Predictors of Nursing Home Placement in a Cohort of European People with Alzheimer’s Disease and Other Dementia Cases Enrolled in SCU-B or Non SCU-B Centers: The RECage Study

Abstract

Background:

Nursing home placement (NHP) can be the final step of patients with Alzheimer’s disease.

Objective:

We aimed to identify NHP predictors among 508 people with dementia with a 3-year follow-up.

Methods:

We analyzed data from the international observational RECage study, involving 508 people with especially Alzheimer’s disease and comparing a cohort enrolled by five centers with a Special Care Unit for BPSD (behavioral and psychological symptoms of dementia) and another one enrolled by six centers lacking this facility. The tertiary objective of the study was to assess the possible role of the SCU-B in delaying NHP. We assessed the relationship of the baseline characteristics with NHP by means of univariate analysis followed by Cox’s multivariate model.

Results:

Patients’ mean age was 78.1 years, 54.9% were women. Diagnosis mean age was 75.4 (±8.32) years; the main diagnosis was Alzheimer’s disease (296; 58.4%). During follow-up, 96 (18.9%) patients died and 153 (30.1%) were institutionalized without a statistically significant difference between the two cohorts (p = 0.9626). The mean NHP time was 902 (95% CI: 870–934). The multivariable analysis without death as a competing risk retained four independent predictors of NHP: age increase (hazard ratio (HR) = 1.023, 95% CI: 1.000–1.046), patient education level increase (HR = 1.062, 95% CI: 1.024–1.101), Neuropsychiatric Inventory total increase (HR = 1.018; 95% CI: 1.011–1.026), and total Mini-Mental State Examination as a favorable factor (HR = 0.948, 95% CI: 0.925–0.971). Gender (females versus males: HR = 1.265, 95% CI: 0.899–1.781) was included in the final Cox’s model for adjusting the estimates for.

Conclusions:

Our data partially agree with the predictors of NHP in literature including the effect of high education level. No caregivers’ factors were statistically significant.

Clinical trial registration:

NCT03507504.

Keywords

Alzheimer’s disease dementia nursing home placement predictive factors

INTRODUCTION

The predictors of institutionalization (nursing home placement, NHP) of the elderly have been repeatedly studied during the last decades. Among many studies, a remarkable 12-year epidemiological study in the USA [1] identified dementia as the single major predictor (hazard ratio (HR)=5.09, 95% confidence interval (CI): 2.92–8.84), measured as a time-dependent variable. In addition, older age (HR = 1.06, 95% CI: 1.03–1.10); Instrumental Activities of Daily Living disability (HR = 1.31, 95% CI: 1.15–1.50); worse/less social support (HR = 1.27, 95% CI: 1.10–1.46); and number of prescription medications (HR = 1.21, 95% CI: 1.11–1.32), measured at baseline were statistically significant.

A recent study [2] on 2013 Belgian health interview of 1,930 people in community-dwelling followed from 2012 to 2018 calculated the NHP risk using a competing risk analysis. Older age, low educational attainment, living alone, and use of home care services were significantly associated with a higher risk of NHP, while a number of need factors (e.g., history of falls, suffering from urinary incontinence, depression or Alzheimer’s disease (AD)) were removed from the final multivariable model.

Since dementia has been proven to be one of the major predictors of NHP, many researchers focused on identification of the predicting factors within the population of people with dementia. In a cohort of 196 community-dwelling patient/caregiver dyads with dementia of various etiology and of moderate severity (mean baseline Mini-Mental State Examination (MMSE) of 15) [3], the variables found to predict institutionalization at 18 months were: a) caregiver health and burden; b) use of services; c) enjoyment of caregiving; d) care receiver cognitive function; e) troublesome behaviors (particularly aggression and incontinence problems); f) caregiver reaction to behaviors.

The risk of institutional care in patients with AD (n = 240) and other dementias (OD, n = 240) and age-matched controls (n = 363) was studied [4] in an enumerated population of USA. The adjusted HR of being admitted to a nursing home compared with controls was 5.44 (95% CI: 3.68–8.05) for AD cases and 5.08 (95% CI: 3.38–7.63) for OD cases.

Another paper [5] studied patient and caregiver characteristics related to NHP in people with dementia. Black ethnicity (HR = 0.60; 95% CI: 0.48–0.74) and Hispanic ethnicity (HR = 0.40; 95% CI: 0.28–0.56) were inversely associated with NHP. Otherwise, living alone (HR = 1.74; 95% CI: 1.49–2.02), 1 or more dependencies in activities of daily living (HR = 1.38; 95% CI: 1.20–1.60), high cognitive impairment (MMSE score≤20: HR = 1.52; 95% CI: 1.33–1.73), and 1 or more behavioral and psychological symptoms (BPSD) (HR = 1.30; 95% CI: 1.11–1.52) were risk factors of NHP. Furthermore, caregiver characteristics associated with NHP were: age (HR = 1.17; 95% CI: 1.01–1.37 and HR = 1.55; 95% CI:1.31–1.84 for the classes of 65–74 and≥75 years, respectively), and high burden, assessed by the Zarit Burden Scale [6] score.

