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

On the best available evidence, advance care planning as it is commonly practiced, a documented conversation about future treatment preferences, does not reliably improve the care people receive at the end of life. Large randomized trials have repeatedly failed to show that it makes the care patients get match the care they said they wanted, which is the outcome the intervention is supposed to produce. That does not mean these conversations are pointless. It means the field measured a document as if it were a result, and built a reimbursed quality metric on a proxy the data do not support.

A widely promoted, reimbursed practice

Advance care planning (ACP) asks patients to think through and record their wishes for future medical treatment, often naming what interventions they would or would not want if they became too sick to speak for themselves. The practice is endorsed across professional guidelines and quality frameworks, and it carries a payment code: since January 2016, Medicare has reimbursed clinicians for these conversations under CPT codes 99497 and 99498, the second an add-on for each additional half hour, according to the Centers for Medicare and Medicaid Services. Once a service is reimbursed and recommended, health systems tend to track it. Documentation that an ACP conversation happened, or that a code was billed, became a convenient marker of good end-of-life care.

The trouble is that the marker and the goal are not the same thing.

What the randomized trials found

In a 2021 JAMA viewpoint titled "What's Wrong With Advance Care Planning?", R. Sean Morrison, Diane E. Meier, and Robert M. Arnold argued that the evidence does not support the core premise. They point to a 2018 review of 80 systematic reviews that found no evidence ACP influenced medical decision making at the end of life or enhanced the likelihood of goal-concordant care, and to five large randomized trials across cancer, nursing home, primary care, serious illness, and heart failure populations that could not identify meaningful differences in health care use, quality of life, or goal-concordant care.

A 2022 systematic review in BMJ Open by Chetna Malhotra and colleagues sharpens the picture. Across 132 randomized controlled trials of ACP, only 12 assessed whether the care patients actually received matched their stated preferences, and only 3 of those found a benefit. None of the 14 trials measuring quality of life showed improvement, and only 4 of 22 trials showed reduced health care use or cost. What did improve, fairly consistently, were nearer-term outcomes: communication with clinicians, reduced decisional conflict, and agreement between patients and their caregivers about preferences. The conversation does something real. It just does not move the distal outcome the metric claims to capture.

Why goal-concordant care resists measurement

Goal-concordant care sounds like a clean outcome, but measuring it honestly requires two things that are genuinely hard to pin down: knowing the patient's goals at the moment a real decision arises, and knowing whether the care delivered actually matched them. Preferences are not fixed. What a person says months in advance, in a calm clinic, may not describe the choice they would make when an acute, specific decision is in front of them. And concordance is usually judged in retrospect, often through a surrogate's recollection, which carries its own biases.

Because the true outcome is so difficult to observe, systems reached for something countable instead: was a conversation documented, was the code billed. That is a process measure. Confirming that a form exists is easy. It tells you almost nothing about whether the care a person received in their last weeks reflected what mattered to them.

The lesson about surrogate quality metrics

This is the part worth generalizing. When an outcome is hard to measure, we substitute a proxy that is easy to count. The proxy gets recommended, reimbursed, and tracked, and over time it quietly becomes the target. The assumption that the proxy tracks the real outcome is rarely re-tested once the metric is in place. ACP documentation is a clean case study: a plausible, well-intentioned surrogate that multiple trials show does not predict the outcome it was meant to stand for.

A useful habit when evaluating any quality metric is to ask two questions. Is this a process measure or an outcome measure? And has anyone demonstrated that the two actually correlate, or is the link merely assumed? Face validity, the sense that a measure ought to work, is not the same as evidence that it does. Reimbursement and endorsement do not settle the question either.

What the evidence still supports

None of this argues for abandoning the conversations. Morrison and colleagues suggest redirecting the effort: help patients name a trusted surrogate decision maker in advance, invest in high-quality shared decision making at the moment real decisions arise rather than far ahead of them, and measure what patients and families actually experience, including symptom burden and whether surrogates felt the patient was heard and understood. Malhotra's group makes a complementary point, proposing that the field reframe the work as advance care preparation, building readiness and communication skills, rather than as planning that promises to predict future choices.

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

This is educational and not medical advice.