The effect of functional status impairment on nursing home admission risk among patients with advanced Parkinson’s disease

Abstract

Aims: To estimate the relationship between functional status (FS) impairment and nursing home admission (NHA) risk in Parkinson’s disease (PD) patients, and quantify the effect of advanced PD (APD) treatment on NHA risk relative to standard of care (SoC).

Materials and methods: PD patients were identified in the Medicare Current Beneficiary Survey (MCBS) (1992–2010). A working definition based on the literature and clinical expert input determined APD status. A logit model estimated the relationship between FS impairment and NHA risk. The effect of levodopa-carbidopa intestinal gel (LCIG) on NHA risk relative to SoC was simulated using clinical trial data (control: optimized oral levodopa-carbidopa IR, ClinicalTrials.gov NCT00660387 and NCT0357994).

Results: Non-advanced PD and APD significantly increased NHA risk when controlling for demographics (p < 0.01). APD status was no longer significant after controlling for FS limitations, implying that FS limitations explain the increased NHA risk in APD patients. Reduced impairment in FS in patients with APD treated with LCIG reduced risk of NHA by 13.5% relative to SoC.

Limitations: This study applies clinical trial results to real-world data. LCIG treatment might have a different effect on NHA risk for the nationally representative population than the effect measured in the trial. Both data sources employ different instruments to measure FS, instrument wording and study follow-up differed, which might bias our estimates. Finally, there lacks consensus on a definition of APD. The prevalence of APD in this study is high, perhaps due to the specific definition used.

Conclusions: Patients with APD experience a higher risk in NHA than those with non-advanced disease. This increased risk in NHA in patients with APD is explained by greater limitations in FS. The relative reduction in risk of NHA for the APD population treated with LCIG is quantitatively similar to doubling Medicaid home care services.

Keywords:

• Long-term care
• disability
• activities of daily living
• functional health status
• Parkinson’s disease
• advanced Parkinson’s disease
• nursing home admission

JEL CLASSIFICATION CODES:

• I00
• I10

Introduction

The number of Americans aged 65 and over is projected to more than double between 2014 and 2060, driving an increased need for nursing home stays and other forms of elderly assistance¹. Spending on long-term care constitutes a sizeable economic burden in the United States (US)². For example, Medicaid fee-for-service spending on nursing home facilities was $49.8 billion in 2014, or 10.1% of total Medicaid spending^3,4. The Congressional Budget Office has estimated that spending on long-term care for the elderly as a share of Gross Domestic Product will nearly double from 1.3% to 3.0% between 2010 and 2050².

Neurodegenerative disorders, such as Alzheimer’s disease (AD) and other dementias, and Parkinson’s disease (PD) are potentially important contributors to nursing home admission (NHA)^5,6. After AD, PD is the most common neurodegenerative disorder in the US⁷, and its prevalence is expected to grow over the next 15 years⁸. PD is a progressive movement disorder characterized by motor fluctuations and periods of muscle stiffness or rigidity⁹. In the advanced stages of PD, limitations in functional status (FS) worsen and the need for assistance with day-to-day activities increases¹⁰. Limitations in FS are a key predictor of NHA rates^11,12. Examples of FS limitations include difficulty in activities of daily living (ADLs), such as walking, dressing, bathing, etc.; and difficulty in instrumental activities of daily living (IADLs), such as preparing meals, shopping, using the telephone, etc. Fong et al.¹¹ studied a cohort of Americans aged 50+ and examined the relationship between NHA risk and ADL limitations, including dressing, walking, bathing, eating, toileting, and getting in and out of bed. They found that individuals with at least one ADL limitation experienced significantly higher NHA risk than those without ADL limitations. Gaugler et al.¹² performed a meta-analysis that examined the predictors of NHA among older individuals in the US. Having three or more ADL dependencies was one of the strongest predictors of NHA.

In this study, the relationship between FS limitations and NHA risk was examined for the non-advanced PD and advanced PD (APD) populations. Because individuals with APD may have a greater number of and more severe FS limitations, they may be at a higher risk of NHA. In addition, the effect of APD treatment relative to the standard of care (SoC) (optimized oral levodopa-carbidopa IR) on reducing NHA risk was quantified. This question is currently salient because new therapies can improve FS among these patients with APD and thus potentially mitigate some of the NHA risk associated with the disease. Duopa™ (levodopa-carbidopa intestinal gel [LCIG]), a treatment that demonstrates improvement in the ability to conduct ADLs (as measured by the Unified Parkinson's Disease Rating Scale [UPDRS] Part II subscale) has recently been approved by the US Food and Drug Administration^13,14. We modeled the determinants of NHA risk for APD patients, including FS limitations, and the potential effect of LCIG treatment relative to the SoC on NHA risk among the APD population.

