Many health care providers and patient advocates have worked diligently to alter the perception of palliative care from something that occurs strictly at the end of life to an ongoing course of action for symptom management and burden reduction.
The focus of research in this area has also shifted, with a renewed emphasis on both its medical and financial efficacy. A recent study published in JAMA found that patients assigned to palliative care following a hematopoietic stem cell transplant reported better overall quality of life than individuals receiving standard care, with decreased rates of posttransplant depression, anxiety, and symptom burden.1 A 2016 study from Health Affairs found that advanced cancer patients who utilized palliative care achieved a reduction in the financial burden of treatment up to 32%.2
Despite increased visibility, palliative care has not been fully integrated into the treatment arena. A study by Epstein and colleagues3 showed that up to 95% of advanced cancer patients surveyed had an insufficient understanding of the state of their disease, rendering them unable to make informed decisions regarding palliative treatment options. Thirty-eight percent of patients reported never having discussed their prognosis or palliative care appropriateness with their oncologists.
Conversations regarding palliative care utilization have proven particularly challenging within the realm of hematologic malignancies, where many advanced diseases remain potentially curable well beyond the threshold often seen with solid cancers.4
Researchers from Indiana University Northwest (Gary, IN), University of South Florida Morsani School of Medicine (Tampa, FL), and H. Lee Moffitt Cancer Center (Tampa, FL) have attempted to better integrate palliative care into long-term treatment discussions with cancer patients using a unique approach: the development of a decision-modeling analytics apparatus based on regret, in which patients choose to continue curative therapy or transition to hospice care based on their perception of regret over making a certain choice.5
A prospective cohort study conducted at Moffitt Cancer Center and Tampa General Hospital, and presented at the 2016 American Society of Hematology Annual Meeting & Exposition, was done to validate the model in a cohort of 178 terminally ill adult patients. The researchers found that the decision model accurately predicted ultimate patient decisions in 72% of cases, with 85% agreeing with the model recommendations for hospice care or continued treatment with a curative intent
(P < .000001).6
Journal of Clinical Pathways spoke with one of the study’s presenters, Benjamin Djulbegovic, MD, PhD, a member of Moffitt’s health outcomes and behavior program and distinguished professor at University of South Florida Morsani College of Medicine (Tampa, FL), about the need for effective entry points to discuss palliative care possibilities with this patient population, as well as the value of emotion-based decision-making tools.
A fair amount of research has suggested that patients with blood cancers face unique difficulties in receiving appropriate palliative care. Why do you think that is?
I would argue that it hinges greatly on a lack of validated tools to help physicians and patients make these decisions. Prognostic tools that provide reliable assessments regarding life expectancy are just not well developed. Many have been developed, but very few have been externally validated, which is an important point to consider. Another common problem is a lack of understanding of how to elicit patient preference in these conversations.
What my colleagues and I argue in our work is that preferences are fundamentally based on emotions: the way we feel influences the way we act. You cannot remove emotion from the decision-making process—especially in the end-of-life setting, which is likely going to be the most difficult situation many of us will find ourselves in. It is a process fraught with emotions, so we should view emotions as a key aspect to guide adequate discussions in this realm.
So this is why you chose to focus on an emotion-based modeling exercise?
Yes. What we are really pointing out here is that patient preferences should reflect the architecture of human cognition. So far, patient preference models have been developed without tapping into this part of the human experience. David Hume7 famously said that without emotions, we can have no goals, and without goals, we cannot adequately make decisions. Modern cognitive science has increasingly shown that human decision-making relies on two systems: one relies on affect and intuition, which is called “Type 1 Processing;” and another relies on deliberate analytical skills, called “Type 2 Processing.” What we argue is that you cannot accurately elicit patient preference without tapping into both of these systems.
Regret is known as a cognitive emotion, one of the rare cognitive processes that links between both of these processing systems. It binds emotion and intuition to logical thinking. So, regret basically serves to ask the question, “What if?” You can feel regret about past choices and future choices. Here, we are asking a patient to anticipate how they will feel in a particular situation in the future if they make a decision now. We ask the patient what they will regret more: choosing to go into hospice, or receiving further treatment that may be unnecessary.
Was regret something you always believed would serve as a good emotion for this type of prognostic tool?
What we actually ended up doing was testing a regret threshold model, which implies that there are some trade-offs we have to make. In the end-of-life setting, a patient faces the decision to cease treatment or continue. The real question we faced was, how do we allow patients to live the remaining days of their lives on their own terms? We needed something to help patients to clarify their choices, and we found that we could do this through the regret process.
