Beyond Mandatory Autonomy: When to Stop Asking

Last week I talked about punctuation on its own terms, on a decision small enough to argue about safely: a PCA coming off. That post ended by saying the period closes the recommendation and not the conversation. This one is about the conversation it leaves open.

Someone's father is in septic shock on two pressors, third admission in six weeks, pancreatic cancer that stopped answering the chemotherapy months ago. His son is in the family room and has been awake for thirty-one hours. And you have to do some talking.

Here is where the training runs out. We teach people to ask beautiful questions. We do not teach them when to stop asking.

I'm going to narrow this down so I can make the strongest case: CPR, offered to a person in whom it cannot produce the thing they told you they wanted.

The Sixty-Eight Percent Ceiling

Start with what we know about the answers. David Shalowitz and colleagues pooled the surrogate accuracy literature and found that surrogates predicted patients' treatment preferences with 68% accuracy. One in three wrong. Neither designating the surrogate in advance nor discussing preferences with them beforehand improved that number. The intervention we prescribe for this problem does not fix this problem, which is a theme I've taken up elsewhere.

Sixty-eight percent is the ceiling. Those studies used hypothetical scenarios, in calm rooms, with people who had slept. Now put the same person in the ICU at hour thirty-one. Grant Shields and colleagues meta-analyzed the effect of acute stress on executive function and found that stress impairs working memory and cognitive flexibility, the two faculties you need to hold a hypothetical future in your head and turn it over. We are asking a person whose cognition is chemically degraded to perform a task they fail a third of the time at their best.

Then we hand them the receipt. David Wendler and Annette Rid reviewed the surrogate outcome literature and found that at least a third are left with a negative emotional burden from making the decision—stress, guilt, and doubt about whether they got it right—often substantial, typically lasting months and sometimes years. The same review reports what reduced that burden: knowing which treatment matched the patient's preferences. The relief lived in the answer, and we have been handing out the question.

Part 1 called the menu a transfer of burden. This is the physiology of the transfer, with an invoice that comes due long after the room has emptied.

What We're Asking Them to Authorize

A question is only as good as the options inside it.

Per the 2025 AHA guidelines, drawing on 2023 Get With The Guidelines–Resuscitation data, 23.6% of adults who arrest in the hospital survive to discharge. That is a better number than most of us have in our heads. Mine is worse than most, because my denominator is selected. I want it on the record before I complicate it, because the case I'm making needs CPR to be specific.

So get specific. Gary Reisfield and colleagues pooled the in-hospital CPR literature in cancer patients. Survival to discharge overall, 6.2%. In the ICU, 2.2%. That last one is the number for our man on two pressors.

Now consider what his son thinks the number is. In 1996, Susan Diem, John Lantos, and James Tulsky watched a full season of ER and Chicago Hope plus fifty episodes of Rescue 911 and counted. Of sixty televised resuscitations, 67% survived to discharge. Two-thirds of the arrests happened to children, teenagers, or young adults.

Nineteen years later, Jaclyn Portanova and colleagues ran it back on Grey's Anatomy and House and got almost the same numbers. Two decades of quality improvement and communication training, and the number on television didn't move. One more finding from that paper deserves its own sentence: across ninety-one episodes, advance directives came up twice.

So the son in the family room has a number in his head. It is roughly thirty times the number in yours. Television also taught him that nobody discusses this in advance, which means your bringing it up at all reads to him as an emergency. Ask him what his father would want, write down his answer, call it a preference, and you have performed stenography.

Some honest fine print, since the numbers are doing a lot of work here. The 23.6% comes from hospitals that participate in a quality registry, which is a selected group. "Favorable neurologic outcome," the figure usually quoted alongside it, includes moderate disability, which is a wide door. And Reisfield's own reading of his improving numbers is that they reflect deselection: the sickest patients getting DNR orders and never entering the denominator. Which is to say, some of what looks like better CPR is better conversation.

And here is where it stops being an arithmetic problem. Amber Barnato and colleagues examined patients hospitalized with COVID-19 across a national sample of community hospitals. Rates of advance care planning were essentially identical across groups, 11 to 12 percent. Rates of DNR orders were not: White 13%, Hispanic 8%, Black 7%. Among those who died, 21% of White patients died receiving maximal life support, compared with 36% of Black patients. The authors' own conclusion is the one I'd have written: greater treatment intensity may have simultaneously narrowed the mortality gap and increased burdensome treatment near death.

Same rate of planning. Different rate of orders. That gap is where this whole essay lives, and the data cannot tell you whether it is preference or practice.

