JOURNAL OF PALLIATIVE MEDICINE Volume 18, Number 11, 2015 ª Mary Ann Liebert, Inc. DOI: 10.1089/jpm.2015.0138

The Impact of Inpatient Palliative Care Consultations on 30-Day Hospital Readmissions Nina R. O’Connor, MD,1 Mary E. Moyer, RN,1 Maryam Behta, PharmD,2 and David J. Casarett, MD, MA1

Abstract

Background: Inpatient palliative care consultations have been shown to reduce acute care utilization by reducing length of stay, but less is known about their impact on subsequent costs including hospital readmissions. Objective: The study’s objective was to examine the impact of inpatient palliative care consultations on 30-day hospital readmissions to a large urban academic medical center. Methods: The hospital’s electronic medical record system was used to identify all live discharges between August 2013 and November 2014. After adjusting for a propensity score, readmission rates were compared between palliative care and usual care groups. Results: Of the 34,541 hospitalizations included in the study, 1430 (4.1%) involved a palliative care consult. After adjusting for the propensity score, patients seen by palliative care had a lower 30-day readmission rate— adjusted odds ratio (AOR) 0.66, 0.55–0.78; p < 0.001. Adjusted rates were 10.3% (95% confidence interval [CI] 8.9%–12.0%) for palliative care and 15.0% (95% CI 14.4%–15.4%) for usual care. Among all palliative care patients, consultations that involved goals of care discussions were associated with a lower readmission rate (AOR 0.36, 0.27–0.48; p < 0.001), but consultations involving symptom management were not (AOR 1.05, 0.82–1.35; p = 0.684). Conclusions: Palliative care palliative care consultations facilitate goals discussions, which in turn are associated with reduced rates of 30-day readmissions.

Introduction

H

ospitals are under increasing pressure to reduce costs by preventing hospital readmissions. Thirty-day readmissions are also included in publicly reported measures of health care quality. Successful strategies to reduce readmissions generally focus on patients at increased risk for readmission, involve multiple diagnostic categories, and utilize a team-based approach.1,2 Growing evidence has demonstrated that palliative care can decrease acute care utilization and inpatient costs. For instance, inpatient palliative care consultations have been shown to lower pharmacy, radiology, laboratory, and intensive care unit costs.3–5 Cost savings are greatest among patients who die in the hospital and increase with earlier palliative care consultation.6,7 Inpatient palliative care consultations might also reduce subsequent acute care utilization by engaging patients, families, and health care providers to align treatment plans

with patient goals. Much less is known, however, about the impact of palliative care on health care utilization after discharge. Previous studies have reported low 30-day readmission rates of approximately 10% following inpatient palliative care consultation,8,9 but two comparative studies found no impact on 30-day readmissions compared to control groups.10,11 One study found that palliative care consultations were associated with a lower rate of 30-day readmissions, but that consultation service included an embedded hospice liaison and readmissions were only reduced among patients referred to hospice.12 It is not known whether palliative care alone can reduce 30-day readmissions. Nor is it known whether palliative care consults have a greater effect on readmissions for some patient populations than others. As health systems face increasing pressure to reduce costs, the answers to these questions can be used to guide the deployment of palliative care resources to reduce readmissions. Therefore, the goal of this study was to examine the impact of

1 Department of Medicine, 2Program for Clinical Effectiveness and Quality Improvement, Perelman School of Medicine, University of Pennsylvania, Philadelphia, Pennsylvania. Accepted July 7, 2015.

956

PALLIATIVE CARE AND 30-DAY READMISSIONS

inpatient palliative care consultation on subsequent 30-day readmissions in a large academic medical center. Methods

