The Doctor Recommended It

In 2010, researchers at Massachusetts General Hospital, a granite complex on the bank of the Charles River that has been treating patients since 1811, enrolled 151 people with stage IV lung cancer in a trial. All were terminally ill. All would die. The question was not whether to treat them. The question was how.¹

Half received standard oncology care. The other half received the same oncology care plus one addition: a palliative care specialist who met with them at least once a month. Not instead of their oncologist. Alongside. The specialist did not administer drugs. Did not perform procedures. The specialist asked questions. What mattered to the patient. What they understood about their illness. What they wanted the remaining time to look like.

The palliative care group stopped chemotherapy sooner. Entered hospice earlier. Reported better quality of life. Were less than half as likely to be clinically depressed at twelve weeks. Sixteen percent versus thirty-eight.²

They also lived longer.

Two point seven months. Thirty percent. Less treatment. Better quality of life. Longer life. All three.³

The intervention was a conversation. Someone sat with them and asked what they wanted. The answer changed what happened next. The patients who were asked chose less aggressive treatment. The patients who were not asked received the system's default. The default was more.

The study was small. A single institution. The survival finding was a secondary outcome the trial was not designed to detect, and subsequent larger studies have not confirmed it in pooled analysis. What has been confirmed, across twenty-four randomized trials and multiple meta-analyses, is the quality of life improvement. Less suffering. Less depression. Less aggressive care at the end. The conversation does not reliably extend life. It reliably makes the remaining life better.⁴

That should have been enough to change the system. It did not. But then, this finding has been available for longer than anyone in that hospital has been alive.

The ward

In 1847, a Hungarian physician named Ignaz Semmelweis noticed something in the maternity ward data at the Vienna General Hospital. The clinic staffed by medical students had a maternal death rate of roughly ten percent. The clinic staffed by midwives ran closer to four. The difference was a single practice: the medical students went directly from performing autopsies to delivering babies. Semmelweis introduced handwashing with chlorinated lime. Maternal mortality fell to roughly one percent.⁵

The medical establishment rejected the finding. The implication that the doctor's hands carried death was an insult to professional standing. Semmelweis was dismissed, spent a decade publishing data nobody wanted to read, and died in an asylum at forty-seven.⁵ The mechanism he identified, that the system's default was causing harm, would take another twenty years to be accepted. By then the women he could have saved were dead.

One hundred and sixty-three years later, a trial at Massachusetts General proved the same structural finding in a different ward. Less treatment. Better outcome. The system has not fully absorbed it now.

The default

The last time you went to the doctor, the appointment lasted roughly fifteen minutes. The average primary care visit in the United States runs thirteen to sixteen minutes.⁶ In that time, the doctor reviewed your chart, asked about symptoms, ordered tests, wrote prescriptions, and moved on. Nobody asked what mattered to you. Nobody needed to. You had a sore throat.

But the fifteen-minute structure is the same structure that operates when the diagnosis is terminal. The system that is adequate for a sore throat is the system that fails when the question stops being how do I fix this and becomes how do I live. The system does not switch modes. It was not designed to. It was designed to do something. The training, the culture, the legal environment, the payment structure. Everything points in one direction.

Fourteen years

Medical school teaches diagnosis and treatment. Fourteen years of training, the bulk of it structured around identifying the problem and intervening. Fewer than one-third of physicians report any formal training in end-of-life discussions.⁷ The curriculum does not include when to stop. It teaches what to do.

The war

Patients fight. Tumors invade. Treatments attack. Survivors are warriors. The dead lost their battle. Nixon signed the National Cancer Act in 1971 and called it a war. In 1978, Susan Sontag argued the metaphor was doing damage. The language did not change.⁸

A team at Lancaster University built a 1.5-million-word corpus from the language of cancer, drawing on patient interviews, online forums, and clinical exchanges. Military metaphors were the most common type. Then they found something the language itself could not accommodate. Patients with curable cancer used war metaphors and found them empowering. Patients with terminal cancer stopped using them. The metaphor worked when it matched reality. When the disease was incurable, when there was no winning, the language of fighting offered no vocabulary for acceptance, for comfort, for a death that could be something other than a defeat.⁹

The language fails exactly when people need it most.

