I met Chris in my first month at a small, hard-partying Catholic high school in north-eastern Wisconsin, where kids jammed cigarettes between the fingers of the school’s lifesize Jesus statue and skipped mass to eat fries at the fast-food joint across the street. Chris and her circle perched somewhere adjacent to the school’s social hierarchy, and she surveyed the adolescent drama and absurdity with cool, heavy-lidded understanding. I admired her from afar and shuffled around the edges of her orbit, gleeful whenever she motioned for me to join her gang for lunch.
After high school, we lost touch. I went east; Chris stayed in the midwest. To pay for school at the University of Minnesota, she hawked costume jewellery at Dayton’s department store. She got married to a tall classmate named Adam and merged with the mainstream – became a lawyer, had a couple of daughters. She would go running at the YWCA and cook oatmeal for breakfast. Then in 2010, at the age of 35, she went to the ER with stomach pains. She struggled to describe the pain – it wasn’t like anything she’d felt before. The doctor told her it was indigestion and sent her home. But the symptoms kept coming back. She was strangely tired and constipated. She returned to the doctor. She didn’t feel right, she said. Of course you’re tired, he told her, you’re raising kids. You’re stressed. You should be tired. Frustrated, she saw other doctors. You’re a working mom, they said. You need to relax. Add fibre to your diet. The problems ratcheted up in frequency. She was anaemic, and always so tired. She’d feel sleepy when having coffee with a friend. Get some rest, she was told. Try sleeping pills.
By 2012, the fatigue was so overwhelming, Chris couldn’t walk around the block. She’d fall asleep at three in the afternoon. Her skin was turning pale. She felt pain when she ate. Adam suggested she see his childhood physician, who practised 40 minutes away. That doctor tested her blood. Her iron was so low, he thought she was bleeding internally. He scheduled a CT scan and a colonoscopy. When they revealed a golf ball-sized tumour, Chris felt, for a moment, relieved. She was sick. She’d been telling them all along. Now there was a specific problem to solve. But the relief was short-lived. Surgery six days later showed that the tumour had spread into her abdomen. At the age of 37, Chris had stage four colon cancer.
Historically, research about the roots of health disparities – differences in health and disease among different social groups – has sought answers in the patients: their behaviour, their status, their circumstances. Perhaps, the thinking went, some patients wait longer to seek help in the first place, or they don’t comply with doctors’ orders.
Maybe patients receive fewer interventions because that’s what they prefer. For Black Americans, health disparities have long been seen as originating in the bodies of the patients, a notion promoted by the racism of the 19th-century medical field. Medical journals published countless articles detailing invented physiological flaws of Black Americans; statistics pointing to increased mortality rates in the late 19th century were seen as evidence not of social and economic oppression and exclusion, but of physical inferiority.
In this century, research has increasingly focused on the social and environmental determinants of health, including the way differences in access to insurance and care also change health outcomes. The devastating disparate impact of Covid-19 on communities of colour vividly illuminates these factors: the disproportionate burden can be traced to a web of social inequities, including more dangerous working conditions, lack of access to essential resources, and chronic health conditions stemming from ongoing exposure to inequality, racism, exclusion and pollution. For trans people, particularly trans women of colour, the burden of disease is enormous. Trans individuals, whose marginalisation results in high rates of poverty, workplace discrimination, unemployment, and serious psychological distress, face much higher rates of chronic conditions such as asthma, chronic pulmonary obstructive disorder, depression and HIV than the cisgender population. A 2015 survey of nearly 28,000 trans individuals in the US found that one-third had not sought necessary healthcare because they could not afford it.
More recently, researchers have also begun looking at differences that originate in the providers – differences in how doctors and other healthcare professionals treat patients. And study after study shows that they treat some groups differently from others.
Black patients, for instance, are less likely than white patients to receive pain medication for the same symptoms, a pattern of disparate treatment that holds even for children. Researchers attribute this finding to false stereotypes that Black people don’t feel pain to the same degree as white people – stereotypes that date back to chattel slavery and were used to justify inhumane treatment. The problem pervades medical education, where “race” is presented as a risk factor for myriad diseases, rather than the accumulation of stressors linked to racism. Black immigrants from the Caribbean, for instance, have lower rates of hypertension and cardiovascular disease than US-born Black people, but after a couple of decades, their rates of illness increase toward those of the US-born Black population, a result generally attributed to the particular racism they encounter in the US.
Black patients are also given fewer therapeutic procedures, even when studies control for insurance, illness severity and type of hospital. For heart attacks, black people are less likely to receive guideline-based care; in intensive care units for heart failure, they are less likely to see a cardiologist, which is linked to survival.