Methods

Sample

Data from the Medicare Current Beneficiary Survey (MCBS) Cost and Use files from 2002–2010 were used to identify PD and APD cohorts. The MCBS is a nationally representative rotating panel of Medicare beneficiaries. Respondents in the MCBS were followed up to 4 years, and new survey respondents were added annually to preserve the representativeness of the sample. The MCBS contains comprehensive self-reported information on the health status, chronic conditions and disabilities, healthcare use and expenditures, and socioeconomic and demographic characteristics of Medicare beneficiaries. The survey data were linked to Medicare claims to track medical care utilization including skilled nursing home services, hospice care, and other medical services. Our sample consists of all individuals documented in the 2002–2010 waves of the MCBS, constituting 100,367 records in the data.

To simulate the effect of LCIG treatment on NHAs, our analysis incorporated results from a 12-week prospective, Phase 3, randomized, double-blind, double-dummy clinical trial that evaluated the efficacy, safety, and tolerability of LCIG treatment in levodopa-responsive APD subjects who continue to experience persistent motor fluctuations despite receiving optimal treatments for PD (ClinicalTrials.gov NCT00660387 and NCT0357994)¹³. The trial compared outcomes of individuals receiving LCIG treatment to individuals receiving SoC, i.e. optimized oral levodopa-carbidopa immediate release. Any individual was eligible to participate in the study if they were at least 30 years old and had a diagnosis of PD according to the UK Brain Banking criteria. Ninety-seven individuals were assessed for eligibility, of which 26 were excluded due to protocol violation, withdrawn consent, or adverse events. The remaining 71 were 1:1 randomized to receive either LCIG treatment (n = 37) or SoC (n = 34). Sixty-six individuals completed the study. Five individuals discontinued (two in the treatment group and three in the control group) due to adverse events, protocol violation, or a lack of efficacy. More details regarding the clinical trial are provided in Olanow et al.¹³.

Study variables

To model the determinants of NHA risk, the dependent variable of interest in our analysis was admission into a nursing home in the next year. Respondents who died between interviews were not included, as their facility status in the second interview would be unknown.

The independent variables of interest were (A)PD status and limitations in FS. In addition, we included demographic characteristics as controls. Individuals were categorized as non-PD, having non-advanced PD, or having APD. Each individual was categorized in the survey year analyzed. Individuals with PD could be directly identified in the MCBS data. To identify individuals with APD within the PD cohort, a working definition was developed based on the APD indicators identified through a Delphi panel study and input from a clinical expert¹⁵. Individuals were classified as having APD if (1) they were diagnosed with PD, and (2a) were unable to walk unassisted or (2b) experienced more than three falls in the last 12 months. Individuals with PD that did not fulfill either criteria 2a or 2b, were classified as having non-advanced PD. As this definition of APD is ambulation anchored, it may be subject to ambulation confounders that are non-motor, such as dementia and hip fractures. We did not exclude individuals with dementia and hip fractures from the study sample as the purpose of this study is to estimate the relationship between FS limitations and NHA risk in a nationally representative PD population. In addition, having “mild” dementia is one of the APD indicators¹⁵, and hip fractures may be due to repeated falls. However, we explored the possible effect of these confounders by doing a sub-group analysis excluding those with dementia or hip fractures.

FS limitations were defined as having difficulty with ADLs and IADLs. Several ADLS and IADLs available in the MCBS were excluded from the analysis due to multicollinearity issues. Activities that were highly correlated with other predictors of NHA risk were examined for multicollinearity by observing the stability of the coefficients when removing individual variables¹⁶. Specifically, we calculated the variance inflation factor (VIF_i) for an activity i by calculating VIF_i = 1/(1–R_i²) where R_i² is the coefficient of determination obtained by regressing activity i on the other predictors in the model. When i is perfectly uncorrelated with the other predictors, VIF = 1. A higher VIF indicates higher correlation between an activity of interest and the remaining predictors in the regression. There is no consensus on what constitutes a problematically high VIF, so we investigated all activities such that with a VIF_i >2. For each of these activities, we compared the coefficients of the remaining predictors when the activity was left in the regression versus when it was removed. If coefficients changed dramatically, the activity was removed from the model. As a result, the model only incorporated five ADLs (dressing, walking, toileting, and getting out of a bed or a chair) and four IADLs (using a telephone, paying bills, doing light housework, and doing heavy housework). All ADLs and IADLs were included as indicator variables for whether or not an individual had difficulty with a certain activity. ADLs and IADLs that were excluded on this basis were bathing, eating, shopping, and preparing meals.

Demographic variables that could be potential determinants of NHA risk were used as covariates. Specifically, the model included age, gender, race/ethnicity (black non-Hispanic, non-black Hispanic, other), education (less than high school, high school graduate, college or more), smoking status, and marital status (widowed, other).

To simulate the effect of LCIG treatment relative to SoC on NHA risk, the effect of LCIG treatment on FS as measured in the clinical trial data was used¹³. Specifically, ADL limitations were measured as a secondary endpoint in the clinical trial by the UPDRS Part II. The UPDRS Part II was administered at treatment initiation and at the conclusion of the 12-week trial. The UPDRS Part II assessment includes individual activities, such as difficulty with walking, dressing, cutting, salivating, turning in bed, bathing, toileting, and swallowing¹⁷. Because the activities included in the UPDRS Part II did not perfectly align with the activities included in the MCBS data, only limitations in walking and dressing were used to simulate the effect of LCIG treatment relative to SoC on NHA risk.