Standard of care

A doctor who treats aggressively and the patient dies followed standard of care. A doctor who does not treat aggressively can be sued for negligence. The legal incentive runs in one direction. Doctors who intervene are defended by the protocol. Doctors who watch and wait are exposed.

The result is defensive medicine. Tests ordered not because the doctor believes the patient needs them but because not ordering them creates legal exposure. In a 2017 survey of over two thousand physicians published in PLOS ONE, eighty-five percent cited fear of malpractice as the primary reason for overtreatment. They estimated that roughly one-fifth of all medical care was unnecessary. The most rigorous cost estimate, from a Harvard study co-authored by the same surgeon who wrote the book on dying, put the defensive medicine bill at $45.6 billion a year. That was 2008 dollars. The number has not shrunk. Average malpractice verdicts have increased seventy-five percent since 2022.¹⁰

The doctor making the recommendation believes, on some level, that the test is justified. The question is whether the recommendation would be the same if the doctor could not be sued. In most cases, the doctor knows it would not.

No billing code

The system has nearly seventy thousand diagnosis codes. It can bill for a lesion on your left pinky finger. Until January 1, 2016, it had zero codes for asking what you wanted.¹¹ For the entire history of modern American medicine until nine years ago, the system had no mechanism to pay for the conversation that the Mass General trial proved was the intervention.

The codes exist now. Medicare pays approximately $80 for the first thirty minutes of advance care planning. Another $78 for each additional thirty minutes. A forty-five-minute conversation about how a patient wants to spend their remaining time pays roughly $157.¹²

A twenty-minute colonoscopy pays the physician roughly $220.¹³

The system pays roughly forty percent more per minute for looking inside someone's body than for asking them what they want done with it. The payment is the clearest signal the system sends about what it values. Insurance reimburses procedures at rates that fund practices. It reimburses conversations at rates that fund compliance. A practice built on conversations cannot survive. A practice built on procedures can.

Over forty percent of oncologists acknowledge offering treatments they believe are unlikely to work.¹⁴ Not because they are dishonest. Because the system around them, the training, the culture, the legal exposure, the payment structure, points in one direction. Doing something is reimbursed. Doing something is defensible in court. Doing something matches the language of fighting and battling and not giving up. Not doing something, watching, waiting, asking, listening, is none of these things.

The bell curve

Even within this system, the assumption is that the care is roughly the same everywhere. A doctor in Cincinnati. A doctor in Minneapolis. Same disease. Same medical knowledge. Same protocols. The patient assumes competence. The patient assumes uniformity. The data says something else.

In cystic fibrosis, 117 accredited centers in the United States follow the same guidelines, undergo the same certification, participate in the same research. You would expect the outcomes to be similar. In 1997, at the average center, patients lived to thirty. At the best center, patients lived to forty-six.¹⁵

Sixteen years. Same disease. Same medical knowledge. Same protocols available to every center. The difference was not what the doctors knew. It was intensity. Aggressiveness about marginal improvement. Refusal to accept a lung function reading that was good enough.

Cincinnati Children's Hospital, among the most respected pediatric hospitals in the country, had lung function results for children under twelve in the bottom twenty-five percent. Not a rural clinic with limited resources. A flagship institution. The bell curve was invisible to the families who brought their children there. The system did not tell them where their center ranked. In many cases the system did not know, or if it knew, it did not publish.¹⁶

Warren Warwick ran the CF center at the University of Minnesota. He had been doing it for decades. His patients lived longer than patients at almost every other center in the country. The reason was not a breakthrough treatment or better funding. It was how he used the exam room. When a patient came in with a lung function reading at ninety-five percent, the textbook said normal. Warwick would lean in and say: that is not your normal. Your normal is a hundred. Let us figure out why you are at ninety-five. He would sit with the patient and adjust, tweak, refuse to accept the number the guideline said was fine. He called national clinical guidelines "a record of the past, and little more." He said they should have an expiration date.¹⁷

What Warwick did was expensive in time and attention, not in money. It was the opposite of the fifteen-minute visit. He treated the patient in front of him, not the protocol on the chart. Most doctors do not do this. Most doctors follow the protocol, deliver the standard of care, produce the average result. The average result, for most conditions, is adequate. Nobody dies from an adequate flu diagnosis. But for cystic fibrosis the gap between average and the doctor who refused to accept average was sixteen years of living.

