- Email: [email protected]
A behavioral rating scale as a predictor for survival of demented nursing home patients
ARCHIVES O F GERONTOLOGY AND GERIATRICS ELSEVIERSCIENCE IRELAND
Archives of Gerontology and Geriatrics 18 (1994) 101-113
A behavioral rating scale as a predictor for survival of demented nursing home patients Pieter T.M. van Dijk, Diederik W.J. Dippel, J. Dik F. Habbema* Center for Clinical Decision Sciences, Department of Public Health, Erasmus University Rotterdam, Postbus 1738, 3000 DR Rotterdam, Netherlands (Received 2 August 1993; revision received 24 December 1993; accepted 27 December 1993)
Abstract As part of a study towards determinants of survival in nursing home patients with dementia, the prognostic value of a behavioral rating scale for nursing home patients is assessed in an 8-year follow-up study. The 2-year survival rate for the entire cohort (n = 569) was 56%. Women (n = 459) had a 2-year survival rate of 62%, and men (n = 110) had a 2-year survival rate of 40%. Items indicating physical impairment, dependency and apathy had most prognostic value. Items measuring aggressive or depressive behavior, and cognitive impairment were less predictive. These results were confirmed in a multivariate proportional hazards analysis. A prognostic model with age, gender and five behavioral items (needs help when walking, occupied in useful activity, restless at night, utters physical complaints, and socializes with other patients) substantially differentiated in survival chances in patients with dementia. The model gives a predicted 2-year survival chance of less than 20% or more than 80% in 80 of the 569 patients. When adjusted for the variables in the model, previous residence had no prognostic value anymore. Possibilities for further work in this area of research are discussed.
Key words." Survival; Dementia; Nursing home; Behavioral rating scale; Proportional hazards
1. Introduction A b o u t 20 000 (1.5%) D u t c h i n h a b i t a n t s over 65 years o f age reside in nursing homes because o f d e m e n t i a (SIVIS, 1990). Their survival after admission ranges between a few days a n d m o r e than 10 years, with a mean o f a b o u t 2.5 years. This wide * Corresponding author. 0167-4943/94/$07.00 © 1994 Elsevier Science Ireland Ltd. All rights reserved. SSDI 0167-4943(94)00537-H
102
P. 7~M. van Di/k et al. / Arch. Gerontol Geriatr 1~¢ (1994) 101-113
range provides only little information to the interested parties. Close relatives of the demented patient are often very interested in his prognosis. A nursing home physician needs prognostic information, for example when he is considering offering his patient an operation because of a senile cataract. Also, health planners combine epidemiological and prognostic information when addressing questions of demand and turnover in nursing homes, and how these depend on admission criteria. Age and gender provide some prognostic differentiation (van Dijk et al., 1991, 1992), but more knowledge is needed. The prognostic value of behavioral rating scales for assessing severity of dementia has been studied before. Measures of dependency in activities of daily life, physical impairment, and inactive behavior appeared to be associated with survival (Barclay et al., 1986; McLaren et al., 1986; Martin et al., 1987; Moran et al., 1990). Most studies, however, were carried out in a univariate or qualitative way. No attempt was made to study the joint effect of prognostic factors on survival. In the present study of survival in dementia patients we use multivariate, quantitative methods in order to assess the information provided by a behavioral rating scale and to identify the subscales and items which particularly have prognostic value. The resulting model can be regarded as a step towards a more individual prognostic evaluation of dementia patients who are admitted to a nursing home. 2. Methods 2.1. Patients
The Psychogeriatric Center Stadzicht in Rotterdam, The Netherlands, is a nursing home which serves about 260 patients, who are suffering from dementia of various etiologies. All 606 patients, 437 women and 169 men, admitted between January 1st, 1982 and December 31, 1988 were included in this study, and followed until death or discharge, or until January 1st, 1990. The mean age at admission was 80.8 years for the whole cohort, 81.3 years (S.D. 6.6) for women and 79.6 years (S.D. 7.3) for men. Dementia was diagnosed according to criteria of the DSM-III-R (American Psychiatric Association, 1980). The mean duration of the dementia before admission was 5.1 years (S.D. 3.2). One-third of the patients was married at admission, and 56% widowed. The place of living before admission was their own house in 42"/,,, a home for the aged in 18%, another nursing home in 17%, and a hospital or other institutions in 23% of the patients. A more detailed description of the study population is presented elsewhere (Van Dijk et al., 1992). 2.2. Methods
The BOP (Beoordelingsschaal voor Oudere Patienten), which is translated as Rating Scale for Elderly Patients, is used in many nursing homes in the Netherlands (Diesfeldt, 1981). The BOP contains 35 items about behavioral and cognitive impairment. Nurses who take care of the patient score each item on a 0-2 scale with higher scores indicating more severe or more frequent disability. The item scores are used to calculate sumscores on six sub-scales: dependency, aggressiveness, physical disability, depression, orientation and communication, and apathy. In this study, the 35 BOP items, the six subscales, and the patients' age and gender are the possible
103
P . T . M . van Dijk et a l . / A r c h . Gerontok Geriatr. 18 (1994) 101-113
predictive variables. In the analysis, we used the BOP-scores measured for the 569 persons (410 women and 159 men) who had a completed assessment at the end of the observation period, about i month after admission. The BOP was not completed in the remaining 37 patients (6°/,0, because of early death [17] or other reasons [20]. Like the rating scale of the CAPE (the Clifton Assessment Procedures for the Elderly) (Pattie, 1988), the BOP is derived from the Stockton Geriatric Rating Scale (Meer and Baker, 1966). These three scales have 14 items in common. This enabled us to assess the prognostic value of that part of the CAPE-rating scale that overlaps with the BOP. Sum scores could be calculated on the apathy and the communication failure subscales of the CAPE. Survival is estimated by the product-limit method. Statistically significant differences in survival between subgroups are identified by the log-rank test (P < 0.05) (Lee, 1980). Multivariate analysis of prognostic variables is carried out using proportional hazards regression models (Kalbfleisch and Prentice, 1980). The log-likelihood is used as a goodness-of-fit statistic: the less negative its value, the more closely the model fits the observed data. When using a proportional hazards model, for each patient the hazard of dying at time t is estimated by the equation Ai(/) = Amean(t)*
exp(/31.zli
+/32 i + • • • +/3n'Zni-
/3mean)
(1)
For each patient the survival chance is estimated according to the equation S~{t)
= Smean(t)exp(/31.
