The Cost-Effectiveness of the Decision to Hospitalize Nursing Home Residents with Advanced Dementia

Abstract

Context

Nursing home residents with advanced dementia commonly experience burdensome and costly hospitalizations that may not extend survival or improve the quality of life. Cost-effectiveness analyses of decisions to hospitalize these residents have not been reported.

Objectives

Estimate the cost-effectiveness of (1) not having a do-not hospitalize order, and (2) hospitalization for suspected pneumonia in nursing home residents with advanced dementia.

Methods

Nursing home residents from 22 nursing homes in the Boston area were followed in the Choices, Attitudes, and Strategies for Care of Advanced Dementia at the End-of-Life (CASCADE) study conducted between February 2003 and February 2009. We conducted cost-effectiveness of aggressive treatment strategies for advanced dementia residents living in nursing homes when they suffer from acute illness. Primary outcome measures included quality-adjusted life days and years (QALDs/QALYs), Medicare expenditures, and incremental net benefits (INBs) over 15 months.

Results

Compared with a less aggressive strategy of avoiding hospital transfer (i.e. having DNH orders), the strategy of hospitalization was associated with an incremental increase in Medicare expenditures of $5,972 and an incremental gain in quality-adjusted survival of 3.7 QALDs. Hospitalization for pneumonia was associated with an incremental increase in Medicare expenditures of $3,697 and an incremental reduction in quality-adjusted survival of 9.7 QALD. At a willingness-to-pay level of $100,000/QALY, the incremental net benefit of the more aggressive treatment strategies were negative and therefore not cost-effective (INB for not having a DNH order, −$4,958; INB for hospital transfer for pneumonia, −$6,355).

Conclusions

Treatment strategies favoring hospitalization for nursing home residents with advanced dementia are not cost-effective.

INTRODUCTION

Dementia is a leading cause of death among Americans,¹ yet patients dying with this disease may not receive optimal end-of-life care.^2–4 Approximately 16% of US decedents dying from dementia die in hospitals.⁵ Nationwide, 20% of nursing home (NH) residents with advanced dementia experience a health care transition, such as a hospitalization, near the end of life.⁶ Recent work emphasizes the need to avoid unnecessary and costly hospitalizations of NH residents with advanced dementia.^6–10 However, cost-effectiveness analyses (CEA) of hospitalization have not been reported.

High-quality advanced dementia care includes decision-making for hospitalizations that are informed and goal-directed from a patient perspective and cost-effective from a societal perspective. NH residents with advanced dementia have profound cognitive and functional disability (i.e., bed-bound, cannot recognize family members, speech limited to < 5 words, incontinent of urine and stool). It is estimated that 75% of hospitalizations for these residents may be avoidable, either because hospital-level care is inconsistent with the goals of care or is unnecessary.¹¹ Over 90% of proxies of NH residents with advanced dementia state that maximizing comfort is the primary goal of care,^{12, 13} Hospitalization seldom achieves this goal, except when the NH cannot provide or adequate palliative care or the level of treatment needed to relieve discomfort (i.e., hip fracture). Prior work has shown that hospitalizations are associated with worse end-of-life outcomes for NH residents with advanced dementia,^{6, 14} and distressing for their families.¹⁵ Moreover, the most common conditions precipitating hospitalization in advanced dementia can often be treated in the NH with similar clinical outcomes.^16–19 Furthermore, hospitalizations and post-hospitalization skilled nursing facility (SNF) care account for 30% and 10% of Medicare expenditures for these NH residents, respectively.²⁰

CEA ascertains the value of added benefits from treatment relative to incremental expenditures. While applying CEA to terminally ill patients is challenging, empirical information about what constitutes cost-effective end-of-life care is essential for improving the health care system.²¹

Leveraging data from a prospective cohort study of NH residents with advanced dementia, The Choices, Attitudes, and Strategies for Care of Advanced Dementia at the End-of-Life (CASCADE) study,²² we conducted CEAs of two hospital-related treatment decisions. The first evaluated the cost-effectiveness of a do-not-hospitalize order (DNH), an advance directive to avoid hospitalization for acute illnesses. The second considered hospitalization for pneumonia, the most common diagnosis precipitating hospitalization.¹¹ These analyses explored whether reducing hospitalization can promote a higher quality end-of-life experience without substantially increasing Medicare expenditures.