The Middle Ground and the Man Who Called It Deceit

In 2007, Chest ran a point/counterpoint framed by a grand rounds vignette: a visiting professor explained that when CPR is unlikely to produce survival with a reasonable quality of life, he tells the patient his assessment, and if the patient doesn't object, he writes the order.

Randall Curtis and Robert Burt took the point. Their argument was that there is a middle ground between wrongful withholding of information and harmful impositions of decision-making burden, and they called that middle ground informed assent: give full information about risks and benefits, convey a specific recommendation, and make explicit that the patient and family are entitled to defer to your judgment.

Alexander Kon—whose shared decision-making continuum I introduced back in Part 1—wrote shortly after to fix the word. In pediatrics, assent means the active affirmation of a minor who lacks legal capacity. What Curtis and Burt described has every element of informed consent except the active agreement. Which makes it non-dissent. Kon's own defense of the technique is the most human sentence in the literature: there is no reason to require families to say let her die.

Constantine Manthous took the counterpoint, and he did not hedge. His position was that in 2007 America it is unethical for physicians acting alone to withhold CPR without seeking consent, on three grounds: no clinician is omniscient, no clinician is infallible, and the clinician's values quietly displace the work of finding out the patient's.

Then he took the silence apart.

Manthous's four silences. When you say "unless you object, I'm going to write this order," and the room says nothing, that silence can mean:

• understanding, with agreement
• understanding, with neither agreement nor disagreement
• understanding, with disagreement
• not understanding

Write the order without knowing which one you're in, he argued, and it is DNR by deceit.

He is right, and the correct response to being told you're right is to say so and then read the next paragraph, because Manthous's conclusion is narrower than it's often made out to be. What he condemns is truly unilateral DNR—a single practitioner withholding CPR without disclosing it—as arbitrary and deceitful. What he grants, in the same breath, is that a system withholding treatment through carefully orchestrated processes, crafted in advance by consensus, is ethical and perhaps inevitable.

So the disagreement between the point and the counterpoint is smaller than the volume suggests. Both men accept that clinicians may decline to provide interventions that cannot work. They are arguing about disclosure and the veto. Curtis said so himself five years later, writing in the AMA Journal of Ethics: when a family asks for CPR, he provides it, and reserves genuine unilateral refusal for resource-intensive therapies like ECMO.

Which gives us the distinction the literature keeps smudging, and the smudge is why this technique frightens people who have only heard it described secondhand:

Informed non-dissent is a recommendation with a real off-ramp. Unilateral DNR is an order without one. Everything below turns on keeping those two things apart.

The professional architecture already draws this line, and draws it well. The 2015 multi-society statement on potentially inappropriate treatment in ICUs—ATS, AACN, ACCP, ESICM, and SCCM together—did the field a service by retiring the word futile for everything except interventions that cannot accomplish the intended physiologic goal. Everything else is "potentially inappropriate," which is honest, because it locates the disagreement in values rather than physics. That statement tells clinicians to communicate and advocate for the plan they believe is right, and to route intractable disagreement through a fair process rather than a unilateral act. The AMA's Opinion 5.5 is consistent: physicians are not required to offer interventions that cannot reasonably be expected to achieve agreed-on goals of care, and disputes get process. Texas has a statute that goes further, with committee review and a transfer window, and I'm not going to litigate it here, because it is the outer boundary and almost nothing reaches the boundary.

We spent thirty years arguing about the endpoint. The endpoint is rare. The conversation is daily. All of that policy architecture exists to handle the cases where the room fails, and the room fails less often when someone in it is willing to say a sentence.

Five Steps, and the One I Left Out

At the bottom of every post in this series there's a playbook. Step 7 says: use informed non-dissent for non-beneficial CPR—elicit values, explain why CPR won't achieve those ends, state a caring assessment, invite disagreement. Four arrows. It's time to complicate them.

So praise or blame can fall where it must: this sequence is mine. Curtis and Burt gave us the concept. Kon gave it the right name and the guardrail. Manthous gave it the objection it has to survive. Nobody has published a stepwise version of this, and nobody has tested the one you're about to read, including me. Take it as a proposal rather than a protocol.

1. Check the gate

Two facts have to be in your hands before you say anything. What this person wants, in their own words. And what CPR does in their specific situation, as a number. Missing either one and you're not doing informed non-dissent.