zli +/32.z2i
+
• • • +/3,.Zni
-
/3mean)
(2)
where A,(t) is the hazard of dying for an individual i at time t, Amean(t) is the hazard of dying for someone with a mean score on all selected variables, S , ( t ) is the chance of survival for an individual i at time t, Smean(/) is the chance of survival at time t for someone with a mean score on all selected variables, /31.zli +/32i.z2i + • • • + / 3 ~ ' z n i is the sumscore for an individual i with scores z I • • • zn on the variables /31 • • • /3n where /31 • • • /3, are the natural logarithms of the rate ratios shown in Table 3, and/3mean is the sum score for someone with a mean score on all selected variables. Two regression models are developed. Model ! uses the BOPitems to predict survival, and Model 2 the BOP-subscales. Because we want to investigate the additional value of the BOP-variables above that of gender and age, the latter ones are forced into the two models before the forward selection of the BOP-variables starts. The reference model is based on age and gender only. The items and scales are entered or removed in a stepwise forward selection mode on the basis of tail probabilities (P < 0.10 and P > 0.15, respectively) from a likelihood ratio test statistic: a variable is only selected into the model if it gives enough additional prognostic information above that provided by the variables already in the model. Model 1 was evaluated in detail. In order to avoid overoptimism, a split-half approach was used. The entire cohort was randomly divided into two equally sized subgroups. The stepwise forward selection was performed on the patients of one subgroup (the training sample), and the resulting model was used for estimating survival
104
P.T.M. van Dijk et al./Arch. Gerontol. Geriatr. 18 (1994) 101-113
chances for the patients in the other subgroup (the validation sample). This process was repeated in such a way that every subgroup (and thus every patient) acted as a validation sample once. The patients were ordered according to their predicted survival chances into four groups with a 'favorable', a 'moderate', a 'poor' and a 'very poor' prognosis. The predicted survival chances in these four groups were compared with the observed survival rates in order to assess the reliability of the model predictions. 3. Results
3.1. Subscales The subscales physical disability, orientation and communication and apathy have one, four and five items in common with the subscale dependency. It is therefore not surprising that they are highly correlated with that subscale (Pearson's r - 0.7). The correlation between apathy and physical disability is 0.6. The correlations between the remaining subscales are weaker and vary between 0 and 0.4. Men score on average somewhat higher on the subscales dependency, apathy and aggressiveness than women (t-test, P <0.05) (see Table 1). 3.2. Survival The 2-year survival rate for the entire cohort (569 patients) is 56"/,,. Except for aggressiveness, scores on the subscales are significantly related with survival (log rank, P < 0.05) (see Table 1). For all subscale scores, survival rates for women are significantly higher than for men. Individual items of the physical disability, the apathy and the dependency subscale are of considerably prognostic value (Table 2). Several clearly behaviorally anchored items, such as 'needs help when eating', 'needs help when walking', 'needs help when dressing', 'incontinent at night' and 'incontinent during the day' show large differences in 2-year survival rates between patients with a score of 0 (no help needed, no impairment) and patients with a score of 2 (much help needed, severe impairment). For instance, the patients who need no help when walking have a 2-year survival rate of 73%, whereas patients who need much help when walking have a 2-year survival rate of 36%. The items of the aggressiveness scale have hardly prognostic value. The items 'responds to his name' (OC4) and 'privileges to leave the ward' (A7) have practically empty categories. Inference regarding these items is therefore limited; they are excluded from further analyses. 3.3. Multivariate prognostic models Table 3 describes the two proportional hazard models for the prediction of survival time. Model 1 contains five BOP-items, age and gender. As expected, the mortality hazard was higher for men, and increased with age. The item 'needs help when walking' was the first behavioral item selected, and had the highest rate ratio. The item 'restless at night' was less significantly associated with survival in the univariate analysis (0.01 < P < 0.05), but nevertheless has been selected: apparently its prognostic information was independent from the information of other items. From the
105
P.T,M. van Dijk et al./ Arch. Gerontol. Geriatr. 18 (1994) 101-113
Table 1 Scores on the behavioral rating (BOP) subscales and 2-year survival rates in 159 male and 410 female institutionalized patients with dementia Subscale
Men (n = 159)
Women (n = 410)
Mean
Mean
All patients Dependency
40% (100%) 21.3
0-16 17-28 29-46 Aggressiveness
Depression
3.9
Apathy
2.9
2.2
62% (55%) 63% (45%) 2.3
48°/,, (67%) * 25% (33%) 2.2
75% (61%) * 43% (39%) 2.4
48% (6Y'/o) * 27% (37%) 4.3
0-4 5-8
0-7 8-14
76% (43%) * 52'¼, (38%) 5I% ( 19%1
39% (40%) 40°/,, (60%)
0-2 3-6 Orientation and communication
19.3
34% (45%) 27'¼, (25%)
0-2 3-6
2-year survival rate (relative frequencyl 62% (lOft¼,)
63% (30%) *
0-2 3-10 Physical disability
2-year survival rate (relative frequency)
68% (58%1 * 55% (42%) 4.3
43% (52%) 37% (48%) 8.0
67% (55%) * 57% (45%) 7.2
58% (40%) * 27°/,, (60%)
73% (52%) * 50'¼, (48%)
*Log-rank test: P < 0.05.
depression subscale the item 'utters physical c o m p l a i n t s ' has been selected. The rate ratio of 1.1 indicates that its predictive value is limited. Two examples show the considerable r e d u n d a n c y a n d overlap in prognostic inform a t i o n between the items a n d thus highlight the additional value of multivariate over univariate analysis. First, 25 BOP-items were significantly related to survival in the univariate analysis (P < 0.05) (Table 2), but only five of them have been selected. Second, of the 3 items of the physical disability scale - - which had all highly significant differences in survival in the univariate analysis - - only the item 'needs help when walking' is selected in Model l; obviously the others do not add sufficient additional prognostic i n f o r m a t i o n . Model 2 contains three BOP-subscales, age, and gender. The rate ratios for age a n d gender do n o t differ essentially from Model I.
P.T.M. van Dijk et al./Arch. Gerontol. Geriatr. 18 (1994) 101-113
106
Table 2 Two-year survival rates for scores on the 35 items of the behavioral rating (BOP) scale. Relative frequencies are given in brackets. Items with statistically significant differences in survival rates are indicated Item
Description of item
2-year survival rate (relative frequencyl Score 0
Score 1
Score 2
66% 67% 64% 56% 57% 59% 56% 61% 68% 68% 63% 60% 59%
(39%) (52'7',) (47%) (8%) (71%) (41%) (48%) (60%) (52%) (51%) (68%) (76%) (78%)
55% (42'7',) 49% (16%) 48% (39'7',) 69% (20'V,,~ 55% (12'7',) 55% (54'7',) 67% (13%) 50% (17%) 50% (28'7`,) 51% (10%) 45% (3%) 47'7,, (14'7',) 50% (14%)
39% (19%) 43% (32%) 56% (14%) 53% (72%) 53% (17'7',) 49'7', (5%) 53% (39%) 49% (23%) 37% (20%) 44% (39%) 42%, (29%) 44% (10%) 38% (8'7',)
57% (65%) 55'7,, (49%) 57% (76'7',) 56% (53%) 58% (41'7`,t
60% (19%) 55'!/,, (21%) 57% (15%) 59% (22%) 60% (32%)
50% (16%) 60% (30%1 46% (9'7',) 54'7', (25%) 49% (27'7',)
61% (86%) 7Y¼, (46'7',) 71% (21%)
21% (5%) 48';4, (28%) 63';4, (36%)
34'7', (9%) 3.6'7,, (26%} 44% (43%)
62% (23%) 63% (43%) 59% (81%)
59% (37%) 58% (22%) 54% (10%)
51% (40%) 47% (35%1 40% (9%)
Knows in which institution he is Knows any of personnel by name Understands others Responds to his name
68'7', (7%) 78% (3'7',) 62% (44%) 57% (87%)
58% 53% 53% 44%
(2Y7',) (4%) (50%) (12%)
55% 56% 45% 88%
(70%) (93%) (6%) (1'7',)
Helps out on the ward Occupied in useful activity Socializes with other patients Helps other patients without being asked Needs supervision outdoors Never starts conversations Privileges to leave the ward
76% (19%) 74% (33%t 74% (40%) 75% (26%) 67'7', (24%) 68% (42%) - (0%)
70% 56% 51% 66% 57% 50% 57%
(18%} (34%) (27%) (23%) (52%) (31%) (99'7',)
47% 39% 39% 43% 46% 44% 0%
(63'7',) (33'7',) (33%) (51%) 124%) (27%) (I%)
Dependency (D) ** ** ** **
** ** ** * *
DI D2 D3 D4 D5 D6 D7 D8 D9 DI0 DI 1 DI2 DI3
Needs help when eating Incontinent during the day Does not make himself understood Unable to find his way around the ward Urinates and defecates at the wrong place Unwilling to do things asked of him Engages in useless repetitive activity Makes repetitive vocal sounds Drowsy during daytime Incontinent at night Needs protection from falling out of bed Objectionable during the night Restless at night
Aggressiveness ( Ag) Agl Ag2 * Ag3 Ag4 * Ag5
Threatens verbally to harm others Accuses others of harming him Hits and kicks other patients Objectionable during the day Angry easily
Physical disability ( PD) ** PDI ** PD2 ** PD3 ~
Needs protection from falling out of chair Needs help when walking Needs help when dressing
Depression (De) * Del ** De2 * De3
Sad Utters physical complaints Weeps easily
Orientation and communication ( OC) OCI a OC2 ~ * OC3 a OC4 ~
Apathy ( A ) ** ** ** ** * ** **
A1 A2 a A3 a A4 a A5 A6 a A7 a
"item is also part of the Dependency subscale. *0.01 < P < 0.05 (log-rank test), **P < 0.01 (log-rank test).
