CHAPTER SEVENTEEN
GETTING IT RIGHT
Much of this book has focused on two singularly tragic cases. Jay and Glenn experienced some of the harshest iterations of our modern healthcare system. I chose these cases to highlight because of the many lessons they offer (and of course because of the remarkable generosity of their families).
But in some ways, Jay’s and Glenn’s cases are atypical of medical error. For one, they represent the catastrophic end of the medical-error spectrum, whereas most medical errors—though equally important to tackle—do not cause such harm. The other way in which their cases are less typical is the degree of inadequate medical care. In fairness, I was not able to interview the medical staff in these two cases, nor was I privy to the full details of the medical records or the negotiated settlements. But from what I could gather as an outsider (and from reviewing these cases with experts), some aspects of the recognition and treatment of these two severe conditions—sepsis and burns—fell glaringly short. I can’t use the legal term “negligence”—again, I don’t have all sides of the stories—but the medical care in both cases seemed wanting.
These cases of flagrant mismanagement and the seemingly uncaring attitude of medical staff are what make headlines and are thus what most people think of when they imagine medical error. But in reality, most errors are made by conscientious and competent staff who care deeply about their patients. These doctors and nurses are not perfect—that’s for sure—and many are not as good as they could be or should be. But when they make an error or somehow harm their patient, most are torn to shreds with remorse.
The thing that is typical about Jay’s and Glenn’s cases is that medical error is rarely one thing only. A bad outcome is most often the result of multiple errors snowballing on top of each other. And what these cases most exemplify, in my opinion, is that tragic outcomes usually result from a combination of specific checklist-able items (like when to remove a catheter or when to transfer to a burn center) and those less tangible qualities, such as clinical reasoning, intellectual humility, effective communication, and a sense of ownership.
As we try to make the medical system safer for patients, we have to consider both aspects. The less tangible qualities are the hardest to imbue. They can only be created by boots on the ground, by people who exemplify these traits, teach these skills, and demand comparable behavior from their colleagues. The concrete items are logistically easier to approach, and that’s where most healthcare organizations focus their efforts. However, it’s not possible to bombard the staff with 3,471 checklists—though some hospitals certainly try. We need to rejigger the system to make it less possible for people to commit errors.
Human-factors engineering is the budding field that examines how people interact with the tools and technologies around them, and uses logical design to make things intrinsically safer. A simple example is the food processor that sits on nearly every kitchen countertop. The whirring blades could easily sever fingertips if a person reached in for a taste while, say, pulverizing cauliflower. Using the principles of human-factors engineering, the designers created a system in which the motor is unable to start unless the cover is locked into place. So if your brain falls under a cruciferous swoon, unable to control its urges to snag some beguiling florets in flagrante delicto, the machine will outsmart you and click off as soon as you open it.
I discussed a comparable example earlier in the book, in which anesthesiologists can no longer mix up nitrogen and oxygen in the operating room because the tubing and connectors are no longer interchangeable. The healthcare system is riddled with opportunities like this, where the system could be revamped to help humans make fewer errors. Often, however, they come to light only after a tragedy.
Heparin is potent blood thinner that is used to treat blood clots. In a far more diluted form, heparin is also used to flush catheters and IVs in order to prevent the tubes from clogging. In 2006, six premature infants in Indiana had their catheters accidentally flushed with the more concentrated heparin. Three of them died.
In examining the error, hospitals realized that the diluted and concentrated heparin came in identically sized vials, and that you had to squint at the label to distinguish them. The manufacturer issued a safety alert and changed the labels so they would be easier to tell apart. Unfortunately, not all hospitals had removed the old vials, and a year later three more infants in California accidentally received the concentrated dose. Luckily, none died, but it pointed out that human-factors engineering has to be thought out carefully.1 Hospitals had to take more drastic steps, like physically removing the vials of concentrated heparin from the general wards. They also switched to prefilled syringes that were of different sizes entirely.
