Quality and cost outcomes of an integrated supportive care program

The LINCC project was conducted as a demonstration program in northeast Ohio operating between February 2015 and February 2018. This program was funded in part by the CMMI, with some additional in-kind funding provided by University Hospitals (UH). A longitudinal comparison of the program with both a concurrent control and an historic control was used to evaluate outcomes.

Patient enrollment occurred over two and a half years from one of four sites (one main campus and three community locations) within the UH Seidman network. Concurrent control patients were selected from candidates who were identified as eligible but not enrolled because of staffing constraints. This control sample was used to evaluate the impact of our intervention on patient experience, utilization, and cost outcomes, including end-of-life measures. Since quality of life data were not available across the entire cancer center, we used an historic control population for evaluating the program’s impact on quality of life. The historic control was identified from a prior NCI-funded trial performed at the same institution from 2008 to 2011 (NIH NRO18717).

The study project was deemed to be exempt, as a federal demonstration project, by the institutional review board at University Hospitals Cleveland Medical Center. Thus, written informed consent was not required.

Eligibility criteria included the following: (1) 18 years of age or older; (2) Medicare or Medicaid beneficiary, and (3) advanced or progressed solid tumor malignancy. Patients who were beneficiaries of commercial insurance were ineligible, unless the commercial insurance was secondary.

Eligible patients were identified by a variety of means, such as tumor board reports, appointment schedules, case reviews of the electronic medical record (EMR), and referrals from the primary oncology teams. The identification of potential candidates for the program was performed by an administrative member of the team with clinical experience. Once identified, one of the nurse care coordinators approached the patient, explained the program, and collected baseline data during that initial patient assessment of needs. Patients identified as clinically eligible but who were unable to be enrolled due to program capacity provided the concurrent control group.

The intervention included three primary components: routine electronic biopsychosocial screening, early access to specialty-level palliative care providers, and nurse care coordination. We developed and used an electronic, tablet-based tool for obtaining patient-reported symptoms and concerns. Patients were screened at routine intervals, monthly during scheduled clinic visits or, if they were not being seen monthly, at midpoint between visits to their oncologist. They underwent a monthly, short screen which included the Edmonton Symptom Assessment Scale (ESAS), Distress Thermometer, and Problem Checklist. Upon enrollment and at months 3, 6, 15, and 24, enrollees completed an extended screen which incorporated the monthly screen and additional measures, including the Functional Assessment of Cancer Therapy (FACT-G). The monthly screens were focused on identifying symptoms which warranted immediate attention, while the extended screen was done for the purpose of gathering more comprehensive data on overall well-being. Nurses reviewed the responses and directed clinically actionable concerns to the patient’s providers, including palliative care providers as appropriate.

A second component incorporated in the program was early referrals to specialty-level palliative care in the outpatient setting. The palliative care providers (MD or nurse practitioner) coordinated with the primary oncology team for both symptom management and advance care planning. The program’s original design was for all enrolled patients to see a palliative care provider. In the first year of the program, only about 60% of enrollees participated in an early consultation with palliative care. This rate was a function of supporting patient choice of whether or not to accept referral to palliative care providers, as well as an attempt to refer all patients in the program to early palliative care, even those with low symptom burden. As the program increased capacity, demonstrated capability and value to the primary oncology teams, and improved the enrollment process to focus increasingly on higher-risk patients, 98% of patients enrolled in the last year of the program participated in early palliative care consults.

The final primary component of the program was providing nurse care coordination for these complex patients. Nurses were aligned with patients in the same way that other supportive care team members (e.g., social workers) are aligned with disease teams within the cancer center. This design was to facilitate acceptance of and ease of communication by the project nurses with the oncology teams and increase the care coordinators’ familiarity with treatment protocols.

Nurse care coordinators on the project also served as the nursing support for the palliative care providers. Coordinators were responsible for tracking the biopsychosocial screening schedule of the patients and contacted the patients for assessment either by telephone or in person (if patient was coming to the cancer center). They connected the patient to any intervention or resources which the screening identified as a need. Nurses tracked admissions and emergency department (ED) visits within our hospital system for enrolled patients and facilitated appropriate follow-up. In doing so, they assured that the primary oncology team was notified of the admission or ED visit. If the patient had not yet been seen by a palliative care provider, they facilitated referral to either the inpatient or outpatient team, as appropriate. In addition, the nurse care coordinators provided clinical follow-up for issues generated by the supportive oncology/ palliative care interdisciplinary team.

The program, through its development of new infrastructure for electronic patient biopsychosocial screening, expansion of specialty-level palliative care providers, and use of nurse care coordinators, gradually became integrated into the fabric of the cancer center. The project team members participated in work to sustain and spread the program within the organization. This included multiple educational sessions for staff, a midpoint survey of oncologists to assess their satisfaction with the program and solicit input on how to improve it, and design and implementation of a comprehensive electronic medical record addition to document end-of-life and goals of care discussions.

Quality of life was measured using the Functional Assessment of Cancer Therapy-General (FACT-G) which is a 27-item screen validated in patients with any form of cancer which measures the four primary QOL domains: Physical Well-Being, Social/Family Well-Being, Emotional Well-Being, and Functional Well-Being [14]. The FACT-G asks patients to report upon how they felt in the past 7 days on a five-point Likert scale from “Not at All” to “Very Much.” For our use, the tool was built within a website and either given to patients on a tablet to fill out or read by the nurse coordinators to the patients over the phone and the nurses entered the responses. The FACT-G was the measure of quality of life incorporated as part of the extended screen, collected at baseline and months 3, 6, 15, and 24 of enrollment.