P.T.M. van Dijk et al./Arch. Gerontol. Geriatr. 18 (1994) 101-113
107
Table 3 Two proportional hazard models for predicting survival, based on age, gender and the behavioral rating scale (BOP). The BOP-items respectively BOP-subscales are entered into the model by stepwise forward selection and are listed by order of entry Variable
Range
Rate ratio a (95'Y,, CI)
Log-likelihood
Model 1 (BOP-items) Gender (female=0, male= 1) Age PD2 Needs help when walking A2 Occupied in useful activity DI3 Restless at night De2 Utters physical complaints A3 Socializes with other patients
0-1 50-100 0-1-2 0-1-2 0-1-2 0 - I-2 0-1-2
1.9 (I.5-2.4) 1.03 (1.02-1,05) 1.4 (1.2-1.6) 1.2 (1.0-1.4) 1.3 (1.1-1.5) 1.1 ( 1.0-1.3) 1.2 (1.0-1.41
-1964
0-1 50-100
2.1 (1.7-2.7) 1.04 ( 1.02- 1.05)
-1966
0-6 0-14 0-6
1.16 ( 1.05- 1.24) 1.06 (1.03-1.12) 1.10 (1.02-1.18)
Model 2 ( BOP-subscales) Gender (female = 0, male = I) Age Physical disability Apathy Depression
aRate ratios express relative mortality hazards. The value for the rate ratio for the covariate age of 1.03 in Model 1 implies, for example, that at every moment after admission the risk of dying for a 90-year-old person is 1.03 l0 = 1.34 times the risk of an 80-year-old person of the same gender and the same scores on the other items. If the latter person had a 2-year survival probability of 0.60, then the 90-year-old person would have a probability of 0.60 T M = 0.50. The rate ratio of 1.9 for gender implies that at every moment the chance of dying for men is almost twice the chance for women. The rate ratio of 1.4 for the item 'needs help when walking' means that the chance of dying for patients who need much help (itemscore 2) is 1.4 times that chance for patients who need some help (itemscore 1), and 1.42 = 2 times that chance for those who do not need any help (itemscore 0). The rate ratio of 1.16 on the physical disability-subscale in Model 2 implies that a patient with a score of 6 on that subscale has a change of dying 1.166= 2.4 times the chance for a patient with a score of 0.
O f t h e five s u b s c a l e s w i t h p r o g n o s t i c v a l u e in t h e u n i v a r i a t e a n a l y s e s ( T a b l e 1), t h e s u b s c a l e s p h y s i c a l d i s a b i l i t y , a p a t h y a n d d e p r e s s i o n a r e selected. T h e d e p e n d e n c y and orientation and communication subscales do not add significantly to survival p r e d i c t i o n o n c e t h e p h y s i c a l d i s a b i l i t y s u b s c a l e is selected.
3.4. Evaluation o f the prognostic models Goodness-of-fit. T h e t w o m o d e l s a r e r e l i a b l e in t h e s e n s e t h a t t h e m o d e l - p r e d i c t e d n u m b e r o f d e a t h s w i t h i n 2 y e a r s a f t e r a d m i s s i o n c o r r e s p o n d s well w i t h t h e o b s e r v e d n u m b e r (X2-test, 5 df, P > 0.3 f o r b o t h m o d e l s ) (see T a b l e 4). A r e f e r e n c e p r o g n o s t i c m o d e l , b a s e d o n a g e a n d g e n d e r o n l y , is a d d e d f o r c o m p a r i s o n . M o d e l 1 seems to be the easiest to implement, because the scores on a limited number of individual items are more readily available than the sum scores of the BOP-subscales. Its p e r f o r m a n c e w a s f u r t h e r a s s e s s e d b y t h e s p l i t - h a l f a p p r o a c h (see M e t h o d s sec-
108
P.T.M. van Dijk et al./Arch. Gerontol. Geriatr. 18 (1994) 101-113
Table 4 Observed and predicted number of deaths during the first 2 years after admission according to Model 1. Model 2 and the Reference model for five subgroups with predicted 2-year survival chances between 0 and 20%, 20 and 40°/,. . . . . . 80 and 100% Predicted 2-year survival
Number of patients Number of deaths Observed
Expected
Model 1 (age, gender, items)
0-20% 20-41Y7`, 40-60% 60-80% 80-100'7,,
27 116 150 223 53
25 66 77 54 5
23 72 69 60 9
31 113 150 228 47
24 67 80 53 3
26 70 69 62 7
4 69 244 245 7
2 36 121 68 0
2.6 42 108 76 1
Model 2 (age. gender, scales)
0-20% 20-40% 40-60% 60-80% 80-100% Reference model (age and gender only) 0-20%
20-40% 40-60% 60-80% 80-100%
tion). The goodness-of-fit was satisfactory, a l t h o u g h the model u n d e r e s t i m a t e d the survival chances in the g r o u p with a very p o o r prognosis (Fig. 1). 3.5. P r e d i c t i v e p o w e r
Once the models have been shown to be r e a s o n a b l y reliable, they can be evaluated for their power in individualizing survival predictions. W e will focus on the prediction o f 2-year survival chances. A 'perfect individualizing' model w o u l d give a 100% predicted survival chance to each patient o f the 56°/,, who have survived the 2-year period, a n d a 0% predicted survival chance to each o f the 44% who died. A totally non-individualizing m o d e l (the 'null' model) will predict for e v e r y b o d y a chance that is equal to the observed 56% 2-year survival rate for the total study p o p u l a t i o n . A n indication o f the predictive p o w e r can be o b t a i n e d from Table 4. M o d e l 1 gives a predicted survival chance between 40% a n d 60% in 150 patients, a n d a chance o f less than 20% o r m o r e than 80% in 80 patients. This m o d e l thus differentiates considerably between patients. M o d e l 2 gives c o m p a r a b l e results. The results for a reference m o d e l are worse. It gives a predicted survival chance between 40°/,, a n d 60% in 244 patients (more than 40% o f all patients), a n d has only a small n u m b e r o f patients with a prognosis o f lesser than 20% or m o r e than 80% (4 and 7 patients, respectively).