Statistical analysis

The analytic approach in this project proceeded in two phases: (1) estimation of the relationship between risk of NHA and limitations in FS for individuals with non-advanced PD vs APD, and (2) using the derived NHA risk from phase 1 to simulate the effect of LCIG treatment relative to SoC on NHA risk. In phase 1, a logit model was developed to estimate NHA risk based on (A)PD status, limitations in FS, and demographics.

In order to simulate the effect of LCIG treatment relative to SoC on NHA risk in phase 2, transition rates between levels of difficulty with walking and dressing were calculated for the control and treatment groups in the clinical trial. To estimate these transition rates, a mapping was created between the limitations in the walking and dressing components of the UPDRS Part II in the clinical trial data and the corresponding MCBS questions. For limitations in dressing, the UPDRS Part II and the MCBS use similar questions, and thus the mapping was straightforward. For limitations in walking, the MCBS uses questions specifying different distances than the questions used in the UPDRS Part II. Hence, individual response categories were matched across the walking questions so that the distribution among categories was similar between the trial respondent sample and the APD cohort in the MCBS data. Once these mappings were developed, transition probabilities in each activity were calculated separately for the treatment and control groups, by starting difficulty state. Specifically, for each activity, we calculated the probability of transiting from state X (i.e. difficulty or no difficulty) to state Y as the fraction of trial subjects that transitioned from state X to Y.

It was assumed that individuals in the APD cohort of the MCBS would respond to LCIG treatment according to the transition rates observed in the clinical trial, and that the effects of LCIG treatment on difficulty in walking and dressing persisted throughout the full year. To identify the particular roles played by improvement to walking, dressing, or both, three “improvement” scenarios were defined in addition to a baseline scenario: (1) improvement to both dressing and walking, (2) only improvement to dressing, and (3) only improvement to walking. For each scenario, the simulation was performed following three steps (as illustrated in Figure 1). First, the baseline levels of difficulty in walking and dressing in the MCBS APD cohort were determined. Second, the future FS states of individuals in the MCBS APD cohort were projected using the FS transition rates from the clinical trial. In the “baseline” scenario, transition rates from the control arm of the trial were used. In the “improvement” scenarios, transition rates from the LCIG treatment arm were used. In the third step, the NHA logit model was applied to predict NHA risk as a function of projected disability. Taken together, these steps allowed us to determine how various treatment scenarios would have altered NHA risk. This process was repeated 100 times in a Monte Carlo simulation to account for the fact that FS impairment transitions in the second stage were random. For instance, if 25% of a sub-group was projected to have a decline in FS, 25% of that sub-group was randomly chosen to experience a decline in FS in the future.

Figure 1. Simulating the effect of levodopa-carbidopa intestinal gel on the Advanced Parkinson’s disease population in the Medicare Current Beneficiary Survey. Notes. This figure was previously published in a poster presentation by Shih et al.³⁴ presented at the American Academy of Neurology 2016 Annual Meeting. Abbreviations. APD, advanced Parkinson’s disease; MCBS, Medicare Current Beneficiary Survey; NHA, nursing home admission.

Sub-group analysis

As our definition of APD is ambulation anchored, the statistical analysis results may be affected by ambulation confounders that are non-motor, such as dementia and hip fractures. To check the robustness of our results, we performed a sub-group analysis that excludes individuals with dementia and hip fractures.

Results

Descriptive statistics

Following our criteria for non-advanced PD and APD status, the unweighted full sample (both non-advanced PD and APD) consisted of 1,826 survey respondents, of which 821 individuals were classified as having APD. Table 1 shows a selection of the baseline characteristics of the weighted samples of non-PD, non-advanced PD, and APD cohorts. Prevalence of APD among the full population (both non-advanced PD and APD) was estimated to be 40.7%, using the MCBS respondent-level weights.

Table 1. Baseline characteristics of the weighted non-Parkinson’s disease, Parkinson’s disease, and advanced Parkinson’s disease populations.

Measure	Non-PD	PD (non-advanced)	APD
Age (mean, 95% CI)	72.95 (72.76–73.15)	76.37 (75.48–77.27)	79.13 (78.01–0.26)***
Male (%)	44	52	45*
Black (%)	10	6	7
Hispanic (%)	8	8	5
Smoker (%)	13	7	7
Functional status
Number of falls in the past year (mean, 95% CI)	2.20 (2.08–2.31)	1.47 (1.36–1.58)	15.04 (9.23–20.86)***
Walk with device (%)	61%	73%	71%
ADLs (mean, 95% CI)	0.37 (0.36–0.38)	0.83 (0.72–0.94)	2.03 (1.93–2.13)***
IADLs (mean, 95% CI)	0.66 (0.63–0.68)	1.40 (1.25–1.56)	2.65 (2.51–2.78)***
ADLs – Difficulty with (%):
Walking	26%	47%	82%***
Bathing	15%	30%	73%***
Dressing	11%	26%	65%***
Eating	5%	13%	40%***
Toileting	5%	15%	41%***
Getting out of bed or chair	14%	34%	61%***
IADLs – Difficulty with (%):
Using a telephone	10%	24%	51%***
Paying bills	13%	26%	65%***
Light housework	14%	29%	61%***
Heavy housework	35%	56%	84%***
Comorbidities
Dementia	6%	20%	43%***
Lung disease	18%	20%	16%
Cancer	18%	19%	12%***
High blood pressure	64%	63%	57%**
Stroke	12%	23%	23%
Body Mass Index
Overweight	62%	57%	56%
Underweight	3%	3%	4%
Resident in a SNF in the past year (%)	4%	8%	14%***
Prior NHA in the past year (%)	6%	8%	39%***