Other studies [7, 8] gave similar results.

Partially at variance from the above papers, de Vugt [9] in a longitudinal, prospective study of 119 community-dwelling people with dementia with a follow-up of over 2 years found that only caregiver distress related to patient behavior was a significant predictor of NHP.

A three-year longitudinal study on 970 patients with MCI or dementia recruited in the Australian PRIME study [10] confirmed that the main predictors of NHP are severity of dementia, lower cognitive ability, more severe BPSD, and greater caregiver burden.

An UK study [11] on 3,075 people with AD showed that predictors of care home or hospital admission and/or associated costs over 6 months were cognition, functional problems, agitation, depression, physical illness, previous hospitalizations, age, gender, ethnicity, living alone, and having a partner.

A systematic review of the literature on the causes of NHP in older people with dementia [12] demonstrated that impaired cognition and BPSD were consistently associated with an increased risk of NHP. Other meta-analyses [13, 14] demonstrated impairments in activities of daily living as a significant predictive factor together with caregiver burden. Furthermore, some studies showed an increased risk of NHP following hip fracture, reduced mobility, and multiple comorbidities.

The more recent MEMORA Group [15] findings suggest that a high proportion of NHP are attributable to caregiver burden and functional limitations in outpatients attending a memory center. Cognitive impairment and BPSD contribute less to NHP.

The predictors of NHP at 2 years in AD were investigated on the follow-up survey from the THERAD study [16] Five independent predictors of NHP were found: a higher patient education level (adjusted HR (aHR) = 6.31; CI 95% : 1.88–21.22), a high caregiver burden (aHR = 3.97; CI 95% : 1.33–11.85), the caregiver being the offspring of the patient (aHR = 2.92; CI 95% : 1.43–5.95), loss of autonomy (aHR = 2.75; CI 95% : 1.13–6.65), and disinhibition as a BPSD (aHR = 2.38; CI 95% : 1.26–4.47). The high patient education level, and a caregiver being the offspring of the patient were unexpected predictors of NHP.

Summarizing, we can say that some NHP predictors are clearly identified by all extant literature (age, ethnicity, severity of dementia, living alone, caregivers’ burden, degree of functional impairment), whereas other ones such as higher education profile, direct versus indirect effect of BPSD, and degree of kinship need further confirmation and/orclarification.

MATERIAL AND METHODS

Study design

We recently run an observational clinical trial on two cohorts of people with dementia and BPSD with three-year follow-up (the RECage: REspectful CAring for AGitated Elderly Clinical Trial [17]. A total of 518 patients were included, divided in two cohorts: 266 patients followed by centers endowed with a Special Care Unit for patients with BPSD (acronym SCU-B) where patients could be temporarily admitted during a behavioral crisis, and 252 patients followed by centers without SCU-B. The SCU-B and the non SCUB centers are reported in the authors’ affiliation. The primary research hypothesis of the trial was to demonstrate the possible superiority of the SCU-B cohort over the non-SCU-B one on the pattern of the total of the Neuropsychiatric Inventory rating scale (NPI) [18, 19] on 3 years of scheduled follow-up. The secondary endpoints were the comparison between the two cohorts of the pattern of several rating scales administered at baseline and at the follow-up visits, performed every six months.

Particularly, we assessed the Cohen Mansfield Agitation Inventory (CMAI), the Quality of Life-Alzheimer’s Disease (QoL-AD), the EQ-5D-5 L (patient proxy-rated and carer self-rated), the Adult Carer Quality of Life Questionnaire (ACQoL) for patients and carers, the Resource Utilization in Dementia (RUD), and the ICEpop CAPability measure for Older people proxy-rated for the health-economic outcomes. In addition, we assessed the Dementia Attitude Scale (DAS), and the Alzheimer’s Disease Cooperative Study ADL Scale (ADCS-ADL), and the Caregiver’s Burden Inventory (CBI). The tertiary endpoint was time to NHP on whose predictive model this paper is based. Since the extensive assessment (clinical and through several rating scales) at baseline, we were able to do a wide search for NHP predictors.

The study protocol was approved by Comitato Etico di Bergamo, reg. sperim, 25/18, 09.02.201.). It was also approved by the institutional review board (IRB) of the other members. Fully informed valid consent was obtained from all patients and caregivers who participated in the study; the written consent was obtained by the patients themselves, when they were competent; for those who were not due to their cognitive status, an informed consent was obtained by the legal representative/family caregiver according to the rules of the country.