Three facts, actually. Last week I conceded that this rule needs a front door, because when Doug White and colleagues showed surrogates two versions of a life-support conversation, 56% wanted a recommendation and 42% did not. Forty-two percent is not noise. So ask. "Some families want me to say what I'd recommend. Others want to work through it together. Which is more useful to you right now?" That question is about my role, not about the medicine. Whether chest compressions can restart a heart in septic shock is not up for a vote. Who says the sentence out loud is.

Last week's rule was that a conclusion you can't defend hasn't earned the period. This is the version one step earlier: a conclusion you never sourced can't be defended in the first place.

The gate is also where equity stops being an abstraction. Daryl Thornton and colleagues audiotaped ICU family conferences. The interpreted conferences ran the same length as the others, 26.3 minutes versus 32, a difference that didn't reach significance. But clinicians spoke for 10.9 minutes in interpreted conferences versus 19.6 in the others. They made fewer statements valuing the family's input, fewer easing emotional burden, less active listening. We didn't leave the room early. We sat in it and said half as much.

If you spoke for eleven minutes through an interpreter, you may not have this person's values.

2. Say why it can't get them there

Part 3 was about telling the truth in general. This is the truth aimed at one object. The number that matters is the one attached to the thing they said they wanted, not the number in the abstract.

"You told me what you want is to get back to your own bed and see the dog. If your heart stops, the chest compressions don't get you there. If they would, I'd say let's do them."

That sentence is only available to you if you did step 1. This is why the order matters.

3. Recommend in a declarative

Then you say it, with a period on the end and your name attached.

"...but they won't, so my recommendation is that we not attempt resuscitation and instead allow you a natural death when that time comes."

I argued that punctuation last week and won't relitigate it. What's different here is the price of the question mark. It hands the weight back to a man who has been awake for thirty-one hours and who is, per Shalowitz, likely to guess wrong.

4. Make dissent free

Curtis and Burt's third clause is that the family is entitled to defer to your judgment. The inverse clause does the ethical work: they are equally entitled to override you, and Kon is explicit that the family retains the power to override any such decision. Say it out loud. "If that's wrong, tell me, and we'll talk about it."

Last week I offered a test for whether a choice is real: if either answer would frustrate you, you weren't offering one. Run it here, before you speak. Imagine the son says no. If what you feel is your jaw tightening, your off-ramp is decorative, and he will read that on your face long before he reads it in your sentence.

Then account for what the off-ramp costs him. Lisa Cooper-Patrick and colleagues surveyed patients about their physicians' participatory decision-making style and found Black patients rated their visits less participatory, and patients in race-concordant relationships rated theirs more participatory than those in discordant ones. The effect is small and the instrument is from 1999. What it does is name a mechanism. A person who has learned across a lifetime that disagreeing with a doctor gets you labeled difficult does not acquire a free off-ramp because you said a nice sentence about it. The exit has a price, and the price is not the same for everyone.

So pay it down. Say the disagreement is welcome. Say it twice. Say what happens if they take it, because the fear underneath the silence is usually not about CPR at all. It's about whether you'll still show up tomorrow. Part 3 called that nonabandonment, and it belongs here more than anywhere.

5. Disambiguate the silence

This is the step that isn't in the playbook, the one that answers Manthous, and the pivot the whole sequence turns on.

You cannot write the order until you know which of his four silences you're standing in. So check—not on the information, which is what Ask–Tell–Ask usually does, but on the recommendation itself. "I just said something big. What are you taking away from it?"

Then write down which one you got. Rochelle Melvin and colleagues read what physicians actually documented after code status discussions across four hospitals. They found the full range, from advice that deferred the decision to explicitly not offering interventions—informed non-dissent, in the wild. They also found that physicians documented a rationale, and then a paucity of documentation supporting how they arrived at it. And some physicians used the phrase "quality of life" to imply the treatment wouldn't help, which is a euphemism doing the work of an argument.

We are doing this. We are not showing our work. Part 2 built the documentation ethic; this is where it gets spent. Record the words they used. Record which silence you were standing in. Record how you know.

And then the guardrail, which is the entire point: if step 5 comes back understanding, with disagreement, you are done. He is full code. You keep showing up. A framework that always produces the same output is a script with extra steps.

Where I Might Be Wrong

The best evidence against me is about regret, and it's about race. Sarah Andersen and colleagues took the PARTNER 2 cohort and measured decision regret at six months. Nineteen percent reported moderate-to-strong regret. Limitations in life support before death were associated with more regret. Surrogate Black race was associated with more regret. And palliative care consultation was associated with more regret. I am advocating a technique that steers toward limitation, and both limitation and my own specialty's fingerprints track with people feeling worse six months later. The model explains 3% of the variance, the confidence interval on race nearly touches zero, and we get consulted for the hardest rooms—all true, all mitigating. But the authors' own read is that the race finding may reflect a breakdown in shared decision-making, and that is my thesis pointing right back at me.

