A recent paper in the Health Services Research journal (“Hospice…“) looks at whether hospice care reduces hospitalizations for elderly terminally ill patients in nursing homes. In the introduction, authors Pedro Gozalo and Susan Miller state that there are two main implications which result from end-of-life hospitalizations:
“At the patient level, hospitalizations of frail NH [nursing home] residents have been shown to include hazards that negatively affect the quality of life (Creditor 1993) and, in many cases, are inappropriate (Saliba et al. 2000). At the policy level, hospitalizations represent the main component of total health care costs, particularly during the last few months of life. In an recent study using both Medicare and Medicaid claims for NH decedents in the state of Florida in 1999, Miller et al. (2004) found hospital expenditures to account on average for 78 percent of all expenditures in the last month of life among those patients that did not receive hospice and 33 percent among NH residents who had any hospice in the last 30 days of life.“
The main problem in determining whether or not hospice care reduces hospitalization is the issue of selection. Individuals who enroll in hospice care may be relatively ‘healthier’ than those who are hospitalized without hospice care. Also, it could be the case that hospice patients are ones that prefer less aggressive treatment methods. Local market idiosyncrasies can also lead to erroneous conclusions.
To take these selection issues into account the authors use an inverse-probability-of-treatment weighting (IPTW) [see Robin, Hernán and Brumback (2000) for more on IPTW methodology].
- Let H=1 if an individual enters hospice care and H=0 otherwise;
- let Y=1 if the individual is hospitalized and Y=0 otherwise.
- Also, let P(W)=P(H=1|W) which is the probability of being hospitalized conditional on W.
- W is a vector of both patient characteristics (X) and a set of hospice provider characteristics (Z).
Using the IPTW methodology, the authors regress Y on H and X and each observation is weighted by the following term:
- H/P(W) + (1-H)/[1-P(W)]
In other words, if the individual enters a hospice, they receive a weighting of P(W)-1, and if the individual does not enter a hospice the observation is weighted by [1-P(W)]-1. One problem with this method is that it assumes that the vector W adequately models the choice of hospice care and that there is no unobservable sorting into hospice compared to non-hospice care. To reduce the endogeneity of the hospice characteristics (Z), the authors wisely decide to use the characteristics of the hospice nearest to the individual, not the actual hospice chosen.
The authors find that the most important determinants of hospice enrollment are: principal diagnosis of cancer and patient preferences for noncurative care. Further, the paper concludes that nursing homes “that choose to contract with hospices may be less likely to hospitalize their residents, even if [the] hospice was not present.” The authors find that one quarter of the hospice effect on hospitalization is due to patient preferences, but the hospice effect on hospitalization is still strong.
- Gozalo; Miller (2007) “Hospice Enrollment and Evaluation of Its Causal Effect on Hospitalization of Dying Nursing Home Patients” Health Services Research 42:2 (April 2007) pp. 587-610.
- Robin, Hernán and Brumback (2000) “Marginal structural models and causal inference in epidemiology” Epidemiology. Sep;11(5): pp. 550-60.