METHODS

Sample

Subjects included NH residents with advanced dementia who participated in CASCADE, a prospective cohort study conducted between 2003 and 2009, the details of which are provided elsewhere.^{13, 22} Residents were recruited from 22 Boston-area NHs. Eligibility criteria included: 1) age > 60; 2) dementia (any type); 3) Global Deterioration Scale (GDS) score of 7,²³ and 4) available English-speaking health care proxy. At GDS stage 7, residents have profound memory deficits, virtually no verbal communication, incontinence, and cannot walk. Proxies provided informed consent. The institutional review board of Hebrew SeniorLife approved the study’s conduct.

Data collection

Resident assessments were conducted at baseline and quarterly for up to 18 months. Independent variables included resident characteristics potentially associated with hospitalization decisions were selected ‘a priori’ from the dataset based on the literature.^{13, 24–27} Baseline resident characteristics obtained from the chart included demographics (gender, race [white vs. other], and age [≤ 85 vs. >85]); co-morbidities (congestive heart failure, active cancer, chronic obstructive lung disease [COPD]); and presence of a percutaneous endoscopic gastrostomy (PEG) tube. Additional resident variables were collected at baseline and follow-up assessments. Cognitive status was measured using the Test for Severe Impairment (TSI) score [range, 0–24, higher scores indicate better cognition; categorized as either equal to 0 or > 0].²⁸ Functional status was quantified by nurse interview using the Bedford Alzheimer Nursing Severity Scale (BANSS) [range 7–28, higher scores signify greater disability].²⁹ The occurrence of an acute major illness within the prior 90 days was ascertained from the chart at each quarterly assessment and included a pneumonia, febrile episode, or any major illness that had the potential to change goals of care (e.g., hip fracture, stroke, myocardial infarction).

Treatment strategies

Health care expenditures associated with two treatment strategies were examined: (1) not having a DNH order; and (2) hospitalization for pneumonia.

We chose to analyze “not having a DNH order” because it reflects a care plan for which hospitalization remains a possible treatment option, whereas having a DNH order reflects a treatment strategy explicitly avoiding hospitalization. DNH status was collected at baseline and each follow-up assessment. For these analyses, residents were considered to have a DNH order if it was present for two consecutive 90-day periods. Residents who died within three months of baseline were excluded to reduce potential bias resulting from including residents who obtained DNH orders just prior to death.

The occurrence and details of suspected pneumonias were based on documentation by a physician, nurse practitioner, or physician assistant, including: date of onset, hospitalization, suspected aspiration, and presence of any unstable vital signs (respiratory rate, > 30/minute; temperature, > 38.3°C; heart rate, > 125/minute; or systolic blood pressure, < 90mm Hg). We only analyzed the first episode among residents with multiple pneumonias.

Medicare expenditures

Utilization of the following Medicare services was abstracted from the chart at each assessment: hospital admissions, emergency department visits, physician and other professional visits in the NH, hospice enrollment, and SNF admission post-hospitalization. Medicare expenditures attributable to these services were determined using publically available sources and based on nationally representative rates from 2007 in US dollars.²⁰

Quality-adjusted survival

Two validated health status measures were collected from nurse interviews.³⁰ The Symptom Management at the End of Life in Dementia Scale (SM-EOLD), ascertained quarterly, quantified the frequency residents experienced distressing symptoms (e.g., pain, depression, fear, anxiety, and agitation) over the prior 90-days.³⁰ The Comfort Assessment in Dying with Dementia Scale (CAD-EOLD), ascertained within 14 days of the death, quantified the frequency residents experienced distressing symptoms during the last week of life. We developed and validated a method that mapped the SM-EOLD and CAD-EOLD to the Health Utility Index Mark 2 (HUI2), detailed elsewhere.³¹ Possible HUI2 scores range from negative 0.025 to 1.00; perfect health is scored 1.00, death is scored 0.00, and a negative score implies a state worse than death. In CASCADE, the residents’ mean HUI2 score was 0.165 ± 0.060 [standard deviation (SD)] (range, −0.005 to 0.215).³¹

For each follow-up period, the resident’s HUI2 score was multiplied by the number of days in the period to derive quality-adjusted life days (QALD) for that period. Total quality-adjusted survival was estimated by summing the QALDs for each period [Quality-adjusted life years (QALY) = QALD/365].

