Wednesday, January 4, 2017

How to Incorporate the Results of a New Trial of Palliative Care in Lung and GI Cancers

by: Eric Widera (@ewidera)

A little over 6 years ago, Jennifer Temel and colleagues published a trial of patients with newly diagnosed metastatic non–small-cell lung cancer showing that early integrated palliative care significantly improved quality of life and other measures like symptom control.   Now we have a follow-up study published this week evaluating early integrated palliative care for patients with newly diagnosed incurable lung or non-colon GI cancers.

A brief overview of the study

350 patients with newly diagnosed incurable cancers were randomized to early palliative care integrated with oncology care compared with usual oncology care.    The patients had to be within 8 weeks of a diagnosis of incurable lung or noncolorectal GI (pancreatic, esophageal, gastric, or hepatobiliary) cancer.   They also couldn’t have a history of therapy for metastatic disease, and had to have a reliatively good functional status (ECOG 0-2).

Those who got palliative care in addition to usual care met with a member of the outpatient palliative care team (a physician or advance practice nurse) within 4 weeks of enrollment and at least once per month until death (or via telephone when an in-person visit was not possible).  For patients who were admitted to their hospital, the inpatient PC team observed them throughout their hospitalization.

The Results

The first interesting result is that there was a fair amount of palliative care going on in the usual group, as the palliative care team meet with 34% of the usual care patients by week 24 weeks.    So it is possible that the effect of the palliative care is greater than what was described below, as many in the control group go the study intervention.

The second interesting finding is that the group randomized to palliative care reported an increase in quality of life scores at 12 weeks compared with usual care patients who reported a decrease.   This difference became statistically significant at 24 weeks.

Other measures such as PHQ-9, HADS-Depression, and HADS-Anxiety scores did't differ significantly between study groups.   However, more intervention patients reported that they had discussed their end-of-life wishes with their oncologist compared with usual care patients (30% vs 15%).

Lastly, an exploratory subgroup analysis showed that there may be a difference differences in study outcomes between patients with lung and GI cancers.   However, take it with a grain of salt, given that there was a lot of cross-over to palliative care and it was not a primary outcome of the study.

The Take-away

It's been a very productive year in regards to palliative care randomized control trials, including these that we have covered on GeriPal:

Notably, the first article in this group was not just a negative trial, but one that showed that maybe the group that got the intervention had greater PTSD symptoms.  All of these trials, both positive and negative, have merit as they help to define how and to whom we should be delivering palliative care.  Putting them together, the picture starts to add weight to an argument on how palliative care should be provided (lifted from an editorial I wrote along with Preeti Malani that discussed the negative ICU trial):
"...this study provides a cautionary tale that care delivered by specialists may not be effective when following a highly structured study protocol that solely aims to provide informational support with little regard to what makes this process meaningful to many patients, families, and clinicians: frequent and longitudinal follow-up, close involvement with the primary clinical team, and a focus on relief of physical and psychosocial distress."


Dio Kavalieratos said...

Eric - thanks for this summary. If I may, I’d like to touch on two points.

First, I’m a bit surprised to see that the Carson et al ICU trial is still being considered a “palliative care” intervention — especially by us within the field. Many (including folks who were involved in the trial) have expressed that they do/did not view the study to be a palliative care intervention, but rather a communication intervention. For example, in our systematic review of palliative care interventions (, we purposefully excluded the trial as it did not meet our (admittedly liberal) definition of "palliative care" (i.e. at least two of the National Consensus Project’s eight domains).

Second, I'd suggest we all be more intentional about the language we use to summarize the outcomes of a trial. Particularly, I’m concerned by the unintended consequences that arise when we confuse a null finding (i.e., no effect) and a negative finding (i.e., harm). For example, on its primary outcomes (anxiety and depression), the Carson ICU trial was null, not negative. Yet we all do this — if a study doesn’t yield the direction we want, we call it a “negative” study when in fact, it’s often null. I’d contend that we (especially in palliative care) should be welcoming of a null finding, and not mislabel it as one that has harm associated with it. In our review, we were very sensitive of this distinction, yet some incorrectly lamented the “negative trials” of palliative care, when in fact (especially if we exclude Carson et al), there were no truly negative trials of palliative care. As the evidence base for palliative care matures to focus on studies that dig into when, why, and how palliative care works, null findings will be critical to our understanding of how to optimally deliver and disseminate palliative care.

Dan Matlock said...


Thanks for this!

A few thoughts:

1) Jen Temel gets it done! I had the opportunity to meet her recently when she was a visiting professor to our institution and she is awesome! This is another helpful trial - this work she does is so so hard.

2) Dio's comment above is nuanced and very smart. His recent review was also extremely helpful.