This study was conducted at a large urban tertiary care medical center. We initially included all live discharges spanning 16 months, from August 1, 2013 to November 30, 2014. We then excluded neonates and patients discharged from obstetrics, ambulatory surgery, and rehabilitation services, because these groups are very unlikely to receive a palliative care consult. At this hospital the palliative care service includes physicians, nurse practitioners, social workers, a chaplain, a pharmacist, and a triage nurse. Primary inpatient teams place consults electronically for one or more of the following reasons: pain, symptom management, goals of care, end-of-life issues, psychosocial distress, and spiritual distress. Approximately two-thirds of consults involve goals of care or end-of-life issues. One third of consults originate in an intensive care unit, and the remainder comes from both medical and surgical floors. Data extracted from the hospital’s electronic medical record system included demographic variables (e.g., age, ethnicity, gender) and severity of illness (coded as minor, moderate, major, or severe). Severity of illness is a commonly used grouping method determined by All Patient Refined Diagnosis Related Group software from 3M (3M APR DRG software, 3M Health Information Systems, Salt Lake City, UT). At the time of discharge, each patient receives a severity of illness category that is calculated from the diagnoses and procedures that were coded for billing during the hospitalization.13 We also included responses to a series of six questions that nurses enter into the electronic medical record during the admission process: ‘‘What is the patient’s cognitive status?’’ (cognitively intact or cognitively impaired); ‘‘How does the patient rate their health at the time of assessment?’’ (excellent, good, fair or poor); ‘‘Describe the patient’s ability to walk’’ (no restrictions, minor restrictions or changes, walks with the help of equipment, requires assistance from another person, or unable to take any steps at all); ‘‘During the past month, has the patient often been bothered by feeling down, depressed, or hopeless?’’ (yes or no); ‘‘During the past month, has the patient often been bothered by little interest or pleasure in doing things?’’ (yes or no); ‘‘How often is a person available to care for the patient?’’ (never, infrequently, occasionally, often, whenever needed). These questions are associated with a patient’s subsequent need for post-acute care services,14 and readmissions may be reduced when these questions are used to electronically alert discharge planners to potential post-acute care needs.15,16 Patients were then classified into two groups by the presence or absence of a palliative care consult order in the electronic medical record. For outcomes, we counted all readmissions to the same hospital that occurred within 30 days, regardless of cause. It can be difficult to determine reliably whether a readmission is related to the index admission. Additionally, many payers use all-cause readmissions for quality evaluation and penalty programs. Analysis

A mixed effects logistic regression model was used to compare characteristics of palliative care and nonpalliative

957

care patients. In this model, a unique patient medical record number was included as a random effect to account for the appearance of some patients several times in the dataset because of multiple hospitalizations. Analysis was performed at the level of hospital discharge rather than patient, because one patient may have been hospitalized several times, with different characteristics and outcomes. The use of a mixed effects model, with the individual as a random effect, allows analysis of the impact of palliative care on these different admissions. We then created a propensity score to account for nonrandom assignment to palliative care consultation. Propensity scores offer a way to create two groups that are balanced with respect to key patient characteristics that are specified a priori. This approach is similar to that of multivariable regression analysis, but offers several advantages.17 First, as in a randomized controlled trial, propensity score adjustment allows the separation of a study’s design (e.g., balancing of groups) from its analysis. Second, propensity score analysis makes it possible to determine when little or no overlap exists in the characteristics of two groups, which in turn suggests that any comparison is likely to be problematic. To create a propensity score, we first used theory-based logistic regression models to examine bivariate associations of patient characteristics with palliative care consultations. Models used mixed-effects logistic regression, and casewise deletion of missing data. We considered variables for inclusion if they reached a moderate level of significance in bivariate analysis ( p < 0.25).18 When two or more models offered similar c-statistics, we chose the model with the best Bayes Information Criterion.19 We used the resulting model to calculate a propensity score that reflects the probability that a patient would receive a palliative care consult during an index hospitalization. We then compared palliative care and usual care patients with respect to the variables in Table 1 across deciles of propensity score. We added additional variables to the propensity model and recalculated the score in order to achieve the smallest possible differences between the two groups.20 Propensity scores may be used to account for nonrandom assignment in at least three ways. For instance, they may be used to create two matched groups from within a larger population (e.g., a matched group of patients receiving palliative versus usual care). Propensity scores can also be used to weight observations according to patient characteristics.17 Finally, propensity scores can be used to adjust for differences in the two groups.21 There is still active debate about which technique is superior, particularly for complex analyses in the setting of small sample sizes.22,23 For this study, we used propensity score adjustment, because there is no consensus around the appropriate use of mixed-effects regression in matching or weighting.21 In addition, matching would have decreased sample size available for analysis to only a fraction of the total, reducing its power to detect associations and limiting the potential for subgroup analysis.24 Finally, weighting is susceptible to bias in small samples like this, particularly when some patients have very high or low scores.22 Therefore, we examined the impact of palliative care consultation on 30-day readmissions after adjusting for the propensity score in a mixed-effects model, with a patient as a random effect. We first analyzed the overall effect, calculating

958

O’CONNOR ET AL.