Data collected

Written informed consent was obtained before inclusion by the patient and/or legal representative as well as by the caregiver. At baseline were collected demographic characteristics (age, gender, ethnicity, and educational level) for the patient as well as for the caregiver, comorbidities and dementia history, vital signs, physical and neurological examination of the patient and evaluation of the MMSE. The rating scales, administered at baseline and six-month follow-up visits, that constitute the primary and the secondary endpoints have been reported previously.

The time of the occurrence of hospital admission, death, and definitive placement to a nursing home was collected. Concomitant medications (particularly psychotropic drugs) were recorded. Data acquisition was managed through an electronic CRF provided by Mediolanum Cardio Research (MCR) Srl.

Statistical analysis

Descriptive statistics (mean, standard deviation, median, first and third quartile) have been calculated for quantitative variables. Absolute and percent frequency have been calculated for qualitative variables.

The relationship between the baseline variables and the time to NHP has been assessed by Cox’s proportional hazard regression [20], followed by the assessment of the proportionality hazard assumptions tested graphically using a plot of the log cumulative hazard, where the logarithm of time is plotted against the estimated log cumulative hazard calculated as ln[-ln (S(t))] [21] or by testing the significance of the interaction between the covariate and the natural logarithm of the follow-up time.

The two not random cohorts have been compared on the NHP event by a bivariate Cox’s model including the propensity score, calculated on all the baseline variables. The cohort, not statistically associated to the NHP event, and the propensity score have not further considered into the NHP predictive model.

Furthermore, baseline variables associated to the time to NHP with a p-value≤0.10 have been inserted into a multivariable Cox’s regression model together with the gender and age to adjust the estimates for. Then, the variables of the starting model have been eliminated according to a backward procedure until a final model with the set of the variables statistically significant independently associated with NHP, in addition to gender and age.

Then, a univariate analysis was performed considering death as a competing risk on each variable statistically associated with the NHP and on the final multivariable Cox’s model. The competing risks analysis has been carried out according to the Fine-Gray method [22] which imposes a proportional hazard assumption on the calculated subdistribution hazard functions. Therefore, the calculated “subdistribution hazard ratios” and the Cox hazard ratios give the increase (decrease) of the “subdistribution hazard” or of the hazard for each unit increase of a covariate), but are estimated on a different risk set. Particularly, the subdistribution hazards are estimated by keeping in the risk set the persons who fail from a competing cause with an assigned maximum possible follow-up value (until the end of the study, for example) so that they can be counted as if they were “cured” by the primary event of interest. Otherwise, the hazard ratios of the Cox’s model are estimated by censoring the persons at the time of the occurrence of a competing event and, consequently, as if the event of interest were the only event that can occur in the at-risk sample (more details are reported in the Supplementary Material together with the Cumulative Incidence function).

Furthermore, the predictive accuracy of the final model of the probability of not having NHP has been assessed by the concordance C-statistics from Harrell [23] and from Uno [24] together with the area under the curve statistics available in the PHREG procedure of SAS® version 9.4 [25].

RESULTS

We analyzed 508 people with dementia and BPSD with at least one assessment from baseline visit. The analyzed people were 78.1 (±7.93, standard deviation) years old, male 229 (45.1%) and female 279 (54.9%), Caucasian 498 (98.0%), and with a formal education of 8.93 (±4.53) years. Age at diagnosis was 75.35 (±8.32) years and AD was the diagnosis (296, 58.4%) more frequent.

The MMSE mean score was 15.5 (±6.25), with moderately severe/severe AD (<15) in 38.0%, moderate AD (16–20) in 37.6%, and mild AD (>20) in 24.4%.

The mean NPI Total Score was 52.4 (±18.92). The Cohen-Mansfield Agitation Inventory Factor 1 – Aggressive behavior – was mean (SD) 12.18 (±4.76), Factor 2-Physically nonaggressive behavior: 17.04 (±8.19), and Factor 3-Verbally agitated – behavior: 16.45 (±7.32).

The primary caregiver was spouse (47.7%) and child (46.0%).

The baseline variables statistically different between the two not random cohorts are reported in Table 1.

Table 1

Baseline characteristics statistically different between the two cohorts

Characteristics	Centers with SCU-B	Centers without SCU-B	p	Total
	N = 262	N = 246		N = 508
Female gender	132 (50.4%)	147 (59.8%)	0.0338	279 (54.9%)
Caregiver educational level, y	11.87 (4.06)	12.69 (3.89)	0.0269	12.24 (4.00)
Cognitive impairment
Frontal	111 (42.4%)	131 (53.3%)	0.0007	242 (47.6%)
Apraxia	103 (39.3%)	125 (50.8%)	0.0014	228 (44.9%)
Patient status
Basic activities	14.77 (5.94)	13.72 (5.72)	0.0434	14.26 (5.85)
Instrumental activities	19.82 (13.61)	22.21 (12.85)	0.0425	20.98 (13.29)
ACQoL total score	73.94 (18.44)	68.49 (17.31)	0.0007	71.28 (18.09)
CBI classes	0.0004
0-24	92 (35.5%)	51 (20.8%)		143 (28.4%)
25-36	63 (24.3%)	59 (24.1%)		122 (24.2%)
>36	104 (40.2%)	135 (55.1%)		239 (47.4%)
CBI total score	32.99 (17.88)	38.07±16.91	0.0006	35.57 (17.67)

Furthermore, the results of all the baseline variables not statistically associated to the NHP event are listed in Supplementary Material 1 and are detailed in the Table 1 of the paper of Mendes et al. [26].

