All Hospice Patients Are Not Equal: Development of a Visit-Based Acuity Index

Methods

The seven participating hospices in this study are part of the CHOICE network (Coalition of Hospices Organized to Investigate Comparative Effectiveness). CHOICE is a research-focused collaborative of hospices that all use Suncoast Solutions Electronic Health Record (EHR) Software and that have agreed to share their data for research purposes. CHOICE projects are defined and approved by a steering committee comprising leaders from all hospices in the network. Hospices participating in this study range in size from 400 to 2700 patients/day. Hospices are located in New Mexico, California, Florida, Pennsylvania, Wisconsin, and Texas, and one cares for patients in both Kansas and Missouri. All are not for profit.

CHOICE uses data extracted from a proprietary platform (Solutions Intelligence) developed by Suncoast Solutions. This platform creates a data warehouse that hospices use for tracking, quality measurement, and benchmarking. Warehouse data reside on a secure server that is managed by Suncoast Solutions. Data elements for this study were defined a priori and then extracted from the warehouse. Extracted data were then stripped of identifiers to create a Health Insurance Portability and Accountability Act (HIPAA)-compliant limited dataset, and then transferred as an encrypted file to the University of Pennsylvania.

We extracted a dataset containing basic demographic variables (age, gender, race), diagnoses (admitting diagnosis and up to three additional diagnoses), and site of care at the time of enrollment (home, long-term care facility, hospital, hospice inpatient unit). We extracted clinical data elements that were markers of the severity of the illness and the complexity of care (e.g., presence of pain, use of oxygen or Foley catheter, presence of intravenous access). Finally, we extracted Palliative Performance Scale (PPS) scores for each patient. The PPS is an 11-point (0–100) scale in which higher numbers indicate better function across several domains (e.g., ambulation, alertness, oral intake).^3–6 For ease of interpretation, we grouped PPS scores into three categories (0–30, 40–60, 70–100) based on previous studies. ⁵

We excluded two groups of patients. First, we excluded patients who received care in inpatient hospice units. These units are staffed around the clock by multiple disciplines, and therefore represent a unique population in terms of visit frequency. Second, we also excluded patients in nursing homes or assisted living facilities because hospice visits for these patients would be influenced by characteristics of each facility (e.g., policies, staffing ratios) for which we did not have adequate data. Therefore, the final sample used for analysis included only those patients who received their first day of hospice care at home.

For these patients, we assessed workload by calculating the number of visits that each patient received during the first 8 days of hospice care (admission day plus next 7 days). We counted visits by nurses (registered nurses, licensed practical nurses, nurse practitioners), social workers, chaplains, and home health aides. We excluded visits from other disciplines (e.g., physicians, physical therapists), because they proved to be uncommon (<3% of total visits) among home care patients in this sample. We also excluded days spent at the continuous home care level because this level of care was very uncommon (<1% of days). Moreover, because continuous care is reimbursed at a higher rate than routine home care is, and because it requires a minimum nursing presence, a simple count of “visits” would not adequately capture time spent with patients as it does for routine home care.

In this study, we focused on care during the first 8 days. Although it will ultimately be very useful to define patient characteristics that predict visit frequency throughout the course of hospice care, we focused on the first 8 days of care because our primary goal was to predict visit frequency based on patient characteristics that are present at the time of hospice enrollment. We assumed that any associations between characteristics and visits—if such associations exist at all—would be strongest for visits early in the course of hospice care. Although it is possible that patient characteristics present at enrollment predict visit frequency at 3 months, 6 months, or later, such associations are less plausible. Therefore, for this initial study, we have focused on the first week, when associations are likely to be strongest and thus easiest to detect.

Some patients died within several days of hospice enrollment, and others were transferred to an inpatient setting during this period. Still others received at least one day of continuous care at home. Therefore, patients had varying periods of time during which they were at home with routine home care. To account for this variation, we calculated a visit index for each patient by dividing the total number of visits each patient received while on routine home care by the number of days (up to 8) they spent on routine home care.

To identify variables that might be used to predict visit frequency, we considered all variables shown in Table 1 for inclusion in a mixed effects regression model in which hospice was a random effect. We used a process of forward stepwise regression, retaining only those variables that were significant. For ease of interpretation, and because Medicare coverage begins for most patients at age 65, we divided the sample at this age. We decided a priori to retain age, race, and gender in the final model because these variables are likely to be markers of other unmeasured characteristics such as insurance and socioeconomic status. We also decided to retain timing of admission (e.g., weekday versus weekend) because we reasoned that weekend admissions might receive a different frequency of visits during the first 1 to 2 days of care. Once the final model was selected, we used the included variables to calculate adjusted average daily visit frequencies for each value of the remaining variables (e.g., for patients with/without intravenous medications). Finally, we used the regression model described above to predict visit frequency for several hypothetical patients that varied by key characteristics.

