Access to preferred skilled nursing facilities: Transitional care pathways for patients with Alzheimer's disease and related dementias

Abstract

Objective

The study aimed to assess whether individuals with Alzheimer's disease and related dementias (ADRD) experience restricted access to hospitals' high‐volume preferred skilled nursing facility (SNF) partners.

Data Sources

The data source includes acute care hospital to SNF transitions identified using 100% Medicare Provider Analysis and Review files, 2017–2019.

Study Design

We model and compare the estimated effect of facility “preferredness” on SNF choice for patients with and without ADRD. We use conditional logistic regression with a 1:1 patient sample otherwise matched on demographic and encounter characteristics.

Data Collection

Our matched sample included 58,190 patients, selected from a total observed population of 3,019,260 Medicare hospitalizations that resulted in an SNF transfer between 2017 and 2019.

Principal Findings

Overall, patients with ADRD have a lower probability of being discharged to a preferred SNF (52.0% vs. 54.4%, p < 0.001). Choice model estimation using our matched sample suggests similarly that the marginal effect of preferredness on a patient choosing a proximate SNF is 2.4 percentage points lower for patients with ADRD compared with those without (p < 0.001). The differential effect of preferredness based on ADRD status increases when considering (a) the cumulative effect of multiple SNFs in close geographic proximity, (b) the magnitude of the strength of hospital‐SNF relationship, and (c) comparing patients with more versus less advanced ADRD.

Conclusions

Preferred relationships are significantly predictive of where a patient receives SNF care, but this effect is weaker for patients with ADRD. To the extent that these high‐volume relationships are indicative of more targeted transitional care improvements from hospitals, ADRD patients may not be fully benefiting from these investments. Hospital leaders can leverage integrated care relationships to reduce SNFs' perceived need to engage in selection behavior (i.e., enhanced resource sharing and transparency in placement practices). Policy intervention may be needed to address selection behavior and to support hospitals in making systemic improvements that can better benefit all SNF partners (i.e., more robust information sharing systems).

What is known on this topic

• Individuals with Alzheimer's disease and related dementias (ADRD) receive postacute services in skilled nursing facilities (SNF) at disproportionate rates and cluster in SNFs that earn lower quality ratings.

• Hospitals target transitional care improvements toward preferred (i.e., high‐volume) SNF partners, meaning that not all patients benefit from these investments.

• Preferred SNFs may be positioned to selectively admit less risky or less complex patients, which could affect placement for patients with ADRD.

What this study adds

• Preferredness of a hospital‐SNF relationship is significantly predictive of where a patient goes to receive care, but this effect is weaker for patients with ADRD.

• ADRD patients may have reduced access to high‐volume referral care sites that are more likely to be receiving transitional care investments from hospitals.

1. INTRODUCTION

Hospitals discharge nearly one in five Medicare patients to a skilled nursing facility (SNF) every year, resulting in over 1.5 million observed transitions. ¹ The transition from hospital to SNF is fraught with coordination challenges that can destabilize care and put patients at risk of adverse outcomes in the SNF as well as rehospitalization. ² , ³ , ⁴ , ⁵ For example, fragmented and delayed information‐sharing practices may cause delays or errors in medication administration or the timely availability of needed equipment. ⁶ , ⁷ , ⁸ This is especially true for patients with complex medical and social needs, for whom environmental needs and treatment plans are more complicated. ⁸ Organizations are increasingly grappling with the particular challenges of caring for the transitional needs of patients with Alzheimer's disease and related dementias (ADRD). ⁹ , ¹⁰ , ¹¹ , ¹² The population of hospitalized patients with ADRD is growing, and these patients are much more likely to be discharged to an SNF for postacute services compared with the general population. ¹³ Recent work suggests that SNF patients with cognitive impairment at the time of transfer are at increased risk of adverse outcomes, ¹⁴ and that an ADRD diagnosis is associated with receiving care in a lower‐quality SNF. ¹⁵ This is consistent with evidence suggesting high‐burden care experiences for these patients and their families. ⁹ , ¹⁰