The primary hypothesis was not confirmed [26]. The results on the secondary endpoints on a different pattern over the time between the two cohorts showed a statistically significant pattern in favor of the non-SCU-B cohort for the three CMAI factors, ACQoL carer, EQ-5D-5 L VAS, EQ-5D-5 L patient proxy and carer self-rated and a statistically significant pattern in favor of SCU-B cohort for the DAS; no statistically significant difference for the previously reported rating scale. More details are in the Mendes et al. paper [26].

During the follow-up, 153 patients were admitted to NH (30.1%) and 96 died (18.9%). The mean follow-up was 741.7 days (standard deviation±370.1, median of 852, first and third quartile: 414–1094 days). The follow-up has been truncated at 1,125 days corresponding to 3 years plus a temporal tolerance window of 30 days. NHP occurred for 31 pts (20.3%) in the first 180 days (20 and 11 in the SCU-B and non SCU-B, respectively) for 59 (38.6%) pts during the first year (37 and 22 in the SCU-B and non SCU-B, respectively), in 97 (63.4%) during the first 18 months (58 and 39 in the SCU-B and non SCU-B, respectively), in 121 (79.1%) during the first 2 years (67 and 54 in the SCU-B and non SCU-B, respectively). Finally, the remaining 32 NHP events (20.9%) occurred during the last third follow-up year (16 and 16 in the SCU-B and non SCU-B, respectively); globally, there have been 83 NPH and 70 NPH in the SCU-B and non SCU-B, respectively. The mean time of no NHP was 902 (95% CI: 870–934) days and the 25th percentile was 664 (95% CI: 551–825) days.

The NHP decision was usually taken by the family supported by the attending physician and/or by the specialist (neurologist, geriatrician, psychiatrist, etc.) treating the patient. It has to be noted that the nursing home costs vary by country and range between 47€ (Italy) and 252€ (Norway) per day reflecting an all-inclusive cost for care, housing, food, etc. In addition, co-payments vary by country and regulations on, for example, income; so, not only the wealthiest patients can afford the NHP.

The baseline variables not associated with NHP were the following: cohort (SCU-B versus non-SCU-B: HR = 1.161, 95% CI: 0.845–1.597, p = 0.3568; SCU-B versus non- SCU-B: HR = 1.008. 95% CI: 0.712–1.429, p = 0.9626 in the bivariate Cox’s model including the propensity score calculated for comparing the two not random cohorts in the RECage study), gender (Male versus Female: HR = 0.877, 95% CI: 0.635–1.211, p = 0.4246), age (for each unity increase HR = 0.999, 95% CI: 0.979–1.019, p = 0.9218), diagnosis of AD versus all other diagnoses of dementia (HR = 0.741, 95% CI: 0.529–1.037, p = 0.0807), age at diagnosis (for each unity increase HR = 0.996, 95% CI: 0.977–1.016, p = 0.6984), and degree of kinship of the caregiver (spouse versus child: HR = 0.941, 95% CI: 0.800–1.108, p = 0.4655), caregiver’s gender (females versus males HR = 0.830 95% CI: 0.589–1.171, p = 0.2885) and caregiver’s age (HR = 1.007 for each year increased, 95% CI: 0.994–1.020, p = 0.3074).

As regards the rating scales, the following baseline values were not statistically associated with the NHP event: CMAI Factor 1 “Aggressive behavior” (on 507 patients, for each unity increase: HR = 1.011, 95% CI: 0.978–1.045, p = 0.5120), Factor 2 “Physically nonaggressive behavior” (on 507 patients for each unity increase: HR = 1.011, 95% CI: 0.992–1.031, p = 0.2686), Factor 3 “Verbally agitated behavior” (on 507 patients, for each unity increase: HR = 1.008, 95% CI: 0.986–1.031, p = 0.4624), DAS total score (on 490 patients, for each unity increase: HR = 0.999, 95% CI: 0.989–1.008, p = 0.0740), ACQol total score (on 505 patients, for each unity increase: HR = 0.993, 95% CI: 0.985–1.002, p = 0.1300), ADCS-ADL Total score (on 507 patients, for each unity increase: HR = 0.993, 95% CI: 0.985–1.002, p = 0.1385), QoL-AD (patient rated) Total Score (on 421 patients, for each unity increase: HR = 0.990, 95% CI: 0.961–1.020, p = 0.5079), and QoL-AD (proxy rated) Total Score (on 502 patients, for each unity increase HR = 0.989, 95% CI: 0.965–1.015, p = 0.4178).