3) In regard to your puzzle (love the picture), we need to push ourselves to define "why" we think aspects of palliative care work. The pall care community responded to the null finding (and the negative press that followed) by saying that this isn't palliative care but rather that palliative care is all about relationships. OK, we'll then let's be explicit about that before we start the trial and articulate the mechanism behind why relationships are helpful. I've also heard people argue that palliative care helps with coping - okay, the health psychology literature is full of coping interventions - should we learn those explicitly? A local leader in health psychology recently told me "they'd never let me say I was just 'doing psychology stuff,' how does palliative care get away with it" (paraphrased). In palliative care, I don't think we can just say that we're doing "palliative care" without pushing ourselves to articulate why. That will help that puzzle come together much much faster.

Eric Widera said...

Great discussion.

Dio. First I'd just like to say that I loved your review paper in JAMA. Well done!

I do have to also question the definitions of "null findings (i.e., no effect) and a negative finding (i.e., harm)." Doesn't null just mean rejection of the hypothesis, and negative mean "negative support for the hypothesis" (see here Neither to me implies a result of zero or nothing, as both refer to the hypothesis. With that said, it's been a while since I took a class on this... Maybe I can use "not significant" but that also may be inappropriately taken as "not being significant as it was a negative trial", rather than it didn't reach statistical significance.


PS. Dan - agree fully with #'s 1-3

Eric Widera said...

Also, I thought the Carson study wasn't included because of timing issue as it was published in July. Now that I look at the inclusion criteria for your study I see my error. I wonder if you can describe a little bit on how much of a domain needed to be covered by an intervention to be deemed a "palliative care" intervention. One could make an argument that the trial meets some of the preferred practices in different domains (#10,11 in domain 1, #18 in domain 4, and maybe some in domain 8 (#33) based on a quick read of the study. It very clearly doesn't meet most of the other preferred practices though.


FYI - The Carson et al main written objectives of of their meetings included the following:
1) Determine the family’s understanding of the patient’s illness, prognosis and treatments
2) Enhance the family’s understanding of chronic critical illness
3) Discuss potential burdens and benefits of continuing intensive care treatment
4) Explore relevant values of the patient and family
5) Elicit treatment preferences that the patient may have expressed
6)  Align family expectations with clinicians ‘ expectations
7) Integrate information previously received from multiple caregivers
8) Discuss expected care needs for the longer term, in light of the patient’s cognitive and functional status and level of dependence on medical and nursing interventions
9) Contribute other information and support as needed by the family for establishing goals of care with the ICU physician

Dio Kavalieratos said...

Eric – thanks for the compliment, and thanks for the Editorial!

Re: null vs negative. We’re wading into nuance –in the strictest sense, “null” is a type of hypothesis. One can reject the null hypothesis, but one doesn’t “accept” the alternate hypothesis. Though, in the work we do, I’ve yet to see a study designed where the null hypothesis wasn’t specified as either no association or no difference; therefore, a “null” study would imply support for the null hypothesis, whereas a negative study would imply an outcome in an unfavorable direction. My argument is a little broader than just the hypothesis, but rather the implications of the study. In that sense, the valence of the words “positive,” “negative,” and “null” allow for nuance between two (generally speaking) two instances where you’d fail to reject the null hypothesis (i.e., a study where there was no effect, and a study where there was an unsavory or harmful outcome) In addition, I worry a bit about the lay interpretation of a “negative” study. In the end, I totally agree with you that regardless of whatever terminology we use, we should strive to be more explicit about what we mean by the term. Perhaps an even better way of discussing studies would be to say something like “there was no statistical support for the hypothesis that [intervention] is associated with [outcome].” Any additional nerdy research semantics chat tonight and I’ll need a beer. :-)

Re: excluding Carson et al. Really good question (and honestly one of the thorniest parts of the project)! For each study, two reviewers independently combed the trials (including protocol papers and entries) to determine how many NCP domains were covered in the intervention. We didn’t use the NQF’s preferred practices, just the NCP domains. Each trial had to include at least two NCP domains, but defining a threshold for whether a domain was met or not was challenging, particularly because many of the NCP domains overlap. In the end, we developed some decision rules to try and distill each domain to one or two distinct foci – for example, Structure and Processes of Care (#1) and Social Aspects of Care (#4) are pretty collinear. But again, this was a challenge. Ultimately, we were confident that the double-blind nature of the review process (which yielded full consensus) was sufficient in categorizing the interventions. We excluded the Carson study because our reviewers concluded that its processes did not uniquely satisfy more than one domain. That said, with any such semi-qualitative process, others may have come up with different conclusions. One of the follow-up analyses I’m hoping to do (as a nights/weekends project) would be to conduct some sensitivity analyses of our results based on tightening or loosening the inclusion criteria. Always looking for extra help!