of care, and it was now eminently clear that something was very wrong. At that point, the hematologists did additional testing, including a bone marrow biopsy, and rapidly made a diagnosis of multiple myeloma.

Ms. Romero’s case has haunted me over the years. For doctors and nurses, it’s devastating to have missed a serious diagnosis, and agonizing to contemplate the additional distress you’ve caused your patient, above and beyond the illness itself. I wish I could rewind my brain those many years to figure out how I missed it. Had I been distracted? Was I running behind? Was I cutting corners to catch up? Was I willfully ignoring the lab data? Was I just having a bad day?

I discussed the case with Hardeep Singh, an internist from Houston who heads up the patient-safety initiatives at the VA hospital there and is an internationally known guru on medical error. Singh pointed out that diagnostic error is a completely different can of worms from procedural error (e.g., surgery on the wrong side of the body or an infection from a central line) because diagnosis is a moving target. He defines diagnostic error as “a missed opportunity to make a correct or timely diagnosis,” even if the patient isn’t harmed by the delay. The question is whether you could have done something different.

“Diagnoses evolve over time,” Singh explained. They also evolve over place and people, as patients sometimes see different doctors or go to different hospitals to continue their care. Autopsies, which used to be the way that the medical system unlocked the answers, have steeply declined because of financial priorities as well as cultural norms. This is partly why diagnostic accuracy is fiendishly difficult to measure precisely. “Doctors often have no way of knowing if they’ve made a diagnostic error or not,” Singh said.

Take the case of abdominal pain. This symptom is a daily occurrence for clinicians who work in general medicine, emergency departments, or urgent-care centers. There are days when the presence of abdominal pain seems as ubiquitous as oxygen, and this ubiquity can sometimes dull the sense of urgency. But the differential diagnosis of abdominal pain is as extensive as it is varied. It ranges from the common (acid reflux) to the rare (porphyria), and from the benign (constipation) to the life-threatening (intestinal rupture).

When I talk with a patient and examine her abdomen, my mind runs through this list, prioritizing which diagnoses seem more likely but still trying to keep in mind the more serious conditions that I can’t afford to miss, even if they are rare. Each additional thing the patient says—no blood in her stool, occasional nausea but no vomiting, no weight loss or fever—rejiggers the list in my head. Her background medical history—absence of cancer, presence of hypertension—as well as the medications she does or does not take, all enter the mix, causing the diagnoses to reshuffle yet again.

There are some situations in which getting the exact right answer isn’t necessary. Just corralling the differential diagnosis into categories of serious versus non-serious is often sufficient, at least initially. If I can separate out those patients who don’t have any “red flag signs,” I can hold off on expensive tests and simply observe to see what happens. Many patients will get better with just the tincture of time, and for those who don’t, I can then consider doing additional tests.

In other cases, it can be reasonable to proceed with a treatment even without a specific diagnosis. An acid blocker, for example, can treat both acid reflux and gastritis, so it often isn’t necessary to do the tests to distinguish between them. Not making a specific diagnosis in this case would not be considered an error.

However, I often never learn the outcome because many patients don’t come back. If they are feeling better, why bother? If they are feeling worse, they may head to an emergency room, or to another doctor. If that next doctor does an endoscopy and discovers a gastric ulcer containing the bacteria H. pylori (which would require antibiotics in addition to acid blockers), then I clearly made an error, right?

Not necessarily, according to Singh. It would depend on whether there had been a clue that I’d missed that should have directed me to order an endoscopy right away—if the patient was feeling full more quickly than usual, for example, or was losing weight or noticed blood in her stool. But there are many cases in which the symptoms are not concerning enough to warrant an invasive test initially, and observation is appropriate. If the symptoms don’t improve, then at the next visit, the endoscopy is warranted. That an ulcer was found at that time wouldn’t necessarily mean that I’d made a mistake initially. It simply illustrates that diagnoses are moving targets and often require a period of time before they can be pinned down. Doing an endoscopy on every single patient who comes to an internist with abdominal pain would certainly catch all those ulcers up front, but it would be a logistical—and financial—impossibility, given that nearly all human beings have abdominal pain at some point in their lives.

Moreover, endoscopies have side effects. The actual risk of major adverse events from endoscopy (infection, bleeding, tearing of the stomach or esophagus, cardiac arrest from the sedation) is low, something in the range of one in a thousand. But if you do endoscopies on the millions of people out there with abdominal pain, these devastating outcomes will start to add up. And if those endoscopies ultimately weren’t necessary for the vast majority of patients, then you’ve simply exposed patients to harm without offering much benefit. That is why, in the absence of red flag signs, most doctors typically observe first and then consider endoscopies only if patients don’t get better with basic treatment.