The baseline variables statistically associated with the time to NHP are reported in Table 2. In addition, the results obtained by considering death as a competitive event are shown in SupplementaryMaterial 2.

Table 2

Cox’s proportional hazard analysis. Baseline characteristics statistically associated with the NHP event with their Hazard Rations (HR), 95% Confidence Interval (95% CI), and statistically significance (p-value)

Variable	HR	95% CI	p
Educational level: for each unity (year) increase	1.050	1.016–1.086	0.0035
MMSE total for each unity increase	0.943	0.920–0.966	<0.0001
*NPI total for each unity increase	1.020	1.012–1.027	<0.0001
Height (on 414 patients) for each unity (cm) increase	0.980	0.961–0.999	0.0374
Weight (on 419 patients) for each unity (Kg) increase	0.972	0.958–0.986	<0.0001
+BMI (on 412 patients) for each unity increase	0.917	0.873–0.963	0.0005
/DBP (on 412 patients) for each unity (mm/Hg) increase	0.917	0.953–0.989	0.0018
Heart Rate (on 395 patients) for each unity increase	1.018	1.001–1.036	0.0431
CBI total score for each unity increase	1.013	1.004–1.021	0.0057

*The NPI (Neuropsychiatric Inventory) score was also calculated on the first 10 items of the scale and turned out to be statistically associated with NHP with HR = 1.024, 95% CI: 1.016-1.033, p < 0.0001. +BMI, body mass index equal to weight (Kg) divided by height (m) squared. /DBP, diastolic blood pressure; The HR for an increase of “x” units can be obtained by raising the HR to “x”.

It has to be noted that the results of the Cox’s statistical analysis without and with death as a competing risk (Supplementary Table 2) are very similar. Height, weight, body mass index, diastolic blood pressure, and heart rate, owing to their relevant prevalence of missing data, have been excluded from the starting multivariable model.

A starting multivariable model has been fitted by including gender, age, educational level, NPI total score, motor deficit, diagnosis of AD versus all other forms of dementia, MMSE total score, and CBI total score. Then, after a backward selection the variables remained in the final model were: educational level, NPI total score, and MMSE total score. In addition, gender, statistically associated with NHP only in the competing risk analysis (p = 0.0174), and age, statistically associated with NHP only in the analysis without competing risk (p = 0.0457), have been kept in the final model for having the estimates adjusted for.

Table 3 shows the descriptive baseline statistics (percentages, mean±standard deviation) of the variables remained in the final multivariable model which allow us to understand that they are statistically associated with the event NHP, owing to the differences in the two groups.

Table 3

Baseline characteristics in the two groups of Yes and No NHP event

Variable	Yes NHP	No NHP
Gender: Females	32.6%	67.4%
Males	27.1%	72.9%
Age	77.5 (±8.2)	78.3 (±7.84)
Educational level	9.82 (±4.68)	8.55 (±4.41)
NPI total	57.5 (±19.89)	50.3 (±18.09)
MMSE total	13.9 (±6.11)	16.1 (±6.21)

Table 4 shows the HR together with their 95% CI and the statistically significance (p-value) of the variables remained in the final predictive Cox’s proportional hazard model. The competing risk analysis is shown in Supplementary Material 2.

Table 4

Predictive factors of the NHP event remained in the final Cox’s predictive model

Variable	HR	95% CI	p
Gender: Females versus Males	1.265	0.899–1.781	0.1770
Age: for each unity (y) increase	1.023	1.000–1.046	0.0457
Educational level: for each unity (y) increase	1.062	1.024–1.101	0.0012
NPI total for each unity increase	1.018	1.011–1.026	<0.0001
MMSE total for each unity increase	0.948	0.925–0.971	<0.0001

The HR for an increase of “x” units (say 5 years for age and educational level, or points of the NPI and MMSE total) can be obtained by raising the HR to “x”.

It has to be pointed out that, owing to missing data the final multivariable analysis has been carried out on 501 patients instead of 508 with 149 NHP (instead of 153) and 95 deaths (instead of 96).

The variable gender and age, not statistically significant at the univariate analyses, were inserted into the final predictive model just for having the estimates adjusted for, as a standard statistical practice. Rather surprisingly, the variable age became statistically significant (p = 0.0457) with a p-value just below the statistical significance threshold. In addition, in the competing risk analysis, the gender became statistically significant (p = 0.0174) and the variable age was no longer statistically significant (p = 0.4192). It has to be noted that, excluding gender and age variables, the results of the two analyses without and with the competing risk on the remaining variables are almost similar (Supplementary Table 4). Particularly, regarding gender, in the 229 males there are 104 (45.4%) censored, 62 (27.1%) NHP, and 63 (27.5%) dead; otherwise, in the 279 females, there are 155 (55.6%) censored, 91 (32.6%) NHP, and 33 (11.3%) dead. Therefore, it is understandable that when the event death is censored in the Cox’s analysis giving NHP event in the 27.1% of males and in the 32.6% of females, the gender is not statistically significant related to the NPH event.

