Impact of hospice on costs for patients in Nursing Homes
May 11, 2015 Leave a comment
A new study by Pedro Gozalo and colleagues from Brown finds that Medicare beneficiaries who die in nursing homes have less intensive treatment in the last year of life, but the overall cost to Medicare of their care in the last year of their life, is higher.
Some key bullet points and issues:
- What did they find? Hospice was associated with less hospitalization and fewer ICU days during the last year of life, but higher overall costs incurred by the Medicare program (around $6,700 more).
- What didn’t they study? The impact of hospice on quality of life of the patients (at least directly; you could normatively infer benefit from less ICU use and hospital days; they discuss this), or families. However, they were using Medicare claims that don’t measure such variables.
- What is new about the study? They use a clever, three part counterfactual strategy to identify the impact of hospice on costs. A key issue in observational assessments of the impact of hospice on cost is that persons who choose hospice differ from those who do not. Past work (our study from 2007; Amy Kelley’s from 2013) has used cross-sectional matching using observed variables to identify the most relevant comparisons to hospice users in order to isolate the hospice impact. Gozalo et al. use traditional observed covariate matching as well (part 1) but within a difference-in-difference approach (part 2) that takes advantage of secular trends in hospice use from 2004 to 2009 (much more likely later) to isolate the impact of hospice on Medicare costs. In short, there are decedents in 2004 who did not use hospice who would have done so under 2009 rates, and vice versa. This allows for isolating the impact of hospice on costs in 2009 on persons who were “new” users due to increasing use rates as compared to those who would have used hospice even without an increase. The same comparison is done in 2004, and the difference in the difference provides the estimated impact of hospice on costs. Finally, because all the subjects lived in Nursing Homes (NH), some variables like DNR or do not hospitalize orders that are unobserved but important are available for matching (part 3).
- Who did they study? All subjects lived in Nursing Homes. This is a key contextual variable, and one that is missed if you say “this study shows hospice increases Medicare costs” because it is found to increase Medicare costs only among persons who died in NH. Most persons receive hospice in their homes, but the subgroup of those receiving it in NHs may be key from a cost standpoint.
- Around 6 in 10 subjects had dementia, the disease associated with the longest period of hospice use in this as well as past studies. This is largely due to the unpredictability of death (as compared to other primary diagnoses) and does not invalidate the study by any stretch, but it helps to highlight the study patients. The hospice benefit has had a per diem payment that was not modified by diagnosis since its inception in 1983. The general trend toward increasing use of hospice, which has meant an increase in dementia as a primary diagnosis, along with a related increase in hospice length of use for dementia that has been widely noted, really signals that a disease specific hospice benefit may be needed if cost is a concern.
- Last year of life costs are an easy concept to grasp, but may not be the best way to assess the impact on hospice on Medicare costs. The limitations of cross-sectional propensity score matching such as what I and colleagues used in a 2007 paper deserve to be noted (Amy Kelley et al added covariates, but has the same general problem), but the focus on measuring cost of cases (hospice) and controls (not hospice) from the point of hospice initiation until death strikes me as a more precise means of estimating the impact of hospice (regardless of how the counterfactual used to identify cases and controls is constructed). Further, the savings identified in past work accrues primarily in the last few weeks prior to death, and it seems conceptually suspect to infer that something used for 90 days should reduce cost over the last 365 days of life. It is a fair point to say that in our paper as well as Amy Kelley’s, all cases of hospice do not match to “statistically good” controls but Medicare has to pay for the care for all patients, but the same can be said for the new paper as well. Pedro and colleagues rightly note that the last year of life is a commonly used time frame, but that doesn’t do away with this conceptual issue. It is an interesting conceptual discussion about the appropriate dependent variable for such a study: full last year of life costs or costs from the date of hospice initiation until death. To push it ahead, I would just like to see more analysis. Pedro and colleagues could update their estimates using the approach that we used and that Amy Kelley etc followed–measuring savings (or not) from the onset of hospice to death and assess their cost findings.
This is an important paper with a very clever estimation strategy, that deserves to be discussed, technically and conceptually (especially my last point above), as well as for what it might mean in terms of policy adjustments. I will have some thoughts on this in a follow up post.