Analyses

Means and proportions were calculated for continuous and categorical variables, respectively. CEAs were conducted based on a 15-month observation period for two treatment strategies: (1) not having a DNH order, and (2) hospitalization following the first suspected pneumonia.

Residents with and without DNH orders, or those who were and were not hospitalized for pneumonia, may have differed in ways that could also explain differences in expenditures, survival, and quality-adjusted survival. Marginal structural modeling (MSM) was used to adjust for possible confounding. MSM was chosen over propensity scores as it produces less biased estimates and facilitates sensitivity analyses.

Two weights were calculated to conduct the MSM. The first was the estimated probability of the observed treatment approach conditional on independent factors identified in a multivariable analysis divided by the unconditional probability of the observed treatment approach.³² Logistic regression was used to estimate the conditional probability, based on independent variables selected ‘a priori’ from the dataset. Age, gender, race, BANS score, TSI score, co-morbidities, and PEG-tube status were included as independent variables for both CEA analyses. Recent acute illness was also considered in the DNH model. Unstable vital signs and suspected aspiration were included in the hospitalization for pneumonia analysis. Bivariable associations between each independent variable and the two treatment outcomes were examined. Variables associated with the treatment at a p-value < 0.10 in the bivariable analyses were included in the multivariate models. The second weight in the MSM, which adjusted for censoring, was the probability of being fully observed over the study period, calculated using Kaplan-Meier estimations.³³

MSM provided estimates of mean Medicare expenditures, survival, and quality-adjusted survival. These were used to calculate the incremental net benefits (INB) of treatment versus non-treatment, calculated as follows: (Willingness-to-pay (WTP)/QALY × Incremental quality-adjusted survival) − (Incremental Medicare expenditures). Standard WTP levels for medical treatments range between $50,000/QALY and $125,000/QALY.³⁴ The INB was determined for three WTP levels: $50,000/QALY, $100,000/QALY, and $150,000/QALY. A negative INB suggests a treatment is not cost-effective. The incremental cost-effectiveness ratio (ICER) (incremental expenditure/QALY) was also estimated. However, because negative ICERs are difficult to interpret, the INB was the primary measure.

Bootstrap methods were used to estimate the standard error of the incremental expenditure and quality-adjusted survival estimates. A plot showing bootstrap estimates of incremental costs against incremental quality-adjusted survival was generated for each treatment strategy. CEA curves (CEACs) were constructed based on the bootstrap plots to display the proportion of positive INBs for a range of WTP levels between $25,000 and $300,000, illustrating the probability of cost-effectiveness at each WTP level.³⁵

Sensitivity analyses were conducted to examine possible bias due to unmeasured confounding.³⁶ Without unmeasured confounding, we assume that average expenditure and quality-adjusted outcomes for the treated and untreated groups would be equivalent if they received the same treatment approach. For the sensitivity analyses, alternative CEACs plots were constructed based on three hypothetical conditions: (1) under the same treatment approach, expenditures for the “treated” group (e.g., hospitalization) would be 30% less than the “untreated” group, (2) no unmeasured confounding for expenditures, and (3) under the same treatment approach, expenditures for the “treated” group (e.g., hospitalization) would be 30% greater than the “untreated” group. For each of these three conditions, five alternative CEACs were plotted based on varying hypothetical levels of unmeasured confounding with respect to quality-adjusted survival, i.e., quality-adjusted survival for the “treated” group was 10%, 20%, 30%, 40% and 50% less than in the “untreated” group. We identified where these alternative curves indicated that the treatment was cost-effective (i.e. 90% of the INBs were positive).

All statistical analyses were conducted using R version 14.0.

RESULTS

Sample

Among the 323 residents in CASCADE, 55 residents who died within three months of baseline were excluded from the CEA of having a DNH order. Characteristics of the remaining 268 residents were similar to the entire CASCADE cohort (Table 1);¹³ ≤ 85 years, 50%; male, 14%; and non-white, 10%. The residents had severe functional impairment (mean BANSS score, 21.2) and cognitive impairment (TSI = 0, 84%). Characteristics of the residents with pneumonia (N=131) were similar (Table 2).

Table 1.

