Abstract
Background:
Information routinely collected during a palliative care consultation request may help predict the level of complexity of that patient encounter.
Objectives:
We examined whether patient and consultation characteristics, as captured in consultation requests, are associated with the number of unmet palliative care needs that emerge during consultation, as an indicator of complexity.
Design:
We performed a retrospective cohort analysis of palliative care consultations.
Setting:
We analyzed quality-of-care data from specialty palliative care consultations contained in the Quality Data Collection Tool of the Global Palliative Care Quality Alliance from 2012 to 2017.
Measurements:
Using 13 point-of-care assessments of quality of life, symptoms, advance care planning, and prognosis, we created a complexity score ranging from 0 (not complex) to 13 (highest complexity). Using multivariable linear regression, we examined the relationships of consultation setting and patient characteristics with complexity score.
Results:
Patients in our cohort (N = 3121) had an average complexity score of 6.7 (standard deviation = 3.7). Female gender, nonwhite race, and neurological (e.g., dementia) and noncancer primary diagnosis were associated with increased complexity score. The hospital intensive care unit, compared with the general floor, was associated with higher complexity scores. In contrast, outpatient and residence, compared with the general floor, were associated with lower complexity scores.
Conclusion:
Patient, disease, and care setting factors known at the time of specialty palliative care consultation request are associated with level of complexity, and they may inform teams about the right service provisions, including time and expertise, required to meet patient needs.
Introduction
Health systems are increasingly calling on specialty palliative care services to provide population-level interventions that improve patients' quality of life (QoL) and the value of care delivered. 1 One accelerator for the tremendous growth in palliative care services across hospital and community-based settings is a strong value proposition that benefits patients and providers, within the context of recent health care reforms, coupled with the strong and convincing message that palliative care services decrease costs while increasing quality. 2
Calls for further integration of palliative care services into usual care have only got louder as risk-based contracts have taken root in the reimbursement landscape. 3 Health systems and care organizations are now striving to understand their patient populations, the services that will need to be provided, and the potential total costs of that care. Armed with such knowledge, health systems and payers can plan for the provisions of care into the future, striving to close the delta between anticipated and actual costs so that shared savings can be achieved.
However, the services provided by palliative care specialists at the individual clinician and patient levels have not been well described. We recently reported the results of a time-in-motion study, which described the administrative and care processes performed by individual clinicians as part of their usual duties. 4 Qualitative reviews of palliative care clinical trial data have also attempted to determine retroactively or during the conducting of the trial itself about what specific processes are being provided, with the ultimate goal of tying particular care processes (e.g., symptom management, goals-of-care planning, building rapport) with outcomes of interest. 5
Those findings can be complemented by analyses of large, population-based datasets that identify what palliative care specialists are faced with during face-to-face consultations. 6 In particular, it is important to characterize the complexity of consultations, and to examine whether information routinely collected or known at the time of referral may help to predict the level of complexity of upcoming consultation, to facilitate palliative care teams' ability to plan, with respect to the time and expertise needed for the visit.
Using a large, national registry of data collected during specialty palliative care consultations, we examined whether patient and consultation characteristics are associated with the number of unmet palliative care needs that emerge during consultation, as well as other complicating factors, to construct an indicator of consultation complexity.
Methods
Data source and population studied
We performed a retrospective cohort analysis of data collected during specialty palliative care consultations by using the Quality Data Collection Tool (QDACT) of the Global Palliative Care Quality Alliance (GPCQA). This study was approved by the Duke Institutional Review Board (IRB) (Pro0003573) and was performed in accordance with the Data Use Agreement and Registry Agreement of the GPCQA.
Briefly, the GPCQA is a quality measurement and improvement organization comprising academic and community-based specialty nonhospice palliative care organizations that agree to submit data on the quality of their care. The data are submitted to the QDACT registry, housed at Duke University. The QDACT registry includes information from consultation requests, including patient demographic and clinical information, and referral source and location of consultation. During inpatient, outpatient, and home-based consultations, palliative care clinicians complete QoL, symptom and prognostic assessments, in addition to assessing patient engagement in advance care planning using QDACT. Thus, data in the registry represent patient-reported, but clinician-entered information. Patients represented in QDACT are diverse with respect to the diagnosis and stage of illness. A detailed description of the data elements and structure of the data collection tool and underlying registry are provided elsewhere.7,8
Patients were eligible for this study if they were 18 years of age or older and received their initial specialty palliative care consultation between October 23, 2012 and January 21, 2017 (N = 8239). We excluded 5118 patients who were missing at least one of the 13 items on which we based our complexity score (the primary outcome in our analysis) and 1449 patients who were missing consult location (a key covariate in our analysis). Therefore, we analyzed initial palliative care consultations from 3121 unique patients (38% of the 8239 who met our inclusion criteria). We applied statistical methods to adjust for potential selection bias due the large number of otherwise eligible patients whom we could not analyze due to missing data.