We estimated that a sample of at least 35,000 home care patients would provide adequate power (1-β>0.80) to detect a small (0.05) difference in the mean number of visits/day (α=0.05) even for patient characteristics that comprise 3% of the population. Although no clear standards exist for the estimation of power for regression models, a rough rule of thumb is to allow at least 10 observations per cell for each variable under consideration. ⁷ Therefore, we reasoned that this study would have adequate power to detect even small effects of a candidate predictor variable on the average number of visits per day that patients received. The University of Pennsylvania's Institutional Review Board approved the use of secondary data for this study. Stata statistical software (Stata MP2 11.0 for Mac; Stata Corp., College Station, TX) was used for all statistical analysis.

Results

Seven hospices contributed data for a total of 69,764 patients admitted between October 1, 2009 and August 30, 2012. Compared with national reports of the hospice population, the entire sample had a slightly shorter length of stay (median 18 days versus 21 days). Sample and national numbers were similar with respect to gender (55% versus 54% female), the proportion with cancer (37% versus 41%) and age >65 (83% versus 83%), but patients in this sample were more likely to be white (89% versus 80%). The analysis described below is based on those patients (n=35,232) who were admitted to hospice at home.

Patients admitted to hospice at home received a mean of 1.51 visits/day during the first 8 days (day of admission plus 7 subsequent days). Patients received a median of 1.32 visits/day (interquartile range: 0.95–1.93). Most of the average 1.51 visits/day were conducted by nurses (0.95 visits per day; 63.0%), followed by social workers (0.25/day; 16.6%), home health aides (0.19/day; 12.6%), and chaplains (0.12/day; 8.0%).

In bivariate regression analysis, several patient characteristics were associated with visit frequency (average number of visits/day of home care). For instance, patients younger than 65 had more visits (β coefficient −0.12, 95% confidence interval [CI] −0.19 to −0.06; p<0.001), and white patients had fewer visits (β coefficient −0.09, 95% CI −0.17 to −0.01; p=0.022). Patients whose primary caregiver was a spouse also received fewer visits (β coefficient −0.02, 95% CI −0.03 to −0.01; p<0.001). Other bivariate associations are described in Table 1.

In a mixed effects regression model, 14 independent predictors of visit frequency were identified (Table 2). For instance, several demographic characteristics were associated with visit frequency, including age (average adjusted visits age <65: 1.65/day, 95% CI 1.60–1.70 versus age ≥65: 1.39/day, 95% CI 1.34–1.44; p=0.013), gender (male: 1.67/day, 95% CI 1.62–1.72 versus female: 1.39/day, 95% CI 1.35–1.43; p=0.007), and primary caregiver (spouse: 1.36/day, 95% CI 1.31–1.41 versus nonspouse: 1.69/day, 95% CI 1.63–1.75; p=0.004). Increased visit frequency was also independently associated with lower (worse) PPS scores (<40: 1.78 visits/day, 95% CI 1.74–1.82 versus 40–60: 1.65 visits/day, 95% CI 1.61–1.69 versus >60: 1.41 visits/day, 95% CI 1.36–1.47; p<0.001), and with the presence of pain (pain: 1.77 visits/day, 95% CI 1.72–1.82 versus no pain: 1.44 visits/day, 95% CI 1.39–1.59; p<0.001). Patients admitted to home hospice from a hospital also received more visits compared with other patients (hospital: 1.73 visits/day, 95% CI 1.67–1.79 versus home: 1.42 visits/day, 95% CI 1.40–1.44; p<0.001).

Many of the variables in Table 2 that were independently associated with visit frequency are also markers of illness severity. Therefore, it is possible that at least some of the observed associations exist because hospice clinicians recognize that a patient is nearing the end of life, and are providing additional services to anticipate increasing needs at the time of death. To explore this effect, we examined these associations for the subset of patients who began their care at home and who also survived for at least 3 weeks (n=20,042; 56.9%). We chose this time point because it is well beyond the one-week time period in which we measured visit frequency. In this sample, all variables in the original multivariable model remained independently associated with visit frequency, with the exception of the use of oxygen.