Much of medical error research focuses on the inpatient setting, when patients are hospitalized. Errors in the inpatient setting are often more dramatic and thus more easily come to light. There’s also the convenience factor for researchers—patients and their medical teams are all tethered to one central location. Plus, there’s a trove of data to work with, since inpatients endure boatloads of tests in concentrated time periods. The majority of medical care, however, takes place in the outpatient setting—in doctors’ offices, clinics, and community health centers. In these settings it’s far harder to identify diagnostic errors because patients may spend only fifteen minutes in the medical setting and then waltz off into their own lives, far from the clutches of researchers and their clipboards.

Slowly, though, the field is turning its lens to this more unwieldy side of medicine. Singh and his colleagues undertook a study to get a sense of the scope of diagnostic error in outpatient, primary care medicine. They quickly learned that it’s much more challenging to track down errors when patients are free to go wherever they please, whenever they please. How would the researchers even know where to look for the errors? Nobody checks off “error” in the chart when there’s a mistake.

Singh and his colleagues reasoned that patients who experience a misdiagnosis are probably more likely to need additional medical care soon, since their problem wasn’t correctly identified at the outset.2 They thus hypothesized that anyone getting a second round of medical care within two weeks of a first visit would be more likely to have experienced a diagnostic error.

The handy part about using this metric was that the researchers could create an automatic trigger in the EMR. For any visit to the doctor, a red flag would automatically go up if—within two weeks—that patient returned for more medical care, such as another doctor’s visit, admission to the hospital, or a visit to the ER or urgent care. Obviously, it was of no help if the patient sought care in another system, but many patients stay within their system for insurance reasons or for simplicity of care.

For every case brought to light by the trigger, researchers would then delve manually into the chart to see if there was anything that pointed to diagnostic error in the first visit, or whether it was just happenstance that there were two medical evaluations within two weeks of each other. Out of 212,165 visits, the trigger uncovered 674 patients who were hospitalized and 669 patients who returned for another outpatient visit (to a doctor, urgent care, or an ER) within two weeks of their initial visit. The researchers compared these to 614 “control” visits that were not followed by additional medical care within two weeks.

The researchers took into account that many diagnoses evolve over time (e.g., symptoms that initially suggested a viral syndrome but after a week progressed to a pneumonia). These they did not consider errors. They looked for cases in which there was some clue in the initial visit that might have been overlooked (e.g., symptoms suggested a viral syndrome, but there was an abnormality on the lung exam). They categorized an error only when there was a missed opportunity to make an earlier diagnosis.

In the group of patients who ended up admitted to the hospital within two weeks of a visit, 21% had evidence of a diagnostic error. In the group that went to the ER, urgent care, or back to the doctor within two weeks, 5% had a diagnostic error. By comparison, only 2% of the control group had an error.

The most commonly missed diagnoses were pneumonia, heart failure, renal failure, cancer, and urinary infection. I was most taken with the observation that nearly all the errors (more than 80%) were related to a problem in the doctor-patient interaction (as opposed to errors related to tests or referrals or patient actions). This means that the problems—and the potential solutions—are grounded in how doctors and patients interact.

Singh and colleagues teased apart this interaction and found that the three elements that contributed the most to errors were the history, the physical, and the ordering of diagnostic tests. (Failure to fully review the patient’s chart—which might have been my error with Ms. Romero—contributed, but to a lesser degree.) Although the history, the physical, and the ordering of tests feel like separate elements, they are really part and parcel of the unified thinking process that goes into evaluating a patient. It reinforces the idea that diagnostic error relates to cognitive errors—how we think. And of course, how we think is much less amenable to the checklist approach we can use in other areas of medical error.

The researchers noted that 80% of visits lacked a differential diagnosis. That is, in only one out of five visits did the doctor deliberatively consider alternative diagnoses in strong enough terms to document them in the chart.

Mark Graber, another leading researcher in the area of diagnostic error, sees this statistic as the core of the problem. “Doctors need the discipline of the differential diagnosis,” he said. “We all learn it, but as we get expert, we stop doing it.” Experienced doctors are so fast at recognizing common medical conditions that we jump to a diagnosis in seconds. And the minute we find a diagnosis that seems to explain the findings, we stop looking for any other explanation. We stop thinking. This simple form of pattern recognition works well for straightforward conditions, but it can be a trap in more complex situations.

Graber and Singh tried to outline the possible ways doctors could improve diagnostic accuracy.3 Recognizing that so many diagnostic errors are cognitive in nature, they grouped the possible interventions into three broad categories: increasing knowledge, getting help, and sharpening the thinking process itself.