These biases affect the quality of many other interactions in clinics. Doctors spend less time and build less emotional rapport with obese patients. Transgender people face overt prejudice and discrimination. The 2015 survey also found that in the preceding year, a third of respondents had had a negative encounter with a healthcare provider, including being refused treatment. Almost a quarter were so concerned about mistreatment that they avoided necessary healthcare. Transgender individuals can therefore face a dangerous choice: disclose their status as trans and risk discrimination, or conceal it and risk inappropriate treatment.
Even though medical providers are not generally intending to provide better treatment to some people at the expense of others, unexamined bias can create devastating harm.
Chris was told that her symptoms, increasingly unmanageable, were not serious. Women as a group receive fewer and less timely interventions, receive less pain treatment and are less frequently referred to specialists. One 2008 study of nearly 80,000 patients in more than 400 hospitals found that women having heart attacks experience dangerous treatment delays, and that once in the hospital they more often die. After a heart attack, women are less likely to be referred to cardiac rehabilitation or to be prescribed the right medication. Critically ill women older than 50 are less likely to receive life-saving interventions than men of the same age; women who have knee pain are 22 times less likely to be referred for a knee replacement than a man. A 2007 Canadian study of nearly 500,000 patients showed that after adjusting for the severity of illness, women spent a shorter time in the ICU and were less likely to receive life support; after age 50, they were also significantly more likely to die after a critical illness.
Women of colour are at particular risk for poor treatment. A 2019 analysis of their childbirth experiences found that they frequently encountered condescending, ineffective communication and disrespect from providers; some women felt bullied into having C-sections. Serena Williams’s childbirth story is by now well known: the tennis star has a history of blood clots, but when she recognised the symptoms and asked for immediate scans and treatment, the nurse and the doctor doubted her. Williams finally got what she needed, but ignoring women’s symptoms and distress contributes to higher maternal mortality rates among Black, Alaska Native and Native American women. Indeed, Black women alone in the US are three to four times more likely to die of complications from childbirth than white women.
There’s also a structural reason for inferior care: women have historically been excluded from much of medical research. The reasons are varied, ranging from a desire to protect childbearing women from drugs that could impair foetal development, via notions that women’s hormones could complicate research, to an implicit judgment that men’s lives were simply more worth saving. Many landmark studies on ageing and heart disease never included women; the all-men study of cardiovascular disease named MRFIT emerged from a mindset that male breadwinners having heart attacks was a national emergency, even though cardiovascular disease is also the leading cause of death for women. In one particularly egregious example, a 1980s study examining the effect of obesity on breast cancer and uterine cancer excluded women because men’s hormones were “simpler” and “cheaper” to study.
Basic to these practices was an operating assumption that men were the default humans, of which women were a subcategory that could safely be left out of studies. Of course, there’s a logical problem here: the assertion is that women are so complicated and different that they can’t be included in research, and yet also so similar that any findings should seamlessly extend to them. In the 90s, the US Congress insisted that medical studies funded by the National Institutes of Health should include women; earlier, many drug studies also left out women, an exclusion that may help explain why women are 50%-75% more likely to experience adverse side-effects from drugs.
As the sociologist Steven Epstein points out, medicine often starts with categories that are socially and politically relevant – but these are not always medically relevant. Relying on categories such as race risks erasing the social causes of health disparities and may entrench the false and damaging ideas that are inscribed in medical practice. At the same time, ignoring differences such as sex is perilous: as a result of their exclusion, women’s symptoms have not been medically well understood. Doctors were told, for example, that women present with “atypical symptoms” of heart attacks. In fact, these “atypical” symptoms are typical – for women. They were only “atypical” because they hadn’t been studied. Women and men also vary in their susceptibility to different diseases, and in the course and symptoms of those diseases. They respond to some drugs differently. Women’s kidneys filter waste more slowly, so some medications take longer to clear from the body.
This dearth of knowledge about women’s bodies has led doctors to see differences where none exist, and fail to see differences where they do. As the journalist Maya Dusenbery argues in her book Doing Harm, this ignorance also interacts perniciously with historical stereotypes.
When women’s understudied symptoms don’t match the textbooks, doctors label them “medically unexplained”. These symptoms may then be classified as psychological rather than physical in origin. The fact that so many of women’s symptoms are “medically unexplained” reinforces the stereotype that women’s symptoms are overreactions without a medical basis, and casts doubt over all women’s narratives of their own experiences. One study found that while men who have irritable bowel syndrome are more likely to receive scans, women tend to be offered tranquilisers and lifestyle advice. In response to her pain and fatigue, my friend Chris was told she should get some sleep.
The doctor who finally ordered the right tests for Chris told her that he’d seen many young women in his practice whose diagnoses had been delayed because their symptoms were attributed to stress. Indeed, studies show that women around the world experience delays in receiving a correct diagnosis for many diseases, including Crohn’s, Ehlers-Danlos syndrome, coeliac disease and tuberculosis. A 2015 UK study of more than 16,000 patients also found delayed diagnoses for many types of cancer – bladder, gastric, head and neck, and lung cancer, and lymphoma, for instance. As Dusenbery argues, the problem is exacerbated by the fact that doctors rarely receive feedback about their misdiagnoses. They never learn where they went wrong.
