Design

This was a 72-month longitudinal study with five assessments. Baseline assessment (A1) took place between November 2004 and January 2005 [56]. The follow-up assessments took place after 12, 31 and 52 and 72 months (A2–A5).

Setting and participants

In total, Norway has 40,000 NH places (beds) [57] for a population of about five million, with about 14% (700,000) of those aged 65 years or older [58]. The jurisdiction for public health care services lies with local municipalities, and local authorities offer social services (such as housing and home services), in-home nursing and institutional care (mainly in NHs), and provide both long- and short-term care and rehabilitation.

This study recruited participants from 26 NHs in 18 municipalities. The selection of small, medium, and large municipalities was made to obtain a wide variety of NH-settings in the sample. NH residents with a stay of at least 14 days were eligible for inclusion, no other inclusion or exclusion criteria were used [56]. In all, 1165 residents were eligible for inclusion and only two declined to participate.

Measures

Psychotropic drugs were grouped according to the Anatomical Therapeutic Chemical (ATC) Classification System into the following groups: antipsychotics (N05A except lithium), antidepressants (N06A), anxiolytics (N05B), hypnotics/sedatives (N05C), and anti-dementia medication (N06D) (yes versus no) [59]. Combination drugs outside ATC NO5B and NO5Cwere not included. The information was collected from the medical record of each resident [56].

Dementia and severity of dementia was assessed using the Clinical Dementia Rating (CDR) scale, covering six domains (memory; orientation; judgment and problem solving; community affairs; home and hobbies; and personal care) with five response categories (0, 0.5, 1, 2, 3) [60, 61]. The total score was calculated using an algorithm that gives priority to memory [60]. Residents with a total score of one or higher were regarded as having dementia. The cut-off CDR ≥1 in defining dementia has been found adequate in previous Norwegian and international studies [62–64]. The categorical scores indicate the severity of dementia: a CDR score of 1 represents mild dementia, a CDR score of 2 represents moderate dementia, and a CDR score of 3 represents severe dementia. The sum-score of the six domains (sum of boxes), ranging from zero to 18, can also be used to measure the severity of dementia, as the categorical and continuous scores correlate highly ≥ 0.9 [65, 66]. The Spearman correlation between the categorical CDR score and the CDR sum of boxes score in the present study was 0.93. Due to a wider range of values, the CDR sum of boxes offers important advantages when analyzing the data [66].

We used the CDR score (CDR ≥1) as an indication of dementia, as it was not possible to perform a standardized dementia work-up for all residents because many were too frail or mentally impaired to take part in examinations such as CT or MRI. A large number of residents with a CDR score of 3 could not be tested with any dementia tests such as the Mini Mental Status Examination [67] or the Clock Drawing Test [68].

Physical health was assessed using the General Medical Health Rating (GMHR) scale [69]. This is a one-item global rating scale with four categories: good, fairly good, poor, and very poor. The rating was based on all available information on physical health and use of drugs. The scale has been used in large studies including older people with and without dementia [70] and has been translated and used in several studies [71, 72].

The Personal Activities of Daily Living (P-ADL) score was assessed with the Physical Self-Maintenance Scale (PSMS), including six items, with a total score ranging from 6 to 30 [73]. High scores indicate a lower level of functioning.

Neuropsychiatric symptoms (NPS) were assessed using the Neuropsychiatric Inventory Nursing Home version (NPI-NH) [74, 75]. The 10-item inventory covers the following symptoms: delusion, hallucination, euphoria, agitation/aggression, disinhibition, irritability/lability, depression/dysphoria, anxiety, apathy/indifference, and aberrant motor behavior (no/yes). For each symptom, severity (score 1–3) multiplied by frequency (score 1–4) provides a score from zero to 12. Based on a previous principal component analysis, psychosis (delusions, hallucination), agitation (agitation/aggression, disinhibition, irritability), and affective (depression, anxiety) sub-syndrome scores were formed by summing the score of the included items [76–79]. The apathy/indifference was analyzed as a single symptom.

Demographic information such as age, gender, marital status, and length of stay in the NH at the time of inclusion was collected from medical records. The type of unit was also recorded from among the following options: regular units (RU), special care unit for people with dementia (SCU), rehabilitation unit (REU), and other units (OU), mainly psychogeriatric wards. The length of stay in a NH before study inclusion was measured in days.