Notes: Results incorporate population sampling weights. (Unweighted) Records for non-advanced PD population (1,826); records for ADP population (821). Variation in health conditions, smoking, and obesity were not significant and are omitted for brevity.

Statistically significantly different from the non-advanced PD cohort at ***p < 0.01; **p < 0.05; *p < 0.1.

Abbreviations. ADL, activities of daily living; APD, advanced Parkinson’s disease; CI, confidence interval; IADL, instrumental activities of daily living; NHA, nursing home admission; PD, Parkinson’s disease; SNF, skilled nursing home facility.

Individuals in the APD cohort were significantly more likely to be older and have dementia than individuals in the non-advanced PD cohort (p < 0.01). However, they were significantly less likely to be male (p < 0.1), have cancer (p < 0.01), or have high blood pressure (p < 0.05) than those in the non-advanced PD cohort. In addition, individuals with APD were significantly more likely to have been admitted to a nursing home or have been a resident in a skilled nursing home facility in the past year (p < 0.01 for both). Looking at FS measures, individuals with APD had significantly more difficulty in ADLs and IADLs (p < 0.01 for both), and were significantly more likely to experience a limitation in any given ADL or IADL (p < 0.01 for all) than individuals with non-advanced PD. Individuals with APD also had a significantly higher mean number of falls in the prior year, 15.04 vs 1.47 falls (p < 0.01), than those with non-advanced PD. The significant increased likelihood in experiencing FS limitations could be due to the definition of APD used in this study, i.e. whether an individual was unable to walk unassisted or experienced more than three falls in the last 12 months. We examined the pairwise correlations between APD status and limitations in FS in Table 2. APD status is positively correlated with all FS limitations, except for walking with a device, ranging from 0.268–1.000. APD status has perfect correlation with having more than three falls in the past year and high correlation with the inability to walk (0.635). Most FS limitations are positively correlated with each other to some degree. The number of falls in the past year, more than three falls in the past year, and walking with a device shows low correlation, and in some cases negative, with all FS limitations.

Table 2. Correlation between advanced Parkinson’s disease status and limitations in functional status.

	Advanced Parkinson’s disease	Difficulty with dressing	Difficulty with walking	Difficulty with toileting	Difficulty getting out of bed or a chair	Difficulty with using a phone	Difficulty with paying bills	Difficulty with light housework	Difficulty with heavy housework	Difficulty with bathing	Difficulty with eating	Number of falls in the past year	At least 3 falls in the past year	Unable to walk	Walking with device
Advanced Parkinson’s Disease	–
Difficulty with dressing	0.426	–
Difficulty with walking	0.312	0.448	–
Difficulty with toileting	0.283	0.572	0.398	–
Difficulty getting out of bed or a chair	0.268	0.578	0.573	0.557	–
Difficulty with using a phone	0.310	0.513	0.240	0.406	0.293	–
Difficulty with paying bills	0.414	0.531	0.320	0.462	0.338	0.517	–
Difficulty with light housework	0.309	0.513	0.464	0.481	0.481	0.382	0.503	–
Difficulty with heavy housework	0.269	0.333	0.482	0.297	0.368	0.223	0.353	0.553	–
Difficulty with bathing	0.448	0.737	0.498	0.567	0.543	0.486	0.571	0.570	0.380	–
Difficulty with eating	0.347	0.571	0.279	0.489	0.378	0.469	0.436	0.331	0.187	0.485	–
Number of falls in the past year	0.439	0.026	0.071	0.039	0.021	0.065	0.037	0.103	0.080	0.116	−0.014	–
At least 3 falls in the past year	1.000	0.048	0.162	0.118	0.130	−0.003	0.110	0.134	0.113	0.101	0.050	0.439	–
Unable to walk	0.635	0.485	0.472	0.364	0.354	0.325	0.429	0.476	0.412	0.527	0.350	0.040	0.042	–
Walking with device	−0.036	−0.057	NA	−0.029	−0.034	−0.114	−0.019	0.053	0.084	0.060	−0.134	0.045	0.032	0.115	–

