Sunday, March 22, 2015
These two concerns motivated me, as a researcher with extensive training in health services research and using claims data and former long-time hospice medical director, to write this post:
1. I have heard three pretty prominent speakers – two at large meetings with big audiences – state that the evidence says that hospice improves survival, and have also seen this stated in key papers and reports. They put this on a par with the Temel RCT (and even Jennifer Temel emphasizes that in her study – survival was a post hoc outcome and shouldn’t be overemphasized) and usually cite the 2007 Connor SR et al study in JPSM.1 (Note: that study is already cited by 40 other papers).
2. One of our Maryland snowbirds recently told me about a perceived poor palliative care experience she had while out of state. There were several issues, but one was that she was told – or at least this is what she heard – that seeing palliative care and not getting chemotherapy would help her live longer, and she was not happy with the way this was communicated. I’ve heard the same thing for palliative care providers talking to patients about hospice care. I have concerns that perhaps people are citing this study (and using it with patients) without having read and evaluated it, and many may not be familiar with the study design, data source and limitations.
This post provides a short critique in journal club form of the Connor study and some relevant literature, including some results from a recent study we published with better-quality data that found the opposite result as a secondary finding.
What did the study find?
Connor et al. found an absolute difference of 8 days longer survival in the hospice compared to the nonhospice group, which was statistically significant (but I think most would say, not clinically significant). Although the regression analysis lengthened this to 29 days, this is a challenging finding and it is always concerning when the regression radically changes the results… especially given all the limitations of this study. Since the average length of stay in hospice is approximately 21 days, and 2 of the cancers (lung and pancreatic) have much shorter lengths of stay, a 29-day improvement in survival with hospice care seems hard to believe. In subgroup analyses, they did find significant differences for 3 of six cancers – CHF, lung, and pancreatic cancer – but not 3 others.
Who were the researchers?
This study was conducted by actuaries – not by PhD health services researchers or statisticians. There were nurses but no physicians or oncologists involved.
Who funded the study?
A hospice industry group – the National Hospice and Palliative Care Organization (NHPCO).
What was the study design?
Basically, studies like this are retrospective case series. They identify people who died and then look backwards. This is the lowest study design possible on the totem pole where the randomized controlled trial is the gold standard. We often use this study design in palliative care for pilot projects where there are not other ways to easily identify palliative care populations…but for major studies and those looking at outcomes as important as survival, I agree with Peter Bach’s JAMA perspective that studies identifying patients who died and then following them retrospectively to evaluate associations of hospice care with survival are retrospective case series and poor study designs and should not be used as high-quality evidence.2
What was the data source?
It is important to note that this is Medicare claims or billing data - which has many, many limitations – for example, you cannot tell small cell from non-small lung cancer, or whether the patient has metastatic disease, and the heart failure codes are notoriously unreliable.
What were some limitations?
There are many issues with this study, as the authors state… For example, they used claims to identify different time points at which patients could “shortly thereafter be advised to consider obtaining hospice care.” For breast cancer, this was switching chemotherapies within one-two quarters of the initial chemotherapy. For colon cancer, this was the initial chemotherapy treatment date if there was no resection. Clearly there are many limitations to these indications and this method, which I don’t think I’ve ever seen used elsewhere. This gives about a year’s worth of data – eg, 337 days in the overall sample – so they’re comparing survival for a period of a year, whereas hospice is usually only for a few days or weeks. It’s also important to note that the data used in this study is 15 years old – and much has changed in the treatment of pancreatic and lung cancer, for example, when there used to be few options.
For those of you who are researchers, there are also a lot of other findings that are strange… again, their regression lengthened the survival difference by four-fold from 8 to 29 days. Whenever I see a finding like this from one of our students that doesn’t make sense, it almost always indicates that there are problems with the way the statistics were run or the regression analysis. The survival increase for CHF was almost 80 days with hospice, which seems particularly improbable. Also, very few CHF patients had any hospice care, and this study does little to try to account for the many differences between patients who would and would not be enrolled in hospice. There is at least one other study that has similar findings to Connor et al, but used many of the same flawed methods.
There was a JAMA study comparing costs of hospice vs nonhospice patients for poor-prognosis cancers published in 2014, which still has many, many limitations but I believe has much better methods in some areas, particularly (1) focusing on poor-prognosis cancers, (2) comparing patients only for the period they were (or for matched patients, would have been) in hospice care –not the whole last year of life, and (3) controlling for other factors that might be different between the hospice and nonhospice groups, including comorbidity, much more rigorously.3
Importantly as well – the Connor et al. study does not account for many patient characteristics, such as preferences. Prospective studies and those with patient-reported data often have very different findings. In a recent study with prospectively collected data including patient-reported information and medical record reviews (not just billing claims data), we included survival time as one covariate in an analysis for hospice use as an outcome in patients with Stage IV lung and colorectal cancer (this is how we would usually analyze hospice in a health services research study – as an outcome). We found that longer survival time was associated with less hospice use in lung and colorectal cancer patients when adjusting for patient preferences for care and accurate understanding of illness (survival time from diagnosis, months, OR 0.99 for any hospice use, p<0.001 for patients living less than 13 months compared to greater than or equal to 13 months). So, controlling for patient preferences and using prospectively collected data with much better-quality cancer data, including cancer stage, we found the opposite result.4
So what is the takeaway message?
I believe that providers should not be telling patients that they should enroll in hospice because it improves survival, and speakers should not be promoting this study’s findings alongside the Temel RCT, at least without reading the study carefully and making their own assessment. The Connor et al. study is very poor-quality evidence and I would state that such retrospective claims studies to look at utilization or costs may be appropriate in small pilot studies, but evaluation of survival should be reserved for high-quality, randomized controlled trials with actual patient data, which are unlikely to happen for hospice care. Regardless, the goal of hospice care is comfort and quality of life, and not to improve or decrease survival. I believe that this should be the goal we strive for and what we should tell patients when informing them about their hospice option.
I appreciate any comments!
Sydney Dy, MD, MSc is a palliative care physician and researcher in the Departments of Health Policy & Management, Oncology and Medicine, and Duffey Pain and Palliative Care Program, Johns Hopkins.
Photo Credit: Antique Clock by Rachel Towne via Flickr Creative Commons License
1. Connor SR, Pyenson B, Fitch K, Spence C, Iwasaki K. Comparing hospice and nonhospice patient survival among patients who die within a three-year window. J Pain Symptom Manage. 2007;33(3):238-246. (Available open access via NHPCO)
2. Bach PB, Schrag D, Begg CB. Resurrecting treatment histories of dead patients: A study design that should be laid to rest. JAMA. 2004;292(22):2765-2770.
3. Obermeyer Z, Makar M, Abujaber S, Dominici F, Block S, Cutler DM. Association between the medicare hospice benefit and health care utilization and costs for patients with poor-prognosis cancer. JAMA. 2014;312(18):1888-1896. (Open Access via PubMed Central)
4. Mack JW, Walling A, Dy S, et al. Patient beliefs that chemotherapy may be curative and care received at the end of life among patients with metastatic lung and colorectal cancer. Cancer. 2015.