Identifying where patients with dementia are placed for SNF care, and why, is important for making targeted policy and practice changes that impact care for this population. One important consideration is that substantial variability exists in the robustness of transitional care practices across SNF settings. Hospitals are motivated to invest in improved handoffs with SNFs given increasing financial penalties associated with readmission and other indicators of poor transitional care quality. ¹⁶ , ¹⁷ These investments include a range of integration activities, such as joint leadership meetings and shared care protocols as well as improved electronic means of sharing information. ¹⁸ But, these investments are uneven and tend to be concentrated among “preferred” SNF partners. Preferred partners are those with relationships that may carry expectations of collaboration—for example, shared care protocols and reporting of quality metrics—in exchange for hospital resources that support transition and acknowledgment as “preferred” in materials provided to patients. ⁸ , ¹⁹ , ²⁰ To maximize the value of these investments, these preferred partners are most often (though not always) the facilities with which hospitals refer the largest volumes of their patients. ¹⁶ , ¹⁹ , ²¹

Preferred status may create implicit or explicit hospital expectations that an SNF will help when needed with accepting patients that have significant care complexities and are considered difficult to place. On the other hand, such high‐volume relationships may enable and encourage SNFs to be more risk‐averse in which referrals they accept, limiting admissions of complex patients to meet preferred partner performance requirements. ²¹ , ²² , ²³ , ²⁴ , ²⁵ If the latter is true, patients with ADRD may not have equitable access to preferred SNFs and the “enhanced” transition pathways they are more likely to offer.

Whether SNFs are agentic in a way that affects the placement of patients with complex conditions is difficult to tease apart from other, more primary factors that drive patient choice (i.e., proximity to the patient's home). Using multiple years of national Medicare claims data, we employ a choice model with a matched sample methodology to assess whether individuals with ADRD experience restricted access to hospitals' high‐volume preferred SNF partners (i.e., where care investments are concentrated). Ultimately, identifying processes that may contribute to greater vulnerability for ADRD patients is a critical step toward developing organizational strategies and policies that more equitably ensure high‐quality hospital‐SNF transitions for this population.

2. METHODS

2.1. Data and study population

Our study uses the 100% Medicare Provider Analysis and Review claim file to construct a census of traditional Medicare beneficiaries experiencing a hospital‐to‐SNF transition between 2017 and 2019. Our primary analytic sample includes all fee‐for‐service (non‐managed care) beneficiary hospitalizations at a general acute care inpatient facility with an observed SNF admission date within 2 days (48 h) of the hospital discharge date. We exclude beneficiaries with a residential ZIP code outside of the contiguous 48 states. ²⁶ The final sample for all choice model analyses totaled 3,024,995 hospital‐to‐SNF transitions between 2017 and 2019, or roughly 1 million hospital‐SNF transitions per calendar year.

2.2. Measures

2.2.1. Outcome

The outcome of our analysis is a patient's selection of an SNF from the full set of SNFs available to that patient, based on the location of their hospitalization. We are interested in whether a patient's chosen SNF is a preferred partner of the hospital from which they were discharged, and how that varies for individuals with and without ADRD. We define a hospital's top SNF partners based on shared referral volume; those who are in the top 50% of a hospital's yearly PAC referral volume in a sorted list are defined as “preferred partners”.

2.2.2. Covariates

Patient “choice” is a function of beneficiary, hospitalization, and SNF characteristics.

Beneficiary characteristics

The main independent variable of interest is beneficiary diagnosis with ADRD. We obtained this measure from the yearly supplemental Medicare chronic condition warehouse file, which includes a field for the date of first diagnosis of ADRD, if applicable. We created a binary indicator of ADRD based on whether the beneficiary had a date of first diagnosis. Other beneficiary‐level characteristics include dual‐eligible status, age, race, and sex.

