The palliative care community has been rallying to defend itself this past week in the aftermath of a scathing attack in the New England Journal of Medicine. A short article written by Scott Halpern, a specialist in intensive care medicine, asserts that none of the approaches used by palliative care clinicians, and he singles out advance care planning tools such as physician orders for life-sustaining treatment (POLST) and videos depicting CPR, have been proven to work. He implies that the only way to know whether they are effective is to use the same criteria as those employed by the FDA in approving a new drug: typically, large, multi-center, randomized clinical trials. Superficially, this seems advisable—or even, to use the language that Medicare invokes in determining what it will cover—reasonable and necessary. It is, after all, a plea for evidence-based medicine, and surely all right-thinking physicians are in favor of using scientifically supported treatment. But to propose that we can learn all we need to know about how to provide excellent palliative care from randomized, double blind, placebo-controlled trials, studies in which two groups are compared that are identical in all respects except for the intervention being tested, is sadly naïve and hopelessly reductionist.
Advance directives are promoted to enable patients to shape their own destiny by choosing, with input from their doctors, how they will be treated when they are seriously ill. Advance care planning seeks to protect patients by safeguarding their autonomy in the event of incapacity; it also aspires to help all patients prepare for the future, even those who retain their cognitive capacity. It’s difficult for anyone to think clearly in the throes of serious illness; prior consideration of the goals of care can provide a useful framework for decision-making under stress. But if the point of advance care planning is to encourage patients to think about their goals and preferences and to allow them to have a say in their care, then readily measurable outcomes—things like how much pain patients experience at the end of life or how many procedures they have or how long they live or what their care costs—are the wrong endpoints.
A randomized trial doesn’t make any sense in this setting: we cannot compare patients who participate in advance care planning (cases) with those who do not (controls) and measure whether the planners are more likely than the non-planners to get the kinds of treatment they want. Such a comparison presupposes that everyone actually knows whether he or she wants attempted CPR or a ventilator or third line chemotherapy in much the same way that people know whether they prefer vanilla or chocolate ice cream. But choosing between vanilla and chocolate requires that people have already tasted both and arrived at a conclusion about which they like more. With end of life medical treatment, patients seldom have any prior experience on which to base a decision. They need help in imagining what they will go through with the various alternatives. And they can’t sample the different possibilities and then decide (though a limited trial is sometimes possible): they will only die once. Nor is it accurate to believe that all patients have well-formed preferences, based on their religious beliefs or values, and that advance care planning simply informs the medical profession of those preferences. In many cases, patients don’t have clear preferences. The essence of good advance care planning is that it elicits preferences—it helps patients figure out what makes sense for them, given the realities of their medical condition.
I’m not suggesting that all formal study of advance care planning is meaningless. Demonstrating that patients who go through a planning process are satisfied with their experience is important. So is comparing different approaches to figure out which one is best received (or perhaps which patients like which strategy). Assessing whether patients who engage in advance care planning are more knowledgeable about their options than those who do not is also useful. So are studies of the implementation of the advance care planning process: finding out whether advance directives (written documentation of the planning process) are available to physicians at the time when decisions need to be made and measuring the concordance between those requests made in advance and what actually happens. All these sorts of studies have been done and collectively make the case that advance care planning is effective.
Although Dr. Halpern longs for a large, randomized controlled study to definitively determine whether advance care planning “works,” he seems to be aware of the inaptness of his FDA analogy. He therefore proposes analyzing electronic medical records using natural language processing as a way to draw conclusions about the efficacy of various palliative care interventions, including advance care planning. Nice idea, but I’m skeptical. It’s not that I’m hostile to natural language processing or totally ignorant of its power: my son has a doctorate in computer science with a specialty in natural language processing and does remarkable work studying how machines can determine the meaning of text. My husband is the co-founder of a tech start up that makes extensive use of natural language processing. It’s a promising technique with many as yet undetermined applications. Maybe it will even prove useful in evaluating advance care planning. But surely we don’t know that. In fact, there are strong a priori reasons for thinking that mining chart notes for words that describe a patient’s experience won’t do anything other than measure how likely a physician is to ask patients about their feelings and to record their answers.
You might think, reading the New England Journal article, that all of medicine, apart from palliative care, has been subjected to the standards demanded by the FDA. You would be mistaken. A great many of the tests and treatments used regularly by physicians have not been tested with any kind of rigor. But even more to the point, physicians still use their judgment in deciding whether to use a given drug in a given patient, even when that drug has been shown in a carefully conducted randomized clinical trials to be superior to the alternatives. Randomized clinical trials are typically carried out in patients who have nothing wrong with them except the condition being studied. Those patients are rarely over age 65. That vaunted FDA-approved pharmaceutical is then used in 85-year-old patients with multiple chronic illnesses. Maybe it will work just as well. But maybe it won’t work at all or will even cause serious side-effects. It might hasten death. Does Dr. Halpern only use treatments in his ICU that have been demonstrated in a randomized trial to be effective—and to be effective in a patient with exactly the combination of heart failure and diabetes and dementia that the patient before him has?
Of course we need to evaluate the tests and treatments that we use in patients. But we should use whatever method is most appropriate for the situation we are studying. Sometimes it will be a randomized clinical trial. Often it will be a combination of other kinds of studies, some of which may be qualitative. Whether we conclude that the test or treatment should be used will depend on the preponderance of the evidence. Only rarely will we know, beyond a reasonable doubt, that we are doing the right thing.