Characteristics of nursing home residents with advanced dementia and their association with not having a do-not-hospitalize order (N=268)

	Proportion/Mean			Odds ratio of not having a DNH order (95% CI)
Characteristic	Total (n=268)	DNH (n=124)	No DNH (n=144)	Unadjusted	Adjusteda
≤85 years (%)	134 (50.0)	57 (46.0)	77 (53.5)	1.4 (0.8–2.2)	-
Male (%)	38 (14.2)	11 (8.9)	27 (18.8)	2.4 (1.1–5.0)b	2.3 (1.1–5.0)b
Non-white (%)	28 (10.4)	4 (3.2)	24 (16.7)	6.0 (2.0–17.8)b	5.6 (1.9–17.0) b
Score on Bedford Alzheimer Nursing Severity Subscale, mean (SD)c	21.2 (2.3)	21.3 (2.5)	21.0 (2.0)	1.0 (0.9–1.1)	-
Score=0 on the Test for Severe Impairmentd (%)	226 (84.3)	107 (86.3)	119 (82.6)	0.8 (0.4–1.5)	-
Percutaneous endoscopic gastrostomy tube (%)	19 (7.1)	3 (2.4)	16 (11.1)	5.0 (1.4–17.7)b	4.0 (1.1–14.5) b
Acute illnesse 90-days prior to DNH decision (%)	74 (27.6)	34 (27.4)	40 (27.8)	1.0 (0.6–1.7)	-
Congestive heart failure (%)	39 (14.6)	15 (12.1)	24 (16.7)	1.5 (0.7–2.9)	-
Active cancer (%)	4 (1.5)	1 (0.8)	3 (2.1)	2.6 (0.3–25.5)	-
Chronic obstructive lung disease (%)	28 (10.4)	10 (8.1)	18 (12.5)	1.6 (0.7–3.7)	-

Abbreviations: DNH, do-not-hospitalize; CI, confidence interval; TSI, Test for Severe Impairment

^aAll variables associated with DNH with P<0.10 in the unadjusted analysis were included in the adjusted analysis

^bP<0.05

^cScores on the Bedford Alzheimer Nursing Severity Subscale range from 7 to 28; higher scores indicate greater functional disability.

^dScores on the TSI range from 0 to 24; lower scores indicate greater cognitive impairment. The TSI was dichotomized as 0 or higher than 0.

^eAcute illness includes febrile episodes, pneumonia, or other major events such as fractures, strokes, or myocardial infarctions.

Table 2.

Characteristics of nursing home residents with advanced dementia experiencing a suspected pneumonia and their association with hospitalization (N=131)

	Proportion/mean			Odds of hospitalization for pneumonia (95% CI)
Characteristic	Total (n=131)	No Hosp (n=113)	Hosp (n=18)	Unadjusted	Adjustedb
≤ 85 years (%)	61 (46.6)	48 (42.5)	13 (72.2)	3.5 (1.2–10.5)c	3.8 (1.1–13.0) c
Male (%)	23 (17.6)	16 (14.2)	7 (38.9)	3.9 (1.3–11.4) c	3.4 (1.0–11.8) c
Non-white (%)	12 (9.2)	9 (8.0)	3 (16.7)	2.3 (0.6–9.5)	-
No do-not-hospitalize order (%)	79 (60.3)	62 (54.9)	17 (94.4)	14.0 (1.8–108.7) c	13.2 (1.6–111.4) c
Score on Bedford Alzheimer Nursing Severity Subscaled, mean (SD)	21.4 (2.0)	21.4 (2.1)	21.5 (1.6)	1.0 (0.8–1.3)	-
Score 0 on the Test for Severe Impairmente (%)	103 (78.6)	87 (77.0)	16 (88.9)	2.4 (0.5–11.2)	-
Percutaneous endoscopic gastrostomy tube (%)	11 (8.4)	9 (8.0)	2 (11.1)	1.4 (0.3–7.3)	-
Congestive heart failure (%)	29 (22.1)	27 (23.9)	2 (11.1)	0.4 (0.1–1.8)	-
Active cancer (%)	1 (0.8)	1 (0.9)	0 (0.0)	0.0 (0.0–∞)	-
Chronic obstructive lung disease (%)	16 (12.2)	11 (9.7)	5 (27.8)	3.6 (1.1–11.9) c	4.4 (1.0–19.0) c
Suspected aspiration (%)	77 (58.8)	63 (55.8)	14 (77.8)	2.8 (0.9–9.0)	-
Unstable vital signsf (%)	48 (36.6)	40 (35.4)	8 (44.4)	1.5 (0.5–4.0)	-

Abbreviations: CI, confidence interval; SD, standard deviation; TSI, Test for Severe Impairment

^aAnalyses are at the level of the pneumonia episode. Odds ratio reflects likelihood of resident being hospitalized versus treatment in nursing home for the episode.