Hospitalization characteristics

The most critical factor shaping where a patient receives SNF care relates to geographic location—proximity of an SNF to the discharging hospital and, more importantly, proximity of an SNF to the beneficiary's home. ²⁷ Therefore, for each hospitalization, we calculated geographic distances from the discharging hospital to the SNF where the patient was referred. We do this using exact physical addresses (street address, ZIP code, county, and state) available in public use files. We also calculated the distance from that SNF to the beneficiary's home residence. These distances are calculated from the exact address of the SNF to the centroid of the 5‐digit ZIP code for the beneficiary's home address, as no exact addresses were available in the claims data files. All distances were calculated using the geodist command in Stata version 17. ²⁸ Other relevant details of a beneficiary's hospital stay include the major diagnostic category (MDC) associated with the hospitalization, whether it was a medical or surgical stay, length of inpatient hospital stay, and the individual's Elixhauser comorbidity index calculated using ICD‐10 diagnoses associated with the claim.

SNF characteristics

SNF organizational descriptors included in analyses are profit status, proportion of facilities within the county that are chain affiliated (two or more facilities), primary rural–urban commuting area code, the total number of licensed beds, occupancy percentage, status as hospital‐based, proportion Medicaid, and overall quality rating. We include these characteristics because they have all demonstrated association with how an SNF operates in the market (i.e., how agentic they might be based on available resources and market leverage). ²⁹ , ³⁰ , ³¹ We also choose to include an indicator of whether the SNF has an ADRD specialty care unit, which often include modified physical environments and/or special staff training, and the benefit of housing a similar subset of patients in a well‐defined space. ³²

2.2.3. Analyses

We first generate summary statistics to compare patient‐, hospitalization‐, and SNF‐level organizational characteristics between all individuals with and without ADRD in our census file of hospital‐to‐SNF transitions. We, then, use multivariate linear probability models to estimate whether a patient's SNF placement is at a preferred facility, as a function of ADRD status and all other characteristics described.

Estimating a patient's SNF selection is challenging because all the SNF attributes and distances influence discharge decisions simultaneously and in interdependent ways. Patients are not evaluating an SNF independently but in reference to other available options and the tradeoffs they present (i.e., longer distance from home but for more specialized services). Therefore, to appropriately compare facility placement between patients with and without ADRD, we want to model how patients rank all available SNFs (i.e., using all available characteristics) and compare how the variable of SNF preferredness differently affects that ranking for patients with and without ADRD. To do this, we use McFadden's choice model approach, ³³ which has been used in other recent empirical analyses to model patient SNF placement. ²⁶ In this model, each patient faces a choice set of SNFs. Each SNF in the choice set is ranked according to the highest estimated likelihood of placement based on known inputs, which allows us to compare the expected discharge location to observed SNF placement.

To execute a choice model, we first create a 1:1 matched sample of ADRD and non‐ADRD patients. We use a matched sample for analysis of patient SNF choice because patients with ADRD are different than those without ADRD across many characteristics (e.g., age, race, and urbanicity) that influence choice in ways that we would not want to attribute to their ADRD diagnosis. We include race as a variable in our descriptive analyses and criteria for matching because of the documented patterns of structural racism that contribute to patients of color, notably Black patients, receiving care at different and lower quality institutions. ³⁴

A conditional logit model does not allow for the inclusion of multiple individual characteristics, so we need matching to control for these other characteristics in our model. ³³ Pairs were first matched on the calendar year of hospitalization, beneficiary residential ZIP code, and discharging hospital ID. From this sample, we retained ADRD and non‐ADRD 1:1 matched pairs if they exactly matched on chronological age ± 3 years, dual‐eligible status, race/ethnicity, sex, Elixhauser score ± 1, hospital length of stay ± 1 day, MDC, and medical versus surgical type of hospitalization. We match without replacement.

With this matched sample, we build out a full SNF choice set for each observed patient discharge in our data. This choice set is all potential SNF options a patient is presented with when discharged from a given hospital. For example, if hospital x has a history of referring patients to 15 unique SNFs for postacute care, we assume that every patient has the potential to end up at any of the 15 SNFs. As noted above, the most high‐volume of these SNF choices are indicated as preferred. We expand the dataset so that the number of observations for an individual is equal to the number of SNFs in their choice set, based on the originating hospital provider number and year of record. We create a binary variable “choice,” which is positively coded when the actual SNF to which the patient was discharged reflects an SNF in the choice set. We model the outcome—choice—using the clogit command in Stata version 17, as a function of preferred partner status, distance variables, and SNF‐level characteristics detailed above (see Appendix Exhibit 1). We interact the dichotomous patient‐level ADRD variable with all other predictors, interpreting the coefficient on the interaction terms as the differential effect of that predictor on choice for ADRD versus non‐ADRD patients. This model clusters standard errors at the beneficiaries' state of residence.

