New MEDPAC study says some past work (mine) overstates hospice savings
July 10, 2015 1 Comment
A MEDPAC contractor report considered the question of “does hospice reduce Medicare costs?” from three perspectives, and concluded that a matched-control approach that only compares costs from the time of hospice initiation to death overstates savings. This is the approach that we used in 2007, and that was replicated (and improved upon in terms of covariates) by Amy Kelley and colleagues in 2013.
The essence of the approach we used is to take hospice decedents, match them to otherwise statistically similar decedents, and then compare costs for the period of hospice use (if used for 10 days, compare only last 10 days of life).
I will blog more about this next week, but a couple of quick thoughts:
- The MEDPAC report notes that the methods we used to pick a control group and that Amy Kelley adapted are not the most appropriate way to address the question “does hospice reduce Medicare spending?” I think the method we used is better than that alternatives, but I want to leave that aside for a second and say that all of the action in a non-experimental study of this type is how you pick the control group. As compared to what? is always the most important policy question.
- In our ongoing work on palliative care, we are planning to use multiple methods of picking a control group, and further plan to think deeply about why the answer to the seemingly simple cost saving question can differ so much by method. There is information in this uncertainty, and non experimental inference is just about the most important thing for policy evaluation that will never be subjected to a randomized trial.
- The answer to the question could differ by market area, as well as across time as an area becomes “fully exposed” to hospice and palliative care concepts. This part of the report (that I have not read carefully yet) looks to be very important. It is hard to think that the early hospice cost assessments that looked at the last year of life–for example–when the median stay was 15 days is the best way to evaluate this question. A market approach might be better.
- I agree with MEDPAC that cost savings is not the most important question to address. Here is a list of many others.
- Non experimental policy evaluation is hard. On January 1, 2016, about halfway through our CMMI project, Medicare will start to finance advance care planning. That means any control group in the last 18 months of our study had ready access to a key palliative care intervention that was not present from the beginning. I am not sure exactly what the answer to all these questions are, but I am confident that more data is better than less.