Procedure

Nurses with extensive clinical experience collected the data. Prior to data collection, all assessors participated in a 2-day course on how to apply the standardized questionnaires. A 1-day training program was carried out prior to each follow-up assessment. The project leader (GS) was available for consultation throughout the data collection period. The nurses collected data from medical records and via a standardized interview with the residents’ primary caregivers, all of whom were registered nurses. All assessment scales used were standard translated Norwegian versions. A pilot study including 41 NH residents was carried out to test inter-rater reliability of CDR prior to the first data collection and the inter-rater reliability was very good. The kappa statistics for the global CDR score were 1 (between geriatric psychiatrist and registered nurse specialized in psychiatry) and 0.86 (between geriatric psychiatrist and registered nurse). More detailed information of the inter-rating reliability test has been published elsewhere [56].

Study information was given to the residents and their family members. The residents and their next of kin were informed that they could refuse to participate at any stage of the study. This procedure was recommended and approved by the Regional Ethics Committee in the south east of Norway, the Data Inspectorate and the Directorate for Health and Social Affairs in 2004 before data collection.

Data analysis

Sample characteristics at baseline were described as means and standard deviations (SD) or frequencies and percentages. Residents with CDR ≥ 1 and CDR < 1 were compared by Independent samples t-test for continuous and χ²-test for categorical variables. Prevalence and persistence of PTD use among those with and without dementia were compared by Z-test for proportions. The persistence in outcomes, use of antipsychotics, antidepressants, anxiolytics and sedatives were assessed with a logistic regression model for hierarchical data (SAS GLIMMIX procedure). Random effects for type of unit nested in a NH were included into the model. The dependent variable was current use of a specific type of PTD drug, while the independent variable was either use of the same drug at the previous time point (lag 1), two time points previously (lag 2) or three time points previously (lag 3). All models were adjusted for a number of relevant covariates measured at the same time point as the independent variable. A similar model was estimated to assess variables associated to persistent use of drugs, where the outcome was defined as 1 in the case of use of drugs at two adjacent time points and 0 otherwise. In addition, exploratory analyses assessing association between change in NPI sub-syndromes and persistent use of drugs were performed. All multivariate models were reduced using Akaike’s Information Criteria, where a lower value means a better model. The results were tabulated as odds ratios (ORs) with corresponding 95% confidence intervals (CI) and presented graphically.

All analyses were performed in SPSS version 22 and SAS v9.3. P-values below 0.05 were considered statically significant. All tests were two-sided.

Sample characteristics

At baseline, the mean (SD) age of the residents was 84.4 (7.8) years and 846 (72.7%) of them were women (Table 1). The mean (SD) baseline CDR sum of boxes was 11.2 (5.3) and 932 (80.1%) residents had CDR ≥ 1 indicating dementia. The mean length of stay at baseline was 938.3 (1013.1) days and the mean (SD) number of prescribed drugs taken regularly was 6.0 (3.1). Of the 1163 residents at baseline, 98 (8.4%) were still alive and available at the fifth follow-up (see Fig. 1). Mean (SD) time of follow-up was 829.5 (690.0) days.

		A1	CDR ≥ 1	CDR < 1	P-valuesa
		1163	932
Socio-demographics
Age	Mean (SD)	84.4 (7.8)	84.5 (7.5)	83.8 (9.0)	0.223
Females	N (%)	846 (72.7)	686 (73.6)	156 (68.7)	0.139
Education < 10 years	N (%)	847 (74.8)	673 (74.1)	170 (76.9)	0.390
Married	N (%)	221 (19.0)	185 (19.8)	36 (15.9)	0.178
Health condition
GMHR
Good   Fairly good   Poor   Very Poor	N (%) N (%) N (%) N (%)	194 (17.2) 386 (34.1) 378 (33.4) 173 (15.3)	142 (15.7) 291 (32.2) 322 (35.6) 150 (16.6)	52 (23.4) 94 (42.3) 53 (23.9) 23 (10.4)	<0.001
PSMS score	Mean (SD)	18.1 (5.4)	18.8 (5.3)	15.4 (4.8)	<0.001
NPI Agitation sub-syndrome	Mean (SD)	5.8 (8.0)	6.5 (8.2)	2.9 (6.1)	<0.001
NPI Psychosis sub-syndrome	Mean (SD)	2.8 (5.1)	3.2 (5.3)	1.3 (3.8)	<0.001
NPI Affective sub-syndrome	Mean (SD)	3.5 (5.3)	3.7 (5.4)	2.9 (4.9)	0.041
NPI Apathy	Mean (SD)	2.2 (3.7)	2.4 (3.8)	1.1 (2.9)	<0.001
No of drugs	Mean (SD)	6.0 (3.1)	5.8 (3.0)	6.9 (3.3)	<0.001
Days in NH	Mean (SD)	938.3 (1013.0)	928.9 (910.1)	975.5 (1359.3)	0.534