Table 1. Comparison of Discharges by Palliative Care Consultation Status

Characteristic Sex Female Male Race White Black Asian American Indian Unknown Missing Age: mean Severity of illness Minor Moderate Major Extreme Missing

Palliative care not consulted

Palliative care consulted

Odds ratio (95% confidence interval)

15,511 (46.9%) 17,600 (53.2%)

736 (51.5%) 694 (48.5%)

– 0.83 (0.75–0.95)

– 0.005

– 0.492

19,979 (60.3%) 9843 (29.7%) 825 (2.49%) 24 (0.1%) 2391 (7.2%) 49 (0.2%)

925 (64.70%) 340 (23.80%) 43 (3.01%) 0 (0%) 121 (8.50%) 1 (0.10%)

– 0.75 (0.66–0.85) 1.13 (0.82–1.54) – – –

– < 0.001 0.459 – – –

– 0.384 0.329 – – –

60.0

1.01 (1.01–1.01)

< 0.001

0.598

28 (2.0%) – 247 (17.30%) 5.03 (3.40–7.44) 629 (44.0%) 15.70 (10.7–23.0) 507 (35.50%) 40.50 (27.6–59.4) 19 (1.33%) –

– < 0.001 < 0.001 < 0.001 –

– 0.272 0.183 0.113 –

9.1 (8.0–10.4)

< 0.001

0.292

– 0.41 (0.34–0.49) –

– < 0.001 –

– 0.689 –

– (1.19–2.55) (1.05–1.86) (1.33–2.12) (1.15–1.77) –

– 0.004 0.020 < 0.001 0.001 –

– 0.847 0.687 0.336 0.547 –

2.23 (1.93–2.58) 2.20 (1.86–2.60)

< 0.001 < 0.001

0.372 0.212

– (0.71–1.16) (0.46–0.71) (0.44–0.67) (0.19–0.29) –

– 0.432 < 0.001 < 0.001 < 0.001 –

– 0.432 0.384 0.204 0.736 –

– 0.55 (0.49–0.64) 0.21 (0.18–0.24) 0.10 (0.07–0.15) –

– < 0.001 < 0.001 < 0.001

– 0.230 0.462 0.538 –

– (1.17–1.80) (1.32–1.98) (1.49–2.27) (1.42–2.29) (1.53–2.46) –

– 0.001 < 0.001 < 0.001 < 0.001 < 0.001 –

– 0.472 0.237 0.221 0.564 0.463 –

56.8 7060 (21.3%) 12,382 (37.4%) 10,125 (30.6%) 3156 (9.53%) 388 (1.14%)

Admitted to an oncology service 3924 (11.9%) 786 (55.0%) Cognitive status Cognitively impaired 2064 (6.23%) 200 (14.00%) Cognitively intact 28,931 (87.40%) 1145 (80.10%) Missing 2116 (6.39%) 85 (5.94%) How often does the patient have a caregiver available? Never 3136 (9.47%) 94 (6.57%) Infrequently 747 (2.26%) 39 (2.73%) Occasionally 2480 (7.49%) 104 (7.27%) Often 6426 (19.40%) 324 (22.70%) Whenever needed 17,763 (53.70%) 760 (53.20%) Missing 2558 (7.73%) 109 (7.62%) Depressions screening Feeling depressed in last month 4357 (13.2%) 352 (24.6%) Lack of interest or pleasure 2915 (8.8%) 245 (17.1%) Ambulatory status Unable to take any steps 1188 (3.59%) 136 (9.51%) Walks with assistance from another person 1398 (4.22%) 145 (10.1%) Walks with the help of equipment 3388 (10.20%) 221 (15.5%) Minor restrictions or changes 4562 (13.80%) 284 (19.9%) No restrictions 20,311 (61.30%) 548 (38.3%) Missing 2262 (6.83%) 96 (6.71%) Patient self-assessment of health Poor 3526 (10.70%) 384 (26.90%) Fair 9649 (29.10%) 584 (40.80%) Good 14,519 (43.90%) 330 (23.10%) Excellent 3045 (9.20%) 34 (2.38%) Missing 2370 (7.16%) 98 (6.85%) Number of comorbidities 0 5220 (15.80%) 142 (9.93%) 1 5585 (16.90%) 221 (15.50%) 2 6593 (19.90%) 290 (20.30%) 3 5095 (15.40%) 255 (17.80%) 4 2869 (8.66%) 141 (9.86%) 5 or more 5211 (15.70%) 275 (19.20%) Missing 2538 (7.67%) 106 (7.41%) a

Adjusted for propensity score.