Other ways that human-factors engineering can be used to reduce medical error include the physical architecture of hospitals and clinics. We know that handwashing is the most critical technology for infection control, but sinks are often inconveniently located, based more on the plumbing geography than anything else. Hospitals could be proactively designed to allow sinks to be closer to patients, where doctors and nurses are more likely to use them. However, if the sinks are poorly designed, they themselves can become vectors for spreading infections. Sinks that gush straight into the not-so-clean drains, for example, can cause splashback, and this has been implicated in outbreaks of infection in several hospitals.
There’s more geographic flexibility with alcohol-based hand sanitizers. They can be placed just about anywhere, but it’s amazing how often they manage to be in inconvenient locations. The one in my exam room, for instance, forces me to turn my back to the patient every time I wash my hands. Besides forcing me to be rude to my patients, it deprives the patients of their ability to accurately monitor the safety situation. If we want patients to be empowered to point out things like doctors forgetting to wash hands, they need to be able to see it. And of course, alcohol-based sanitizers are not a complete substitute for handwashing. While they inactivate microorganisms, they don’t necessarily kill them. Notably, they don’t work for C. diff, the organism responsible for much of hospital-acquired diarrhea. Additionally, as anyone who uses sanitizers eighty-five times per day can tell you, their desiccating power is prodigious, and most of us walk around with appendages that resemble prunes imported from the Kalahari.
Human-factors engineering for infection control can strategize beyond facilitating handwashing and look for ways to decrease the squatting rights of hospital microbes. Simple things such as shortening the sleeves on lab coats and outlawing neckties could keep dangly material from sweeping bacteria from patient to patient. Bed rails—perhaps the prime dalliance site for hands and bacteria—could be made of antimicrobial materials such as copper. The privacy curtains that surround beds and exam tables—as well as lab coats and scrubs—could be made of material that is less hospitable to microorganisms. Though handwashing remains the primary way to prevent spread of infection, these types of designs can act as a check on cognitively overloaded humans who occasionally forget to wash their hands.
An idea that shows promise for decreasing medication error is the creation of what are known as medication safe zones in pharmacies and nursing stations. These safe zones would include revolutionary features such as bright lighting and magnifying glasses to read fine print. They would also organize medications in cognitively logical ways. Rather than just alphabetize everything, for example, the system could be set up so that commonly used medications are easier to access, or so that everything needed for an IV setup is located together.
Most importantly, though, these safe zones acknowledge the need for mental focus. In a typical nursing station, you can hardly maintain a train of thought to successfully scratch your left ear, given the blizzard of phones ringing, alarms blaring, staff crisscrossing, housekeepers mopping, patients requesting ginger ale, wandering visitors looking for their loved ones, clueless medical students who can’t work the electronic thermometer, post-call surgery residents hoping to score leftover donuts, a transporter waiting impatiently to bring Mr. Atkins to radiology, and then the nursing supervisor swooping in to remind everyone to be shipshape because of the upcoming Joint Commission inspection. Oh, and the EMR will be down for the next few hours because of a new roll-out, for which there will be a required inservice instruction session, but there aren’t enough staff to cover, so every third nurse will be floated to a different unit. Oh, and the staff restroom is out of order again, so you’ll have to use one on a different floor; please stagger your bathroom breaks accordingly.
While I exaggerate perhaps a touch, this is not so far off from the reality of a typical nursing station in a busy hospital ward. It would be challenging to invent an environment less conducive to the dogged thoroughness that is required to get every medication right every time for every patient, which is, of course, what is expected. The goal of medication safe zones, therefore, is to create secluded, quiet places that are free from distractions. A nursing station could be designed with a recessed room that physically barricades the nurse from the bedlam of the ward. Nurses preparing medications might even wear snazzy safety vests of fluorescent orange, so that everyone else knows to back off and shut up.
When it comes to diagnostic error—perhaps the toughest nut to crack—I think about the need for similar distraction-free zones in order to think critically about a patient’s diagnosis. The crush of EMR documentation, however, snuffs out any possible semblance of contemplation. Our current model of indentured EMR servitude saps the vigor of even the most committed clinician. If we want medical staff to apply Graber and Singh’s cognitive discipline about creating a differential diagnosis, questioning data that don’t fit, examining our thought processes for bias, doubling back to ensure we didn’t miss anything crucial—well, our neurons will need some time and space to strut their stuff. Something will have to give. Either we decide that there needs to be more time in a visit (which is ultimately a financial decision) or we have to radically scale back the documentation demands to allow clinicians time to actually think.