Complexity measure
We based our complexity measure on 13 QDACT assessments, 1 involving QoL, 9 addressing physical symptoms (pain, dyspnea, depression, anxiety, appetite, constipation, drowsiness, nausea, and tiredness), and 3 addressing advance care planning (presence of advance directive, code status, and surrogate decision maker on record). QoL and symptoms were assessed by using the Edmonton System Assessment Scale (ESAS).9,10 The ESAS scores range from 0 to 10, with higher scores representing poorer outcomes. QoL scores of 6–10 and symptom scores of 4–10 were considered “more complex” in our study, consistent with established ESAS cut-offs for “moderate” and “severe” severity.9–11 Patients were considered to be “more complex” if they did not have an advance directive, had a code status other than do not resuscitate (DNR), or had not indicated a surrogate decision maker.
After all thirteen outcomes were dichotomized, we weighted each “more complex” item equally with a value of one point; thus, a summed score resulted in a complexity score ranging between 0 and 13. The decision to weight each item equally stems from the National Consensus Project Clinical Practice Guidelines for Quality Palliative Care, which consider unmet needs as uniform in clinical importance. 12 Further, we did not identify conclusive evidence or recommendations that associate specific palliative care assessments or patient needs with different requirements for clinician time. To assess how well our complexity score performed for identifying complex cases, we evaluated the strength of association between the palliative performance scale (PPS) and our complexity score. The PPS is frequently calculated and documented upstream to consultations (e.g., oncologists documenting fitness for therapy) or by palliative care clinicians during specialty visits. We used PPS as the comparator, as lower scores are associated with a need for more interprofessional collaboration and care navigation, due to greater reliance on assistance with activities of daily living, more frequent care transitions, 6 and higher prevalence and intensity of certain refractory symptoms.10,13
Data analysis
Summary statistics for patient characteristics, including complexity score, were tabulated by location of consultation, and they were compared by using Chi-squared or Kruskal–Wallis tests. Our primary analyses were based on an inverse probability weighted linear regression model for the complexity score based on patient, disease, and consultation characteristics. To accomplish this, we first fit a logistic regression model to estimate predicted probabilities of being included in the study based on the following covariates: continuous age, race (white, black, or other), diagnosis (cancer, noncancer, or neurological), and sex. Missing covariates (≤11% for all variables in our dataset) were imputed by using 11 multiple imputation datasets before fitting the logistic regression model. The inverse of the predicted probabilities from this model was then used as a weight in our linear regression models that evaluated the association between complexity score and consultation location (residence, hospital general floor, hospital intensive care unit [ICU], or outpatient), age at consultation, sex, race and primary diagnosis, and PPS. We reported the weighted estimates from these models with 95% confidence intervals (CIs) based on bootstrap re-sampling (Technical Appendix). 14
Results
Characteristics of patients across consultation locations are shown in Table 1. Patients seen in the residence (median age, interquartile range: 82.1, 71.9–88.9 years) or on the hospital general floor (78.3, 68.4–85.9 years) were older than those seen in the hospital ICU (73, 63.9–81.2 years) and outpatient care settings (65.1, 55.0–74.1 years). The hospital ICU saw the largest proportion of patients with poor PPS (74.3%) compared with the general floor (39.0%), residence (16.1%), and outpatients (0%), and outpatient clinics saw the largest proportion of patients with cancer (85.4% compared with 12%–30% at other locations) (p < 0.0001 for all comparisons).
Demographic Features
Kruskal–Wallis test.
Chi-square test.
ICU, intensive care unit; PPS, palliative performance scale; SD, standard deviation.
The mean complexity score in our cohort was 6.7 (standard deviation [SD] = 3.7). As we expected, there was an inverse relationship between the complexity score and PPS, suggesting that more complex patients had lower performance status. After adjustment for age, race, diagnosis, gender, and consult location, we estimated that 38% of the variation in complexity scores was due to PPS (R 2 = 0.38).
Table 2 compares complexity score and the proportion of patients with “needs met” versus “complex” by consultation location. We found that hospital ICU had the largest proportions of patients who were more complex with respect to QoL, pain, dyspnea, depression, anxiety, appetite, constipation, drowsiness, nausea, and tiredness. Patients seen in their residence had the highest proportion of patients with more complexity with respect to advance directive completion (82.1% patients without an advance directive). Patients seen in outpatient clinics had a higher proportion of complexity with respect to other aspects of advance care planning. Specifically, 13.9% had no surrogate decision maker, and 68.8% had a code status other than DNR. Patients admitted into the hospital ICU had the highest average complexity score (mean = 9.0; SD = 3.5).