Choice models are utility maximization models (i.e., within‐person ordering of preferences), which makes it difficult to interpret coefficients in the traditional way—for example, to state the average effect of preferredness on choice. We, therefore, use simulation‐based analyses, a modified approach to marginal effects, to aid interpretation of our findings. ³¹ Our goal is to estimate whether the effect of preferredness on patient choice is weaker for patients with ADRD compared with those without. To do so, we start with each patient's closest SNF option (based on the centroid of the patient's home zip code). Based on the estimated choice model, we predict probabilities of going to each SNF in the choice set twice—first forcing every patient's most proximate SNF to be preferred, and then again forcing every patient's most proximate SNF to be nonpreferred. We calculate the difference in probability of going to this close‐by SNF, based on preferredness, for patients with ADRD. We compare this value to the same estimation done for patients without ADRD. The comparison of these two values estimates the differential effect of preferredness on patient choice for patients with versus without ADRD. Because patients often have multiple nearby SNF options, and thus the likelihood of going to a nearby SNF is often distributed across a few local options, we replicate this simulation considering the two closest and then the three closest SNFs to patients' homes.

We then run a second simulation where we estimate differences in the effect size of “preferredness” for ADRD versus non‐ADRD patients across a range of relationship strengths. We do this by simulating different scenarios of how high‐volume the relationship is between a discharging hospital and the patient's closest SNF (based on their home ZIP code). We ran the choice model six times, each time modifying the percent of the discharging hospital's SNF referrals that went to that SNF: 0%–10%, 11%–19%, 20%–29%, 30%–39%, 40%–49%, and ≥50%. We calculate and compare the marginal effect of preferredness on the predicted probability of discharge to the closest SNF across these six simulated models.

Sensitivity analyses

By using only a cutoff for high‐volume referral status as an indicator of preferred, there is some concern that the SNFs we are identifying may not actually have any explicit preferred status—that is benefitting from meaningful investment or partnership with the discharging hospital. We, therefore, run an alternate specification where we only identify the single most high‐volume SNF partner as preferred, as we are the most confident that this SNF is receiving some attention as a top receiving partner.

Second, ADRD is a degenerative disease with disease progression requiring increased levels of care. To assess whether the effect of preferred partner status on patient choice is attenuated by the duration of diagnosis with ADRD (a proxy for condition severity), we replace the binary ADRD indicator in our main model with a categorical ADRD variable indicating length of time with diagnosis (<1 year, 1–3 years, and 4+ years).

3. RESULTS

3.1. Descriptive results

Table 1 compares patients with and without ADRD in the full census of 2017–2019 Medicare hospitalizations resulting in an SNF transfer (N = 3,019,260) and our matched sample (N = 58,190). In the complete data, ADRD patients are older than their non‐ADRD counterparts, more often female, and overwhelmingly live in metropolitan ZIP codes. They are less likely to be white and more likely to have complex hospitalizations. Individuals with ADRD are also, on average, discharged to an SNF farther from their residential ZIP code and closer to the origin hospital than those without ADRD. Matching, by design, almost entirely removes the differences in characteristics of the non‐ADRD and ADRD samples.

TABLE 1.

Descriptive statistics of individual characteristics by ADRD diagnosis and matched versus full sample.