With all the talk of how the digital revolution will rescue healthcare, and how creative disruption will bulldoze us into a technology-infused nirvana, I look at the idea of medication safety zones and find it mildly ironic that one of the biggest “innovations” in patient safety is the concept of peace and quiet. But there you have it: one of the key ways to decrease medical errors is to allow nurses and doctors time and space to think. Uninterrupted. (And while we’re at it, such peace and quiet would be highly salutary for patients, who can hardly get two consecutive winks of sleep in the hospital on account of all of our sophisticated technology.)
Improving patient safety ultimately requires a shift in the culture. Our traditional definition of medical error is things that go palpably wrong, like operating on the wrong side of the body or giving the wrong medication. But many things that we once took for granted as “regrettable but expected” side effects of medical treatment—like catheter-associated infections and pressure ulcers—are now viewed as preventable.
In the traditional fee-for-service system, insurance companies would pay for whatever medical care was delivered. Increasingly, though, insurers don’t want to pay for things that shouldn’t have happened. In 2008, Medicare initiated a “nonpayment program” in which it declined to reimburse hospitals for conditions it viewed as preventable. These hospital-acquired conditions included urinary infections, bloodstream infections, falls, blood clots, transfusion of incorrect blood type, and uncontrolled blood glucose levels. In a before-and-after analysis of more than 800,000 patients in 159 hospitals, the years following the implementation of this nonpayment program showed a distinct decline in these conditions, especially infections.2
There is no doubt that squeezing hospitals in the pocketbook gets them to focus harder on these preventable harms. One hopes that hospitals achieve these improvements by investing in the appropriate resources—adequate nurse staffing, an EMR that doesn’t lobotomize its users—but of course there’s always the human instinct to game the system. When Medicare began penalizing hospitals when too many patients were readmitted quickly (suggesting inadequate treatment), hospitals developed “observation units” where readmitted patients could stay for up to 23 hours. These observations units were considered “outpatient,” so, technically, patients were not actually admitted to the hospital (even if they were lying in hospital beds, wearing hospital gowns, and eating the same unpalatable hospital food). While observation units have many legitimate purposes, there is no doubt that some hospitals use them to artificially tamp down their readmission numbers and avoid the fines.
Another aspect of the necessary culture shift is in the vantage point we use to fix medical error. Most often, we approach errors with hindsight—something bad happens, and we address it after the fact. This is a laborious process, as investigations can take months, and implementing changes can take even longer. Moreover, this process usually relies on someone voluntarily reporting the error. Even if a medical professional is able to overcome the discomfort of owning up to an error, the reporting systems can be so arduous to use—please create a temporary password with at least three Cyrillic numerals, two capitalized Latin declensions, plus a fungal species of your choice—that even the best Samaritans are defeated. Most experts estimate that less than 10% of adverse events or errors ever come to light, though admittedly this is impossible to measure.
What if we could set up a system that could detect errors automatically? Even better, what if the system could detect errors in real time? Put another way, if we are all going to be manacled to the EMR anyway, we may as well harness its power and use it to catch errors as they are happening. Infection-control teams have been doing a version of this for years. Microbiology labs track culture results to uncover clusterings that suggest incipient infection outbreaks. However, there are only a finite number of variables that humans can track.
With the EMR, it’s possible to track all patients with infection—every single place in the hospital they’ve been, every staff member they’ve been in contact with, every medication they’ve taken, who their roommates were, exactly when their samples turned positive, what diet they were prescribed—and compare all of these data to patients who did not get infected to figure out which factors are spreading infections.