CDR Clinical Dementia Rating scale

GMHR General Medical Health rating

PSMS Physical Self-Maintenance Scale

NPI Neuropsychiatric Inventory

NH Nursing home

^aCalculated by using t-test for Independent samples for continuous or χ²-test for categorical variables

Prevalence of psychotropic drugs over time

At baseline, the use of any PTD occurred more frequently in residents with dementia, while at the last follow-up any PTD was more often used in residents without dementia. Atypical antipsychotics were used more frequently in residents with dementia at A1–A3, whereas residents without dementia used anxiolytics more frequently at A2 and A3 and sedatives more frequently at A1–A4, respectively.

Persistent use of psychotropic drugs

The proportion of PTD use at two consecutive time points was high (>50%) throughout the period for all types of PTD, except for use of anti-dementia drugs (Table 2). Persistent use of anxiolytics at two consecutive time points was higher for residents without dementia than those with dementia. Among those who completed all assessments (n = 98), 10.4% used antipsychotic drugs, 19.8% used antidepressants, 11.5% used anxiolytics and 9.4% used sedatives at all assessments. Unadjusted and adjusted odds for use of PTD at one time point, given use of the same type of PTD at an earlier time point was estimated for all types of PTD except for anti-dementia drug use (Table 3 and Fig. 2). Both in unadjusted and adjusted analyses, the odds for persistent use of antipsychotics, antidepressants, anxiolytics and sedative were high. The odds for persistent use of these PTD were highest when compared with use at the closest earlier assessment time point (Lag 1) and fell successively when the distance between the assessment time points increased (Lags 2 and 3), with one exception for antipsychotics where the odds slightly increased when there were three time points between assessments (Lag 3) compared to two time points (Lag 2). All results were highly significant (p < 0.001).

Variables	Unadjusted		Adjusted
	OR (95% CI)	P-value	OR (95% CI)	P-value
Lag 1, N = 1406 observations
Antipsychoticsa	30.7 (21.8; 43.1)	<0.001	28.8 (20.31; 40.8)	<0.001
Antidepressantsb	32.6 (23.8; 44.8)	<0.001	34.1 (24.6; 47.4)	<0.001
Anxiolyticsc	33.1 (23.7; 46.2)	<0.001	32.2 (23.1; 45.0)	<0.001
Sedativesd	22.6 (15.6; 30.8)	<0.001	23.1 (16.8; 31.7)	<0.001
Lag 2, N = 654 observations
Antipsychoticse	13.3 (8.3; 21.4)	<0.001	13.5 (8.1; 22.3)	<0.001
Antidepressantsf	15.5 (9.9; 24.2)	<0.001	16.0 (10.2; 25.3)	<0.001
Anxiolyticsg	21.4 (12.8; 35.7)	<0.001	23.4 (13.9; 39.6)	<0.001
Sedativesh	9.3 (5.9; 14.4)	<0.001	9.4 (5.9; 14.8)	<0.001
Lag 3, N = 288 observations
Antipsychoticsi	16.5 (7.5; 36.3)	<0.001	17.9 (7.5; 42.4)	<0.001
Antidepressantsj	10.0 (4.9; 20.4)	<0.001	12.6 (5.8; 27.2)	<0.001
Anxiolyticsk	14.4 (6.5; 32.1)	<0.001	14.3 (7.0; 29.2)	<0.001
Sedativesl	7.0 (3.6; 13.9)	<0.001	6.4 (3.2; 12.6)	<0.001

Lag 1: two consecutive assessment time points

Lag 2: one time point between selected time points

Lag 3: two time points between selected time points

^aAdjusted for: PSMS score, NPI Agitation sub-syndrome, NPI Psychosis sub-syndrome, NPI Affective sub-syndrome, NH size age and duration in NH