Characteristic	Full sample			Matched sample
	N = 3,019,260			N = 58,190
	Non‐ADRD (N = 1,464,730)	ADRD (N = 1,554,530)	p‐value	Non‐ADRD	ADRD	p‐value
Demographics
Age (mean [SD])	76.1 (11.1)	81.9 (10.3)	<0.001	84.1 (7.2)	84.4 (7.1)	<0.001
% Female	59.9%	60.9%	<0.001	79.6%	80.3%	0.04
Race
Unknown	0.8%	0.4%	<0.001	0.01%	0.01%	0.996
Non‐Hispanic White	83.3%	81.4%		98.0%	98.0%
Black	9.0%	10.6%		1.3%	1.2%
Other	0.6%	0.6%		0.01%	0.01%
Asian/Pacific Islander	1.6%	1.9%		0.2%	0.2%
Hispanic	4.2%	4.8%		0.5%	0.5%
American Indian/Alaska Native	0.6%	0.5%		0.03%	0.03%
Urbanicity
Metro	78.9%	80.8%	<0.001	83.7%	83.8%	0.880
Micro	12.1%	11.1%		13.5%	13.5%
Small town	5.3%	5.0%		2.2%	2.1%
Rural	3.7%	3.1%		0.7%	0.6%
Preceding hospitalization
Elixhauser Index	4.5 (2.3)	4.6 (2.1)	<0.001	3.6 (1.8)	3.5 (1.7)	0.269
Length of stay	7.1 (6.5)	7.1 (7.4)	<0.001	4.2 (1.9)	4.2 (1.8)	0.994
Distance measures (median [IQR])
Distance between hospital and beneficiary zip code (miles)	7.48 (16.60)	6.79 (14.84)	0.030	4.05 (5.81)	4.00 (5.68)	0.204
Distance between SNF and beneficiary zip code (miles)	24.79 (41.22)	23.28 (39.74)	<0.001	21.10 (35.02)	20.98 (34.83)	0.877
Distance between SNF and hospital (miles)	19.31 (35.30)	18.45 (32.58)	<0.001	20.88 (38.12)	20.82 (37.92)	0.091

Note: p‐value based on t‐test for continuous variables and χ² for categorical variables.

Abbreviations: ADRD, Alzheimer's disease and related dementias; IQR, interquartile range; SD, standard deviation; SNF, skilled nursing facility.

We present the absolute and adjusted differences in SNF characteristics between non‐ADRD and ADRD patients in Table 2. In the full data, ADRD patients are more likely to be placed in SNFs that are larger, for‐profit, and have a greater percentage of Medicaid patients. These differences persist in our matched sample. Admitting SNFs for ADRD patients in our matched sample are also less likely to be hospital based, more likely to be chain affiliated, and more likely to have a dedicated ADRD unit.

TABLE 2.

Comparison of admitting SNF characteristics by ADRD diagnosis.

SNF characteristic	Non‐ADRD	ADRD	Absolute difference	Adjusted difference
Full sample (N = 3,019,260)
Number of certified beds	120.1 (64.1)	122.6 (66.1)	2.53***	0.15**
Average total residents	99.1 (58.3)	101.6 (60.7)	2.55***	0.23***
Percent of facilities that are hospital based	2.8%	2.6%	0.0023***	−0.0048***
Payer mix‐Medicaid percentage	54.5 (24.7)	54.8 (24.7)	0.29***	0.16***
Percent of facilities that are for‐profit	72.9%	73.9%	0.020***	−0.0007
Percent of facilities that are multifacility (i.e., chain)	60.4%	59.4%	0.0096***	−0.00057
Percent of facilities with Alzheimer's unit	14.8%	14.3%	0.0047***	0.00076**
Facility quality rating
0–3	30.1%	30.9%	0.0121***	0.0016
4–5	69.9%	69.1%
Matched sample (N = 58,190)
Number of certified beds	121.5 (73.61)	124.2 (72.9)	2.73***	2.55***
Average total residents	101.9 (67.2)	104.2 (66.3)	2.32***	2.14***
Percent of facilities that are hospital based	6.6%	4.8%	0.018***	−0.031***
Payer mix‐medicaid percentage	37.3 (26.2)	39.9 (26.0)	2.56***	2.68***
Percent of facilities that are for‐profit	58.3%	61.3%	0.05***	−0.059***
Percent of facilities that are multifacility (i.e., chain)	57.2%	58.3%	0.011**	0.013***
Percent of facilities with Alzheimer's unit	14.8%	15.8%	0.009***	0.0090***
Facility quality rating
0–3	24.5%	24.9%	0.036***	−0.039***
4–5	75.5%	75.1%