Diagnostic errors, it is estimated, cause 80,000 deaths a year in the US. Cognitive factors are estimated to play a role in 75% of these cases. What could have been done in Chris’s case? Certainly, it’s essential that doctors increase their awareness of their own capacity for biased decisions, and their motivation to overcome it. We know that biases are more likely to arise when people are mentally taxed. Meaningful, collaborative contact with those in other social groups can also help. But there’s another approach to reducing bias that can support all these efforts, providing another layer of protection against the risk of interpersonal bias.
Elliott Haut is a trauma surgeon at Johns Hopkins hospital in Baltimore. Affable and baby-faced, he looks happiest when talking about safety. The desk in his office is scattered with books about preventable deaths. A note taped over his computer reads “reduce system errors”.
In other parts of the country, the trauma unit might see farm accidents or motorcycle crashes. At Hopkins, many trauma patients are victims of gunshots or stabbings. One patient arrived with the shard of a beer bottle still lodged in his neck, the entire word “Budweiser” perfectly legible along the length of jagged glass.
About 15 years ago, Haut was asked to oversee efforts to improve the Hopkins trauma department. The goal was to create better outcomes for the patients by improving the performance of the doctors. When Haut dived into the hospital’s data, he found that patients were developing blood clots at a strikingly high rate.
Blood clots – the condition that threatened Serena Williams’s life when she was in the hospital giving birth – are gelatinous globs of stuck-together blood cells that can travel through blood vessels and block blood flow to the lungs. They kill about 100,000 people a year in the US – more than breast cancer, Aids and car crashes combined. Many of these clots are preventable, if doctors prescribe the right clot prevention. In some cases, this means blood thinners; in others, mechanical “squeezy” boots that inflate and deflate around the legs to get the blood moving. But at Hopkins, only a third of the highest-risk patients were getting the right blood-clot prevention, Haut found. “We’d get a patient into surgery – a routine surgery – and a week later they’d die of a pulmonary embolism,” he told me as we sat in his office in east Baltimore, near a pile of wooden puzzles. And this problem wasn’t specific to Hopkins: at hospitals around the country, patients were getting proper clot prevention only about 40% of the time – a problem that the American Public Health Association was calling a crisis.
Haut wasn’t sure why doctors were failing to prescribe the right interventions. Maybe, he thought, they overestimated the risks of blood-clot prevention because patients who had developed complications from blood thinners sprang to memory more easily than those who were treated successfully. Haut wasn’t thinking about disparities; his goal was to improve clot prevention for everyone.
To do so, Haut and his team sought out an approach that had been developed by Peter Pronovost, another Hopkins doctor, whose own father had died because of a cancer misdiagnosis. Pronovost had formulated a technique for improving medical care by adapting an approach used in aviation: the humble checklist. A checklist is just what it sounds like – a reminder of all the mandated steps a clinician should take. It plugs memory holes and hangs a safety net under human errors so they don’t add up. Proper ICU care, for instance, requires nearly 200 separate actions each day. Complications can arise from missing even one or two.
Pronovost showed that using a checklist in intensive care units reduced infections simply by ensuring that doctors adhere, each time, to a predetermined set of tasks. In one trial, a five-step checklist reminding workers in more than 100 ICUs to do things such as washing their hands and cleaning the patient’s skin with antiseptic led to a 66% drop in catheter-related bloodstream infections. The drop held steady over the 18 months of the study.
Haut and his team decided to try developing a checklist for blood-clot prevention. In their version, whenever a healthcare provider admitted a patient to the hospital, a computerised checklist would pop up on screen. The checklist would walk the doctor step by step through risk factors for blood clots and for bleeding from blood-thinning medication. After the checklist was complete, the system would suggest a recommended treatment – a blood thinner, for instance, or a mechanical squeezy boot to move the blood. If doctors didn’t choose the suggested treatment, they had to document their reasons.
The approach worked. After introducing the checklist, the percentage of patients getting the right clot prevention surged, and preventable clots in trauma and internal medicine were close to eliminated. One study of a month of hospital admissions found that the number of internal medicine patients who returned to the hospital with blood clots within 90 days of discharge fell from 20 to two. And after the introduction of the checklist, the rate of fatal pulmonary embolism was cut in half.
