Sunday, March 10, 2019

Decision-Making in the ICU - The Problem Is Us

by Drew Rosielle (@drosielle)

Annals of Internal Medicine has published a fascinating trial of a web-based surrogate decision-making tool aimed at improving decisions for patients receiving prolonged mechanical ventilation in an ICU.

It's one of the most fascinating trials I've read in a long time, and also somewhat of a monster (in size/scope of data presented) - there are 4, lengthy online supplements attached to it (!), which is daunting, and so I'm mostly just going to write here about what I find most interesting about it.

Which is that I think this may be the trial which should convince us all that what's 'wrong' with ICU decision-making is not a lack of sharing good information -- what is wrong is our entire approach to surrogate decision making.

First the trial, briefly summarized. This was a randomized trial of a personalized, web-based decision aid for surrogate decision makers of patients with prolonged mechanical ventilation. It took place in several ( mostly academic medical center) ICUs in the US. The subjects were primary or secondary surrogates of actual ICU patients (~280 patients in total), all of whom had been mechanically ventilated for 10 days at least, and weren’t anticipated to be immediately extubatable.

The main intervention was that the surrogates received and were asked to interact with a personalized, web-based decision aid tool, which is a fascinating thing that I hope to be able to see in action one day (they have many screen shots in the supplement, but because the tool is connected to real patients there’s no way to access it publicly)

The tool involves giving the surrogates prognostic information about their loved one (based on a prognostic model) which gave an estimation of their loved one’s chance of 1 year survival; educational information about prolonged ventilation, the role of surrogates, and different care goals (eg life prolongation at any cost, maximizing comfort regardless of longevity, etc), and asked the surrogates questions about the patient’s values/preferences. Then the aid presented to the surrogate a ‘goals of care recommendation’, based on those reported values/preferences. The surrogates had an opportunity to disagree with what was presented, and indicate what they thought the actual care goals should be.

(If you’re curious, the prognostic model was probably pretty accurate. The model predicted 1 year survival for all subjects on average to be ~50%, and the actual 6 month mortality--that’s as far out as they have data--was 40%.)

The ‘output’ of the decision-aid was given to the treating clinicians, who were encouraged to incorporate it in the family meeting which was scheduled for day 2 after enrollment for all patients. Control patient surrogates received no decision aid or other information, and just received 'usual care', apart from that everyone was supposed to have a family conference on day 2 post enrollment (and nearly everyone did).

They measured many, many things here, but the major outcomes they were looking at were clinician-surrogate concordance about prognosis, surrogate understanding of prognosis, surrogate well-being both acute and long-term (anxiety, trauma symptoms, etc), patient outcomes like LOS, in-hospital and long-term mortality, etc.

The simple way to summarize their findings is that the intervention had nearly no measurable effect on anything: most importantly on what happened to the patients (eg, ICU and hospital LOS, ICU and hospital mortality, long-term mortality, % who received tracheostomies, % who had discontinuation of life-prolonging technologies, etc), as well as what happened to the surrogates (anxiety and trauma symptoms, satisfaction with care and communication, etc). One of the numerous pre-specified secondary outcomes (changes in a decisional-conflict scale), was slightly better in the intervention group - by 0.4 points on a 5 point scale. Note that the family meetings which were part of the protocol for everyone were recorded and similar content was discussed at these meetings, regardless of randomized group.

The very, very interesting findings were, at least to me, the ‘concordance’ and prognosis findings (if you want to look at this yourself, the best place to go is Online Supplement 3, by the way).

Concordance (meaning surrogate and clinician ‘agreement’ on prognosis which here was measured as chance of 1 year survival) was not improved by the intervention. Fundamentally, that seems to be because surrogate’s assessment of prognosis was essentially impervious to information about prognosis as either presented by the prognostic model in the decision aid or a clinician in the family meeting.

Notably, the physician’s prognostic estimations were quite accurate on average (I don’t know if they had access to the prognostic model prognosis or not). For the entire study the the physicians and model both predicted a ~50% 1 year survival for the patient group on average. The physicians also did a pretty good job of estimating what the surrogates thought the prognosis to be (which was around 70% 1 year survival on average). The surrogates who had the decision aid did a little better at articulating what they thought the physicians’ prognostic estimation was (they guessed around 58% 1 year survival, when it was actually around 49%). Control surrogates, who did not get the decision aid, did a worse job of articulating the physicians’ estimation (they guessed ~67%). However, regardless of what group they were in, the surrogates in both groups articulated a prognosis of around 70% 1 year survival. Ie, the intervention didn’t improve the surrogates’ own prognostic estimation, even though they knew it was significantly different than the physicians’ estimate of survival.