^bAll variables associated with hospitalization with P<0.10 in the unadjusted analysis were included in the adjusted analysis

^cP<0.05

^dScores on the Bedford Alzheimer Nursing Severity Subscale range from 7 to 28; higher scores indicate greater functional disability.

^eScores on the TSI range from 0 to 24; lower scores indicate greater cognitive impairment. The TSI was dichotomized as 0 or higher than 0.

^fUnstable vital signs defined by presence of any of the following: respiratory rate > 30/min; temperature > 38.3°C; heart rate > 125/min; or systolic blood pressure < 90mm Hg.

Cost-effectiveness of not having a DNH order

There were 124 (46%) and 144 (54%) residents who did and did not have DNH orders, respectively. Resident characteristics independently associated with not having a DNH order were: male, AOR, 2.3 (95% CI, 1.1–5.0); non-white, AOR, 5.6 (95% CI, 1.9–17.0); and PEG tube, AOR, 4.0 (95% CI, 1.1–14.5) (Table 1).

The estimated incremental increase in average Medicare expenditures among residents not having a DNH order was $5,972 (SD $1,569) and the incremental gain in quality-adjusted survival was 3.7 QALD (SD, 4.1) or 0.01 QALY (SD, 0.01) (Table 3). At WTP levels of $50,000/QALY and $150,000/QALY, the INB of not having a DNH order was −$5,465 (SD, $1,718), and −$4,451 (SD, $2,316), respectively. These negative INBs suggest that not having a DNH order was not cost-effective. The estimated ICER of not having a DNH order was $589,130/QALY.

Table 3.

Cost-effectiveness analyses of not having a do-not-hospitalize order and hospitalization for suspected pneumonia among nursing home residents with advanced dementia

Outcome	No do-not-hospitalize order	Hospitalization for suspected pneumonia
Incrementala expenditure (SD)	$5,972 ($1,569)	$3,697 ($5,981)
Incrementala survival (days) (SD)	20.9 (22.0)	−15.1 (54.5)
Incrementala quality-adjusted survival (days) (SD)	3.7 (4.1)	−9.7 (7.1)
Incremental net benefit at selected levels of willingness-to-payb
$50,000 per year	−$5,465 ($1,718)	−$5,026 ($5,859)
$100,000 per year	−$4,958 ($1,956)	−$6,355 ($5,928)
$150,000 per year	−$4,451 ($2,316)	−$7,683 ($6,144)
Incremental cost-effectiveness ratio:
Incremental expenditure/Quality-Adjusted Day	$1,614/day	−$381/day
Incremental expenditure /Quality-Adjusted Year	$589,130/year	−$139,114/year

Abbreviations: SD, standard deviation

^aIncremental comparison of treatment group (no do-not-hospitalize order; hospitalization for pneumonia) relative to non-treatment group (do-not-hospitalize order; no hospitalization for suspected pneumonia).

^bIncremental net benefit = (willingness-to-pay × Incremental quality-adjusted survival (years)) − Incremental expenditure. Negative values indicate treatment strategy not cost-effective at corresponding willingness to pay level

The CEAC in Panel A of Figure 1 shows the proportion of bootstrap samples with positive INBs for not having a DNH order at WTP levels ranging from $25,000 to $300,000 per QALY. The proportion of positive INBs was below 20% for WTP up to $300,000. At WTP amounts less than $125,000, less than 3% of the bootstrap samples show a positive benefit.

Figure 1. Cost-effectiveness analysis bootstrap and cost-effectiveness analysis curves (CEAC).

Panel A presents the cost-effectiveness results for not having a DNH order. Panel B presents the cost-effectiveness results for the hospitalization for pneumonia. In both panels, the left plot shows the bootstrap estimates of incremental Medicare expenditure ($1000 increments) (Y-axis) against and quality-adjusted survival (days) (X-axis), demonstrating the variance of the joint estimates of expenditure and quality-adjusted survival. Two willingness-to-pay lines are presented: $50,000/quality-adjusted life year (QALY) (dotted line) and $300,000/QALY (dashed line). All points below and to the right of these willingness-to-pay lines are considered cost-effective (i.e., positive INB).