1.74 1.40 1.68 1.42

0.91 0.57 0.54 0.24

1.45 1.62 1.84 1.81 1.94

Adjusted P value P Valuea

PALLIATIVE CARE AND 30-DAY READMISSIONS

Table 2. Variables Included in Propensity Score for Palliative Care Consultation Variable associated with palliative care consult

Odds ratio (95% confidence interval)

P-value

Male sex Age Admission to an oncology service Admission occurred within 30 days of a previous discharge Illness severitya Intact cognitive status Positive depression screen Self-assessment of healthb Ambulatory statusc

0.72 (0.64–0.81) 1.00 (1.00–1.00) 6.97 (6.16–7.90)

< 0.001 0.976 < 0.001

5.84 (5.23–6.28)

< 0.001

2.35 0.73 1.35 0.68 0.81

< 0.001 0.001 0.001 < 0.001 < 0.001

(2.17–2.54) (0.61–0.88) (1.12–1.62) (0.63–0.74) (0.77–0.84)

a Greater illness severity was associated with increased likelihood of consult. b Better self-assessment of health was associated with decreased likelihood of consult. c Increasing independence with ambulation was associated with decreased likelihood of consult.

adjusted readmission rates and 95% CIs. Finally, we looked at readmission rates among several patient subgroups that we hypothesized a priori would be associated with 30-day readmission rates. The University of Pennsylvania’s institutional review board granted an exemption for this secondary use of existing data. Stata statistical software version 11.0 (StataCorp, College Station, TX) was used for all statistical analysis. Results

Between August 2013 and November 2014 there were 42,689 discharges. Of these, 1212 (2.8%) were deaths; and 6936 (16.7%) were patients discharged from obstetrics, neonatal care, physical medicine and rehabilitation, and ambulatory surgery. These deaths and discharges were excluded, leaving 34,541 discharges in the study sample. Of these discharges, 20,904 patients (60.5%) were white, 18,294 (60.0%) were male, and 4710 (13.3%) were admitted to an oncology service. Of the 34,541 discharges included in the study, 1430 (4.1%) involved a palliative care consult sometime during the hospitalization. On average, patients were discharged 1.4 times in the study period, or approximately one time per

959

calendar year, so this sample represents 23,524 unique patients. Of these patients, 1112 (4.7%) were seen once by palliative care, and 318 (1.4%) were seen twice or more. Subsequent analysis uses discharge as the unit of analysis, with mixed-effects logistic regression to account for repeat discharges by the same patient. Patient characteristics are described in Table 1. Patients who received palliative care consultation were less likely to be male (OR 0.83, 95% CI 0.75–0.95) and less likely to be black (OR 0.75, 95% CI 0.66–0.85). Palliative care patients were much more likely to be admitted to an oncology service (OR 9.10, 95% CI 8.0–10.4) and had higher severity of illness. Palliative care patients were less likely to be cognitively intact (OR 0.41, 95% CI 0.34–0.49) and more likely to self-report depression in the previous month (OR 2.23, 95% CI 1.93–2.58). Palliative care patients required more assistance with ambulation, had worse self-reported health, and had a greater number of comorbid medical conditions (see Table 1). Table 2 describes the results of the mixed-effects logistic regression used to establish the propensity score. The final propensity score included nine variables, summarized in Table 2. These included sex, illness severity, admission to oncology, and whether the index admission occurred within 30 days of a prior discharge. The final model also included ratings of several characteristics that are associated with readmission risk: cognitive impairment, depression, selfrating of health, and difficulty walking, based on a nurse’s assessment at the time of admission. Age was not a significant predictor of palliative care utilization but was added to balance the two groups. The area under the curve for the final model was 0.863. This indicates good ability to discriminate between patients who did and didn’t receive a palliative care consult but is low enough to permit some overlap between groups. For all live discharges in the sample (34,541), the raw unadjusted 30-day readmission rate was 16.0% (n = 5,541). For hospitalizations that involved palliative care, this rate was significantly higher than for usual care patients (325/ 1430, 22.7% versus 5216/33,111, 15.8%; OR 1.57, 95% CI 1.35–1.82; p < 0.001) (see Table 3). However, after adjusting for the propensity score, patients seen by palliative care had a lower 30-day readmission rate (AOR 0.66, 95% CI 0.55– 0.78; p < 0.001). Adjusted readmission rates were 10.3% (95% CI 8.9%–12.0%) for palliative care, and 15.0% (95% CI 14.4%–15.4%) for usual care. To understand how palliative care consultations might reduce readmissions, we examined subgroups of patients according to their use of post-acute services. Of the 34,541 patients in the sample, 850 (2.46%) were discharged to