Aggressiveness of care was measured with multiple metrics, categorized into end-of-life measures and active treatment. End-of-life measures included death in the hospital, death in the ICU, ICU admission in the last 30 days of life, hospitalization in the last 30 days of life, emergency department visit in the last 30 days of life, chemotherapy in the last 14 days of life, rates of hospice admission, hospice length of stay, and healthcare costs within the last 30 and 90 days of life. Active treatment measures included normalized rates, on a per member per month basis, of inpatient admissions (emergent, urgent, and total), total patient days, patient days in excess of CMS MSDRG geometric length of stay, emergency department visits, and total cost. These indicators have been commonly used to reflect aggressiveness of care at the end of life [15, 16].

Healthcare utilization data were extracted from an internal financial decision support system providing detailed billing data for claims paid to wholly owned University Hospitals’ entities. This source did not include access to oral pharmaceutical costs and acute healthcare utilization outside of the University Hospitals network. Full claims datasets from CMS for similar populations at UH suggest that the costs outside our wholly owned entities are about 10% of total claims. Hospice utilization data, including admission dates and length of stay, were obtained through a partnership with two hospice programs which combined represented 85% of hospice admissions from our system.

SPSS Version 24.0 was used for all analyses with p < 0.05 used as the criterion for statistical significance. Linear regressions were conducted for all cost analyses. Data met the assumptions for linear regression and no transformations were required. Estimated marginal means were used for comparison. Prior to conducting analyses, outlier cases whose average cost per member per month was > 2.5 standard deviations above the total group mean were removed. This represented 13.2% of total cases (n = 215 LINCC and n = 113 control). Next, covariates were entered in step 1 with the grouping variable (LINCC vs control) entered in step 2. The following covariates were included in the analyses because there were statistically significant differences between groups (see Table 1): patient race, patient Medicaid status, cancer type, cancer stage, treatment status, weighted Hierarchical Condition Categories (HCC) system code, and number of palliative care visits. The HCC risk adjustment methodology is used by CMS to predict healthcare costs, establish Medicare advantage rates, and adjust performance calculations for value-based models such as Accountable Care Organizations (ACOs) [17].

For non-cost-related analyses (hospice, palliative care), there were no cost outliers removed for analytic purposes. Chi-square analyses were employed for categorical analyses, analysis of variance analyses were employed for continuous analyses, and linear regression was used for analyses that employed covariates. In order to examine the association of palliative care with hospice use patterns, palliative care visits were dichotomized into 2 categories: 0–1 visits and > 1 visits. When examining the total number of palliative care visits per patient (which ranged from 0 to 25), the 75th percentile was 1.0 visit, thus guiding the decision to group palliative care visits.

For comparison of quality of life scores over time, using the Functional Assessment of Cancer Therapy-General (FACT-G) form, repeated measures ANOVA was used.

The demographic characteristics of the sample are shown in Table 1. Consistent with the hospital’s population, the total sample was 71% Caucasian. Twenty percent of the sample utilized Medicaid as the only health insurance. Gastrointestinal cancers were the most common diagnosis. One-third of the participants were receiving no cancer-directed therapy during the study period; such patients had likely successfully completed therapy or may have exhausted all cancer-directed treatment options. Although patients in both groups all had advanced cancer, there were significant differences in many variables when comparing patients enrolled in the LINCC project and those not enrolled, and these were controlled through statistical analysis.

Table 2 presents the total average cost of care during the study period. Cost, emergency room visits, and admissions were normalized to “per member per month” in both cohorts. Resource use in the periods prior to death (30 days and 90 days) is shown for patients who died during the study period.

In terms of care at the end of life, almost all measures reflect a lower aggressiveness of care for the LINCC group. Use of IV chemotherapy in the last 2 weeks prior to death was so low as to be clinically insignificant for both groups. The differences in all other metrics except total cost in the last 30 days were statistically significant.

The use of hospice among patients who died was examined in two ways (Table 3): comparing patients in the control group with those who received LINCC services and comparing those who received 0 or 1 palliative care visit with those who received more than 1 visit. While the length of stay in hospice was not different, a significantly greater proportion of each group (LINCC: 36.6% vs control: 25.4%) were enrolled in hospice prior to death. However, the receipt of formal palliative consultation visits was not associated with either referral to hospice or length of stay in hospice.

Finally, we examined quality of life measures over time (T1 = enrollment, T2 = 3 months; T3 = 6 months) using total scores from the FACT-G. Because the control group had no data measuring quality of life, we compared LINCC patients with an historical group from an earlier study of advanced cancer patients. That study enrolled patients with newly diagnosed advanced gastrointestinal, lung, and gynecologic cancers and provided supportive services from a team of advanced practice nurses (APNs), social workers, and a spiritual care counselor; although the APNs focused on symptom management, palliative care referrals were not provided.

In order to control for variations associated with time since diagnosis and type of cancer, we included only LINCC patients with GI, lung, and gynecologic cancers who had been diagnosed within the previous 6 months. While this reduced the sample size for LINCC to only 26, there was a significant difference in the pattern of change over time (see Fig. 1). While the LINCC patients had a slightly lower total FACT-G score at enrollment, they had a marked improvement by time 2, which was significantly different than the comparison study group and this difference was maintained at time 3.