Increasing knowledge is the traditional manner of bettering ourselves in medicine. The years of medical training, the licensing exams, the recertification process—these are the brute-force ways in which we doctors pound more facts into our heads. In general, this is reasonably successful (it’s how we all became doctors, after all), but it’s limited by how much stuff you can pack into a given brain.

There are many programs to increase knowledge, but the ones that work best target specific areas and involve intensive training. Periodically, my hospital—and most health organizations—will turn its attention to a single hot-button issue. A few years back, we had an all-hands-on-deck effort to improve the diagnosis of depression. There were workshops and programs and task forces and quality-improvement projects. At times it was overwhelming, but it did significantly raise the profile of depression diagnosis and treatment. Another time there was a similarly intense focus on diagnosing hospital-acquired infections. Another time it was a major initiative toward increasing the accuracy of every patient’s medication list.

All of these are valuable initiatives directed at important clinical conditions. It’s probably true that our patients benefited, but it’s always difficult to prove this. You have to weigh the benefits against the costs in money and time, and also consider what falls by the wayside when everyone converges on one high-profile issue.

Even with the benefits, I’m always left with mixed feelings after each of these full-frontal onslaughts. It’s as though we focus the microscope lens of our medical practice on one narrow sliver of medicine, briefly but intensely. For the moment, you become an expert in that one disease. It’s in the differential diagnosis for every single patient you see for the next few months. It’s the super-number-one priority in everyone’s daily life until it starts to fade into the background. Then some other “quality-improvement project” puts a bee in everyone’s bonnet and we’ve completely forgotten about the first one.

It’s not possible to practice medicine that way. You can’t keep lasering in on this issue and then that issue and then that one over there. It’s discombobulating and ultimately exhausting. The field is just too vast—especially in primary care, where every possible malady of every possible organ is fair game. You can’t know—and tackle—everything in medicine.

Recognizing the impossibility of knowing everything, Graber and Singh recommend “getting help” as another approach to reduce diagnostic error. Getting a second opinion is something that patients often do, but doctors and nurses can do it too. Curbsiding a colleague with a question is common practice but we could also pursue a formal second opinion the same way patients do. The caveat, though, is that a second opinion is just that—another opinion. While second opinions can help clarify, they can just as easily muddy the waters. And of course, any opinion also has the potential to introduce errors into the situation.

Sharpening the thinking process—the third option—is a much tougher nut to crack. I’ll delve into this more in chapters 5 and 15, but even a relatively straightforward case can illustrate the wide range of diagnostic thinking techniques that clinicians can use.

A 57-year-old patient of mine once came to see me about tingling in her hands. She had mild hypertension and elevated cholesterol but was otherwise healthy. Her symptoms weren’t exactly in the classic distribution of carpal tunnel syndrome (thumb and first two fingers), but most primary care doctors stick with the dictum—initially, at least—that common things happen commonly. Experience has taught us that an atypical presentation of a common illness is far more likely than a rare illness. (As they say in medical school, “If you hear hoofbeats, think of a horse, not a zebra.”) So even if her symptoms didn’t fit the textbook exactly, it still was much more likely that she had carpal tunnel syndrome (or some benign variant thereof) than multiple sclerosis or stroke.

I told her to purchase over-the-counter wrist splints, take some ibuprofen, and come back if her condition didn’t improve in a few weeks—a fairly standard primary-care response. Looking more critically at my diagnostic process, I know that I did not enumerate a differential diagnosis in the chart. I probably jotted something along the lines of “no red flag signs, most likely carpal tunnel syndrome.” If my chart were analyzed by researchers like Graber and Singh, I’d be dinged for not documenting an explicit differential diagnosis. Of course I’d actually done one in my head, unconsciously, as I interviewed and examined her, mentally eliminating diagnoses that seemed unlikely.

I also knew that the stakes were low; that is, I didn’t worry about her dropping dead within a week from a missed diagnosis. I could have ordered electrodiagnostic studies to document carpal tunnel syndrome, but that didn’t seem worthwhile. If her symptoms got better within a week or two of her wearing wrist splints, frankly, that’s good enough for primary care. And if they didn’t, well, then, maybe that would be the time to start ordering tests.

Well, my patient wasn’t so happy with my wait-and-see approach, so she promptly saw another doctor, who happened to be a cardiologist. He immediately sent her for electrodiagnostic studies, but he also ordered a nuclear stress test, an echocardiogram, venous dopplers of her legs to look for blood clots, and an arterial-brachial index test to evaluate for arterial disease—all of which were normal.