One hospital used this to track an outbreak of the energetically contagious C. diff bacteria, the bug mentioned earlier that isn’t killed by hand sanitizers. C. diff (Clostridium difficile) infection provokes massive diarrhea and can be deadly. The bacteria sheathe themselves in cleanser-resistant spores, which hop happily from hands to beds to sheets to instruments to sinks as patients shuttle through the various departments of the hospital, sending epidemiologists and housekeepers on ceaseless chases with gallons of bleach. (The name “difficile” was given because culturing it in the lab was so difficult, back when the bacillus was first identified in 1935. The appellation remains apt because it’s so damn difficult to eradicate.)
In this hospital, the voluminous EMR churned data for more than 86,000 admissions, including more than 400,000 patient-location changes.3 A sophisticated analysis was able to pinpoint the source of the outbreak to a single CT scanner located in the emergency room. Digging deeper, the team uncovered the reason for the outbreak: the radiology department had updated its protocol for cleaning CT machines, but apparently the news hadn’t reached the ER. The ER radiology suite was still employing the older technique, which was clearly no match for the tenacious C. diff.
The glut of data generated by EMRs thus offers the tantalizing possibility of picking up errors as they are happening. The Global Trigger Tool, developed by the Institute for Healthcare Improvement, is one such algorithm that uses the EMR to identify potential trouble spots. It’s important to note that these are “potential” trouble spots and must be investigated to determine if something untoward is actually occurring. For example, use of diphenhydramine (Benadryl) is considered a trigger because it’s given in response to allergic reactions and anaphylaxis. However, it’s also given as a sleep aid or if a patient suffers from seasonal allergies. Whenever diphenhydramine is used, a red flag would snap to attention in the EMR, but a human being would have to check to see whether the patient was having an allergic reaction to her medications or whether a visitor was wearing a cardigan knitted from cat hair.
Triggers, then, aren’t automatic indicators of errors or adverse events, but they suggest that there is a good possibility and thus should be investigated in real time. Deployment of the code team or a rapid response team is a trigger, as is unplanned dialysis. Other triggers include a patient returning to the emergency room within 48 hours, or returning to the operating room a second time, or being readmitted to the hospital within 30 days. All of these are clues that something amiss might be happening.
Triggers include certain lab values—very low or very high blood sugar, elevated lactate (suggesting sepsis), or rapid decline of kidney function or blood counts. Other triggers are blood clots noted on CT scans or ultrasounds, administration of vitamin K (an antidote for the blood thinner warfarin), patient falls, the need for physical restraints, and the development of pressure ulcers.
In hospitals that have piloted the Global Trigger Tool, about ten times as many “safety events” are uncovered compared with hospitals that use the standard voluntary reporting method.4 Typically, a nurse is assigned to review the triggers generated each day. This person can investigate the case—read the chart, talk to the staff, examine the patient—to see if an error or adverse event occurred. If the events are actually in progress, such as sepsis, the nurse can ensure that the appropriate treatment is being instituted.
Sometimes triggers will not necessarily show an error or adverse event but will identify patients who are at higher risk. A forward-thinking hospital might then assign these patients a higher staff ratio, or place them closer to the nursing station, or have a pharmacist review the medications, or arrange for a visiting nurse after discharge. These interventions can decrease future adverse events, which of course is the goal.
Many people are looking to artificial intelligence (AI) as a means to decrease medical error and improve patient safety. Eric Topol, a cardiologist in California, probed these issues in his book Deep Medicine.5 AI could be particularly helpful with diagnostic error. As I discussed in chapter 5, AI does especially well with visuals such as chest X-rays and rashes, because you can feed endless images into the system until it learns to recognize patterns and match them to the diagnosis.
Much ink has been spilled about whether AI will surpass doctors in accuracy or even put them out of business altogether. From Topol’s perspective, these are the wrong questions to debate. Take the case of skin rashes. Even if AI is no better than a board-certified dermatologist, it could still increase overall diagnostic accuracy because the vast majority of rashes are evaluated and treated by non-dermatologists. For all the primary care doctors, nurse practitioners, ER staff, and pediatricians who take care of the majority of skin disease, an AI system could be immensely useful in cutting down diagnostic error.