Note: To calculate absolute differences, we provide results of a mean comparison t‐test grouped by ADRD status. Adjusted differences were calculated via an ordinary least squares (OLS) model with origin hospital zip code fixed effects. SNF characteristics were the outcome variables and patient‐level characteristics were incorporated as predictors; including ADRD status, age, race, gender, hospitalization Elixhauser, hospital length of stay, and major diagnostic category (MDC) type. Status as hospital based, for‐profit or multisite facility, and quality rating were binary—the rest are continuous.

Abbreviations: ADRD, Alzheimer's disease and related dementias; SNF, skilled nursing facility.

^* p < 0.10;

^** p < 0.05;

^*** p < 0.01.

3.2. Regression results

Our ordinary least squares (OLS) regression results for the full census data are shown in Appendix Exhibit 2. Using postestimation margins, we estimate a 2.4 percentage point reduced probability of a patient with ADRD being discharged to a preferred SNF, relative to patients without ADRD (54.4% compared to 52.0%, see Figure 1).

FIGURE 1.

Predicted probability of discharge to a preferred SNF based on ADRD diagnosis. ADRD, Alzheimer's disease and related dementias; CI, confidence interval.

The results of our conditional logistic regression, using the matched sample, are included as Appendix Exhibit 3. While these models offer limited opportunity for direct interpretation of coefficients (i.e., because they predict utility related to choice), we highlight only that the effect of preferredness is significantly more predictive of choice for non‐ADRD versus ADRD patients.

3.3. Simulation analyses

When we use simulation analyses to estimate the effect of preferredness on choice, we find a consistent and significant difference between patients with ADRD and those without (see Figure 2). Considering just a patient's closest SNF, simulated results show that patients with ADRD have an estimated 34.2% predicted probability of entering this SNF if it is preferred, compared with 14.4% if not preferred (difference: 19.8 percentage points). For patients without ADRD, results estimate a 35.4% predicted probability of entering this SNF if it is preferred, compared with 13.3% if not preferred (difference: 22.1 percentage points). These findings indicate that the effect of preferredness on choice is 2.4 percentage points lower for patients with ADRD compared with those without.

FIGURE 2.

Simulated effects of preferredness on SNF choice, by ADRD status. Preferred facility indicates a hospital's highest volume SNF referral partners. ADRD, Alzheimer's disease and related dementias; SNF, skilled nursing facility.

These numbers reflect the fact that patients often have multiple options in close proximity, and thus, the probability of admission to any one SNF is not particularly high. When we run the simulation and consider patients' three closest SNFs, we find that the cumulative probability of a patient with ADRD going to any one of these SNFs is 73.7% (24.6% probability per SNF) if they are all preferred and 42.8% if not preferred (difference: 30.9 percentage points). For patients without ADRD, the cumulative probability of going to any one of these SNFs is 75.4% (25.1% probability per SNF) if they are all preferred and 42.8% if not preferred (difference: 35.1 percentage points). These findings indicate that the effect of preferredness on choice is 4.2 percentage points lower for patients with ADRD compared with those without.

We were also interested in simulating changes in “how preferred” an SNF may be with their discharging hospital (i.e., not treating preferredness as a binary indicator). We find that the effects of preferredness on choice only differ between patients with and without ADRD if the preferred SNF is getting at least 20% of the hospital's discharges (Figure 3). For example, if the closest SNF is getting 10%–19% of a hospital's discharges, patients with and without ADRD are equally likely to go to that facility (27% vs. 26.6%). However, if the closest SNF has 40%–49% of a hospital's total discharges, patients with ADRD are 4.4 percentage points less likely to be placed at that facility compared with patients without ADRD (38.4% vs. 42.8%).

FIGURE 3.

Choosing the closest SNF across a simulated range of preferredness Strengths. Preferred facility indicates a hospital's highest volume SNF referral partners. ADRD, Alzheimer's disease and related dementias; SNF, skilled nursing facility.