Regarding age, there is no simple direct interpretation except in the univariate Cox’s regression where the mean age is 78.3 years in censored or dead patients and 77.5 years in patients with NHP event, which leads to an HR lower than 1 (actually 0.992; 95% CI: 0.973–1.012) and a not statistically significant association. Finally, the mean age is 77.4 years in censored patients, 77.5 years in patients with NHP event, and 80.8 years in dead patients.

The proportionality test for the variables in the final Cox’s model without considering gender and age, turned out to be not statistically significant: p = 0.9791 and p = 0.4347 for the model without and with the competing risk, respectively. Harrell’s C (concordance) statistics was 0.6858 (95% CI: 0.6448–0.7267) a little greater than Uno’s Concordance Statistic: 0.6624 (95% CI: 0.6071–0.7177).

Figure 1 shows that the time-dependent area under the curve of the probability of not having the NHP event remains at a satisfactory level of about 0.7 in the central portion of the follow-up apart from the first and the last part where the event NHP are only a few. Particularly, the estimate of the “integrated time-dependent Area Under the Curve” is 0.7231.

Fig. 1

Time-dependent Area Under the Curve of not having the NHP event during the follow-up.

Figure 2A shows the adjusted “probability curve of not having NHP” for: (i) a patient male 70 years old with NPI total score of 45, MMSE total score of 15, and educational level of 13 years (first curve from the top); (ii) a male 80 years old with NPI total score of 50, MMSE total value of 15, and educational level of 10 years (third curve from the top), and, finally, (iii) a 90-year-old male with NPI total score of 65, MMSE total score of 15, and educational level of 18 years (fourth curve from the top). Intermediate between the first and the third curve from the top, there is the adjusted cumulative probability curve of no NHP event for a patient with the mean values of all the independent variables. Values greater than the mean of the favorable (unfavorable) prognostic covariates increase (decreases) this curve. Since the mean values are 1.55 for gender (from 1 for males and 2 for females), 78.08 years for age, 52.51 for NPI, 14.44 for MMSE, and 8.91 years for educational level, the curve is practically superimposed on the third curve with gender equal to 1 and very similar covariate values. For illustrative purposes, the two curves have been separated (at least for the second part of the follow-up) by attributing a value of 1.3 (instead of the mean value of 1.55) to gender and keeping the mean values of the other covariates equal; it has, consequently, reduced the influence of the female gender coded as 2 and made the protective effect of the male gender more evident.

Fig. 2

Adjusted functions of not having the event (NHP) for males with different values of age, NPI total, years of educational level and MMSE total. Left-2A: There is also the “function of not having the event of interest for a patient with the mean value of the covariates” (second curve from the top). Right-2B: The curve at the top is for a MMSE total value of 30 instead of 15 for the curve at the bottom; age is 90 years, NPI is 65 and Educational (Ed) years is 18 for both curves.

Figure 2B shows a curve for a 90-year-old male with a NPI total score of 65, a MMSE total score of 15, and educational level of 18 (at the bottom) and a curve for a male 90 years old with a NPI total score of 65, educational level of 18, but with a MMSE total score of 30 just for showing the impressive protective effect that an increase of 15 points of the MMSE total score can give.

DISCUSSION

Our study, concerning 508 patient/caregiver dyads well studied and followed-up for three years, confirms some data already known: females have a greater hazard than males of being institutionalized, as well as older people, patients with less severe cognitive decline (lower MMSE score), and patients with behavioral disorders (higher NPI scores).

Neither the degree of functional impairment nor the caregiver burden correlated with NHP. Furthermore, no association with the degree of kinship of the caregiver was found.

Important to remark, as in the THERAD study, we found a positive correlation between NHP and higher educational level. This finding is at variance with some previous studies [27, 28] which showed a protective effect of the education level.

Probably the socio-cultural context plays a role and, due to the composite structure of our population (pertaining to six different European countries) and the social attitude toward the nursing homes, often more positive in the Northern European countries than in Southern Europe. A significant factor might be the ability to pay for the institutions (generally correlated with the educational level) and the relatively lower costs of paid caregivers at home in some countries.

It has to be underlined that the shifting from the not statistically significant association between the gender and the NHP in the Cox’s model to the statistically significance in the Fine-Gray model for the competing risks due to the fact that death has not considered a censored event allows to stress the relevance of considering competing risks in survival analysis, particularly when they have a non-negligible prevalence (say more than 10%, actually 18.9% in this case). Finally, the different pattern of the statistically significance of age between the multivariable Cox’s model and the Fine-Gray model, owing to its relationship with the events (NHP and dead) and the other covariates, suggests the usefulness of a competing risks analysis for having a better understanding of the influence of the covariates on a predictive model. However, it has to be taken into account that the Cox’s regression censors the patients with a competing risk while the Fine-Gray model keeps indefinitely at risk the patients who have experienced a competing event.