Topol illustrates a similar situation in ophthalmology. Diabetic retinopathy is one of the leading causes of preventable blindness. Most patients with diabetes do not have their retinopathy diagnosed early enough to prevent visual loss. (A delayed diagnosis falls under the category of diagnostic error.) It’s not that ophthalmologists have poor diagnostic skills, but rather that most patients aren’t able to get to an eye doctor. As with skin rashes, the key to decreasing overall diagnostic error is to make it possible to get an accurate diagnosis where patients actually are—usually at their family doctor or internist. By feeding hundreds of thousands of retinal images into a computer, there is now an AI algorithm that can make the diagnosis based on a picture taken by a non-ophthalmologist or even a nondoctor. It could be the medical assistant taking the blood pressure who snaps the photo, or even the clerical staff checking in the patient. At some point it could even be the patient taking the picture at home.
But what about diagnosis on a wider scale, something more complex than identifying a clear-cut pattern such as a rash or a retinal picture? When I interviewed Topol, I’d just finished a bruising clinic session that day. My patients came with all sorts of nonspecific complaints—aches, pains, fatigue, dizziness—overlaid upon a flotilla of possibly related or possibly unrelated circumstances—middling triglycerides, “social” alcohol use, a mildly abnormal bone-density scan from six years ago, a cousin with colon cancer, a uric acid level at the upper limits of normal, an EKG with “nonspecific changes,” a penchant for McDonald’s, a noxious odor at work, gallstones that a doctor thirty years ago mentioned, and so on and so forth. How was AI going to help an internist like me pluck out a correct and actionable diagnosis from the real-world cauldron of confusing possibilities? (And do it while I wrestled with an hour’s worth of EMR documentation in the measly fifteen minutes that I was supposed to be using for the history and physical.)
“Well,” Topol said, “imagine if you didn’t have to chase all of that down. Imagine if the EMR presented you with all the data for the patient—all the scans, all the labs, all the history—all in one place and you didn’t have to hunt for everything. And it also had the patient’s genome and access to all medical research. And the chart had been edited by the patient for accuracy.
“You wouldn’t have to do anything,” he continued. “You could spend your entire time talking to the patient. The EMR would use ‘natural language processing’ technology to listen to your conversation and condense it into a concise clinical note. And if you ordered an incorrect or unnecessary test, it would provide you with a reference to explain why you might consider something else.”
I had to admit that sounded appealing—using the entire visit to interact with the patient rather than with the EMR. That alone would probably decrease my error rate (and my daily spasms of splenic ravings). But I’m especially tantalized by the possibility of AI to cross-connect things that I don’t have the bandwidth to keep in my brain—the blood tests from last year, how effective statins are if taken only intermittently, how heritable gastric cancer is, how accurate an MRI is for detecting lymph node disease, the false-positive rate of mammograms in postmenopausal women, which kind of vasculitis affects the intestines, when immunity from the measles vaccine begins to wane, how effective protein restriction is in advanced kidney disease, which rare illnesses arise from the patient’s country/city/village of origin, and so on.
Topol’s point is not that AI would necessarily be better than me—although it certainly sounds like it has a fighting chance!—but that it could allow me to practice medicine at the top end of my abilities, rather than squandering 97% of my time and focus as a frazzled data-entry clerk. It would allow me to do the things that I’m better at—teasing out a complicated history, picking up on emotional and nonverbal cues, weighing complex treatment options that don’t have a right answer, problem-solving the real-world issues that patients face when coping with disease. The other tasks—surveying the 2 million research papers published every year, flipping through the 3,000 pages of my internal medicine textbook, digging through 80,000 bytes of data entered into the EMR per patient per year—could all be left to AI, which can surely do it more comprehensively and more briskly than I can. Without getting a migraine. Or an ulcer. Or carpal tunnel syndrome. Or so burned out that it considers quitting medicine and getting an MBA instead. (Okay, only kidding on that last one . . .)
A major part of tackling medical error is coaxing the issue out of the healthcare world and into the larger society. Patient safety has to permeate the national discourse the way, say, automobile safety did in the 1970s, launching a raft of government regulations and industry initiatives that ultimately led to a precipitous decline in car-related deaths and injuries.