Table 3. Results by Palliative Care Consultation Status Outcome Readmitted within 30 days Referred to hospice Discharged with DNR order a

With palliative care consultation 22.7% (325/1430) 29.6% (423/1430) 39.8% (569/1430)

Adjusted for propensity score.

Without palliative care consultation 15.8% (5216/33,111) 1.3% (427/33,111) 4.1% (1355/33,111)

Unadjusted odds ratio (95% CI) 1.57 (1.35–1.82, p < 0.001) 32.15 (22.72–37.29, p < 0.001) 15.49 (13.74–17.44, p < 0.001)

Adjusted odds ratio (95% CI)a 0.66 (0.55–0.78, p < 0.001) 16.32 (13.23–20.14, p < 0.001) 8.44 (4.02–12.67, p < 0.001)

960

inpatient or home hospice, 13,759 (39.83%) were discharged to home care, 15,080 (43.66%) were discharged to home without services, and 4852 (14.05%) were discharged to an institution (e.g., skilled nursing facility, long-term acute care hospital). Palliative care patients were much more likely to be referred to hospice at the time of discharge (unadjusted analysis: 423/1430, 29.58% versus 427/33,111, 1.29%; OR 32.15, 95% CI 27.72–37.29; p < 0.001) (see Table 3). After adjusting for the propensity score in a mixed-effects model with patient as a random effect, palliative care patients still had higher hospice utilization at discharge (AOR 16.32, 95% CI 13.23–20.14; p < 0.001). Palliative care patients had an adjusted hospice referral rate of 14.9% (95% CI 12.7%– 17.4%) versus 1.1% with usual care (95% CI 1.0%–1.2%). After adjusting for hospice utilization, palliative care patients actually had a trend toward a higher 30-day readmission rate (AOR 1.20, 95% CI 0.99–1.45; p = 0.058), which suggests that increased hospice utilization among palliative care patients is an important factor in reducing 30-day readmissions. Compared to usual care patients, those seen by palliative care were more likely to be discharged with a DNR order (569/1430, 39.8% versus 1355/33,111, 4.1%; OR 15.49, OR 13.74–17.44; p < 0.001). Of 569 palliative care patients discharged with a DNR order, 240 (42.2%) had a DNR order entered after the palliative care consult. An additional 329 (57.8%) had a DNR order in the medical record prior to the consultation. Palliative care patients for whom a new DNR order was written (n = 240) had a significantly lower 30-day readmission rate compared to palliative care patients who either had a DNR order prior to the consultation or no DNR order at all (22/240, 9.2% versus 303/1190, 25.5%; OR 0.53, 95% CI 0.34–0.82; p = 0.004). Palliative care patients for whom a new DNR order was written after the consult also had a lower 30-day readmission rate compared to palliative care patient for whom no DNR order was written (22/240, 9.2% versus 285/861, 33.1%, OR 0.20, 95% CI 0.13–0.32; p < 0.001). This hospital uses a three-tiered DNR order system. DNRA orders indicate no CPR but continuation of all other treatment. DNR-B orders specify no CPR and no escalation of life-prolonging treatment beyond what is already being provided. DNR-C orders indicate comfort care with discontinuation of any treatments that do directly contribute to comfort. Of the 240 patients with a new DNR order after palliative care consultation, 127 had a DNR-A (52.9%), 34 had a DNR-B (14.2%), and 79 (32.9%) had a DNR-C order. The 30-day readmission rate for palliative care patients without DNR orders was 33.1% (285/861). For patients with a DNR-A order at discharge, the rate was 17.3% (22/127). The 30-day readmission rate was zero for patients with a DNR-B order (0/34) or a DNR-C order (0/79). The most common reasons for palliative care consultation were symptom management and goals of care. Of 1430 discharges seen by palliative care, 703 (49.2%) involved a request for symptom management, 557 (40.0%) involved a request for assistance establishing patient or family goals, and 255 (17.8%) included both. One hundred seventy consultations (11.8%) involved neither symptom management nor goals of care. Readmission rates were highest among patients for whom consultations involved symptom management (254/703; 29.1%) and lowest when consultations