The patient does not know where on the bell curve their doctor falls. The system does not tell them. The system barely knows.

After the Cystic Fibrosis Foundation began publishing center-level outcome data, something unexpected happened. The gap between the best and the average did not close. It widened. The top centers improved fastest. They had, above all else, a capacity to learn and change, and to do it faster than everyone else. The transparency helped. But it helped the best more than it helped the rest.¹⁸

The question

Sara Thomas Monopoli was thirty-four and pregnant when she was diagnosed with stage IV lung cancer. She had never smoked. Over the following months she went through one treatment after another, each less likely to work than the last. She was admitted to the hospital again with pneumonia, and by the time the team recognized she was dying, she was on a ventilator with a tube in her chest. She died in the ICU. Nobody had asked her what she wanted until there was nothing left to offer.¹⁹

When someone finally asks, the answers are not clinical. They are human. One man wanted to eat chocolate ice cream and watch football on television. One woman wanted to stay in her own home with her dog. Another wanted to be well enough to attend a wedding. These are not treatment goals. They are life goals. No protocol captures them. No diagnosis code contains them. No billing system reimburses the conversation that surfaces them.²⁰

If the diagnosis were yours, you would have answers like these. Specific. Personal. Nothing the system could generate from your chart. The system knows your blood pressure, your lab work, your imaging. It does not know what you would trade six months for. It does not know what you would refuse. It does not know what would make the remaining time feel like yours. That knowledge lives in you and nowhere else, and the system has no mechanism to ask for it.

Seventy-one percent of Americans say they would prefer to die at home. Nine percent prefer a hospital.²¹

In 2017, 29.8 percent of Americans died in hospitals. Down from 39.7 percent in 2003. The number is moving in the right direction. It is still not close to what people say they want.²²

Eighty percent of Americans say they would want to talk to their doctor about end-of-life wishes if they were seriously ill. Seven percent have actually had that conversation.²³ Only eleven percent have ever discussed end-of-life wishes with a doctor at all. Not seven percent of the terminally ill. Seven percent of everyone. The conversation that the Mass General trial proved was the most effective intervention available for dying patients happens almost never.

In 1995, the largest study of dying in American hospitals followed 9,105 adults across five teaching hospitals. Nearly half had a six-month mortality rate. The findings, published in JAMA, were blunt. Only forty-seven percent of physicians knew when their patients preferred to avoid CPR. Forty-six percent of do-not-resuscitate orders were written within two days of death. Thirty-eight percent of patients who died spent ten or more days in an ICU, in a coma, or on a ventilator. Half of conscious patients who died had families who reported moderate-to-severe pain at least half the time.²⁴

That study was thirty-one years ago. The system has not fundamentally changed.

In 2012, the New England Journal of Medicine published a study of 1,193 patients with metastatic cancer. Sixty-nine percent of those with lung cancer and eighty-one percent of those with colorectal cancer did not understand that chemotherapy was not at all likely to cure their disease.²⁵

The patients were being treated for a terminal illness. Most of them did not know their treatment could not cure them.

The oncologist did not lie. The oncologist presented the treatment options. The patient heard options and inferred hope. The system's language, its culture of fighting, its reimbursement of intervention, its legal bias toward action, all of it created a context in which the patient understood "we can try this treatment" as "this treatment might save me." Nobody corrected the inference because correcting it requires a conversation the system does not pay for, does not train doctors to have, and does not protect doctors who attempt it.