In Denmark, the Patient Safety Act of 2003 led to broader societal changes in how medical error is viewed. In the United States, a Patient Safety Act was passed in 2005, but it was nowhere near as comprehensive as the Danish version. The US law has increased awareness, but it has not initiated a wholesale restructuring of the healthcare and liability systems. Most glaringly, the law did not create any sort of national incident reporting system. This is partly from an American aversion to centralized actions but more practically because no money was allotted to create and administer such a program.
The job of carrying out the Patient Safety Act in the US fell to a small government agency, AHRQ (affectionally called Arc by those in the know, but which stands for Agency for Healthcare Research and Quality). AHRQ helps healthcare systems set up their own Patient Safety Organizations (called PSOs in this acronym-addicted field). These PSOs act locally, promoting the voluntary reporting of adverse events. To encourage medical professionals to report, the Patient Safety Act specifies that this information is confidential and cannot be used to discipline the person filing the report. AHRQ manages a network of patient-safety databases, but it doesn’t have the resources to integrate everything, nor the teeth to enforce anything.
AHRQ’s biggest success is on the educational front. It set up the Patient Safety Network, an online resource that houses a trove of information for the public as well as for medical professionals and administrators. The website (www.psnet.ahrq.gov) offers online learning modules, how-to guides, podcasts, updates about the latest research, and case reports—submitted by the public—that are analyzed by experts. (The cases, by the way, can be submitted anonymously.) The Patient Safety Network attempts to fill the yawning gaps in medical and nursing schools that leave most medical professionals unprepared to recognize, handle, and prevent medical error.
It’s not clear if the United States will ever be able to solidify a comprehensive national reporting system, as Denmark has. The closest it has come is in the Veterans Administration (VA) medical system. The VA itself is larger than the entire Danish medical system, but in some ways it has more in common with Danish socialized medicine than the American privatized system. The VA has a single payment system for its more than 1,200 healthcare facilities that cover nearly all the medical needs of veterans. All the facilities use the same EMR, and most of its operational systems are standardized. I once visited a VA hospital in California and everything was eerily identical to our VA hospital in Manhattan, down to the binders for the medical charts, the odd-sized paper they used, and the color scheme of the patients’ pajamas.
The VA Patient Safety Reporting System was spearheaded in 1999 by Jim Bagian, a former astronaut who sought to transplant the NASA “systems approach” to healthcare. (The Danish incident reporting system, in fact, is based on this VA reporting system.) Bagian, who is also a physician, argued strongly to shift the focus from “Whose fault is this?” to “What happened, why did it happen, and what can we do to prevent it in the future?”
Importantly, Bagian focused on the reasons that medical professionals don’t report errors. “If you ask doctors,” he told me, “they’ll say it’s because they’re worried about malpractice.” But when the VA did an anonymous survey, the overwhelming reason that doctors did not report errors was because of the profound shame.
When I was a resident, I once botched the diagnosis of an intracranial bleed. Somewhere in the handoff of care, somebody had said, “Radiology fine,” and so I hadn’t bothered to look at the CT scan myself. I was ready to sashay the patient back home, but luckily the bleed was caught by someone else. The patient went straight to the operating room to have the blood drained from his skull, and he did fine. Technically, my error would be called a “near miss,” since the patient’s care was ultimately not impacted by my oversight.
But “near miss” hardly captures the essence of what had transpired. “Near miss” just means that the patient got lucky. My error—relying on a verbal report rather than looking at the scan myself—was still an error. I had still put my patient’s life at risk. I was so ashamed that I had given substandard care that I did not tell a soul. It took me more than twenty years to talk about it publicly.