Risk of nursing home admission in individuals with advanced Parkinson’s disease

Table 3 shows the results of the logit model examining the determinants of NHA risk. The results are presented with and without controlling for demographic characteristics and FS limitations to demonstrate the role played by FS limitations. Non-advanced PD and APD had a significant effect on NHA risk when controlling only for demographic characteristics. The odds of being admitted to a nursing home was ∼2.6-times (OR = 2.55; 95% CI = 1.79–3.62) and almost 3-times (OR = 2.99, 95% CI = 1.84–4.87) larger for non-advanced PD and APD patients, respectively, compared to individuals without PD (both significant at p < 0.01). However, the effect of APD status on the risk of NHA was no longer significant when controlling for FS limitations (OR = 0.95; 95% CI = 0.58–1.55). Among difficulty in ADLs and IADLs, having difficulty with walking (OR = 1.68; 95% CI = 1.41–2.01), paying bills (OR = 3.15; 95% CI = 2.58–3.83), light housework (OR = 1.39; 95% CI = 1.17–1.66), or heavy housework (OR = 1.51; 95% CI = 1.26–1.82) significantly increased NHA risk (p < 0.01 for all). By contrast, non-advanced PD still significantly increased NHA risk after controlling for difficulty in ADLs and IADLs (OR = 1.76; 95% CI = 1.21–2.55; p < 0.01). As such, the risk of NHA for the non-advanced PD population would be under-estimated when only using difficulty in ADLs and IADLs and demographics as predictors. As earlier results showed (Table 1), FS impairment is less prevalent among the non-advanced PD population and may, thus, be a relatively less important predictor to NHA risk than other factors, like family circumstance and other unobservables.

Table 3. Results of the nursing home admission model.

	PD status only	Adding demographics	Adding activities
	OR (95% CI)	OR (95% CI)	OR (95% CI)
Demographics^a	N	Y	Y
PD, non-advanced	2.78*** (2.13–3.63)	2.55*** (1.79–3.62)	1.76*** (1.21–2.55)
APD	6.32*** (4.61–8.66)	2.99*** (1.84–4.87)	0.95 (0.58–1.55)
Difficulty with:^b
Walking			1.68*** (1.41–2.01)
Dressing			1.10 (0.90–1.35)
Going to the toilet			1.11 (0.91–1.35)
Getting out of bed/chair			1.10 (0.92–1.32)
Using a phone			1.05 (0.88–1.25)
Paying bills			3.146*** (2.58–3.83)
Light housework			1.391*** (1.17–1.66)
Heavy housework			1.514*** (1.26–1.82)
N^c	156,403	118,183	117,818

Notes. ^aIncluded demographics are indicators for male, non-Hispanic black, Hispanic, high school graduate, college graduate or more, current smoker, widow. In addition, age is included as a covariate.

^bHaving difficulty with a certain activity is a yes/no question.

^cN is the weighted sample size.

Statistically significant at ***p < 0.01; **p < 0.05; *p < 0.1.

Abbreviations. APD, advanced Parkinson’s disease; PD, Parkinson’s disease; OR, Odds ratio; CI, Confidence interval.

Table 4 shows the absolute and relative risks of NHA for the non-PD, non-advanced PD, and APD populations. Individuals with APD had an absolute risk (AR) of 5.3%, compared to 3.6% and 1.2% for individuals with non-advanced PD and individuals without PD, respectively. This translates to the NHA risk of the APD population being 1.5- and 4.4-times higher than the non-advanced PD and non-PD populations, respectively. In turn, the non-advanced PD population had a NHA risk 3-times higher than the non-PD population.

Table 4. Risk of nursing home admission in individuals with advanced Parkinson’s disease and Parkinson’s disease in the Medicare Current Beneficiary Survey.

Disease status	Absolute NHA risk (%)	Risk relative to non-PD	Risk relative to non-advanced PD
Non-PD	1.2%	–	–
Non-advanced PD	3.6%	3.0	–
APD	5.3%	4.4	1.5

Abbreviations. APD, advanced Parkinson’s disease; NHA, nursing home admission; PD, Parkinson’s disease.

Effect of levodopa-carbidopa intestinal gel on limitations in functional status and nursing home admission risk

To simulate the effect of LCIG treatment relative to SoC on FS limitations and risk of NHA among the APD population, transition rates were calculated for having difficulty with dressing and/or walking. Figure 2 displays the transitions observed from the clinical trial between having no difficulty vs any difficulty in walking and/or dressing, for the treatment and control groups. For walking, individuals using LCIG treatment were less likely to develop difficulties with walking or continue to have difficulties with walking relative to individuals in the control group. For dressing, individuals using LCIG treatment were less likely to continue to have difficulties with dressing relative to individuals in the control group.

Figure 2. Transition rates for difficulty with walking and dressing from levodopa-carbidopa intestinal gel randomized controlled clinical trial^a.

Notes: ^a Transition rates from difficulty with walking and dressing estimated from a 12-week, prospective, Phase 3, randomized, double-blind, double-dummy, clinical trial that evaluated the efficacy, safety, and tolerability of levodopa-carbidopa intestinal gel treatment in levodopa-responsive advanced Parkinson’s disease subjects who continue to experience persistent motor fluctuations despite receiving optimal treatments for Parkinson’s disease¹³.