109
P.T.M. van Dijk et al./Arch. Gerontol. Geriatr. 18 (1994) 101-113 survival 100%
observed 80%
expected
60%
~
prognosis
40% favorable
20%
moderate poor very poor
0% 0
1
2
3
4
5
6
7
years
Fig. 1. Observed and predicted survival according to Model I for four equally sized [142] subgroups of patients with a favorable, moderate, poor and very poor prognosis according to their survival probabilities after the split-half approach on Model I. The mean predicted 2-year survival chances for these four groups are 0.24 (range 0.00-0.39), 0.50 (range 0.39-0.60), 0.66 (range 0.60-0.73) and 0.81 (0.73-0.94).
3.6. Help in survival prediction A prognostic index, and henceforth survival chances can be calculated for every person (second equation, Methods section). The average 1-, 2-, and 5-year survival rates for the entire cohort were 72%, 56%, and 20%, respectively. The prognostic index and the corresponding survival chances are easily obtainable from Fig. 2 (see the legend for examples). 3.7. Implications for users of the CAPE The mean sumscores at admission on the CAPE-subscales apathy and communication failure were 5.0 (±2.6) and 1.3 ( ± 1.2), respectively. When these two scales were put into a proportional hazards model with gender and age, the communication failure subscale had no predictive value and the apathy subscale had a hazard rate ratio of 1.17 for each point one scored higher (on a scale from 0 to 10). In a multivariate analyses of the 14 items, gender and age, five items were selected. The seven variables and their rate ratios were: gender (1.9), age (1.03), needs help when walking (1.4), helps out on the ward (1.2), socializes with other patients (1.3), unwilling to do things asked of him (0.9), and objectionable during the night (1.3). For these seven variables a prognostic chart similar to Fig. 2 is constructed (see Table
110
P. I'.M. van Dijk et al. / Arch. Gerontol, Geriatr. IS (1994) 1(t1-113
Table 5 Prognostic chart for estimating survival chances, according to the results with the CAPE Score Gender Age
Needs help when walking Helps out on the ward Socializes with other patients Objectionable at night Unwilling to do things asked of him
if man 51-60 61-70 71-80 81-90 some much sometimes never sometimes never sometimes often sometimes never
6 0 3 6 9 3 7 2 4 2 5 2 5 1 3
Add relevant scores:
[]
Subtract 2
[]
Use: circle relevant scores, add them to a sumscore, and subtract 2 points to get the prognostic index.
5). For every (CAPE-)prognostic index, survival chances can be read directly from Fig. 2. 4. Discussion Gender and age, and patient behavior as measured on a behavioral rating scale (the BOP) were found to be helpful in estimating survival chances in institutionalized patients with dementia. Subscales measuring physical disability, apathy and depressive behavior carried independent prognostic value. The first two scales were mentioned before as being of prognostic importance (Diesfeldt, 1975; Jacobs et al., 1978). The association between the orientation and communication subscale' and survival was only small (Diesfeldt, 1975) or even absent (Jacobs et al., 1978). Observations in non-institutionalized patients also emphasized the prognostic value of scales measuring physical impairment (Martin et al., 1987) and impairment of ADL activities (Knopman et al., 1988). On the item-level our results are consistent with Diesfeldt (1980), who found that all items of the Physical Disability scale and seven items of the dependency scale were associated with l-year survival. The five selected items in Model 1 are all easily interpretable, and most of them are also incorporated in other rating scales. Table 4 and Fig. 1 showed that the fit of Model 1 was satisfactory. We would be very interested to see how adequately this 'internally validated'
111
P.T.M. van Dijk et al./Arch. Gerontol. Geriatr. 18 (1994) 101-113
survival probability 100%
80%
Calculation of prognostic index "-,~
_
gender
if male
age
61 - 70 71 - 80 81
"..~ %. 60%
"& x\
"...
5 year"....2 "Q..
~
X year',\ ~ year 4\ ~
"-", '... "~ ~..
40%
\'\
X
\\ & "\
'~ ~
needs help when walking occupied in useful activity restless at night utters physical complaints socializes with ~ otherpatients
""-. \\ X Add relevant X...........Q.. X& ~scores
20%
0%
~
,
I
10
,
i
20
....fh.........© 30 prognostic index
-~ 40
7 3 7
- 90
10
some 3 much 6 sometimes 2 never 4 sometimes 2 often 5 sometimes 1 often 3 sometimes 1 never 3 pro.gnostic [ ~ Inoex
1
50
B
60
Fig. 2. Prognostic index according to Model 2 and the predicted 1-, 2- and 5-year survival chances. Examples: a woman of 65 years needs much assistance when walking and is sometimes 'restless at night'. She has favorable scores on the other three items. Her prognostic index is 3 + 6 + 2 = 11. In Fig. 2 it can be seen that her predicted 1-, 2-, and 5-year survival chances are 83%, 78'Y,,and 48%, respectively. A man of 85 years who sometimes "utters physical complaints" and is only sometimes occupied in useful activity. and who has a score of two on the other three items has a prognostic index of 7 + 10 + 6 + 2 + 5 + 1 + 3 = 34, and his survival chances are only 23%, 7'7,,and 0%.
m o d e l c a n predict survival in o t h e r n u r s i n g homes. W e t h i n k o u r clientele is rather representative for p s y c h o g e r i a t r i c n u r s i n g h o m e s in the N e t h e r l a n d s . T h e m e a n age o f d e m e n t e d p a t i e n t s a d m i t t e d in the N e t h e r l a n d s in 1986 (78 years for m e n a n d 80 years for w o m e n ) was slightly lower t h a n o u r clientele, a n d relatively m o r e m e n (33%) were a d m i t t e d (SIVIS, 1986). Scores o n the B O P - - r a t i n g scales on d e m e n t i a patients were n o t essentially different f r o m figures f r o m o t h e r n u r s i n g h o m e s in the N e t h e r l a n d s , n e i t h e r did the 2-year survival rates (Diesfeldt, 1975; J a c o b s et al., 1978; v a n Dijk et al., 1991). D u t c h n u r s i n g h o m e s are r a t h e r c o m p a r a b l e to A m e r i c a n s k i l l e d - n u r s i n g facilities ( C o o l s a n d v a n der Meer, 1988), a l t h o u g h the availability a n d k i n d o f m e d i c a l care in D u t c h n u r s i n g h o m e s is different from that a b r o a d (Cools, 1993). Differences in case-mix m a y decrease the g e n e r a l i z a b i l i t y o f the m o d e l to other settings. W e therefore investigated the i m p a c t o f p r e v i o u s residence o n the results. W h e n a d d e d to M o d e l 1, n e i t h e r the v a r i a b l e ' c o m i n g f r o m o w n house" ( i n d i c a t i n g p a t i e n t s with less c o - m o r b i d i t y ) , n o r the v a r i a b l e ' c o m i n g from a h o s p i t a l ' (in-
112
P . T . M . van Di/k et al. / Arch. Ger¢mtol. Geriutr. 18 (1994) 101-113
dicating patients with more co-morbidity) had rate ratios significantly different from 1: their rate ratios were 1.1 (95% confidence interval 0.9-1.4) and 0.9 (0.7-1.1), respectively. This suggests that for an estimation of survival chances the severity of impairment (in combination with gender and age) is much more important than previous residence; this finding corroborates the possibility of using our results in other nursing homes. Variables such as reason for admission and problems in adaptation to the sudden change in living situation are not considered in the BOP. Furthermore medical symptoms and diagnoses undoubtedly also contain prognostic information. The fact that the item 'utters physical complaints' (because of physical origin or not) was selected is already an indication for this. Although co-morbidity and behavioral items are mutually related - - many diseases result in physical disability and dependency, and an increased severity of dementia may cause a higher vulnerability for acquiring all sorts of diseases-, co-morbidity might also give independent prognostic information. For instance, the 2-year survival rate for patients with a very poor prognosis according to Model 2 was 30%. Patients in this group with parkinsonism, a heart failure, or a respiratory tract infection had 2-year survival rates of about 15%. This rates was 32% in the absence of these diseases. Thus, clinicians must be aware that the prognosis of the patient may be considerably worse when severe co-morbidity is present. Our results can also be of help for health planners. Patients in our nursing home had a 2-year survival rate of 56%. This gives a direct indication of turnover in nursing homes. Together with demographic and epidemiologic data, future demand and capacity for the current admission strategy can be assessed. It is also possible to estimate 2-year survival rates, when the admission policy is changed, e.g. in such a way that only more severely demented patients are admitted. If, for instance, the mean scores on all BOP-items at admission increased with 0.5 points, the mean prognostic index according to Model 1 would have such an increase, that the expected 2-year survival rate would decrease from 56 to 41%. In conclusion, behavioral items (which in many nursing homes are assessed for other purposes also) contain interesting prognostic information and can be combined in multivariate prediction models. Using these models, health care planners might be able to estimate more adequately the need of beds in the future for patients with dementia. Also, these models can support the nursing home physician in getting the best possible insight into the prognosis of a patient. The more accurate this prognosis, the more well-founded the nursing home physician can decide to perform a diagnostic evaluation or a surgical procedure, which may be of great risk and inconvenience to the patient initially, but improve the quality of life afterwards. A better life-expectancy increases the chance that the expected long-term benefit for the patient exceeds the instantaneous risk and inconvenience of the diagnostic or surgical procedure, and thus weights in favour of an active approach (Rango, 1985).