At the time, I was so devastated that I could hardly function. As a fledgling physician, I felt that I had failed my patient so gravely that I was ready to quit medicine altogether. I wasn’t surprised, years later, to read of a nurse who took her own life after an accidental miscalculation of a medication dose killed an infant in her care.6
The famed French surgeon René Leriche wrote, “Every surgeon carries about him a little cemetery, in which from time to time he goes to pray, a cemetery of bitterness and regret, of which he seeks the reason for certain of his failures.” This patient with the intracranial bleed resides in my cemetery, as does Ms. Romero—whose anemia and cancer I missed—and, regretfully, many others. Like all doctors and nurses, I visit that cemetery periodically, painfully, even though I’m well aware that the effect of the errors on me is secondary to the effects on the patient. Even with the passage of time and the smoothing over of the graves, the stinging sadness and shame of having harmed your patient never recedes.
This reinforces the bedrock principle that medical error and patient-safety issues will never be fully resolved with a blitzkrieg of mandates, checklists, and zero-tolerance policies. The human element is paramount. In fact, Jim Bagian doesn’t even use the term “medical error.” “The term ‘error,’” he wrote to me in an email, “implies a person-centered approach rather than a more effective systems-based approach that includes the person but generally yields solutions that are more sustainable. Experienced safety professionals usually choose the term ‘adverse medical events’ since you are trying to prevent the event that harms a patient.”
Bagian was careful to acknowledge that some situations are indeed “blameworthy.” Obvious ones would be when a staff member is inebriated or has committed an actual crime. The other category of blameworthy events would be when a doctor or nurse knowingly and deliberately performs an unsafe action. These events are rare but should be disciplined accordingly. For all the rest, the person who makes the error should take responsibility, but the institutional focus should be on how the system made that error possible and how it could be prevented in the future.
When I started writing this book, I set out to determine whether or not medical error truly was the third-leading cause of death, as the headlines had proclaimed. Now that I’ve spent several years digging through the research and interviewing patients, families, doctors, lawyers, nurses, administrators, researchers, and patient-safety advocates, I’ve concluded that the question itself isn’t as relevant as I’d thought it would be.
In terms of the paper from the British Medical Journal that drew this conclusion, there was a fair amount of academic criticism. The first issue was that it wasn’t a primary-source study, in which researchers sifted through medical charts trying to figure out whether and where medical errors occurred. Rather, it was a reanalysis of previously published data. The authors certainly didn’t present it as anything other than that, but in the public eye it came across as a “brand-new study” that “proved” that medical error was the third leading cause of death. Reanalyzing data, per se, isn’t against the rules (the Institute of Medicine used the same technique in its report To Err Is Human), but it can introduce flaws and biases. In particular, it is problematic to take results from a narrow slice of life and then extrapolate them to the entire population. Statistically speaking, this is risky business.
For example, one of the four studies the analysis relied upon drew its data from ten hospitals in North Carolina.7 It’s not that these data are necessarily wrong, but they may not accurately represent the entire population. This is particularly precarious when you are dealing with infrequent events. In these ten hospitals, the researchers of this original study concluded that fourteen people died as a result of medical error (out of 2,344 cases reviewed). Fourteen is a very small number. If three more patients had died, the death rate in the study would have jumped up by 20%. When you extrapolate 2,344 cases to a nation of 330 million people, a difference of 20% represents almost a half million extra deaths. So you can see that tiny shifts in a small sample size can exert outsize effects when you try to draw a conclusion about the general population.
Then, of course, there’s the challenge of deciding whether an error, in fact, caused the death in question. A patient with end-stage colon cancer may have been given an incorrect medication dose. This is clearly a medical error, but the patient may have died simply from the colon cancer. Parsing causality is not at all straightforward. In the studies upon which the BMJ analysis was based, researchers could not always agree whether a particular error caused (or hastened) death. The fourteen deaths due to medical error reported in the North Carolina study might, in reality, have been seventeen deaths, or twelve (which, when extrapolated to the general population and its affection for aviation metaphors, would result in entirely different numbers of jumbo jets crashing). It remains remarkably difficult to calculate precisely how many deaths are caused by medical errors.
When I interviewed Martin Makary, the lead author on the BMJ paper, he emphasized that pinpointing any cause of death is challenging: “Every cause of death is an estimate. It’s not a perfect science.” Unfortunately, death certificates require doctors to pick one primary cause of death, which is often impossible. “How do you take a complex series of factors,” Makary asked, “and identify a single cause?”