Outcomes by Consult Location
Chi-square test.
Kruskal–Wallis test.
There were a total of 30 encounters that were missing information on health care proxies.
QoL, quality of life.
Table 3 shows the adjusted associations of consultation location and patient characteristics with complexity score. Complexity scores for patients in the hospital ICU were higher, on average, than for patients seen on the hospital general floor (adjusted mean difference [AMD] = 2.33, 95% CI: 1.98 to 2.77 for ICU). Patients seen in the outpatient or residence were less complex, on average, than patients seen on the hospital general floor (AMD = −0.37, 95% CI: −0.78 to −0.02 for outpatient; AMD = −1.44, 95% CI: −1.63 to −1.02 for residence). For the demographic features, primary diagnosis was also associated with complexity score. Specifically, patients with a neurologic condition were, on average, more complex than patients with cancer (AMD = 3.14, 95% CI: 2.62 to 3.37). Finally complexity scores for female versus male patients were slightly higher on average (AMD = 0.39, 95% CI: 0.03 to 0.51).
Results of Linear Regression Predicting Complexity Score Based on Patient and Consultation Characteristics
Adjusted for age, race, gender, disease status, and location of consultation.
CI, confidence interval; PE, point estimate.
Discussion
In a retrospective analysis of a quality measurement registry, we found that several patient, disease, and care setting factors are associated with more complex consultations. Most notably, we observed high complexity associated with female sex, neurologic diagnoses, (e.g., dementia), and hospital ICU care settings. We also found that the outpatient and residence care settings were associated with complex advance care planning, with large proportions of patients not having a documented proxy, code status addressed, or advance directive compared with hospital care settings.
When planning for palliative care consultative services, advance knowledge of patient, disease, and setting factors that may lead to greater complexity can help for care team planning. With the current workforce shortage,15–17 palliative care teams must strategize about how to best meet the needs of the patients and families while also caring for members of the interdisciplinary team. For example, since complexity scores were highest, on average, in the ICU, teams may consider planning differently when the ICU consultation caseload is high. This may entail planning for increased time per consult, or fewer consults per day in that setting. For outpatient and residence settings, where patients tended to have unmet advance care planning needs, the use of a social worker to help with goal setting and advance directive completion could be an effective strategy. Being aware of these patient factors and care setting characteristics can help palliative care teams work more effectively and efficiently.
Our findings support reports from the field of existing efforts to triage consultation requests as demand for services increases. A recent survey of hospital-based consult teams named “triaging consults” (63%) as the leading strategy for managing workload. 18 Other suggested strategies include creating protocols to redirect consultations to hospice liaisons or pain consultation teams when appropriate, and to create “office hours” for high utilizers of services to receive answers to curbside questions. 19 A notably successful strategy comes from Johns Hopkins Bayview. There, authors implemented a quality improvement project to have one team member review consultation requests, liaise with the referring team, and perform a chart review. Then, based on a standardized protocol, a recommendation was made to include particular interprofessional members in the clinical encounter. In this approach, the authors showed increases in total completed visits, physician interactions in consultations, and work relative value units. 20 We believe our population-level findings complement existing individual, patient-level approaches to match professionals and services to patients with appropriate needs.
Further, with a transition to accountability-based payment programs and increasing focus on population health, understanding upfront the needs of a population of patients with incurable illnesses can better inform the appropriate payment programs that will sustain the services needed. This modeling of unmet needs to proxy how palliative care services can be rendered and tracked by setting is a key step in aligning the value of services that can be measured. The Medicare Care Choices Model 21 is a pilot demonstration focused on aligning a $400 per member/per month additive supportive care service for hospice eligible patients who are not ready for hospice election. Key questions remain regarding how any per member/per month can support the full provision of services required. For example, it is unknown what reimbursement amount would support the necessary expertise and time required to address the distress of an ICU patient with nine or more unmet needs and high complexity.
A better understanding of services provided by palliative care specialists to patients with serious illness can inform more modern and tailored reimbursement programs for such services. Medicare Part B will pay for physician provision of palliative care services, but the existing reimbursement rates that are available and used for such care are based on payments consistent with routine ambulatory care. Palliative care visits typically take longer and are more complex than are routine ambulatory care encounters. We find that patients with a lower PPS score, therefore more limited functioning and poorer prognosis, have significantly more unmet needs as compared with patients with a higher level of functioning at a given encounter. This issue must be addressed to expand access to palliative care outside of hospital settings, where facilities are often willing to cross-subsidize palliative care given the overall business case for hospitals.