5. Acknowledgements We thank the Psychogeriatric Center Stadzicht and its medical and psychological staff for their generous permission to use the medical records of their patients. This
P.T.M. van Dijk et al./Arch. Gerontol. Geriatr. 18 (1994) 101-113
113
study was supported by a grant from the Netherlands' Health research promotion programme (SGO). 6. References American Psychiatric Association (1980): Diagnostic and Statistic Manual of Mental Disorders (DSMIll). Washington: American Psychiatric Association. Revised version: 1987. Barclay, L.L., Zemcov, A., Blass, J.P. and McDowell F.H. (1985): Factors associated with duration of survival in Alzheimer's disease. Biol. Psychiatry, 20, 86-93, Cools, H.J.M. and Van der Meer, J.W.M. (1988): Infection control in a skilled nursing facility: a 6-year survey. J. Hosp. Inf., 12, 117-124. Cools, H.J.M. (1993): Adding life to years (in Dutch). AVO, The Hague. Diesfeldt H.F.A. (1975): Predicting survival and longevity of mentally impaired elderly patients (in Dutch). T. Soc. Geneesk,, 57, 343-350. Diesfeldt H.F,A. (1980): Activities of daily living, cognitive disturbances and survival in psychogeriatric patients (in Dutch). Gerontologie, 11, 205-212. Diesfeldt H.F.A. (1981): The BOP, a report on ten years' experience with a Dutch geriatric rating scale (in Dutch). Gerontologie, 12, 139-147. Jacobs, M., Trommel, J. and Gips, C.H. (1978): A rating scale for geriatric patients: need of care, age groups and one-year survival of psychogeriatric patients (in Dutch). Ned. Tildschr. Gerontol., 9, 29-34. Kalbfleisch, J.D. and Prentice, R.L. (1980): The Statistical Analysis of Failure Time Data. John Wiley and Sons. Knopman, D,S., Kitto, J., Deinard, S. and Heiring, J. (1988): Longitudinal study of death and institutionalization in patients with primary degenerative dementia. J. Am. Geriatr. Soc., 36, 108-112. Lee, E.T. (1980): Statistical Methods for Survival Analysis. Lifetime Learning Publications. Martin, D.C., Miller, J.K., Kapoor, W., Arena, V.C. and Boiler, F. (1987): A controlled study of survival with dementia. Arch. Neurol., 44, 1122-1126. McLaren, S.M., Barry, F., Gamsu, C.V. and McPherson, F.M. (1986): Prediction of survival by three psychological measures. Br. J. Clin. Psychol., 25, 223-224. Meer, B. and Baker, F. (1966): The Stockton Geriatric Rating Scale. J. Gerontol., 21, 392-403. Moran, S.M., Cockram, L.U, Walker, B. and McPherson, F.M. (1990): Prediction of survival by the Clifton Assessment Procedures for the Elderly (CAPE), Br. J. Clin. Psychol., 29, 225-226. Pattie, A. (1988): Measuring levels of disability - - the Clifton Assessment Procedures for the Elderly. In: Psychological Assessment of the Elderly, pp. 61-81. Editors: J.P. Wattis and 1. Hindmarch. Churchill Livingstone, London. Rango, N. (1985): The nursing home resident with dementia. Clinical care, ethics and policy implications. Ann. Intern. Meal., 102, 835-841. SIVIS Jaarboek 1986. SIVIS Utrecht. SIVIS Jaarboek 1990. SIVIS Utrecht. Van Dijk, P.T.M., Dippel, D.W.J. and Habbema, J.D.F. (1991): Survival in patients with dementia. J. Am. Geriatr. Soe., 39, 603-610. Van Dijk, P.T.M., Van de Sande, H.J., Dippel, D.W.J. and Habbema, J.D.F. ( 1992): The nature of excess mortality in nursing home patients with dementia. J. Gerontol.: Med. Sci., 47, M28-34.
Archives of Gerontology and Geriatrics 18 (1994) 101-113
A behavioral rating scale as a predictor for survival of demented nursing home patients Pieter T.M. van Dijk, Diederik W.J. Dippel, J. Dik F. Habbema* Center for Clinical Decision Sciences, Department of Public Health, Erasmus University Rotterdam, Postbus 1738, 3000 DR Rotterdam, Netherlands (Received 2 August 1993; revision received 24 December 1993; accepted 27 December 1993)
Abstract As part of a study towards determinants of survival in nursing home patients with dementia, the prognostic value of a behavioral rating scale for nursing home patients is assessed in an 8-year follow-up study. The 2-year survival rate for the entire cohort (n = 569) was 56%. Women (n = 459) had a 2-year survival rate of 62%, and men (n = 110) had a 2-year survival rate of 40%. Items indicating physical impairment, dependency and apathy had most prognostic value. Items measuring aggressive or depressive behavior, and cognitive impairment were less predictive. These results were confirmed in a multivariate proportional hazards analysis. A prognostic model with age, gender and five behavioral items (needs help when walking, occupied in useful activity, restless at night, utters physical complaints, and socializes with other patients) substantially differentiated in survival chances in patients with dementia. The model gives a predicted 2-year survival chance of less than 20% or more than 80% in 80 of the 569 patients. When adjusted for the variables in the model, previous residence had no prognostic value anymore. Possibilities for further work in this area of research are discussed.