That could have been the end of the story. But Haut’s office was, at the time, two doors down from the office of Adil Haider, a doctor who studies gender and racial disparities in health care. Their conversations prompted Haut to wonder whether there had been disparities in blood-clot prevention. The team hadn’t sliced the data that way, but when they went back over the numbers, an alarming pattern appeared. While 31% of male trauma patients had failed to get treatment, the rate was 45% for women. In other words, women had been nearly 50% more likely to miss out on blood-clot prevention than men, and in greater danger of dying of this particular cause.
It’s possible that factors other than gender might have been at play. Most patients who arrive with gunshot wounds, for instance, are men; perhaps doctors prescribed more prevention for more severe injuries. But as the researcher who analysed the data put it, the disparities in treatment fit a consistent, large and well established pattern of women receiving suboptimal care.
Looking at the numbers after the checklist was introduced, Haut and the team found that it had eliminated the gender disparities. Women and men received the right clot prevention at exactly the same rates. The gap had disappeared.
In 2008, the University of Chicago economist Richard Thaler and the legal scholar Cass Sunstein, co-authors of Nudge: Improving Decisions about Health, Wealth, and Happiness, coined the term “choice architecture” to describe a powerful phenomenon: the context within which we make a choice has a profound influence on the way we choose. Just as the design of a physical environment can influence our behaviour (as seen in coffee shops that skimp on electric outlets to discourage people from sitting with laptops), the design of a process can also shape our behaviour. It, too, can be thought of as a kind of architecture.
For instance, researchers at the University of Minnesota discovered that they could coax students into eating more vegetables simply by redesigning their lunchtime routine. In a typical lunch line, students encounter carrots next to more tempting options such as fries and pizza. Instead, the researchers gave kids a cup of carrots the moment they arrived in the cafeteria, when they were at their most hungry. It worked: kids ate a lot more carrots. The key was to put the carrots “in a competition they actually can win” – a contest not against fries, but against being really hungry. To change how students ate, it wasn’t necessary to sell them on the virtues of vitamin A. What changed was the choice architecture.
The Hopkins checklist is a kind of choice architecture, too – a way of shaping a doctor’s behaviour not through persuasion, but through design. It doesn’t ask doctors to think more carefully about their biases; it simply interrupts the process by which they make decisions. The Hopkins checklist forces doctors to disentangle the thinking that goes into a medical decision. In a way, it acts like a prism, reverse-engineering a holistic judgment into its constituent parts, the way a prism separates white light into its rainbow colours.
The checklist also supports that human judgment. It is meant to remind doctors of steps they might forget, but bias isn’t really about forgetting. It’s about using assumptions to judge and evaluate, without necessarily being aware of the presence of those assumptions. Some doctors resist the intrusion, pointing out that these mandated checklists aren’t perfect. As one hospitalist told me, they may not take into account the full range of factors a doctor might consider. While a clot checklist asks questions to assess risk at one moment in time, an experienced doctor might note that a patient with pain could have a procedure tomorrow that could change the risk profile over time. The checklist does not have the capacity to account for this nuance. As the medical scenarios become more complex, the checklist may be best considered a failsafe for decision-making, not a substitute.
But checklists have been shown to reduce bias elsewhere. After a structured decision-making tool was introduced in the state of Illinois, the disparities between psychiatric hospitalisation of young, low-risk Hispanic and Black patients and white patients shrank. When the Mayo Clinic instituted a system of automatic referrals for cardiac rehabilitation after heart attacks, the gender gap between men’s and women’s referral rates disappeared.
Using principles of behavioural design to reduce bias dates back to 1952, when the Boston Symphony Orchestra began changing the way it auditioned musicians. Instead of having musicians play in full view of a panel of judges, a screen was set up to divide them. Women musicians were asked to remove their shoes, so clacking high heels wouldn’t be a tell. Instead, a man standing onstage created fake footsteps with his shoes.
In the following decades, curtained auditions rippled through American orchestras – a heavy cloth was hung from the ceiling, or a room divider was stretched like an accordion across the stage. By the 90s, most had adopted the practice. When the economists Claudia Goldin and Cecilia Rouse studied the differences between orchestras that did and did not use this approach, they found stark evidence that masking gender changed judges’ assessment of women’s skills. They found that concealing musicians’ identities increased women’s odds of advancing to the next round of auditions by 50%. Today, women make up almost 40% of orchestras.
Relying on a blunt tool like blurring out a person’s social identity is problematic, and can veer toward erasure, which is itself a form of discrimination. But in the case of a hiring decision, it can also shield a person’s evaluation from harmful stereotyping – or unfair advantages.
Of course, the masked approach isn’t possible in medicine, which usually depends on face-to-face interactions between doctor and patient. Structured decision-making tools such as checklists is a close cousin. These steer people in positions of power from using assumptions and preconceptions, so they rely instead on official criteria. That alone can unleash powerful changes.
This is an edited extract from The End of Bias: How We Change Our Minds by Jessica Nordell, published by Granta on 23 September and available at guardianbookshop.co.uk