Figure 3 shows the results on absolute risk of NHA from the different treatment scenarios for APD patients receiving LCIG treatment relative to SoC. The absolute risk of NHA for APD patients on oral treatment was 5.2% in the baseline scenario (e.g. when we apply transition probabilities from the control group of the clinical trial), compared to 4.5% when LCIG treatment affected both difficulty in walking and dressing. This corresponds to a relative NHA risk reduction of 13.5%. Incorporating only the effect of LCIG treatment on improvement to dressing led to a 5.3% absolute risk of NHA, slightly higher than the baseline risk. When incorporating only the effect of LCIG treatment on improvement to walking, the absolute NHA risk was estimated to be 4.5%, equivalent to the risk when both walking and dressing were improved.

Figure 3. The effect of levodopa-carbidopa intestinal gel on nursing home admission risk under different treatment scenarios.

Notes: The baseline scenario applies to patients with APD on standard of care (optimized oral levodopa-carbidopa IR), and is based on the transition rates of control subjects provided in Figure 2. The other scenarios apply to patients with APD on levodopa-carbidopa intestinal gel.

Abbreviations: APD, advanced Parkinson’s disease; NHA, nursing home admission.

Sub-group analysis

Supplementary Table 1 shows the results of the logit model examining the determinants of NHA risk for the sub-population where individuals with dementia or hip fractures were excluded. Similar to the main results, non-advanced PD and APD status significantly increase NHA risk when controlling for demographics (both p < 0.01), but the magnitude of these increases differ. The odds of being admitted to a nursing home decreased from being 2.6-times larger to 2-times larger for non-advanced PD patients compared to those without PD (OR = 1.94; 95% CI = 1.23–3.06); and the odds for patients with APD increased from being 3-times larger to being 5-times larger compared to those without PD (OR = 5.09, 95% CI = 2.92–8.89). APD no longer has a significant effect when controlling for difficulties with ADLs and IADLs (OR = 1.63; 95% CI = 0.89–2.99). In contrast to the main analyses, non-advanced PD status also no longer has a significant effect when controlling for FS limitations (OR = 1.43; 95% CI = 0.86–2.37). Among difficulty in ADLs and IADLs, having difficulty with walking (OR = 1.73; 95% CI = 1.43–2.08), paying bills (OR = 2.59; 95% CI = 2.05–3.27), light housework (OR = 1.51; 95% CI = 1.22–1.86), or heavy housework (OR = 1.49; 95% CI = 1.23–1.81) also significantly increased NHA risk in the sub-group analysis (all significant at p < 0.01). In contrast to the main analysis, going to the toilet (OR = 1.32; 95% CI = 1.05–1.65) and getting out of a bed or chair (OR = 1.24; 95% CI = 1.02–1.51) significantly increased the odds of being admitted to a nursing home (both p < 0.05). The sub-group analysis thus shows that difficulties with ADLs and IADLs determine NHA risk among both the non-advanced PD and APD populations.

Supplementary Table 2 shows the absolute and relative risks of NHA for the non-PD, non-advanced PD, and APD sub-group populations excluding those with dementia or hip fractures. The absolute risk of NHA was smaller for individuals without PD (0.9% vs 1.2%) and individuals with non-advanced PD (2.2% vs 3.6%) compared to the main analysis, and larger for individuals with APD (5.9% vs 5.3%). In addition, the relative risk of individuals with non-advanced PD to individuals without PD was smaller than in the main analysis (2.4 vs 3.0). For individuals with APD, the relative risk to individuals without PD (9.3 vs 4.4) and individuals with non-advanced PD more than doubled. As our definition of advanced disease is ambulation anchored, excluding the individuals that might be compromised in their ambulation due to other reasons than PD increased both the absolute and relative risk of NHA for the advanced disease population.

Supplementary Figure 1 shows the results on absolute risk of NHA patients from the different treatment scenarios for APD patients receiving LCIG treatment relative to SoC, excluding those with dementia and hip fractures. The absolute risk of NHA increases for all scenarios compared to the main analysis. The absolute risk for patients on oral treatment was 8.2% in the baseline scenario compared to 5.2% in the main analysis. Three scenarios were implemented for patients receiving LCIG: (1) the treatment affected both difficulty with dressing and walking, (2) the treatment affected only difficulty with dressing, and (3) the treatment affected only difficulty with walking. When the treatment affected both difficulty with walking and dressing, the absolute risk of NHA was 7.0% compared to 4.5% in the main analysis. When only affecting dressing, the absolute risk of NHA was 8.2% compared to 5.3%; and when only affecting walking 7.0% compared to 4.5%. APD individuals, excluding those with dementia and hip fractures, receiving LCIG treatment that affected both difficulty with walking and dressing experienced a relative risk reduction of 14.6% compared to the baseline (13.5% risk reduction in main analysis). Although the sub-group experienced an increased absolute risk of NHA when LCIG treatment affects both walking and dressing, the reduction in relative risk is bigger.

