AJMC

Survival and Cost-Effectiveness of Hospice Care for Metastatic Melanoma Patients | Page 2

Published Online: May 20, 2014
Jinhai Huo, PhD, MD, MPH; David R. Lairson, PhD; Xianglin L. Du, MD, PhD; Wenyaw Chan, PhD; Thomas A. Buchholz, MD; and B. Ashleigh Guadagnolo, MD, MPH
To minimize potential selection bias, we used propensity score–based 1:N match (1 case matched with N controls) in the survival and cost models. Since a 3-group propensity score–matching algorithm is not available, and survival for patients with no hospice care was similar to that of patients who used 1 to 3 days hospice, we combined these 2 groups into 0 to 3 days of hospice use and further matched with patients who used 4 or more days of hospice care by applying a propensity score–based 1:N match algorithm developed by Parsons.13 In this algorithm, all the demographic variables were included in the propensity score logistic model to generate the predicted probability that is used for matching. To maximize the sample size from a 5-matching scenario (1:N, N is 1 to 5), we used a 1:5 match-optimized cohort by using an 8-to-1-digit matching algorithm.13 In the matched cohort, a Cox proportional hazards model stratified by matched pair evaluated the associations between 4 or more days of hospice care or 0 to 3 days of hospice care and overall survival time in months.

To conduct the economic analysis, we divided the total cost of care after diagnosis into 3 phases based on the phase-of-care approach developed by Riley and colleagues.14 The majority of resources are typically consumed in the initial phase, when a patient’s disease is diagnosed and treated, and during the final (end-of-life) phase, when extensive efforts are employed to extend the patient’s life or to improve quality of life. Thus, the costs calculated from this method would follow a U-shaped pattern, with the highest costs on the 2 end points. In our study, the initial phase, which lasts an average of 3 months, was defined as the period during which medical intervention was implemented for advanced melanoma and might include the times of diagnosis, surgery, chemotherapy, and radiation therapy. The end-of-life phase is defined as the last 3 months immediately preceding death. The interim months of continuing care after the initial phase include surveillance and routine therapy costs.

We calculated the cost difference by comparing the total Medicare payments incurred by patients receiving 4 or more days of hospice care with those incurred by patients not receiving hospice care prior to death and those patients receiving 1 to 3 days of hospice care. The total cost of care for patients was calculated as the sum of reimbursements authorized by Medicare. Medicare claims reimbursements were adjusted for inflation to 2009 dollars using the Prospective Pricing Index for Part A claims and the Medicare Economic Index for Part B claims.15 Costs were adjusted for geographic variation using the geographic adjustment factor for Part A claims and the geographic practice cost index for Part B claims.15 These adjusting factors are acquired from direct communication with the National Cancer Institute’s Health Services and Economics Branch of the Applied Research Program. These indices were matched via the state and county codes for each patient and then multiplied with the costs from each file in the database. Since the median survival time for metastatic melanoma patients is less than 1 year, discounting was not applied to cost or survival time. Costs were further analyzed in a generalized linear model with a gamma distribution controlling for patient demographic and clinical covariates.16

The cost-effectiveness analysis utilized the mean of costs from all 3 phases of cancer care and survival. The incremental cost-effectiveness ratio (ICER) = (C1 - C2) / (E1 - E2) = ΔC / ΔE, where Cx equals cost of group x and Ex is effectiveness at group x, with the quotient representing cost per life-year gained. In the cost-effectiveness model, a bootstrap simulation analysis was implemented to assess the statistical uncertainty. We performed an analysis with 1000 bootstrap estimates of the ICER in both the original cohort and the 1:5 matched cohort. Statistical analysis was conducted using SAS version 9.3 (SAS Institute, Inc, Cary, North Carolina).

RESULTS

Patient and Tumor Characteristics


Characteristics of the entire cohort and matched cohort as well as univariate analysis of hospice use and patient characteristics are shown in Table 1. Of 862 patients, 225 (26%) had no hospice care after diagnosis, 523 (61%) had 1 to 3 days of hospice care, and 114 (13%) had 4 or more days of hospice care. All covariates were evenly balanced in the matched cohort.

Overall Survival

At the end of the 60-month study period, the unadjusted survival curves for the entire cohort categorized by hospice use are shown in Figure 1A. The median survival time was 6.1 months for patients who did not enroll in hospice, 6.5 months for patients who enrolled in hospice for 1 to 3 days, and 10.2 months for patients who enrolled in hospice for 4 or more days. The survival curves for the propensity score–matched cohort after combining the groups of patients with no hospice use or only 1 to 3 days of hospice use are shown in Figure 1B. The overall survival rates at all-time points for the patients enrolling in 4 or more days of hospice care were significantly better than those for the comparison group (log-rank test, P <.001). In Cox proportional hazards models, 4 or more days of hospice care was associated with an improvement in survival when adjusting for other characteristics (Table 2). The estimated improvements in survival for 4 or more days of hospice use were similar in the original-cohort Cox proportional hazards model (HR, 0.63; 95% CI, 0.52-0.77, P <.0001) and propensity score–matched model (HR, 0.66; 95% CI, 0.54-0.81, P <.0001). Patients enrolled in 4 or more days of hospice care had 3.9 months longer median survival time in the unmatched cohort model (P <.0001), and 3.3 months longer median survival time in the propensity score–matched cohort model (P <.0001). The findings were similar across various models and cohorts, suggesting that the overall association between 4 or more days of hospice use and reduced mortality was not affected by statistical modeling methods.

Cost Analysis

PDF is available on the last page.

Issue: May 2014
More on AJMC.COM