O’CONNOR ET AL.

were for goals discussions (71/557; 12.8%). Among all palliative care patients, consultation involving a goals discussion was associated with a lower readmission rate (OR 0.36, 0.27–0.48; p < 0.001), but consultation involving symptom management was not (OR 1.05, 0.82–1.35; p = 0.684). Discussion

Hospitals are increasingly focused on reducing readmissions to avoid financial penalties under the federal Hospital Readmissions Reduction Program.25,26 Readmissions also negatively impact performance benchmarks used by Medicare to determine a hospital’s payment rates.27 This study’s finding that palliative care consultations may reduce readmissions suggests that palliative care may offer an important tool for hospitals and health systems, which could complement other interventions. In addition, this study provides new insights into mechanisms by which palliative care reduces readmissions. For instance, we found that specialist-level symptom management is not associated with a lower rate of hospital readmissions. In contrast, goals discussions by palliative care specialists were associated with a significant decrease in 30day readmission rates. Similarly, when a palliative care consultation resulted in a new DNR order, patients had a lower rate of readmissions. Together, these results suggest that palliative care’s impact on 30-day readmission rates may be driven largely by goals discussions that let patients and health care providers choose a less aggressive plan of care. DNR orders may be an objective proxy measure for goals discussions. At least some of palliative care’s impact on 30-day readmissions may also be due to hospice referrals. Patients in hospice generally have much lower rates of acute care utilization,28 and a previous study demonstrated reduction in 30day readmissions when a hospice liaison was embedded in an inpatient palliative care team.12 Our study also suggests that hospice may be an important mechanism by which palliative care reduces readmissions, even when the palliative care service does not include a hospice liaison. Palliative care teams assist with determining hospice eligibility and raise hospice as an option with patients, families, and care teams. Palliative care consultants can also allay fears and misconceptions about hospice. Even without the inclusion of a hospice representative on the team, we found that palliative care discussions were associated with a substantial increase in hospice utilization and a decreased rate of readmissions. This study has several limitations. First, we studied only one palliative care team in a single academic hospital. Palliative care programs vary widely in their training, staffing, and acceptance within a hospital’s culture, and these differences likely affect a given palliative care team’s ability to impact 30day readmissions. Second, we only included readmissions to the study hospital. It is possible that patients had readmissions to other hospitals, and that might have changed the results of this study. Finally, propensity score adjustment cannot control for unmeasured differences. For example, it is possible that patients who received palliative care had less aggressive treatment preferences and stronger preferences against rehospitalization than usual care patients. However, we were able to adjust for variables that are markers of preferences, such as a history of previous 30-day readmissions.

PALLIATIVE CARE AND 30-DAY READMISSIONS

Although preliminary, these results provide important evidence to support the value of palliative care in reducing 30-day readmissions. These results may be of considerable interest to payers, since many other efforts to achieve this goal rely on interventions such as home care or skilled nursing facilities, which can increase costs.29,30 In contrast, a palliative care intervention that aligns care with patients’ goals and creates a less aggressive care plan may offer a unique opportunity to improve patient-centered care while reducing total costs. Author Disclosure Statement