3.4. Sensitivity analyses

Our sensitivity analyses first assess whether results change when we restrict preferredness to only the single highest volume SNF partner for each discharging hospital. We find consistency in the direction and magnitude of our findings (see Appendix Exhibit 4). We next considered whether results differed when we modeled ADRD diagnosis using our categorical measure (i.e., based on length of time with ADRD diagnosis) rather than a binary indicator. Relative to those without ADRD, patients with ADRD become increasingly less likely to be discharged to a preferred SNF following hospitalization as the duration of their diagnosis increases (p < 0.001) (see Appendix Exhibit 5).

4. DISCUSSION

Using Medicare data that captured 100% of hospitalizations between 2017 and 2019, we examined placement decisions for patients who subsequently required postacute services at an SNF. Patients have multiple options and tend to choose an SNF based on geographic proximity to home. However, there is some evidence to suggest that patients might be better served at preferred SNF partners (i.e., high‐volume referral partner) of the discharging hospital where efforts to improve transitional care are concentrated. ¹⁶ , ¹⁸ , ¹⁹ , ²⁰ We found that patients diagnosed with ADRD were less likely to be admitted to a preferred facility compared with patients without ADRD. Adjusted OLS regression models estimate a 2.4 percentage point reduced probability of a patient with ADRD being discharged to a preferred SNF, relative to patients without ADRD (54.4% compared with 52.0%). We interpret this effect size as meaningful in that it was similar in our model to that of dual eligibility status, which has been demonstrated in other literature to be a significant patient‐level predictor of SNF placement. ³¹

Using a choice modeling approach that then more rigorously compares facility placement between a matched cohort of ADRD and non‐ADRD patients, we find consistent evidence that the effect of preferredness on patients' choice of SNF is weakened for patients diagnosed with ADRD. Considering a patient's closest SNF options, preferredness dramatically increased the likelihood that a patient will go to one of these facilities. However, that effect size (change in probability) is 2.4–4.2 percentage points lower for patients with ADRD compared with those without. These results are notable, first, in consistency with our linear probability model which signals robustness to different modeling approaches. Second, the magnitude of this finding is similar to recent findings that estimate the difference in the likelihood of admission to a top‐quality (4‐ or 5‐star) SNF based on ADRD status. ¹⁵ This similarity adds strength to our assumption that high‐volume relationships may proxy for care quality and necessitates a deeper examination of the consequences of preferred network structures on care for patients with clinical and social complexity.

While preferred status may create some obligation for an SNF to broadly accept patients from partnering hospitals, our results reinforce earlier findings that strong referral relationships reduce the likelihood of those facilities admitting patients with more complex profiles. ²¹ , ²² , ²³ , ²⁴ Our study is the first to consider this selection behavior specific to patients experiencing ADRD, a set of conditions that do pose some specific challenges to providing high‐quality postacute care. ⁹ , ¹⁰ , ¹¹ , ¹² We control as best we can for factors that would suggest patients are choosing nonpreferred facilities due to their preferences such as closer distance or presence of a memory care unit. It may be that hospitals maintain distinct preferred SNF relationships specific to placement for ADRD patients. Existing literature demonstrates that ADRD patients are more likely to go to lower‐quality SNFs, which is discouraging, but those quality measures are not specific to care for ADRD patients. ¹⁵ Identifying whether and how certain facilities are achieving better outcomes and are a choice care option for ADRD patients, even and especially if they do not have designated memory care units, presents a critical opportunity for future research.