Key words." Survival; Dementia; Nursing home; Behavioral rating scale; Proportional hazards
1. Introduction A b o u t 20 000 (1.5%) D u t c h i n h a b i t a n t s over 65 years o f age reside in nursing homes because o f d e m e n t i a (SIVIS, 1990). Their survival after admission ranges between a few days a n d m o r e than 10 years, with a mean o f a b o u t 2.5 years. This wide * Corresponding author. 0167-4943/94/$07.00 © 1994 Elsevier Science Ireland Ltd. All rights reserved. SSDI 0167-4943(94)00537-H
102
P. 7~M. van Di/k et al. / Arch. Gerontol Geriatr 1~¢ (1994) 101-113
range provides only little information to the interested parties. Close relatives of the demented patient are often very interested in his prognosis. A nursing home physician needs prognostic information, for example when he is considering offering his patient an operation because of a senile cataract. Also, health planners combine epidemiological and prognostic information when addressing questions of demand and turnover in nursing homes, and how these depend on admission criteria. Age and gender provide some prognostic differentiation (van Dijk et al., 1991, 1992), but more knowledge is needed. The prognostic value of behavioral rating scales for assessing severity of dementia has been studied before. Measures of dependency in activities of daily life, physical impairment, and inactive behavior appeared to be associated with survival (Barclay et al., 1986; McLaren et al., 1986; Martin et al., 1987; Moran et al., 1990). Most studies, however, were carried out in a univariate or qualitative way. No attempt was made to study the joint effect of prognostic factors on survival. In the present study of survival in dementia patients we use multivariate, quantitative methods in order to assess the information provided by a behavioral rating scale and to identify the subscales and items which particularly have prognostic value. The resulting model can be regarded as a step towards a more individual prognostic evaluation of dementia patients who are admitted to a nursing home. 2. Methods 2.1. Patients
The Psychogeriatric Center Stadzicht in Rotterdam, The Netherlands, is a nursing home which serves about 260 patients, who are suffering from dementia of various etiologies. All 606 patients, 437 women and 169 men, admitted between January 1st, 1982 and December 31, 1988 were included in this study, and followed until death or discharge, or until January 1st, 1990. The mean age at admission was 80.8 years for the whole cohort, 81.3 years (S.D. 6.6) for women and 79.6 years (S.D. 7.3) for men. Dementia was diagnosed according to criteria of the DSM-III-R (American Psychiatric Association, 1980). The mean duration of the dementia before admission was 5.1 years (S.D. 3.2). One-third of the patients was married at admission, and 56% widowed. The place of living before admission was their own house in 42"/,,, a home for the aged in 18%, another nursing home in 17%, and a hospital or other institutions in 23% of the patients. A more detailed description of the study population is presented elsewhere (Van Dijk et al., 1992). 2.2. Methods
The BOP (Beoordelingsschaal voor Oudere Patienten), which is translated as Rating Scale for Elderly Patients, is used in many nursing homes in the Netherlands (Diesfeldt, 1981). The BOP contains 35 items about behavioral and cognitive impairment. Nurses who take care of the patient score each item on a 0-2 scale with higher scores indicating more severe or more frequent disability. The item scores are used to calculate sumscores on six sub-scales: dependency, aggressiveness, physical disability, depression, orientation and communication, and apathy. In this study, the 35 BOP items, the six subscales, and the patients' age and gender are the possible
103
P . T . M . van Dijk et a l . / A r c h . Gerontok Geriatr. 18 (1994) 101-113
predictive variables. In the analysis, we used the BOP-scores measured for the 569 persons (410 women and 159 men) who had a completed assessment at the end of the observation period, about i month after admission. The BOP was not completed in the remaining 37 patients (6°/,0, because of early death [17] or other reasons [20]. Like the rating scale of the CAPE (the Clifton Assessment Procedures for the Elderly) (Pattie, 1988), the BOP is derived from the Stockton Geriatric Rating Scale (Meer and Baker, 1966). These three scales have 14 items in common. This enabled us to assess the prognostic value of that part of the CAPE-rating scale that overlaps with the BOP. Sum scores could be calculated on the apathy and the communication failure subscales of the CAPE. Survival is estimated by the product-limit method. Statistically significant differences in survival between subgroups are identified by the log-rank test (P < 0.05) (Lee, 1980). Multivariate analysis of prognostic variables is carried out using proportional hazards regression models (Kalbfleisch and Prentice, 1980). The log-likelihood is used as a goodness-of-fit statistic: the less negative its value, the more closely the model fits the observed data. When using a proportional hazards model, for each patient the hazard of dying at time t is estimated by the equation Ai(/) = Amean(t)*
exp(/31.zli
+/32 i + • • • +/3n'Zni-
/3mean)
(1)
For each patient the survival chance is estimated according to the equation S~{t)
= Smean(t)exp(/31.
zli +/32.z2i
+
• • • +/3,.Zni
-
/3mean)
(2)
where A,(t) is the hazard of dying for an individual i at time t, Amean(t) is the hazard of dying for someone with a mean score on all selected variables, S , ( t ) is the chance of survival for an individual i at time t, Smean(/) is the chance of survival at time t for someone with a mean score on all selected variables, /31.zli +/32i.z2i + • • • + / 3 ~ ' z n i is the sumscore for an individual i with scores z I • • • zn on the variables /31 • • • /3n where /31 • • • /3, are the natural logarithms of the rate ratios shown in Table 3, and/3mean is the sum score for someone with a mean score on all selected variables. Two regression models are developed. Model ! uses the BOPitems to predict survival, and Model 2 the BOP-subscales. Because we want to investigate the additional value of the BOP-variables above that of gender and age, the latter ones are forced into the two models before the forward selection of the BOP-variables starts. The reference model is based on age and gender only. The items and scales are entered or removed in a stepwise forward selection mode on the basis of tail probabilities (P < 0.10 and P > 0.15, respectively) from a likelihood ratio test statistic: a variable is only selected into the model if it gives enough additional prognostic information above that provided by the variables already in the model. Model 1 was evaluated in detail. In order to avoid overoptimism, a split-half approach was used. The entire cohort was randomly divided into two equally sized subgroups. The stepwise forward selection was performed on the patients of one subgroup (the training sample), and the resulting model was used for estimating survival
104
P.T.M. van Dijk et al./Arch. Gerontol. Geriatr. 18 (1994) 101-113
chances for the patients in the other subgroup (the validation sample). This process was repeated in such a way that every subgroup (and thus every patient) acted as a validation sample once. The patients were ordered according to their predicted survival chances into four groups with a 'favorable', a 'moderate', a 'poor' and a 'very poor' prognosis. The predicted survival chances in these four groups were compared with the observed survival rates in order to assess the reliability of the model predictions. 3. Results
3.1. Subscales The subscales physical disability, orientation and communication and apathy have one, four and five items in common with the subscale dependency. It is therefore not surprising that they are highly correlated with that subscale (Pearson's r - 0.7). The correlation between apathy and physical disability is 0.6. The correlations between the remaining subscales are weaker and vary between 0 and 0.4. Men score on average somewhat higher on the subscales dependency, apathy and aggressiveness than women (t-test, P <0.05) (see Table 1). 3.2. Survival The 2-year survival rate for the entire cohort (569 patients) is 56"/,,. Except for aggressiveness, scores on the subscales are significantly related with survival (log rank, P < 0.05) (see Table 1). For all subscale scores, survival rates for women are significantly higher than for men. Individual items of the physical disability, the apathy and the dependency subscale are of considerably prognostic value (Table 2). Several clearly behaviorally anchored items, such as 'needs help when eating', 'needs help when walking', 'needs help when dressing', 'incontinent at night' and 'incontinent during the day' show large differences in 2-year survival rates between patients with a score of 0 (no help needed, no impairment) and patients with a score of 2 (much help needed, severe impairment). For instance, the patients who need no help when walking have a 2-year survival rate of 73%, whereas patients who need much help when walking have a 2-year survival rate of 36%. The items of the aggressiveness scale have hardly prognostic value. The items 'responds to his name' (OC4) and 'privileges to leave the ward' (A7) have practically empty categories. Inference regarding these items is therefore limited; they are excluded from further analyses. 3.3. Multivariate prognostic models Table 3 describes the two proportional hazard models for the prediction of survival time. Model 1 contains five BOP-items, age and gender. As expected, the mortality hazard was higher for men, and increased with age. The item 'needs help when walking' was the first behavioral item selected, and had the highest rate ratio. The item 'restless at night' was less significantly associated with survival in the univariate analysis (0.01 < P < 0.05), but nevertheless has been selected: apparently its prognostic information was independent from the information of other items. From the
105
P.T,M. van Dijk et al./ Arch. Gerontol. Geriatr. 18 (1994) 101-113
Table 1 Scores on the behavioral rating (BOP) subscales and 2-year survival rates in 159 male and 410 female institutionalized patients with dementia Subscale
Men (n = 159)
Women (n = 410)
Mean
Mean
All patients Dependency
40% (100%) 21.3
0-16 17-28 29-46 Aggressiveness
Depression
3.9
Apathy
2.9
2.2
62% (55%) 63% (45%) 2.3
48°/,, (67%) * 25% (33%) 2.2
75% (61%) * 43% (39%) 2.4
48% (6Y'/o) * 27% (37%) 4.3
0-4 5-8
0-7 8-14
76% (43%) * 52'¼, (38%) 5I% ( 19%1
39% (40%) 40°/,, (60%)
0-2 3-6 Orientation and communication
19.3
34% (45%) 27'¼, (25%)
0-2 3-6
2-year survival rate (relative frequencyl 62% (lOft¼,)
63% (30%) *
0-2 3-10 Physical disability
2-year survival rate (relative frequency)
68% (58%1 * 55% (42%) 4.3
43% (52%) 37% (48%) 8.0
67% (55%) * 57% (45%) 7.2
58% (40%) * 27°/,, (60%)
73% (52%) * 50'¼, (48%)
*Log-rank test: P < 0.05.