In the model, we ask two questions regarding levels of regret. We ask the patient what they would regret more: going into hospice care or continuing treatment. We then employ the threshold model. For example, a patient may be indifferent towards going to hospice or not going to hospice. If, according to the model, the patient learns that the probability of death is above the certain threshold, theoretically the preference would be to go hospice. If it was below the threshold, the likelihood is that the patient would be inclined to continue treatment.
How did you validate the threshold model?
We tested two aspects. First, we tested whether patients agreed with the recommendations. Then we linked the threshold model with a life expectancy prognostic calculator, based on performance scale. Then we asked our regret-based questions about hospice or treatment. Basically, we are linking patient preferences with a predictive model about death and dying. Finally, we asked patients whether they agreed with what the model recommended. We found that more than 85% of patients agreed with the model recommendations, and almost everyone we included found the entire process very helpful.
The question then falls to something known as the normative vs descriptive gap. That is, what people actually should be doing is not what people actually end up doing. So, our next step was to figure out what the actual choices were. People may have told us that they agreed with the model but then went on to change their minds. So we went on to test whether patients followed through with the choices they affirmed, and we found that people followed through with their initial choice 71% of the time. Although lower than the initial overall agreement, this is a pretty good figure. We consider the model both descriptively and prescriptively validated.
Can you talk a bit about the concept of “acceptable regret”?
We developed a theory of acceptable regret that looks at whether someone can make a mistake and not feel regret about it. We wanted to identify what the probability of death would be where a patient would not regret making a decision that turned out to be the wrong decision. Basically, if the probability of death is greater than 96%, a patient would not regret going into hospice, even if that turned out to be the wrong decision. Similarly, a patient with a greater predicted chance of survival would not regret seeking more treatment, even if it turned out to be unnecessary.8 Most doctors will remember [the title of a] famous paper published in The New England Journal of Medicine: “Our stubborn quest for diagnostic certainty.”9 People really require almost absolute certainty before they are comfortable making a decision, without feeling regret. You may make a decision, but you will still feel regret up to a certain point. The concept of acceptable regret centers around a level of comfort with a decision, even if it turns out to be the wrong one.
It has been said that we spend more money on raising the probability of something from 95% to 100% than we do in getting from zero to 95%. We do so much testing to increase the level of certainty. But 100% certainty of diagnosis or death is theoretically impossible. So we have to figure out how much uncertainty we can tolerate without feeling regret.
Is the model being used regularly now at your institution?
Our initial prospective trial was not randomized, so the next phase we would like to test it out in a randomized trial. I personally use it, and some other colleagues do, as well. But ideally, I would like to see a randomized trial done, and I think it would be helpful to integrate it with electronic health records. But even without a randomized trial, I think we have a lot to offer with this model.
References
1. El-Jawahri A, LeBlanc T, VanDusen H, et al. Effect of inpatient palliative care on quality of life 2 weeks after hematopoietic stem cell transplantation: a randomized controlled trial. JAMA. 2016;316(20):2094-2103.
2. May P, Garrido MM, Cassel JB, et al. Palliative care teams’ cost-saving effect is larger for cancer patients with higher numbers of comorbidities. Health Aff (Millwood). 2016;35(1):44-53.
3. Epstein AS, Prigerson HG, O’Reilly EM, Maciejewski PK. Discussions of life expectancy and changes in illness understanding in patients with advanced cancer. J Clin Oncol. 2016;34(20):2398-2403.
4. Odejide OO, Cronin AM, Condron NB, et al. Barriers to quality end-of-life care for patients with blood cancers. J Clin Oncol. 2016;34(26):3126-3132.
5. Djulbegovic B, Tsalatsanis A, Mhaskar R, Hozo I, Miladinovic B, Tuch H. Eliciting regret improves decision-making at the end of life. Eur J Cancer. 2016;68:27-37.
6. Djulbegovic B, Tsalatsanis A, Mhaskar R, Hozo I, Miladinovic B, Tuch H. Improving hospice referral: application of a regret-based decision modeling at end-of-life care. Presented at: American Society of Hematology Annual Meeting & Exposition; December 2-6, 2016; San Diego, CA.
7. Hume D. A Treatise of Human Nature. Oxford, UK: Clarendon. 1978 (1738).
8. Tsalatsanis A, Hozo I, and Djulbegovic B. Acceptable regret model in the end of life setting: Patients require high level of certainty before forgoing management recommendations. Eur J Cancer. 2017; In press.
9. Kassirer JP. Our stubborn quest for diagnostic certainty. N Engl J Med. 1989 320(22):1489-1491.
