Context Clinicians have observed various patterns of functional decline at the end of life, but few empirical data have tested these patterns in large populations. Objective To determine if functional decline differs among 4 types of illness trajectories: sudden death, cancer death, death from organ failure, and frailty. Design, Setting, and Participants Cohort analysis of data from 4 US regions in the prospective, longitudinal Established Populations for Epidemiologic Studies of the Elderly (EPESE) study. Of the 14456 participants aged 65 years or older who provided interviews at baseline (1981-1987), 4871 died during the first 6 years of follow-up; 4190 (86%) of these provided interviews within 1 year before dying. These decedents were evenly distributed in 12 cohorts based on the number of months between the final interview and death. Main Outcome Measures Self-or proxy-reported physical function (performance of 7 activities of daily living [ADLs]) within 1 year prior to death; predicted ADL dependency prior to death. Results Mean function declined across the 12 cohorts, simulating individual decline in the final year of life. Sudden death decedents were highly functional even in the last month before death (mean [95% confidence interval {CI}] numbers of ADL dependencies: 0.69 [0.19-1.19] at 12 months before death vs 1.22 [0.59-1.85] at the final month of life, P=.20); cancer decedents were highly functional early in their final year but markedly more disabled 3 months prior to death (0.77 [0.30-1.24] vs 4.09 [3.37-4.81], PϽ.001); organ failure decedents experienced a fluctuating pattern of decline, with substantially poorer function during the last 3 months before death (2.10 [1.49-2.70] vs 3.66 [2.94-4.38], PϽ.001); and frail decedents were relatively more disabled in the final year and especially dependent during the last month (2.92 [2.24-3.60] vs 5.84 [5.33-6.35], PϽ.001). After controlling for age, sex, race, education, marital status, interval between final interview and death, and other demographic differences, frail decedents were more than 8 times more likely than sudden death decedents to be ADL dependent (OR, 8.32 [95% CI, 6.46-10.73); cancer decedents, one and a half times more likely (OR, 1.57 [95% CI, 1.25-1.96]); and organ failure decedents, 3 times more likely (OR, 3.00 [95% CI, 2.39-3.77]). Conclusions Trajectories of functional decline at the end of life are quite variable. Differentiating among expected trajectories and related needs would help shape tailored strategies and better programs of care prior to death.

This paper profiles Medicare beneficiaries' costs for care in the last year of life. About one-quarter of Medicare outlays are for the last year of life, unchanged from twenty years ago. Costs reflect care for multiple severe illnesses typically present near death. Thirty-eight percent of beneficiaries have some nursing home stay in the year of their death; hospice is now used by half of Medicare cancer decedents and 19 percent of Medicare decedents overall. African Americans have much higher end-of-life costs than others have, an unexpected finding in light of their generally lower health care spending.

The surprise question is a simple, feasible, and effective tool to identify patients with cancer who have a greatly increased risk of 1-year mortality.

9588 Background: In patients with advanced cancer, failure to accurately estimate and communicate prognoses can lead to overly aggressive care at the end of life with less attention to important palliative care issues such as pain and symptom management and patients’ values and goals for care. The “surprise” question—would I be surprised if this patient died in the next year?—has been recognized as an innovation to improve end-of-life care in the primary care population by identifying patients with a poor prognosis who are appropriate for palliative care. It has not been previously tested in cancer patients. The purposes of this study were to determine the feasibility and outcomes of the use of the “surprise” question in a cancer center population. Methods: Between July and November 2007, oncologists prospectively classified consecutive breast, lung, and colon cancer patients being seen at the Mary Babb Randolph Cancer Center of West Virginia University into “Yes” and “No” groups based on the surprise question. Patients were followed and their status at the end of one year-alive or dead-was determined along with patient demographics, type of cancer, and stage at presentation. A multivariate Cox proportional hazards regression analysis was used to identify variables associated with patient death. Results: Oncologists classified 826 of 853 prospective patients (97%), with 131 (16%) classified into the “No” group and 695 (84%) into the “Yes” group. At the end of the year, 71 patients had died; 41% of the “No” patients compared to 3% of the “Yes” patients (P <.001). The surprise question ‘No‘ response was more predictive of patient death than stage of cancer, cancer type, or age (hazard ratio 7.53, P value <.001). The “Yes” patients lived longer than the “No” patients (359.8±32.8 days versus 270±131.5 days, P <.001). The sensitivity of the surprise question “No” response was 75% and the specificity was 90%. Conclusions: We conclude that the surprise question is a simple, feasible, and effective tool to identify cancer patients with the worst prognoses who should receive the highest priority for palliative care interventions, particularly advance care planning. No significant financial relationships to disclose.