Discussion

This study examined the role of limitations in FS on NHA risk for the patients with non-advanced PD and APD in the Medicare population. To our knowledge, this is the first study to quantify this risk in a nationally representative US PD population (both non-advanced PD and APD). Prior studies have demonstrated that limitations in FS increase the risk of NHA for the general elderly population. As mentioned previously, Gaugler et al.¹² found having three or more ADL dependencies to be among the strongest predictors of NHA among the US elderly population. Our NHA model indicates that, aside from demographic characteristics, FS limitations significantly increase NHA risk among both non-advanced PD and APD patients. APD status is no longer significant when controlling for FS limitations, whereas non-advanced PD status remains significant. This result means that FS limitations play a dominant role in the excess NHA risk of patients with APD, and account for part of the excess NHA risk for patients with PD, relative to individuals without PD. Specifically, by incorporating FS limitations, the NHA model predicted that patients with APD were 1.5-times as likely to be admitted to a nursing home as patients with non-advanced PD, who were in turn 3-times as likely to be admitted to a nursing home relative to patients without PD. Because the MCBS data are nationally representative, these results are generalizable to the broader elderly population, with the caveat that the impact of treatment is based on clinical trial results in a controlled setting, not observational data^18,19. A follow-on analysis after sufficient real-world LCIG data are available would be a welcome addition to these findings.

This study further demonstrates that treatments that improve FS may reduce the risk of NHA in the non-advanced PD and APD populations. Specifically, when simulating the effect of LCIG treatment on difficulty in walking and dressing, NHA risk reduced by 13.5% within the APD population relative to SoC (optimized oral levodopa-carbidopa IR). Guo et al.²⁰ studied the relationship between NHA risk and Medicaid home care expenditures, which are intended to facilitate continued home residence for the elderly. They found that an increase of 100% in prior monthly Medicaid home expenditures reduced the risk of nursing home entry in the following year by 14.4%. Hence, our results showed that the relative effect of LCIG treatment on difficulty walking and dressing on NHA risk is similar in magnitude to doubling Medicaid-financed home care services.

Improvement in walking due to LCIG treatment led to reductions in NHA risk similar in size to improvements in both walking and dressing due to LCIG treatment. Thus, the effect of LCIG treatment relative to SoC on reducing NHA risk was primarily generated due to a reduction in difficulty in walking. This result is likely due to (1) the insignificant effect of difficulty in dressing on NHA risk in the NHA model, (2) the ambiguous effect of LCIG treatment relative to SoC on the transition probabilities for difficulty in dressing, and (3) the lower prevalence of difficulty dressing vs difficulty walking within both the non-advanced and advanced PD populations.

Admission to a nursing home has important implications for quality-of-life and healthcare costs^21–23. For example, Kaltenboeck et al.²³ found that PD patients who had received their first claim for a skilled nursing facility had excess Medicare claims of $102,750 over five years relative to their non-PD counterparts. By comparison, patients with PD receiving their first claim for an ambulatory assistance device had relatively lower excess claims, $50,923 over 5 years when compared to their non-PD counterparts. Reduced NHA risk among advanced PD patients due to treatments like LCIG, may therefore offset the overall healthcare burden associated with APD. Kowal et al.²⁴ found that the PD population had $8.1 billion ($12,800 per capita) excess medical expenses compared to the non-PD population, of which 57% were due to higher use of nursing home services.

This study had several notable limitations. This study applies clinical trial results to real-world data. The population used in the trial does not match the nationally representative population, and hence LCIG treatment might have a different effect on NHA risk for the nationally representative population than the effect on NHA risk as measured in the trial. Further, the LCIG clinical trial and the MCBS employed different instruments to measure FS limitations. As a result, some of the ADLs examined in the clinical trial were not available in the MCBS. This led to four major challenges. First, the effect of LCIG treatment relative to SoC could only be estimated for difficulty in walking and/or dressing, rather than all FS limitations available in the data. Since the clinical trial results showed that LCIG treatment also reduced limitations in other activities, our approach may have under-estimated the effect of LCIG relative to SoC treatment on NHA risk. Second, the walking question used in the clinical trial differed from the walking question used in the MCBS. This disparity was addressed by re-categorizing the levels of walking difficulty used in the MCBS so that the resulting distribution for the APD population matched the distribution among the clinical trial subjects. However, this mapping between questionnaires likely resulted in some error. Third, the estimation of the effect of LCIG treatment relative to SoC on difficulty in a particular ADL incorporated the broad categories of “difficulty” or “no difficulty” from the clinical trial. As a result, it was not possible to estimate progress or decline in difficulty for patients that experienced difficulty throughout the trial period. The simulated effect of LCIG treatment relative to SoC on improvement to walking and dressing may, thus, be under-estimated. Fourth, variations in the survey instruments may have generated different responses. Potential differences in the format of the survey, the order of questions, etc., may lead to error between our mappings of the outcomes from the trial to the MCBS. Such error would generate bias in our estimates of the effect of LCIG on nursing home admission. Finally, the trial data had a 3 months follow-up, whereas the MCBS data was measured annually. In our simulation, the effect of LCIG treatment was assumed to last a full year. This assumption may lead to some error, but was supported by literature demonstrating persistence of the benefits from LCIG treatment for a year or longer^25–27.