It must be underlined that the predictive model has been fitted on an adequate sample size while some of the other predictive models have been fitted on sample sizes less than two hundreds of patients; it is well known that the sample size is a limiting factor of a predictive model validity [29].

It is well known that the clinical validity of a predictive model must be confirmed by an external validity model on a new cohort of patients with the same disease, but with similar characteristics.

However, the moderate level of our internal validity assessment allows to conclude for a clinical relevance of our results.

Conclusion

Our data are in partial accordance with the literature in identifying the predictors of NHP. Particularly, in the multivariable model no factors related to the caregiver remained as independent statistically significant predictors of NHP. Furthermore, a rather surprising result was the protective effect of the increase of the MMSE total, maybe due to a selective effect from confounder factors.

AUTHOR CONTRIBUTIONS

Bruno Mario Cesana (Conceptualization; Data curation; Formal analysis; Methodology; Validation; Visualization; Writing – original draft; Writing – review & editing); Sverre Bergh (Supervision; Validation; Visualization; Writing – review & editing); Alfonso Ciccone (Supervision; Validation; Visualization; Writing – review & editing); Emmanuel Cognat (Supervision; Validation; Visualization; Writing – review & editing); Andrea Fabbo (Supervision; Validation; Visualization; Writing – review & editing); Sara Fascendini (Supervision; Validation; Visualization; Writing – review & editing); Giovanni B. Frisoni (Supervision; Validation; Visualization; Writing – review & editing); Lutz Froelich (Supervision; Validation; Visualization; Writing – review & editing); Ron Handels (Supervision; Validation; Visualization; Writing – review & editing); Maria Cristina Jori (Supervision; Validation; Visualization; Writing – review & editing); Patrizia Mecocci (Supervision; Validation; Visualization; Writing – review & editing); Paola Merlo (Supervision; Validation; Visualization; Writing – review & editing); Oliver Peters (Supervision; Validation; Visualization; Writing – review & editing); Magda Tsolaki (Supervision; Validation; Visualization; Writing – review & editing); Carlo Alberto Defanti (Supervision; Validation; Visualization; Writing – review & editing).

Footnotes

ACKNOWLEDGMENTS

The authors thank all the members of the RECage study group who, throughout the study, contributed to the accomplishment of this work. Also, we wish to thank all the patients, family members and caregivers taking part in this study.

FUNDING

RECage study has been funded by the European Union’s Horizon 2020 research and innovation programme under grant: H2020 GA No: 779237 H2020-SC1-2017-Single-Stage-RTD.

CONFLICT OF INTEREST

The authors have no conflict of interest to report.

DATA AVAILABILITY

The data that support the findings of this study are available on request from C.A. Defanti: carloalberto.defanti@ferbonlus.com. or after authorization of Prof. C.A. Defanti from: jori@mcr-med.com.

The supplementary material is available in the electronic version of this article: .

References

Bharucha

, Pandav

, Shen

, Hodge

, Ganguli

(2004) Predictors of nursing facility admission: A 12-year epidemiological study in the United States. J Am Geriatr Soc 52, 434–439.

Berete

, Demarest

, Charafeddine

, De Ridder

, Vanoverloop

, Van Oyen

, Bruyère

, Van der Heyde

(2022) Predictors of nursing home admission in the older population in Belgium: A longitudinal follow-up of health interview survey participants. BMC Geriatr 22, 807.

Cohen

, Gold

, Shulman

, Wortley

, McDonald

, Wargon

(1993) Factors determining the decision to institutionalize dementing individuals: A prospective study. Gerontologist 33, 714–720.

Eaker

, Vierkant

, Mickel

(2002) Predictors of nursing home admission and/or death in incident Alzheimer’s disease and other dementia cases compared to controls: A population-based study. J Clin Epidemiol 55, 462–468.

Yaffe

, Fox

, Newcomer

, Sands

, Lindquist

, Dane

, Covinsky

(2002) Patient and caregiver characteristics and nursing home placement in patients with dementia. JAMA 287, 2090–2097.

Zarit

, Reever

, Bach-Peterson

(1980) Relatives of the impaired elderly: Correlates of feelings of burden. Gerontologist 20, 649–655.

O’Donnell

, Drachman

, Barnes

, Peterson

, Swearer

, Lew

(1992) Incontinence and troublesome behaviors predict institutionalization in dementia. J Geriatr Psychiatry Neurol 5, 45–52.

Gilley

, Bienias

, Wilson

, Bennet

, Beck

, Evans

(2004) Influence of behavioral symptoms on rates of institutionalization for persons with Alzheimer’s disease. Psychol Med 34, 1129–1135.