The plots on the right of each panel display the proportion of bootstrap estimates with positive incremental net benefits (INBs) (i.e., cost effective) for each level of willingness-to-pay from $25,000 to $300,000. Any point along the CEAC where the proportion of positive INBs is 90% or greater can be considered cost-effective. For example, in Panel A, only 20% of the INBs are positive at a willingness-to-pay of $300,000.

Cost-effectiveness of hospitalization for pneumonia

Among residents with pneumonia, 113 (86%) were not hospitalized and 18 (14%) were hospitalized. Resident characteristics independently associated a greater likelihood of hospitalization included: age ≤ 85, AOR 3.8, (95% CI 1.1–13.0); male, AOR 3.4 (95% CI 1.0–11.8); no DNH order, 13.2 (95% CI 1.6–111.4); and COPD, AOR, 4.4 (95% CI 1.0–19.0) (Table 2).

In the adjusted analyses, the incremental increase in average Medicare expenditures of $3,697 (SD, $5,981) for residents who were hospitalized and an incremental loss in quality-adjusted survival of 9.7 QALD (SD, 7.1) or 0.03 QALY (SD 0.02) (Table 3). At WTP levels of $50,000 and $150,000, the estimated INBs of hospitalization were −$5,026 (SD $5,860) and −$7,683 (SD $6,144), respectively. These negative INBs suggest that hospitalization for pneumonia was not cost-effective. The estimated ICER was negative.

The CEAC in Panel B of Figure 1 shows the proportion of bootstrap samples with positive INBs for hospitalization at WTP levels ranging from $25,000 to $300,000 per QALY. The proportion of positive INBs was below 20% for WTP up to $300,000.

Sensitivity analyses

Sensitivity analyses suggest that not having a DNH order remained not cost-effective at lower levels of WTP assuming low to moderate levels of unmeasured confounding (Figure 2, Panels A1–A3). For example, Panel A3 presents the hypothetical condition whereby expenditures for residents without DNH orders would be 30% greater than those with the order. We further assumed that quality-adjusted survival for residents without a DNH order ranged from 10% to 50% greater (five lines on plot) compared those with the order. At a WTP level of $75,000, the treatment approach was not cost-effective (i.e., < 90% of the INBs were positive) when confounding related to quality-adjusted survival was 30% or less. At a WTP of $100,000, the treatment approach remained not cost-effective when unmeasured confounding related to quality-adjusted survival was 20% or less. When assuming expenditures for those without DNH orders are 30% less than those with DNH orders (Panel A1) or expenditures between the two groups are the same (Panel A2), the plots show that not having a DNH order is not cost-effective at higher levels of WTP and assuming higher levels of unmeasured confounding with respect to quality-adjusted survival. Taken together, at levels of WTP less than $150,000 and unmeasured confounding with respect to quality-adjusted survival limited to 30%, not having a DNH order does not appear to be cost-effective.

Figure 2. Sensitivity of CEAC estimates to unmeasured confounding.

Without unmeasured confounding, the assumption is that average expenditure and quality-adjusted outcomes for the two groups (treated and untreated) would be the same if they received the same treatment. In the sensitivity analysis, the assumption was relaxed, and alternative plots of the CEACs were constructed based on three different sets of hypothetical conditions: (1) less costly residents tend to seek treatment, (2) no unmeasured confounding for expenditures, and (3) more costly residents tend to seek treatment. For each level of unmeasured confounding with respect to expenditures, five alternative CEACs were plotted based on different levels of unmeasured confounding with respect to quality-adjusted survival, ranging from 10% to 50%. For example, Panel A3 presents the hypothetical condition whereby expenditures under treatment for residents without DNH orders would be 30% greater than those with the order. At a WTP level of $75,000, the treatment approach was not cost-effective (i.e. less than 90% of the INBs were positive) if unmeasured confounding related to quality-adjusted survival was 30% or less.

The sensitivity analyses suggest that hospitalization for pneumonia remains not cost-effective (Figure 2, Panels B1–B3). For all WTP levels, and all levels of unmeasured confounding related to expenditures and quality-adjusted survival, hospitalization was not cost-effective (i.e. < 90% of INBs were positive).

DISCUSSION

This study found that more aggressive treatment strategies leading to hospitalization are not cost-effective for NH residents with advanced dementia when compared to approaches that avoid hospitalization. This suggests that clinical and policy initiatives aimed at avoiding hospitalization may reduce Medicare spending and, at the same time, promote a higher quality end-of-life experience for these residents.