We also find that effects are stronger when we consider more versus less strongly preferred SNFs and when we consider patients with more advanced dementia. These aspects of our approach/findings increase our confidence that effects are being driven by selection behavior among SNFs. Such behavior is reasonable to expect given that SNFs struggle under high regulation, chronic underresourcing, and associated rates of turnover in staff and leadership. ³⁵ However, our findings are concerning because patients with ADRD could disproportionately benefit from the presumably stronger coordination processes that can be implemented and systematized in handoffs between institutionalized settings that share a high volume of patients. ¹⁶ , ¹⁸

Hospitals and SNFs share the responsibility for patients with cognitive impairments receiving optimal discharge planning and high‐quality transitional care. The needs of these patients are often underrepresented or even excluded when organizations develop strategies to improve care transitions. ¹² , ³⁶ From the hospital perspective, this amplifies general challenges that discharging physicians experience when referring patients to postacute care with little knowledge of the SNF care environment or feedback about the appropriateness of placement decisions. ⁴ Hospitals can leverage the existing infrastructure of integrated care relationships (i.e., regular joint leadership meetings) to better understand and support the specific needs that SNFs have when caring for patients with cognitive impairment. If hospitals can better meet these needs—for example, by providing memory care support staff who spans the transitional care window—SNFs may be less hesitant to accept these patients. Hospitals might also need to consider modifying existing measurement efforts that track the performance of partnering SNFs to capture these behaviors. Finally, there are some actions that can be taken to rectify uneven investments by hospitals across SNF partners in transitional care activities. For example, hospital information sharing practices are generally lacking and are even less robust with nonpreferred SNFs. ⁶ , ⁷ , ⁸ Hospitals can revise standard policies and EHR‐supported discharge tools to provide more timely, detailed, and usable information to all SNF partners to reduce the potential gaps in care that result from poor information sharing.

From a policy perspective, this type of selection behavior suggests that risk‐adjusted payment methods may be insufficient to ease facility concerns. One consideration is that robust care for patients with ADRD requires larger fixed‐cost investments (i.e., designated staff) not easily covered by incremental additions to fixed payments. Policymakers may need to consider other mechanisms to disseminate needed resources (e.g., specialized staffing) across facilities. In addition to these enabling activities, oversight of selection behavior may be needed as part of regulatory or accreditation activity if future research continues to show equity concerns in where patients receive care and the quality of that care.

4.1. Limitations

This study has several limitations. First, we cannot determine the actual mechanism by which ADRD patients are less likely to be referred to preferred SNFs, as we did not conduct any interviews with SNF or hospital staff. We also had to infer preferred relationships based only on the shared volume of patient referrals. We feel somewhat reassured that misidentifying SNFs as preferred that do not have a meaningful hospital relationship would only bias our results toward the null. And, our sensitivity analysis suggests that results are consistent when we consider only the most high‐volume SNF partner (i.e., the partner most likely to be benefiting from any volume‐directed transitional care investments). However, this work would be strengthened by the availability of data that explicitly confirms the presence and nature of these relationships, including the availability of shared financial incentives, cross‐covering providers and staff, facilitated information sharing, etc. ⁸ , ¹⁶ , ¹⁸

Second, while we were able to match patients on several characteristics, there may still be unobserved characteristics that identify differences between the patients that an admission coordinator may see in a referral note or through a warm handoff with a discharge planner. Finally, there may be specialized staffing at SNFs that are accepting ADRD patients that are not observable in facility‐level survey data. Whether there are unobserved characteristics that make some facilities preferred specifically for ADRD patients is a critical opportunity for future work.

5. CONCLUSION

Using a matched sample of the observed hospital‐to‐SNF discharges, we find that the effect of preferred organizational relationships on patient placement is weaker for patients with ADRD. This possible restriction of choice complicates an already fraught transition for patients and their families. Identifying these patterns is a first step toward guiding more intentional investments aimed at reducing the need for SNFs to engage in selection behavior. Policy intervention may be needed to address selection behavior, to align specialized resources with where ADRD patients are concentrated, and to support hospitals in making systemic improvements that can better benefit all SNF partners.

FUNDING INFORMATION

This study was funded by the National Institute on Aging of the National Institutes of Health under grant no. R03AG072215 (MPI: McHugh, Cross).

Supporting information

Data S1: Supporting Information.

Cross DA, Bucy TI, Rahman M, McHugh JP. Access to preferred skilled nursing facilities: Transitional care pathways for patients with Alzheimer's disease and related dementias. Health Serv Res. 2024;59(2):e14263. doi: 10.1111/1475-6773.14263