Along these lines, they were also able to show that even after answering questions about what the patient’s values would be, and then being told by the decision aid some sort of conclusion (eg your loved ones care goals fit best with eg, balancing longevity with quality), a large number of surrogates actively adjusted that conclusion (‘disagreed’ with that conclusion so to speak, although the surrogates themselves were the one who gave the answers to the questions about patient values), and almost all of them who did that adjusted the care goals towards being more aggressive (see the figure - this was about 40% of the group).



To summarize: surrogates substantially overestimated patient’s survival, even when presented with the decision aid, and even when they understood that the physician thought otherwise. Many surrogates also disagreed with the goals of care as summarized by their own statements of their understanding of patients’ own values, and wiped that all away to state that the plan of care should be aggressively prolonging life.

There is this idea that what we need to do is somehow say the right words to these terrified, grieving, desperate families, and once we figure out the right words, spoken by the right person/presented in the right way (like a decision aid which focuses on clearly stating prognosis and prompting surrogates to reflect on the patient's values/preferences) that will lead the surrogate/s to make the 'right' decision. 'Right' decision meaning, I guess, the one purportedly the patient would have made ('substituted judgment'), acknowledging that that is typically unknowable. The problem with this model of surrogate decision making is that it does not at all seem to be what surrogates actually do, and data from this study basically show even when surrogates know things about a patient's values and preferences, many of them are unwilling to 'enact' those. Indeed, a substantial number of the surrogates in this study seem impervious to 1) objective data-based prognostic information, 2) accurate prognostic information shared by the patient's treating physician, and 3) acknowledgement of the patient's own values and preferences when they make decisions. And, in fact, it appears that a substantial number of surrogates don't really make any decision at all, and you wonder if they even perceive there to be any decision to be made, because the only 'decision' they are going to endorse is 'do everything.’ I’m really grateful for this study for really showing us that this approach may not be helpful.

And my question to us, collectively, as a medical community, is: so what? Do we think this is a problem? And if so, why? Why is anyone surprised or perturbed that fearful grieving families don't make the same decision that an ethical robot would? The only reason we think it's a problem is because there's been this collective decision the last several decades that, in fact, the right people to be making decisions about what technologies to be applying to a dying body are not the highly trained health care professionals, most of whom went to school/trained for over a decade (and sometimes much longer) just so they can know their way around an ICU, but the shocked, grieving, fearful families, who hear us demand from them, again and again, and in ways that make them feel that they have their loved one’s life in their hands, What do you want us to do? I ask this because if you listen closely to your colleagues, and even palliative people sometimes, you get a sense from some of them is that they believe the problem here is the grieving families, who are ‘in denial’, or ‘unrealistic’, or ‘just don’t trust us’. And, fundamentally, I think that’s the wrong take here.

We’re the problem, not the families.

It also bears noting that in the entire history of our species up until the mid 20th Century, virtually no one had to make end of life decisions for their loved ones, because there wasn't anything that could even be done. Now, it's not at all uncommon for us to have conversations with people about highly technical decisions about where tubes go (or don't go) into their dying relatives' bodies. No one should be surprised it often doesn't go well, and that it's a 'problem' that can't be fixed with communicating data more artfully.

I get it; this is complicated.

There are, for instance, a certain number of patient/families for whom this very nice idea I described above, the happy version of how surrogate decision making is supposed to work, does seem to ‘work.’ Honestly, I’m not too worried about those families - we just need to get them good, realistic information, and they’ll make patient-centered decisions. I also get how unrealistic (and unwise) it is to think we are going to go back to the authoritarian-paternalism days, where an individual physician's personal judgment, subject to all its biases and blind spots, is the unassailable measure of a good decision.

The only way out of this is forward, although I don't myself here have any specific 'policy' proposal to fix this, and I worry all the focus on patient choice (which originally was meant to be a way to empower patients in saying "No" to things, not a way of us in medicine of giving up our role in decision making) just makes things worse, by amplifying this idea that families get to choose medical treatments off a menu at the end-of-life, when there is nowhere else in medicine where we really do that. All the things we are taught to do in palliative care - focus the conversation on the patient, emphasize long-term health and functional outcomes and discuss care goals in that context as opposed to focusing on technical questions, careful, interprofessional attention to the grief and loss the family are experiencing, and actively making recommendations about what we think the best plan is (based on what we know of the patient) - help, undoubtedly, to an extent. But it’s not enough.

For more Pallimed posts about journal article reviews.
For more Pallimed posts by Drew click here.

Drew Rosielle, MD is a palliative care physician at the University of Minnesota Health in Minnesota. He founded Pallimed in 2005. You can occasionally find him on Twitter at @drosielle.

References

1 Cox CE, White DB, Hough CL et al. "Effects of a Personalized Web-Based Decision Aid for Surrogate Decision Makers of Patients With Prolonged Mechanical Ventilation: A Randomized Clinical Trial" Ann Int Med. 2019. Vol 170, 285-297.

Pallimed | Blogger Template adapted from Mash2 by Bloggermint