De Vugt

, Stevens

, Aalten

, Lousberg

, Jasoers

, Verhey

FRJ

(2005) A prospective study of the effects of behavioural symptoms on the institutionalization of patients with dementia. Int Psychogeriatr 17, 577–589.

10.

Brodaty

, Connors

, Xu

, Woodward

, Ames

; PRIME study group (2014) Predictors of institutionalization in dementia: A three year longitudinal study. J Alzheimers Dis 40, 221–226.

11.

Knapp

, Chua

K-C

, Broadbent

, Chang

C-K

, Fernandez

, Milea

, Romeo

, Lovestone

, Spencer

, Thompson

, Stewart

, Hayes

(2016) Predictors of care home and hospital admissions and their costs for older people with Alzheimer’s disease: Findings from a large London case register. BMJ Open 6, e013591.

12.

Toot

, Swinson

, Devine

, Challis

, Orrell

(2017) Causes of nursing home placement for older people with dementia: A systematic review and meta-analysis. Int Psychogeriatr 29, 195–208.

13.

Cepoiu-Martin

, Tam-Tham

, Patten

, Maxwell

, Hogan

(2016) Predictors of long-term care placement in persons with dementia: A systematic review and meta-analysis. Int J Geriatr Psychiatry 31, 1151–1171.

14.

Gaugler

, Yu

, Krichbaum

, Wyman

(2009) Predictors of nursing home admission for persons with dementia. Med Care 47, 191–198.

15.

Dufournet

, Dauphinot

, Moutet

, Verdurand

, Delphin-Combe

, Krolak-Salmon

, the MEMORA group (2019) Impact of cognitive, functional, behavioral disorders, and caregiver burden on the risk of nursing home placement. J Am Med Dir Assoc 20, 1254–1262.

16.

Villars

, Gardette

, Frayssignes

, Deperetti

, Perrin

, Cantet

, Soto-Martin

(2022) Predictors of nursing home placement at 2 years in Alzheimer’s disease: A follow-up survey from the THERAD study. Psychiatry 37, doi: 10.1002/gps.5724.

17.

Poptsi

, Tsolaki

, Bergh

, Cesana

, Ciccone

, Fabbo

, Frisoni

, Frolich

, Lavolpe

, Guazzarini

, Hugon

, Fascendini

, Defanti

(2021) Rationale, design, and methodology of a prospective cohort study for coping with behavioral and psychological symptoms of dementia: The RECage Project. J Alzheimers Dis 80, 1613–1627.

18.

Cummings

(1997) The Neuropsychiatric Inventory: Assessing psychopathology in dementia patients. Neurology 48(5 Suppl 6), S10–S16.

19.

Cummings

, Mega

, Gray

, Rosenberg-Thompson

, Carusi

, Gornbein

(1994) The Neuropsychiatric Inventory: Comprehensive assessment of psychopathology in dementia. Neurology 44, 2308–2314.

20.

Cox

(1972) Regression models and life tables. J R Stat Soc Series B Methodol 34, 187–220.

21.

Kleinbaum

, Mitchel

(2012) Survival Analysis: A Self-Learning Text. Statistics for Biology and Health, Third Edition. Springer, New York.

22.

Fine

, Gray

(1999) A proportional hazards model for the subdistribution of a competing risk. J Am Stat Assoc 94, 496–509.

23.

Harrell

Jr , Lee

, Califf

, Pryor

, Rosati

(1984) Regression modelling strategies for improved prognostic prediction. Stat Med 3, 143–152.

24.

Uno

, Cai

, Pencina

, D’Agostino

, Wei

(2011) On the C-statistics for evaluating overall adequacy of risk prediction procedures with censored survival data. Stat Med 30, 1105–1117.

25.

SAS Institute Inc. (2017) SAS/STAT® 14.3 User’s Guide. SAS Institute Inc., Cary, NC.

26.

Mendes

, Bergh

, Cesana

, Handels

, Ciccone

, Cognat

, Fabbo

, Fascendini

, Frisoni

, Froelich

, Jori

, Mecocci

, Merlo

, Peters

, Tsolaki

, Defanti

(2023) Respectful Caring for the Agitated Elderly (ReCAGE): A multicentre, prospective, observational study to evaluate the effectiveness of Special Care Units for people with dementia. J Alzheimers Dis 96, 1083–1096.

27.

Luppa

, Luck

, Weyerer

, König

, Brähler

, Riedel-Heller

(2010) Prediction of institutionalization in the elderly. A systematic review. Age Ageing 39, 31–38.

28.

Smith

, Kokmen

, O’Brien

(2000) Risk factors for nursing home placement in a population-based dementia cohort. J Am Geriatr Soc 48, 519–525.

29.

Cesana

, Scolletta

(2019) Predictive models in clinical practice: Useful tools to be used with caution. Minerva Anestesiol 85, 701–704.