My anchor citation doesn't say what I want it to say. Reisfield's meta-analysis concludes that cancer patients' CPR survival compares favorably to unselected inpatients. I used his subgroup numbers to make a point he declined to make. The 2.2% is real and I'll defend it. The framing around it is mine, and you should know that.

The differential-application worry is a worry rather than a finding. I cannot show you a study demonstrating that "potentially inappropriate" gets applied more readily to Black patients. The multi-society statement names the risk—patients affected by these determinations are vulnerable by definition, have little choice of clinician, and limited ability to seek treatment elsewhere—but nobody has measured the pattern. I'm reasoning from adjacent data: the DNR gap, the interpreter gap, the participatory-style gap. That's an inference and it isn't a result. If someone has the study, I want it, including if it points the other way.

Structured communication has a mixed record on the outcomes that matter most. Douglas White's PARTNER trial randomized a family-support intervention and missed its primary endpoint; anxiety, depression, and PTSD symptoms at six months were the same in both arms. What improved was the quality of communication and the length of ICU stay, and that last one only among patients who died, 4.4 days versus 6.8. I read that as the intervention shortening the part of dying that happens in an ICU, which is exactly the target. A skeptic reads it as me calling a process measure a win because the outcome measure didn't cooperate. The skeptic has a case.

My framework has a resourcing problem I can't solve from here. Step 1 demands a professional interpreter, unhurried values elicitation, and a clinician with the time to do both. Those are line items. The settings least likely to have them are, reliably, the settings serving the patients with the most to lose, which means the disciplined version of this technique gets practiced where the resources are and a stripped-down version gets practiced everywhere else. A stripped-down informed non-dissent, with the gate unchecked and the silence undisambiguated, is unilateral DNR wearing a lanyard. I've written a framework whose safety features are the first things a short-staffed night shift will drop. I don't have a fix for that inside the conversation, because the fix lives on a budget line rather than in a playbook.

And the mistrust story is too easy. The reflex, when we see the DNR gap, is to reach for Tuskegee and call it explained. Catheryn Koss and Tamara Baker went looking for that in the Health and Retirement Study and found that controlling for medical mistrust didn't shrink the Black–White gap in advance directive completion, and neither did measures of perceived discrimination. Kimberly Johnson's review found something the mistrust story misses entirely: that intensive treatment preferences among many African American patients are anchored in affirmative belief—that only God decides, that the doctor is God's instrument—rather than fear of us. Explaining a patient by their historical trauma is its own condescension. It also lets us off the hook for the version of the problem we can actually fix, which is that we spent nine fewer minutes talking.

Safe to Refuse

The son has been awake for thirty-one hours. You asked whether he wanted your recommendation, and he said yes. You know what his father wanted, because you asked in a room with an interpreter and enough time. You know what CPR does to a man in septic shock with metastatic disease, and the number is 2.2%. You've said why it can't get him where he wanted to go. You've made your recommendation, with a period.

Now the ninety seconds after the sentence, when you stay in the room, find out what he actually heard, and mean it when you tell him he can say no.

The three posts before this one asked you to stop doing things. Stop handing over the menu. Stop pretending you're neutral when you aren't. Stop tiptoeing around the prognosis. This one asks you to do something, and to keep doing it after it stops feeling good.

The asking is the easy part. It's the move that lets you leave the room with clean hands and a documented preference and a man who will get chest compressions he never wanted, because his son had thirty-one hours of adrenaline and a lifetime of television and forty-five seconds to answer.

The hard part is becoming someone whose recommendation is safe to refuse. That takes everything the first three posts were about: knowing the person well enough to speak for them, being honest enough that they believe you, and being trustworthy enough that "no" costs them nothing.

Last week I wrote that the period closes the recommendation and not the conversation. Here is the harder version. The period only closes the recommendation if the room believes it can be reopened.

This is the final installment of Beyond Mandatory Autonomy. Start at Part 1, then Part 2 and Part 3. The grammar argument underneath all of it is in The Question That Ends With a Period.

I am a palliative care physician, educator, and professional strategery expert. Known for turning rounds into rants and rants into teaching points. Rounds & Rants represents my views — not those of any institution or professional membership organization where I hold a role. I don't write on their behalf and they don't vet what I publish.