Another important limitation is the lack of consensus around a clinical definition of APD. This study applied recent results from the literature and clinical expert input to define APD¹⁵. APD was defined as (1) individuals who were diagnosed with PD, and (2a) were unable to walk unassisted or (2b) experienced more than three falls in the last 12 months. While prior studies had similarly employed inability to walk unaided or the use of an ambulatory assistance device to categorize PD patients as having “advanced” disease, our results might have changed under alternative definitions of APD^23,28. In particular, we estimated a rather high prevalence (41.7%) of advanced disease among the full sample (both non-advanced PD and APD), which may be driven due to the definition of APD. In addition, our definition of APD may make our results subject to ambulation confounders that are non-motor. To check the robustness of our results, we performed a sub-group analysis that excludes individuals with dementia and hip fractures. The sub-group analysis shows that difficulties with ADLs and IADLs determine NHA risk among both the non-advanced PD and APD populations. In addition, both the absolute and relative risk of NHA were bigger for the APD population compared to the main analysis. Excluding the individuals that might be compromised in their ambulation due to other reasons than PD increased the relative risk reduction for those receiving LCIG treatment that affected both difficulty with walking and dressing to those receiving SoC compared to the main analysis.

Future research should consider the comprehensive effect of APD treatments on ADLs and IADLs in reducing NHA risk and estimate the sensitivity of NHA risk to alternative APD definitions. A full understanding of the various contributors to NHA risk may help patients and practitioners assess individual risks better²⁹. In particular, Gaugler et al.¹² found cognitive impairment, in addition to having at least three ADL dependencies, to be a strong predictor of NHA. Further, studies showed an association between cognitive impairment and limitations in ADLs and IADLs^30,31. Cognitive impairment was not directly incorporated in our analysis. However, the significant effect of IADL limitations on NHA risk suggests that cognitive impairment could be an important determinant of NHA. Specifically, difficulty with paying bills had the largest effect on NHA risk among all IADLs in the current study. These results suggest that the relationship between cognitive impairment and NHA among the non-advanced PD and APD populations is an important direction for future research. Results from other studies suggest a relationship between cognitive impairment or dementia and NHA^32,33. As a related matter, future studies might also assess the extent to which treatments reduce the risk of NHA by reducing the symptoms of cognitive decline in non-advanced PD and APD patients.

Conclusions

Spending on nursing homes presents a major fiscal challenge as the population ages. Patients with non-advanced PD and APD are at particular risk for nursing home entry, due in large part to the impairment of FS: the prevalence of impairment in (I)ADLS is higher in the PD (both non-advanced and advanced) population than the overall Medicare population, and still higher among only the APD patients. Mitigating these impairments might help reduce the overall risk of nursing home entry and the size of the nursing home population. Treatments such as LCIG that have been shown to improve functional status in clinical trials may thus potentially reduce the risk of NHA. The relative reduction in risk of NHA for the APD population treated with LCIG is quantitatively similar to doubling Medicaid homecare services.

Transparency

Declaration of funding

This work was supported by AbbVie (https://www.abbvie.com). The design, study conduct, and financial support for the study were provided by AbbVie.

Declaration of financial/other relationships

Kavita R. Sail, Thomas S. Marshall, and Yash J. Jalundhwala are employees of AbbVie and may own AbbVie stock or stock options. Jeffrey Sullivan and Emma van Eijndhoven are employees of Precision Health Economics (PHE), a consulting firm providing services to the life sciences industry. PHE received compensation from Abbvie to conduct this study. Tiffany M. Shih was a former employee of PHE during the time of the study. Darius N. Lakdawalla holds equity in Precision Medicine Group, parent company of PHE. Cindy Zadikoff currently is an employee of AbbVie and may own AbbVie stock or stock options. When the study was being conducted, Cindy Zadikoff was affiliated with Feinberg School of Medicine, Northwestern University, Chicago, IL, and has previously received honoraria for consulting and lecturing from AbbVie. In addition, she has received consulting honoraria from various biopharmaceutical companies. JME peer reviewers on this manuscript have received an honorarium from JME for their review work, but have no other relevant financial relationships to disclose.

Previous presentations

This work was previously presented at the 2016 AAN Annual Meeting in Vancouver, BC, Canada on April 15–21, 2016.

Acknowledgements

We would like to thank Suepattra May-Slater, employee of PHE, and Warren Stevens, Jennifer Benner, and Shalak Gunjal, former employees of PHE, for their support of this study. Financial support for their services was provided by AbbVie. In addition, we would like to thank Dr Kapil Sethi for his clinical advice.

Data availability

Data was used from the Medicare Current Beneficiary Survey and a 12-week prospective, Phase 3, randomized, double-blind, double-dummy, clinical trial that evaluated the efficacy, safety, and tolerability of LCIG treatment in levodopa-responsive APD subjects who continue to experience persistent motor fluctuations despite receiving optimal treatments for PD (ClinicalTrials.gov NCT00660387 and NCT0357994). The MCBS data is DUA-restricted and the trial data is confidential.