Our analyses highlight the inherent challenges to registry-based data collection and aggregation in busy specialties such as consultative palliative care. High missingness of data, particularly in nonphysical aspects of care delivery such as spiritual and social assessments, limit exploring relationships between unmet needs in these domains and outcomes. Gaps in data may stem from a few causes. First, clinicians may prioritize symptom and advance care planning assessments in their clinical practice, or at minimum in their documentation within the registry. Second, clinicians may be reflecting that most consultations requests do not involve focusing on spiritual or social domains of care, and gaps in data reflect necessary consultation etiquette. Lastly, our use of particular instruments to assess these constructs may not align with clinicians' instrument preferences, and thus assessments are occurring but do not fit within the questions of the registry. Future analyses will explore these hypotheses further.
We have important limitations to highlight. Most notably, we created a de novo measure of complexity based on commonly used measures of patient palliative care needs or distress that have not been externally validated. However, we did compare our complexity measure with another, well-accepted measure of complexity that is typically collected upstream of the initial palliative care consultation and saw reasonable correlation between the two. Thus, we feel our study offers some valuable insight into the association between care setting/patient characteristics and complexity. Nonetheless, our approach does not place semantic value on any specific complexity score; that is, we cannot define what a single point increase (or decrease) in complexity score means specifically, other than it reflects a more complex case than a lower score. In addition, we recognize that an evidence base that creates relationships between specific unmet needs (e.g., high pain scores) and clinician time or management complexity is needed. When developed, comparisons between two clinical needs (e.g., high pain scores vs. spiritual distress) can be more accurately made so differential weights can be applied.
Further, our study was retrospective and observational, thus we limit our inference to associations and not causations. This is the inherent nature of a clinician-entered, longitudinal registry. Finally, we acknowledge that unmet need is not the only way to conceptualize complexity, and that our unmeet need cutoffs for assessment scores (e.g., symptom score of 4 and higher) may differ from what others clinically use. Our original intent was to include other relevant outcomes in our complexity measure, especially the presence of a caregiver, being at peace, spiritual concerns, and family burden. We were unfortunately not able to include these factors in our assessment due to nontrivial missingness in the QDACT registry.
Finally, our results are based on a majority white, non-Hispanic population that, although consistent with current expectations regarding palliative care utilization, presents the possibility that there exist additional patterns of complexity in minority populations eligible for palliative care that were not measured in our study. This has important implications related to the broader utility of our complexity score in managing the wide variety of patients who have real need for palliative care. Therefore, as the palliative care community continues to work toward improved awareness and understanding of palliative care among patients and physicians, and to increase access to palliative care for all patients, it will be important to continue re-evaluating the way we triage complex patients and adapt accordingly.
Conclusions
Our data indicate that there are key patient, disease, and consultation setting factors that may clue a palliative care team into budgeting the appropriate amount of time and assigning the appropriate resources. Further, in a reimbursement environment that is rapidly progressing to rewarding for health care expenditures that do not exceed planned costs, organizations will benefit from real-world models that provide clues into patient complexity, and associated time and resource costs, and expected outcomes. We have provided some of the first data to elucidate factors related to complex consultation in the palliative care setting. This information can guide palliative care providers to plan accordingly before consulting with patients.
Footnotes
Authors' Contributions
A.H.K.: owner of Prepped Health LLC, advisory relationships with Compassus Hospice, Astra Zeneca, Jannsen Oncology, New Century Health, United Health care, Pfizer, Heron Therapeutics, and Carevive, Inc. D.K.C., S.W., G.S., J.T., and M.H.: none. J.B.: principal of Four Seasons Consulting Group. Research funding Astra Zeneca, Tabular Rosa. J.M.N.: owner of Prepped Health LLC. D.H.T.: Advisory relationship with Islands Hospice and Vigilant Health. Unpaid Board member of Carolina's Center for Hospice and Palliative Care. Owner of Pork Barrel Research, LLC.
Disclaimer
The content of this article is solely the responsibility of the authors and does not necessarily represent the official views of the Agency for Health care Research and Quality or the U.S. Department of Health and Human Services. Results may or may not be consistent with or confirmed by the findings for the independent evaluation contractor.
Funding Information
Funding for this project was provided by the Agency for Health care Research and Quality (K08 HS023681-A1 PI: A.H.K.) and Grant 1C1CMS331331 from the U.S. Department of Health and Human Services, Centers for Medicare and Medicaid Services.
Author Disclosure Statement
No competing financial interests exist.