To our knowledge, formal CEAs of treatment options for patients with terminal conditions, and specifically advanced dementia, have not been reported.²¹ Prior studies of NH residents with advanced dementia have demonstrated the following: acute care accounts for a large proportion of their Medicare expenditures;^{20, 37, 38} their mortality rate is high, especially following hospitalization or pneumonia;^{13, 19, 39} and they experience substantial discomfort near the end-of-life.^{13, 14, 19} This report links these findings by jointly estimating Medicare expenditures and quality-adjusted survival, demonstrating that decisions to hospitalize NH residents with advanced dementia are not cost-effective.

We examined two decisions pertaining to hospitalization: advance directives to avoid future hospital transfers in the event of an acute illness, and decisions not to hospitalize when an acute illness (i.e., pneumonia) occurred. These concepts are clearly linked. In CASCADE, not having a DNH order was the factor most strongly associated with hospitalization for an acute illness.¹¹ The presence of advance directives is also the most consistent factor associated with other markers of high-quality palliative care in advanced dementia, including: lower feeding tube use,^{40, 41} fewer terminal hospitalizations,^{6, 11} better family satisfaction and mental health outcomes,^{42, 43} and greater hospice use.⁴⁴ The prolonged period of severe disability in advanced dementia punctuated by predictable complications (e.g., infections),^{13, 45} provides optimal circumstances for advance care planning. However, in 2000, only 7.1% of US NH residents with advanced dementia had DNH orders,⁷ much lower than in CASCADE. Thus, there is a great opportunity to promote more cost-effective care by engaging their proxies in advance care planning about future hospitalizations.

Pneumonia is the most frequent diagnosis precipitating hospitalization and often a terminal event among NH residents with advanced dementia. ^{13, 39, 46} Observational studies suggest that treating pneumonia in the hospital, compared to the NH, does not improve their survival.^{16, 18, 19, 47} The negative incremental survival and quality adjusted survival found here suggest that hospitalizations may actually be detrimental in advanced dementia. These patients commonly undergo uncomfortable interventions in the hospital that have associated risks and are of questionable clinical benefit (e.g. feeding tube insertions)).^{14, 39, 48, 49} Care transitions also place them at risk for medical errors and adverse drug events.^{50, 51} NH residents with advanced dementia living in regions with higher health care transitions have been shown to experience worse end-of-life care.^{6, 49}

This study has several limitations. CASCADE was an observational study; the possibility of unmeasured confounding (e.g., from comorbidities not included in the analyses) remains despite the robustness of the adjusted and sensitivity analyses. The relatively small sample size resulted in large variance estimates for some cost and quality-adjusted survival measures, particularly for the hospitalization analysis. Medicare expenditures were calculated from public sources,²⁰ an approach that likely underestimated costs.^{27, 37} Utility-based quality of life measures were estimated by mapping to health status measures,³¹ albeit using previously validated methods. There also may have been inaccuracies in the ascertainment of utilization data, however such errors are likely to be non-differential between treatment groups. Medicaid expenditures were not examined, but are much less variable than Medicare expenditures.⁵² Finally, generalizability outside the Boston area is uncertain.

In this era of fiscal restraint and health care reform, much attention has focused on reducing hospitalizations of NH residents for whom acute care is both costly and often unnecessary.^{8, 10, 53, 54} While these concerns extend to residents with advanced dementia, there are additional compelling motivations to limit hospital transfers in this population. These residents are very near the end-of-life, have profound cognitive and functional disability, may be particularly traumatized by hospitalization, and most often desire comfort as their goal of care. Our findings help quantify these concerns. In this very frail and terminally ill population, the added expenditures of hospitalizations are not sufficiently counterbalanced by additional quality-adjusted survival to justify such treatment. Policy strategies that change incentives to support caring for residents in the facility and avoiding hospitalizations are warranted. At the individual level, providers should be encouraged to engage proxies in advance care planning to establish whether hospitalization aligns with the goal of care. At the NH level, enhanced resources to provide both on-site palliative and acute care, and broader quality improvement initiatives that aim to reduce transfers are needed.^{8, 18} Finally, policies that incentivize cost-effective care could have significant implications for the millions of Americans dying with dementia by promoting care that is less burdensome and costly, and more consistent with preferences.