^bAdjusted for: PSMS score, NPI- Affective sub-syndrome, NPI Apathy and duration in NH

^cAdjusted for CDR sum of boxes, NPI- Affective sub-syndrome and level of education

^dAdjusted for PSMS score, NPI Apathy and duration in NH

^eAdjusted for PSMS score, NPI Psychosis sub-syndrome, NPI Apathy, Type of nursing home and age

^fAdjusted for: CDR sum of boxes, PSMS score and NPI- Affective sub-syndrome

^gAdjusted for CDR sum of boxes, PSMS score and level of education

^hAdjusted for CDR sum of boxes, PSMS and age

ⁱAdjusted for PSMS score and age

^jAdjusted for: CDR sum of boxes, NPI- Affective sub-syndrome and NPI Apathy

^kAdjusted for NPI Agitation sub-syndrome, NPI Psychosis sub-syndrome, NPI Affective sub-syndrome, NPI Apathy, level of education and Type of nursing home unit

^lAdjusted for CDR sum of boxes, NPI Psychosis sub-syndrome and Nursing home size

Factors associated with use of psychotropic drugs at two consecutive time points

In a subsequent analysis where sub-syndrome scores of NPS at the first time point were replaced by change in sub-syndrome score between the two assessments, we found no association between change in the sub-syndrome scores of NPS and persistent use of antipsychotic drugs, antidepressants, anxiolytics or sedatives.

Use of antipsychotics

The study found that the prevalence of atypical antipsychotics was higher in residents with dementia than in residents without, but only for the first three time points. Use of conventional antipsychotics did not differ between residents with or without dementia at any time point. The persistence in use of both atypical and conventional antipsychotics at two consecutive time points was high (>50%) during the entire follow-up period of 72 months for both groups of residents. A small 6-month follow-up study of newly arrived NH residents in Australia has previously reported the persistence of antipsychotics to be equally high [49]. Given the strong evidence on the increased mortality risk associated with use of conventional antipsychotics in people with dementia it is rather surprising to see that they are still used to that extent. These findings are alarming, since the duration of such treatment should be as short as possible [32]. In the present adjusted logistic regression analysis for persistent use of antipsychotic drugs at two consecutive time points, higher severity of psychosis increased the risk for persistent use of antipsychotics. However, it is surprising that in the exploratory analysis, a change in NPS was not associated with persistent use of antipsychotics. Clinical recommendations have highlighted the need for clinicians to monitor NPS closely and consider discontinuing treatment with antipsychotics when an obvious treatment effect does not occur or the residents have side effects due to the treatment [32]. It could be that people receiving PTD had more severe symptoms prior to drug initiation. Also, it could be that antipsychotics were described as unspecific sedatives. Very few studies have examined variables associated with persistent use of antipsychotics in NH residents [46, 52], but none of these studies have explicitly explored the importance of NPS in the persistent use of antipsychotic drugs. However, cross-sectional studies have explored the association between NPS and use of antipsychotic drugs and found that higher total symptom load [80] and more severe psychosis sub-syndromes symptoms were associated with use of antipsychotics [11, 14, 81, 82].

In the adjusted logistic regression analysis of use of antipsychotics, younger age and male gender increased the risk for persistent use of antipsychotics. In contrast, a small Swedish 6-month follow-up study in NH residents with dementia did not find age, gender or other personal characteristics of the residents such as P-ADL or cognitive functioning important for persistent use of antipsychotics [52]. Our finding may partly be explained by age and gender-based expressions of behavioral symptoms [80] not captured by the NPS. Male and younger residents may be experienced as more threatening in their verbal or physical expressions and are physically stronger compared to women and older residents, and for this reason may be put on antipsychotic drugs. In line with our results, cross-sectional studies of use of antipsychotics in NH residents have found that younger residents [4, 8, 15, 22, 80, 82, 83] and male residents [5, 80, 84] are more likely to receive antipsychotics.

Persistent use of antipsychotics has been reported to be more frequent in regular care units than in SCU [46]. In the present study we did not find an association between the type of care unit and persistent use of antipsychotics. In Norway, SCUs have residents with more severe NPS than other units, which could explain our result [30]. However, we found that residents with a longer stay in NH at baseline had an increased likelihood of being persistent users of antipsychotics. We do not have a firm explanation for this, but it may be that newly arrived NH residents receive more attention from the staff or respond better to care that is intended to reduce stress, strain and symptoms related to dementia.

