Assessing End-of-Life Care across Settings in an Integrated Health Care Delivery System

Methods

We explored end-of-life care experiences during the last month of the patient's life with 11 survey questions (see Supplementary Material). A panel of subject matter experts identified topics central to end-of-life care quality: communication with the family, communication with the patient, concordance between patient preferences and care, symptom management, and emotional support. We adopted and developed items to address each topic. Five items were from the FATE and/or BFS surveys, the latter of which is endorsed by the National Quality Forum, ¹⁴ and three were from Kaiser Permanente regional surveys. We created three items to address topics for which we could not locate appropriate items in validated surveys.

Response options were scaled for seven items (“always/usually/sometimes/never” or “a lot/some/only a little/not at all”) and dichotomous (“yes/no”) for three items; these also included “don't know/unsure” as a response option. Assessment of overall quality had only scaled response options (“excellent/very good/good/fair/poor”). Two items assessing concordance with patient preferences included options indicating that they were not applicable because the patient's health status precluded making decisions about care or because the patient's wishes were unknown.

All decedents in five Kaiser Permanente regions in April to June or June to July 2017 were identified. They received care in home health, hospice, inpatient, and outpatient settings and skilled nursing facilities. Using administrative data, a surviving survey recipient was identified for each decedent in decreasing order of preference: primary health care decision maker, primary caregiver, emergency contact, and resident or mail recipient at decedent's last known address. The survey and a cover letter were mailed to identified recipients, with nonresponse follow-up of up to 10 phone call attempts. Members receiving follow-up phone calls could complete the survey by phone.

We characterized decedents whose identified recipients did and did not respond to the survey with descriptive statistics, testing differences with the chi-square test for proportions and the Student's t-test for averages. We also used descriptive statistics to characterize response patterns to survey items, focusing on the top two options for scaled items and “yes” responses for dichotomous items. Although the literature conventionally reports top-box scores, we sought to identify opportunities in quality of end-of-life care across our organization for the first time by distinguishing between positive (top two boxes) and negative responses. We tested the statistical significance of differences in response patterns for patients who did and did not receive palliative care or hospice care with the chi-square test. This quality improvement study did not meet criteria for Institutional Review Board oversight.

Results

Mailed materials were delivered to 10,308 identified surviving recipients; 1921 (19%) returned the mailed survey and 710 (7%) completed the survey by phone, yielding an overall response rate of 2631 (26%). Most patients had died at home (1239, 47%), followed by inpatient settings (656, 25%), skilled nursing facilities or nursing homes (220, 8%), assisted living facilities (163, 6%), and hospice (157, 6%).

Item response patterns were characterized by small proportions of “don't know/unsure” responses (4–6%) and indecipherable or missing responses (4–9%). However, the proportion of “don't know/unsure” was 14% for an item assessing how much the respondent thought the care team tried to accommodate the patient's preferred location of death. The proportions of missing responses were 10%, 11%, and 15%, respectively, for items assessing whether the patient received unwanted care, the patient died in his or her preferred location, and the care team helped the patient and/or family with the things that worried the patient the most.

Respondents and nonrespondents differed. Decedent characteristics associated with higher response rates included receiving hospice care, palliative care, or both, a cancer diagnosis, completing an advanced directive, and age 18–64 or ≥85 years (Table 1).

Among all respondents, 80% reported that overall care in the last month of life was excellent or very good. The proportion of excellent and very good responses within regions ranged from 74% to 84%, with statistically significant differences between the lowest performing region and three others (p < 0.05, data not shown). The proportion of respondents rating overall care in the last month of life as excellent or very good was 69% for patients who received neither hospice nor palliative care and 78–82% for those who received palliative care, hospice care, or both (p < 0.05 for all, data not shown).

The proportion of respondents selecting positive response options varied across items (Table 2). It ranged from 69% (1750) of respondents selecting “yes” for “Do you think that the patient passed away where he or she wanted to?” to 89% (2172) of respondents selecting “a lot” or “some” for “In your opinion, how much did the doctors and other staff do to ease the patient's pain and suffering?” Five items had positive responses lower than 80% (Table 2). They included items assessing concordance between care and preferences (items 8 and 9), emotional support (item 11), and communication (item 2), in addition to the overall rating of care (item 1). Conversely, the proportion of positive responses was 80–89% for five items assessing communication with patients, concordance between patient preferences, discussions, and care received, and symptom management. In response to the single item with reverse wording, just 7% of respondents indicated that patients had received unwanted medication or medical treatments.

Discussion

Using a short survey consisting primarily of validated items from existing instruments, we assessed patient and family experiences of end-of-life care across settings. The reported proportion of positive experiences varied across regions and items, suggesting that our approach identified patient- and family-centered measures that could be used across organizational settings to drive quality improvement in end-of-life care. In addition, the relationship between supportive services (i.e., palliative care, hospice care, or both) and improved patient and family satisfaction is well-documented, ¹⁵ but organization-specific data confirming this association across participating regions pinpoints potential quality gains and can drive the spread of these services.

Strengths of our project include use of a single survey across settings to understand the landscape of end-of-life care at Kaiser Permanente. The use of administrative data to identify respondents for decedents in decreasing order of preference has not been previously reported. We assessed concordance between treatment preferences and goals with multiple items.

Limitations include a 26% response rate despite the association of mixed-mode survey administration with higher response rates on care experience surveys. ¹⁶ We attribute our response rate to identifying survey recipients through administrative data. A Kaiser Permanente region that identified caregivers as bereavement survey recipients in hospice workflows had a 30% response rate, equivalent to that of the CAHPS Hospice survey administered to caregivers of decedents identified in hospice records.^17,18 Our experience suggests that it is feasible to use administrative data to identify proxy respondents for patients' end-of-life care experiences and that the results should be interpreted judiciously. Differences between respondents and nonrespondents could have affected our results, but nonresponse bias for the mailed version of the BFS is minimal. ⁸

We did not adjust for patient or regional characteristics that may have affected our findings. Most respondents were decedents' spouses or children, but we did not analyze response rates by survey recipient type (e.g., primary caregiver vs. emergency contact). The proportion of missing and “don't know” responses was >10% for four items. Proxies for decedents find it more difficult to respond to items about experiences or events they cannot directly observe, such as the care team's attempts to accommodate the patient's preferred location of death. ¹⁹ In addition, patients may not communicate fully with proxies due to decreased levels of consciousness or coping strategies such as denial and concealing symptoms, making it difficult for proxies to respond to some items.^13,19

The survey achieved our aim of understanding care in the last month of life across settings. The organization-specific positive association between end-of-life care experience and palliative or hospice care led to expansion and optimization plans for both programs. Insights from segmentation analyses helped formulate strategies to increase access to palliative and hospice care. One region compared care experience across palliative care delivery settings (i.e., inpatient, outpatient, and home) and between in-network and outside hospice care providers. Another region used survey data to inform and validate strategic planning and implementation and to better understand care for patients at the end of life by assessing service-level and patient-specific circumstances. Survey data and clinical data have been linked to understand relationships between care delivery and care experience. Regular survey administration is under consideration to enable tracking performance over time, a critical capability in quality improvement. In addition, we are also focusing on gathering data from real-time patient-reported outcome measures to better meet patients' and families' immediate needs and drive high-quality end-of-life care.