When someone does ask the question, the answer changes everything. The patients in the Mass General trial who received palliative care were less likely to receive aggressive end-of-life treatment. Thirty-three percent versus fifty-four. They entered hospice with a median of eleven days remaining, not four. They had documented their resuscitation preferences. Fifty-three percent versus twenty-eight. The conversation did not give them different medical options. It gave them the information to choose among the options that already existed.²⁶

The intervention was a question and the patience to hear the answer.

The counter-case

Medicine also produced the strongest example of measurement working exactly as intended.

In 2009, the World Health Organization published the results of a surgical safety checklist tested across eight hospitals on four continents. Nineteen items. Three pause points. Two minutes. Before anesthesia: confirm the patient. Before incision: confirm the site. Before the patient leaves: confirm the count. The rate of major complications fell thirty-six percent. Deaths fell forty-seven percent.²⁷

The checklist did not measure whether the patient survived. It measured whether the team communicated. It measured the preconditions for good outcomes, not the outcomes themselves. The distinction matters. When measurement is upstream of the outcome, when it tracks the process rather than the result, it works. The Apgar score transformed obstetrics. The checklist transformed surgery. Both measured the right thing.

The survival metric in end-of-life care measures the wrong thing. It counts days alive. The patients it is measuring want something else. The system optimized for survival so thoroughly that, in the Mass General trial, the patients who received less treatment lived longer. The metric did not merely replace the thing it was supposed to represent. It degraded it.

The same doctor who invented the checklist wrote the book about dying. He held both truths at the same time. Measurement saves lives. And measurement, pointed at the wrong thing, can destroy them.²⁸

The strongest counterargument is that the system works. For most visits, it does. You break your arm. The system sets it. You have an infection. The system treats it. The training is real. The expertise is real. For problems that are acute, identifiable, and fixable, the system's bias toward action is the right bias. Nobody wants a doctor who watches and waits while a bone heals crooked.

The problem is not the system's capacity. It is the system's default. The default is more. More tests. More treatment. More intervention. And for most visits, the default and the patient's interest align closely enough that nobody notices. The arm gets set. The infection clears. The patient goes home.

For the visit that matters most, the one where the question is not how do I survive but how do I live, the default and the patient's need point in opposite directions. The data says less treatment produces a better outcome. The system defaults to more. The patient receives aggressive care not because the doctor concluded that aggressive care was best but because the insurance covers it, the metrics incentivize it, the legal environment protects it, the payment structure rewards it, and the training produced a physician who knows how to do it. All of that was in place before the patient entered the room.²⁹

The cost of this default is not paid by the system. It is paid by the person in the bed. A 2008 study found that terminally ill patients who received the most aggressive interventions, ventilators, defibrillation, chest compressions, ICU admission near death, had a substantially worse quality of life in their last week. Their caregivers were three times as likely to suffer major depression six months later.³⁰ The cost is invisible to the system because the system measures survival, not suffering. The patient who dies in the ICU after four rounds of chemotherapy is, by the system's metrics, a patient who received the full standard of care.

One economist had a name for this structure. When the person making the decision does not bear the cost of being wrong, the decision does not self-correct. The doctor who over-treats bears no cost from the unnecessary test. The patient bears no visible cost because insurance covers it. The insurer bears the cost and distributes it across all premiums. The cost is socialized. The decision is individual. Nobody in the chain is the residual claimant of the patient's actual wellbeing.³¹ The structure is not unique to medicine. It operates wherever the person making the decision is insulated from the consequence of being wrong. Medicine is where you feel it in your body.

Hospice, when patients finally arrive, has a median stay of seventeen days. Twenty-eight percent of hospice patients stay fewer than seven days. One in six cancer patients enters hospice in their final three days of life.³² The system does not deny hospice. It delays the conversation that leads to it. By the time the conversation happens, the remaining time is measured in days, not months. The palliative care group in the Mass General trial entered hospice with eleven days remaining. The standard care group entered with four. Both numbers are small. One of them is three times larger than the other.

You went to the doctor. The doctor examined you. The doctor recommended a treatment. You did what the doctor said. The expertise was real. The recommendation was real.

The system that shaped the recommendation was also real. And it was in place before you entered the room.