Use of antidepressants

We found, in line with other studies, that antidepressants were the most frequently used PTD in NH residents [48, 49] and that the frequency did not differ between residents with or without dementia [85]. The persistence of use of antidepressants was also high [49]. As we could expect, those with more severe affective symptoms had higher odds for persistent use of antidepressants. This result is in line with cross-sectional studies of associations for use of antidepressants [11, 15, 81]. Even so, the efficacy of antidepressants in treating depression in persons with dementia is uncertain. The high prevalence and persistence of antidepressants use may indicate that these drugs also are used for the treatment of agitation. Better performance in P-ADL was more likely to be associated with persistent antidepressant users. We do not have a firm explanation for this, but it may be that those with better P-ADL also have a better ability to express their emotional state.

Use of anxiolytics, sedatives and anti-dementia drugs

The prevalence of anxiolytics in the NH residents varied between 20 and 40%, and was significantly higher in residents without dementia, while the persistent use of anxiolytics in adjusted analysis was not explained by the severity of dementia. As for antidepressants, persistent use of anxiolytics was explained by more severe affective symptoms and better P-ADL functioning, but not by change in affective sub-syndrome symptoms between time points. The size of the NH was associated with persistent use of anxiolytics. We speculate that staff distress [81], registered nursing hours per resident [86] and other organizational factors [5] that we have not measured may be related to NH size, quality of care and use or persistent use of anxiolytics.

Use of sedatives was higher among residents without dementia than residents with dementia and varied between 19 and 26% in residents with dementia and between 32 and 40% in residents without dementia. A possible explanation may be that residents without dementia are more vulnerable to disturbances in the NH environment and at the same time may be more able to ask for sedatives than residents with dementia.

The prevalence of anti-dementia drugs use was at baseline 11.3% for all residents and as expected the prevalence of such use declined at each of the follow-ups and at the fifth assessment no one used anti-dementia drugs. At the four first assessment time points between 1.9 to 2.2% of residents with CDR less than 1 were prescribed anti-dementia drugs. This may indicate that the NH physicians have prescribed the anti-dementia drugs without having a dementia diagnosis (P70), but the physicians may have used the diagnosis cognitive problems (P20) as an indication for prescribing the drug. This is quite common in Norway. However, in a recent Norwegian study it was found that a substantial number of persons who purchased anti-dementia drugs had no diagnosis of dementia or cognitive problems registered in the primary health care service system [87].

Strength and limitations

The study has significant strengths. Firstly, all nurses participating in the data collection participated in a 2-day educational course to ensure that they had adequate knowledge prior to collecting data for this study and participated in a 1-day educational course before the follow-up data collections. This ensured high data quality. Secondly, a large sample size allowed us to adjust for many potentially important variables, such as health and demographic factors. Furthermore, this study benefits from the inclusion of NHs located in a large part of the country. However, we cannot guarantee that the sample is representative for Norwegian NH residents since inclusion was not based on random selection from all NHs in Norway.

The study has some limitations. Firstly, the data from the present study is quite old since data collection started in 2004, and thus, may not represent medication use patterns in Norwegian NHs of today. Secondly, a high drop-out rate mainly due to death might have introduced some bias into the results. However, this methodological problem is inherent to most longitudinal NH studies. Also, we used linear mixed models that include all available data (data from drop-outs as well) in the analysis. Due to the low number of participants at the end of the follow-up period, the analysis of persistence was limited to three lags only. Thirdly, the time intervals between the assessments varied somewhat among participants. However, this affects the results only minimally, since the time intervals between assessments were quite long and not used in the models explicitly, only as lags. In the present study the time intervals between the assessments were 1 year or more and due to the study design, we do not know whether there were changes in PTD use between assessments.

Fourthly, there is a limitation on the accuracy of dementia diagnoses, since dementia and degree of dementia are based on the CDR rating of several assessors and the fact that a CDR assessment was used, and not a standardized dementia diagnosis, including neuropsychological tests. A large number of residents with a CDR score of 3 could not be tested with any dementia tests such as the Mini Mental Status Examination or the Clock Drawing Test. However, CDR is an accepted assessment tool and is commonly used in epidemiological NH studies to identify dementia and measure the severity of dementia [65, 88], and the agreement between CDR and a diagnostic assessment according to the ICD-10 is high [62].