depression subscale the item 'utters physical c o m p l a i n t s ' has been selected. The rate ratio of 1.1 indicates that its predictive value is limited. Two examples show the considerable r e d u n d a n c y a n d overlap in prognostic inform a t i o n between the items a n d thus highlight the additional value of multivariate over univariate analysis. First, 25 BOP-items were significantly related to survival in the univariate analysis (P < 0.05) (Table 2), but only five of them have been selected. Second, of the 3 items of the physical disability scale - - which had all highly significant differences in survival in the univariate analysis - - only the item 'needs help when walking' is selected in Model l; obviously the others do not add sufficient additional prognostic i n f o r m a t i o n . Model 2 contains three BOP-subscales, age, and gender. The rate ratios for age a n d gender do n o t differ essentially from Model I.
P.T.M. van Dijk et al./Arch. Gerontol. Geriatr. 18 (1994) 101-113
106
Table 2 Two-year survival rates for scores on the 35 items of the behavioral rating (BOP) scale. Relative frequencies are given in brackets. Items with statistically significant differences in survival rates are indicated Item
Description of item
2-year survival rate (relative frequencyl Score 0
Score 1
Score 2
66% 67% 64% 56% 57% 59% 56% 61% 68% 68% 63% 60% 59%
(39%) (52'7',) (47%) (8%) (71%) (41%) (48%) (60%) (52%) (51%) (68%) (76%) (78%)
55% (42'7',) 49% (16%) 48% (39'7',) 69% (20'V,,~ 55% (12'7',) 55% (54'7',) 67% (13%) 50% (17%) 50% (28'7`,) 51% (10%) 45% (3%) 47'7,, (14'7',) 50% (14%)
39% (19%) 43% (32%) 56% (14%) 53% (72%) 53% (17'7',) 49'7', (5%) 53% (39%) 49% (23%) 37% (20%) 44% (39%) 42%, (29%) 44% (10%) 38% (8'7',)
57% (65%) 55'7,, (49%) 57% (76'7',) 56% (53%) 58% (41'7`,t
60% (19%) 55'!/,, (21%) 57% (15%) 59% (22%) 60% (32%)
50% (16%) 60% (30%1 46% (9'7',) 54'7', (25%) 49% (27'7',)
61% (86%) 7Y¼, (46'7',) 71% (21%)
21% (5%) 48';4, (28%) 63';4, (36%)
34'7', (9%) 3.6'7,, (26%} 44% (43%)
62% (23%) 63% (43%) 59% (81%)
59% (37%) 58% (22%) 54% (10%)
51% (40%) 47% (35%1 40% (9%)
Knows in which institution he is Knows any of personnel by name Understands others Responds to his name
68'7', (7%) 78% (3'7',) 62% (44%) 57% (87%)
58% 53% 53% 44%
(2Y7',) (4%) (50%) (12%)
55% 56% 45% 88%
(70%) (93%) (6%) (1'7',)
Helps out on the ward Occupied in useful activity Socializes with other patients Helps other patients without being asked Needs supervision outdoors Never starts conversations Privileges to leave the ward
76% (19%) 74% (33%t 74% (40%) 75% (26%) 67'7', (24%) 68% (42%) - (0%)
70% 56% 51% 66% 57% 50% 57%
(18%} (34%) (27%) (23%) (52%) (31%) (99'7',)
47% 39% 39% 43% 46% 44% 0%
(63'7',) (33'7',) (33%) (51%) 124%) (27%) (I%)
Dependency (D) ** ** ** **
** ** ** * *
DI D2 D3 D4 D5 D6 D7 D8 D9 DI0 DI 1 DI2 DI3
Needs help when eating Incontinent during the day Does not make himself understood Unable to find his way around the ward Urinates and defecates at the wrong place Unwilling to do things asked of him Engages in useless repetitive activity Makes repetitive vocal sounds Drowsy during daytime Incontinent at night Needs protection from falling out of bed Objectionable during the night Restless at night
Aggressiveness ( Ag) Agl Ag2 * Ag3 Ag4 * Ag5
Threatens verbally to harm others Accuses others of harming him Hits and kicks other patients Objectionable during the day Angry easily
Physical disability ( PD) ** PDI ** PD2 ** PD3 ~
Needs protection from falling out of chair Needs help when walking Needs help when dressing
Depression (De) * Del ** De2 * De3
Sad Utters physical complaints Weeps easily
Orientation and communication ( OC) OCI a OC2 ~ * OC3 a OC4 ~
Apathy ( A ) ** ** ** ** * ** **
A1 A2 a A3 a A4 a A5 A6 a A7 a
"item is also part of the Dependency subscale. *0.01 < P < 0.05 (log-rank test), **P < 0.01 (log-rank test).