When she brought me the stack of tests at our next visit some months later, I was floored at the extensive workup she’d received. To me, none of the additional cardiac and vascular tests were indicated. The most charitable explanation I could think of was that yes, sometimes heart disease can present with pain that radiates to the left arm and hand, and yes, this patient’s hypertension and cholesterol put her at risk for heart disease. But that was an enormous stretch, given her symptoms of tingling in both hands. The least charitable explanation was that the doctor took a look at her insurance and ordered every feasible, billable test. The middle-of-the-road explanation is that when you are a hammer, everything looks like a nail. If you are a cardiologist, any blip of physiology makes you think of the heart and the vascular system.

Still, I wondered about our diagnostic approaches and which of us—if any—was making an error. The cardiologist did order the appropriate test for carpal tunnel syndrome. The electrodiagnostic study, incidentally, came out negative; the patient did not have carpal tunnel syndrome, or at least not to any significant degree. So my diagnosis had been wrong. Of course, her hand tingling had improved after using the wrist splints, so maybe I’d been right. Or her hand tingling improved despite the wrist splints, so maybe I’d been wrong.

When I put our two approaches to Hardeep Singh, he described it as “the tension between under-diagnosis and over-testing.” With my wait-and-see approach, I was running the risk of missing the diagnosis. With the cardiologist’s order-every-test-in-the-book approach, he exposed the patient to the risks of over-testing, which include false positives, harms from the tests, and whopping medical bills. “This pendulum of under- and over-testing is what we experience as physicians every day and there are several factors that make physicians swing from one end to the other,” Singh said. “Some of us are more tolerant of uncertainty than others. The cardiologist likely doesn’t manage uncertainty well, but there could be other reasons he pushed for all those tests, including just giving reassurance to the patient.”

In textbooks, diagnoses exist as pristine, self-contained entities, with crisply itemized criteria and clearly enunciated testing strategies. In real medicine, however, uncertainty reigns. Symptoms can be vague, or patients may exhibit only a select few of the symptoms of a particular diagnosis. Patients’ symptoms can overlap with a dozen other diagnoses and they can evolve and change over time. Additionally, symptoms can be obscured, such as a fever being masked because the patient took acetaminophen. Nailing a diagnosis amid all this uncertainty has been described as trying to find a “snowball in a blizzard.”

If making a diagnosis is tough due to uncertainty, teasing out diagnostic error in this ocean of uncertainty is downright daunting. Did a doctor make a mistake in diagnosis, or was she just navigating reasonable uncertainty? This task is made even harder because most research in diagnostic error is done using the EMR. Doctors are forced to select a specific diagnosis code, dictated, of course, by billing purposes. Insurance companies reimburse for diagnoses, not for uncertainty. So there is no legitimate place in the EMR to indicate uncertainty.

Uncertainty also segues into the all-important issue of context. Primary care doctors and cardiologists practice in starkly different contexts. As a primary care doc, I take care of everyone from the worried well to the truly sick; there is no filter in my practice. This is one of the things that I love about primary care, but my front-and-center task, then, is to separate out the aches and pains of daily life—of which there are many!—from the graver conditions that require more urgent intervention. This dizzying range of variation means that my practice is awash in uncertainty.

A cardiologist, however, has a practice that has already been filtered. A patient is referred to a cardiologist generally because another doctor strongly suspects heart disease. The amount of uncertainty is therefore much less. There might, then, be value in running an extensive panel of tests on these patients because the baseline prevalence of cardiac disease is so much higher.

By contrast, if I did that same panel of tests on my primary care patients, I’d easily make a host of diagnostic errors. In a group of patients whose illness severity is “diluted out” by all the 18-year-olds with acne and the 35-year-olds with tendinitis, the overall rate of heart disease will be much lower. Any positive results on those tests could very well be false positives.

When my patient came to me with tingling in her hands, I didn’t know exactly what her diagnosis was, but I suspected that it would end up in the broad category of minor-symptoms-that-usually-get-better-and-don’t-cause-serious-harm. While that isn’t a standard medical term you’d find in a textbook or in the EMR, it’s a useful way of thinking for primary care doctors on the front line. It might not be the best approach, however, for cardiologists or critical-care doctors, who practice in very different environments with different patient populations.

Diagnostic tests—and diagnostic approaches—do not exist in a vacuum. Context and uncertainty always matter. As Jay’s story continues, you will see how different doctors, nurses, and family members navigate uncertainty and how interpretations of context can often be conflicting. I’ll discuss more about the diagnostic thinking process itself in chapter 5.