Does Advance Care Planning Improve End-of-Life Care?

The short answer, on the best evidence available, is no, at least not in the way the practice is usually sold. Advance care planning, meaning a documented conversation in which patients record their wishes for future treatment, has not been shown in large randomized trials to make the care people actually receive at the end of life match the care they said they wanted. That gap matters, because matching wishes to care is the entire point of the exercise. The conversations are not worthless. What the data undercut is a quieter assumption baked into policy and payment: that documenting a plan is close enough to delivering on it.

Key points#

• Randomized trials have repeatedly failed to show that advance care planning (ACP) improves goal-concordant care, meaning care that matches what a patient said they wanted.
• What ACP conversations do improve is nearer-term and real: clearer communication, less decisional conflict, and better agreement between patients and their families.
• Medicare has paid clinicians for these conversations since 2016, which turned a documented plan into an easy thing to count and track.
• The wider lesson is a classic appraisal trap: when the true outcome is hard to measure, systems substitute a countable proxy, and the proxy slowly becomes the target.

A plan you can bill for is a plan you can count#

Advance care planning asks people to think ahead about what they would and would not want if illness left them unable to speak for themselves, and to write those wishes down. It is endorsed across guidelines and quality frameworks, and it comes with a price tag attached. According to the Centers for Medicare and Medicaid Services, since January 2016 clinicians can bill for these conversations under CPT codes 99497 and 99498, the second an add-on for each additional half hour of discussion.

Reimbursement changes behavior in a predictable way. Once a service is paid for and recommended, it becomes something health systems watch. A billed code or a completed form is easy to log, easy to audit, and easy to put on a dashboard. Before long, the presence of an ACP note starts to stand in for the quality of someone's final months of care. The problem is not that anyone chose a bad proxy on purpose. The problem is that the note and the outcome are two different things, and the trials show the distance between them.

What the randomized evidence actually shows#

A 2021 viewpoint in JAMA, pointedly titled "What's Wrong With Advance Care Planning?", laid out the case that the core promise is not supported by data. Its authors cited a 2018 overview pulling together 80 systematic reviews that found no evidence ACP influenced medical decision making at the end of life or raised the likelihood of goal-concordant care. They also pointed to five large randomized trials spanning cancer, nursing home, primary care, serious illness, and heart failure populations, none of which found meaningful differences in health care use, quality of life, or whether care matched stated goals.

A 2022 systematic review in BMJ Open put numbers on it. Across 132 randomized controlled trials of ACP, only 12 bothered to measure whether the care patients received actually matched their preferences, and just 3 of those found a benefit. Not one of the 14 trials that measured quality of life showed improvement. Only 4 of 22 trials showed any reduction in health care use or cost. What did improve, and fairly consistently, were the proximal things: communication with clinicians, less decisional conflict, and closer agreement between patients and their families about what they wanted. In other words, the conversation genuinely helps the people having it. It just does not reliably reach the distant outcome the metric claims to capture.

Why goal-concordant care is so hard to pin down#

Goal-concordant care has the ring of a tidy outcome, but honestly measuring it demands two things that are slippery in practice. First, you have to know the patient's goals at the exact moment a real decision lands. Second, you have to know whether the care delivered actually met them.

Both are harder than they sound. Preferences are not carved in stone. What someone states in a calm clinic visit, months before anything happens, can differ from the choice they would make when a specific, urgent decision is in front of them and the stakes are concrete rather than hypothetical. And concordance is almost always judged after the fact, frequently through a grieving family member's memory of what the patient would have wanted, which carries its own distortions.

Faced with an outcome this elusive, systems reached for what they could see: was a conversation documented, was a code billed. Think of it as checking that a map exists rather than confirming the traveler reached the destination. Verifying the form is easy. It tells you almost nothing about whether a person's last weeks reflected what mattered most to them.

The real lesson is about proxies#

This is the part worth carrying to other topics. When an outcome is difficult to measure, we reach for something countable that seems to track it. That proxy gets recommended, reimbursed, and monitored, and over time it stops being a stand-in and becomes the goal itself. The original assumption, that the proxy moves with the true outcome, is rarely re-examined once the metric is entrenched. ACP documentation is a clean example: a reasonable, well-meant surrogate that several trials show does not predict the result it was chosen to represent.

A durable habit when sizing up any quality metric is to ask two plain questions. Is this a process measure (something was done) or an outcome measure (something got better)? And has anyone actually shown the two correlate, or is the link simply assumed? Face validity, the intuition that a measure ought to work, is not the same as evidence that it does. Neither endorsement nor a billing code settles the matter.

What the evidence still backs#

None of this is an argument for dropping the conversations. The JAMA authors suggest redirecting the effort rather than abandoning it: help patients name a trusted decision maker ahead of time, invest in strong shared decision making at the moment real choices arise instead of far in advance, and measure what patients and families actually live through, including symptom burden and whether a surrogate felt the patient was heard. The BMJ Open group makes a complementary case, proposing that the field rename the work advance care preparation, building readiness and communication skills, rather than planning that claims to forecast future choices.

The correction is modest and honest. It keeps what the trials show helps, the communication and the preparation, and stops crediting a document with a result it does not deliver.