Using the regression model described above, and including all patients who began their care at home (n=35,232), we predicted the average visit frequency for four hypothetical patients (Table 3). For instance, the predicted visit frequency was only 1.02 visits/day for an 84-year-old woman with dementia who was admitted to hospice from an outpatient clinic and who received care from her husband (Patient #1). On the other hand, we predicted 2.40 visits/day for a 60-year-old woman with cancer, who was discharged from a hospital with oxygen, intravenous morphine for dyspnea and pain, and a jejunostomy tube for nutritional support, and who was cared for by her daughter (Patient #4).

Discussion

Hospices need to make clinical decisions about workload and assignments to ensure patients receive the visits and care they need. They must also make strategic decisions in ways to ensure they can meet the needs of the populations they serve. This is the first study to help guide these decisions by predicting visit frequencies during the first week of hospice care, and it offers three key results.

First, this study identifies several patient characteristics that are independently associated with average visit frequency during the first week of hospice care. Hospices could assess some of these (e.g., gender, diagnosis, referral site) before a patient is seen and evaluated. These characteristics could be particularly useful in guiding staffing decisions and could be used to assign staff to patients based on the likely number of visits, insuring a more even caseload among hospice staff.

These results also have important implications for operations and strategic planning. In fact, even small differences in visit frequency among patient groups could have a relatively large impact on staffing needs. For instance, a patient with a feeding tube who has an average of 0.35 more visits per day (compared with patients without a feeding tube) would receive 2.45 additional visits in the first week of care alone. For every 100 patients with a feeding tube admitted, that would be 245 additional visits in the first week or (assuming 5 visits per day), an additional 49 days of staff time, all of which must be covered by a fixed per diem payment.

Second, this study defines associations between patient characteristics and visits and offers a novel perspective on the way home hospice care is delivered. On one hand, many of the characteristics that are associated with visit frequency are expected. For instance, the use of intravenous medications and the presence of a pressure ulcer or pain are all indicators of more severe illness, which requires more intensive support.

Other results are less intuitive. For instance, we found a greater visit frequency for men and for patients with nonspouse caregivers. These characteristics may be surrogate measures of a lack of social support, but they might also be markers of higher caregiver burden. For instance, some studies have found higher levels of burden and emotional distress among caregivers of male patients.^8,9 However, other studies have found the reverse to be true. ¹⁰ It is possible, too, that these associations reflect unmeasured differences, particularly if spousal caregivers' burden is more closely associated with patients' physical limitations as opposed to other needs. ¹¹ Regardless of the explanation, this finding suggests that meaningful models to predict hospice visit frequency will need to include a consideration of caregiver factors, as Home Health Resource Group scores do for skilled home nursing care. ¹²

Third, one of the most significant results of this study is the finding that patients admitted to home hospice following hospital discharge have significantly greater visit frequencies during the first week of care. This is particularly relevant to hospices that are entering into relationships with ACOs that create population-based models of care.^13,14 If a hospice makes a commitment to provide services for a given population, it will be essential to understand in advance what that population's needs are likely to be. Specifically, if patients discharged from the hospital to home hospice require more visits, this information could help hospices to better anticipate the costs of relationships with hospitals through ACOs and bundled payments.

This study has three limitations that should be noted. First, it is possible that some of these associations are the product of hospice-specific culture or policies. For instance, one hospice's staff might believe that patients with certain characteristics need more frequent visits. If this is the case, it is possible that some of the associations described here are idiosyncratic.

A second limitation, related to the first, is that these data are derived from only seven hospices. Different patterns might be observed in other hospices with different cultures, staffing, and policies. Therefore, further research is needed to determine whether these associations exist in other hospice populations.

Third, this study only examined predictors of visits. We were not able to determine why patients needed the visits, or even whether these visits were truly necessary. It is possible that visit frequency is driven in part by factors other than clinical need, such as patient or family expectations. Moreover, it is not yet known whether and how visits are associated with quality of care and patient and family outcomes. Therefore, in future research it will be essential to clarify the association between visits and outcomes of care.

Nevertheless, this study is an important first step in determining what factors may influence the allocation of staff resources among hospice patients. Furthermore, these results will lay the foundation for future studies to determine whether the right patients are receiving optimal visit frequencies. Once we understand which patients receive the most visits, we can begin to determine whether those visits are allocated in a way that produces the best outcomes. Similarly, future research will need to explore the relative impact on outcomes of home visits versus other types of contact such as telephone contact, which were not included in the current study, and which may reflect different care need (e.g., emergency triage).

As hospice care continues to evolve, hospices will increasingly enter into partnerships with health care systems, physician practices, and managed care organizations. In developing these relationships, it will be important to better understand how visits translate into improved outcomes overall, and among patient subgroups. It will be essential to ensure that hospices can accurately estimate the needs of the patients they will be serving so that reimbursement structures are adequate to meet those needs.