The intense media focus on errors as a “leading cause of death” shifted attention away from all the errors that are happening but aren’t leading to death (a far larger number). While these errors may not kill patients, they can cause severe harms—amputation, kidney failure, debilitating pain, paralysis, anaphylaxis, financial ruin. These cases won’t appear in stats that count only deaths, but they are serious enough to warrant being grouped together in the severe-harms-from-medical-error category when we are trying to assess the overall toll of medical error. And then, of course, there are all the errors that cause no harms at all, which get completely ignored when the focus is solely on the death rate. These smaller errors are equally important to address, because they represent the most extensive pool of harm just waiting to happen.
So it’s likely that the “third leading cause of death” claim is not accurate. Medical error probably ranks lower on the list of the things that mow us down. But the fact that it’s lower on the list doesn’t make the issue any less critical. Any harm to patients that is preventable is something we should be aggressively investigating.
My takeaway is that medical error and adverse events (even ones that aren’t errors per se) are much more prevalent than we think. The majority may not cause significant harms, but enough do, enough that the issue needs to be front and center in healthcare today. So even if the BMJ paper was not correct in its final numbers—numbers which, in any case, may not even be possible to calculate accurately—it did serve to bring public attention to the issue.
Makary pointed out that there is currently no way to indicate medical error on a death certificate. Death certificates are one of the primary ways we collect epidemiological data about the population. That’s how we know that heart disease and cancer are the leading causes of death in the United States. (The third leading cause of death, incidentally, is accidents, followed by emphysema and stroke.) Makary proposes that in addition to the standard cause of death, death certificates ought to have an additional field asking whether a preventable complication contributed to the death. This way we might actually know how many errors are occurring—at least those ones that caused or hastened death. As Florence Nightingale pointed out back in 1850, you can’t start fixing anything until you gather the data to establish the current state of affairs.
Would doctors actually check such a box? That’s a debatable question, given that fear of lawsuits—in the US at least—remains pervasive. But there’s no doubt that we need some sort of unified national database for medical errors, adverse events, and any preventable harm. Mandatory reporting would likely backfire, and voluntary reporting catches only a small subset of errors. The only way to make such a system workable and accurate is to create a culture in which reporting an adverse event is a routine and ordinary event for medical professionals, just like—ahem—washing your hands before touching a patient. Obviously, this would require a culture shift of colossal proportions, but that is the goal we should be striving toward.
A generation or two ago, the medical system was viewed with unmitigated reverence. Now the pendulum has swung so far the other way that many view the medical world with such suspicion that they avoid even routine and well-validated medical care. The truth settles somewhere in between. Medical science has made enormous strides in the past century and has objectively decreased mortality and suffering on a grand scale. Our great-grandparents would give their eyeteeth for the medical benefits that we take for granted today—vaccinations, anesthesia, cancer treatments, heart transplants, dialysis. But there’s no doubt that medical care also causes harm, a good deal of which is likely preventable. Patients and families—as well as medical staff—should cast a careful, questioning eye on all medical tests and treatments.
Medicine is a team sport, and that team isn’t just the doctors and nurses but also the patients, families, and closest friends. Too often it can feel like we are on opposing teams, or at least on teams with opposing agendas. But really, there is just that one goal: helping the patient get better. There’s a plethora of technology out there to assist the patient in getting better, but the responsibility for making sure it all works falls to the humans.
The IOM certainly picked the title of its groundbreaking report well. To err is definitely human. It’s also human, however, to care about what happens when error occurs. This holds for the immediate aftermath with an individual patient, as well as for thinking forward, more broadly, about how to minimize errors for all patients. We all enter the medical system with the expectation that we will come out better at the other end or, at least, not worse off. We certainly have the right to expect that the medical care itself does not make us worse off. Almost 2,500 years ago, Hippocrates offered some advice to his fellow Greek healers, writing in his treatise Of the Epidemics: “As to diseases, make a habit of two things—to do good, or at least to do no harm.” No one has said it better since.