No competing financial interests exist. References

1. Burke RE, Coleman EA: Interventions to decrease hospital readmissions: Keys for cost-effectiveness. JAMA Intern Med 2013;173:695–698. 2. Dharmarajan K, Hsieh AF, Lin Z, et al.: Hospital readmission performance and patterns of readmission: Retrospective cohort study of Medicare admissions. BMJ 2013;347:f6571. 3. Penrod JD, Deb P, Luhrs C, et al.: Cost and utilization outcomes of patients receiving hospital-based palliative care consultation. J Palliat Med 2006;9:855–860. 4. Morrison RS, Penrod JD, Cassel JB, et al.: Cost savings associated with US hospital palliative care consultation programs. Arch Intern Med 2008;168:1783–1790. 5. May P, Normand C, Morrison RS: Economic impact of hospital inpatient palliative care consultation: Review of current evidence and directions for future research. J Palliat Med 2014;17:1054–1063. 6. Morrison RS, Dietrich J, Ladwig S, et al.: Palliative care consultation teams cut hospital costs for Medicaid beneficiaries. Health Aff 2011;30:454–463. 7. Hanson LC, Usher B, Spragens L, et al.: Clinical and economic impact of palliative care consultation. J Pain Symptom Manage 2008;35:340–346. 8. Enguidanos S, Vesper E, Lorenz K: 30-day readmissions among seriously ill older adults. J Palliat Med 2012;15:1356– 1361. 9. Fromme EK, Bascom PB, Smith MD, et al.: Survival, mortality, and location of death for patients seen by a hospitalbased palliative care team. J Palliat Med 2006;9:903–911. 10. Paris J, Morrison RS: Evaluating the effects of inpatient palliative care consultations on subsequent hospice use and place of death in patients with advanced GI cancers. J Oncol Practice 2014;10:174–177. 11. Sidebottom AC, Jorgenson A, Richards H, et al.: Inpatient palliative care for patients with acute heart failure: Outcomes from a randomized trial. J Palliat Med 2015;18:134–142. 12. Tangeman JC, Rudra CB, Kerr CW, et al.: A hospicehospital partnership: Reducing hospitalization costs and 30day readmissions among seriously ill adults. J Palliat Med 2014;17:1005–1010. 13. Averill RF: The evolution of case-mix measurement using DRGs: Past, present and future. Stud Health Technol Inform 1994;14:75–83. 14. Bowles KH, Holmes JH, Ratcliffe SJ, et al.: Factors identified by experts to support decision making for post acute referral. Nurs Res 2009;58:115–122.

961

15. Bowles KH, Hanlon A, Holland D, et al: Impact of discharge planning decision support on time to readmission among older adult medical patients. Prof Case Manag 2014;19:29–38. 16. Bowles KH, Chittams J, Heil E, et al.: Successful electronic implementation of discharge referral decision support has a positive impact on 30- and 60-day readmissions. Res Nurs Health 2015;25:1–13. 17. Austin PC: An introduction to propensity score methods for reducing the effects of confounding in observational studies. Multivariate Behav Res 2011;46:399–424. 18. Mickey J, Greenland S: A study of the impact of confounderselection criteria on effect estimation. Am J Epidemiol 1989;129:125–137. 19. Wu J, Roy J, Stewart WF: Prediction modeling using EHR data: Challenges, strategies, and a comparison of machine learning approaches. Med Care 2010;48:S106–S113. 20. Belitser SV, Martens EP, Pestman WR, et al.: Measuring balance and model selection in propensity score methods. Pharmacoepidemiol Drug Saf 2011;20:1115–1129. 21. Weitzen S, Lapane KL, Toledano AY, et al.: Principles for modeling propensity scores in medical research: A systematic literature review. J Pharmacoepidemiol Drug Safety 2004;13:841–853. 22. Lunt M, Solomon D, Rothman K, et al.: Different methods of balancing covariates leading to different effect estimates in the presence of effect modification. Am J Epidemiol 2009;169:909–917. 23. Sturmer T, Joshi M, Glynn RJ, et al.: A review of the application of propensity score methods yielded increasing use, advantages in specific settings, but not substantially different estimates compared with conventional multivariable methods. J Clin Epidemiol 2006;59:437–447. 24. Seeger JD, Kurth T, Walker AM: Use of propensity score technique to account for exposure-related covariates: An example and lesson. Med Care 2007;45:S143–S148. 25. Joynt KE, Jha AK: A path forward on Medicare readmissions. N Engl J Med 2013;368:1175–1177. 26. Joynt KE, Jha AK: Characteristics of hospitals receiving penalties under the Hospital Readmissions Reduction Program. JAMA 2013;309:342–343. 27. Chen C, Ackerly DC: Beyond ACOs and bundled payments: Medicare’s shift toward accountability in fee-forservice. JAMA 2014;311:673–674. 28. Woolley DC, Old JL, Zackula RE, et al.: Acute hospital admissions of hospice patients. J Palliat Med 2013;16:1515– 1522. 29. Chandra A, Dalton MA, Holmes J: Large increases in spending on postacute care in Medicare point to the potential for cost savings in these settings. Health Aff 2013;32:864–872. 30. Kuo YF, Goodwin JS: Association of hospitalist care with medical utilization after discharge: Evidence of cost shift from a cohort study. Ann Intern Med 2011;155:152–159.

Address correspondence to: David J. Casarett, MD, MA 3615 Chestnut Street Philadelphia, PA 19104 E-mail: [email protected]