P.T.M. van Dijk et al./Arch. Gerontol. Geriatr. 18 (1994) 101-113
107
Table 3 Two proportional hazard models for predicting survival, based on age, gender and the behavioral rating scale (BOP). The BOP-items respectively BOP-subscales are entered into the model by stepwise forward selection and are listed by order of entry Variable
Range
Rate ratio a (95'Y,, CI)
Log-likelihood
Model 1 (BOP-items) Gender (female=0, male= 1) Age PD2 Needs help when walking A2 Occupied in useful activity DI3 Restless at night De2 Utters physical complaints A3 Socializes with other patients
0-1 50-100 0-1-2 0-1-2 0-1-2 0 - I-2 0-1-2
1.9 (I.5-2.4) 1.03 (1.02-1,05) 1.4 (1.2-1.6) 1.2 (1.0-1.4) 1.3 (1.1-1.5) 1.1 ( 1.0-1.3) 1.2 (1.0-1.41
-1964
0-1 50-100
2.1 (1.7-2.7) 1.04 ( 1.02- 1.05)
-1966
0-6 0-14 0-6
1.16 ( 1.05- 1.24) 1.06 (1.03-1.12) 1.10 (1.02-1.18)
Model 2 ( BOP-subscales) Gender (female = 0, male = I) Age Physical disability Apathy Depression
aRate ratios express relative mortality hazards. The value for the rate ratio for the covariate age of 1.03 in Model 1 implies, for example, that at every moment after admission the risk of dying for a 90-year-old person is 1.03 l0 = 1.34 times the risk of an 80-year-old person of the same gender and the same scores on the other items. If the latter person had a 2-year survival probability of 0.60, then the 90-year-old person would have a probability of 0.60 T M = 0.50. The rate ratio of 1.9 for gender implies that at every moment the chance of dying for men is almost twice the chance for women. The rate ratio of 1.4 for the item 'needs help when walking' means that the chance of dying for patients who need much help (itemscore 2) is 1.4 times that chance for patients who need some help (itemscore 1), and 1.42 = 2 times that chance for those who do not need any help (itemscore 0). The rate ratio of 1.16 on the physical disability-subscale in Model 2 implies that a patient with a score of 6 on that subscale has a change of dying 1.166= 2.4 times the chance for a patient with a score of 0.
O f t h e five s u b s c a l e s w i t h p r o g n o s t i c v a l u e in t h e u n i v a r i a t e a n a l y s e s ( T a b l e 1), t h e s u b s c a l e s p h y s i c a l d i s a b i l i t y , a p a t h y a n d d e p r e s s i o n a r e selected. T h e d e p e n d e n c y and orientation and communication subscales do not add significantly to survival p r e d i c t i o n o n c e t h e p h y s i c a l d i s a b i l i t y s u b s c a l e is selected.
3.4. Evaluation o f the prognostic models Goodness-of-fit. T h e t w o m o d e l s a r e r e l i a b l e in t h e s e n s e t h a t t h e m o d e l - p r e d i c t e d n u m b e r o f d e a t h s w i t h i n 2 y e a r s a f t e r a d m i s s i o n c o r r e s p o n d s well w i t h t h e o b s e r v e d n u m b e r (X2-test, 5 df, P > 0.3 f o r b o t h m o d e l s ) (see T a b l e 4). A r e f e r e n c e p r o g n o s t i c m o d e l , b a s e d o n a g e a n d g e n d e r o n l y , is a d d e d f o r c o m p a r i s o n . M o d e l 1 seems to be the easiest to implement, because the scores on a limited number of individual items are more readily available than the sum scores of the BOP-subscales. Its p e r f o r m a n c e w a s f u r t h e r a s s e s s e d b y t h e s p l i t - h a l f a p p r o a c h (see M e t h o d s sec-
108
P.T.M. van Dijk et al./Arch. Gerontol. Geriatr. 18 (1994) 101-113
Table 4 Observed and predicted number of deaths during the first 2 years after admission according to Model 1. Model 2 and the Reference model for five subgroups with predicted 2-year survival chances between 0 and 20%, 20 and 40°/,. . . . . . 80 and 100% Predicted 2-year survival
Number of patients Number of deaths Observed
Expected
Model 1 (age, gender, items)
0-20% 20-41Y7`, 40-60% 60-80% 80-100'7,,
27 116 150 223 53
25 66 77 54 5
23 72 69 60 9
31 113 150 228 47
24 67 80 53 3
26 70 69 62 7
4 69 244 245 7
2 36 121 68 0
2.6 42 108 76 1
Model 2 (age. gender, scales)
0-20% 20-40% 40-60% 60-80% 80-100% Reference model (age and gender only) 0-20%
20-40% 40-60% 60-80% 80-100%
tion). The goodness-of-fit was satisfactory, a l t h o u g h the model u n d e r e s t i m a t e d the survival chances in the g r o u p with a very p o o r prognosis (Fig. 1). 3.5. P r e d i c t i v e p o w e r
Once the models have been shown to be r e a s o n a b l y reliable, they can be evaluated for their power in individualizing survival predictions. W e will focus on the prediction o f 2-year survival chances. A 'perfect individualizing' model w o u l d give a 100% predicted survival chance to each patient o f the 56°/,, who have survived the 2-year period, a n d a 0% predicted survival chance to each o f the 44% who died. A totally non-individualizing m o d e l (the 'null' model) will predict for e v e r y b o d y a chance that is equal to the observed 56% 2-year survival rate for the total study p o p u l a t i o n . A n indication o f the predictive p o w e r can be o b t a i n e d from Table 4. M o d e l 1 gives a predicted survival chance between 40% a n d 60% in 150 patients, a n d a chance o f less than 20% o r m o r e than 80% in 80 patients. This m o d e l thus differentiates considerably between patients. M o d e l 2 gives c o m p a r a b l e results. The results for a reference m o d e l are worse. It gives a predicted survival chance between 40°/,, a n d 60% in 244 patients (more than 40% o f all patients), a n d has only a small n u m b e r o f patients with a prognosis o f lesser than 20% or m o r e than 80% (4 and 7 patients, respectively).
109
P.T.M. van Dijk et al./Arch. Gerontol. Geriatr. 18 (1994) 101-113 survival 100%
observed 80%
expected
60%
~
prognosis
40% favorable
20%
moderate poor very poor
0% 0
1
2
3
4
5
6
7
years
Fig. 1. Observed and predicted survival according to Model I for four equally sized [142] subgroups of patients with a favorable, moderate, poor and very poor prognosis according to their survival probabilities after the split-half approach on Model I. The mean predicted 2-year survival chances for these four groups are 0.24 (range 0.00-0.39), 0.50 (range 0.39-0.60), 0.66 (range 0.60-0.73) and 0.81 (0.73-0.94).
3.6. Help in survival prediction A prognostic index, and henceforth survival chances can be calculated for every person (second equation, Methods section). The average 1-, 2-, and 5-year survival rates for the entire cohort were 72%, 56%, and 20%, respectively. The prognostic index and the corresponding survival chances are easily obtainable from Fig. 2 (see the legend for examples). 3.7. Implications for users of the CAPE The mean sumscores at admission on the CAPE-subscales apathy and communication failure were 5.0 (±2.6) and 1.3 ( ± 1.2), respectively. When these two scales were put into a proportional hazards model with gender and age, the communication failure subscale had no predictive value and the apathy subscale had a hazard rate ratio of 1.17 for each point one scored higher (on a scale from 0 to 10). In a multivariate analyses of the 14 items, gender and age, five items were selected. The seven variables and their rate ratios were: gender (1.9), age (1.03), needs help when walking (1.4), helps out on the ward (1.2), socializes with other patients (1.3), unwilling to do things asked of him (0.9), and objectionable during the night (1.3). For these seven variables a prognostic chart similar to Fig. 2 is constructed (see Table
110
P. I'.M. van Dijk et al. / Arch. Gerontol, Geriatr. IS (1994) 1(t1-113
Table 5 Prognostic chart for estimating survival chances, according to the results with the CAPE Score Gender Age
Needs help when walking Helps out on the ward Socializes with other patients Objectionable at night Unwilling to do things asked of him
if man 51-60 61-70 71-80 81-90 some much sometimes never sometimes never sometimes often sometimes never
6 0 3 6 9 3 7 2 4 2 5 2 5 1 3
Add relevant scores:
[]
Subtract 2
[]
Use: circle relevant scores, add them to a sumscore, and subtract 2 points to get the prognostic index.
5). For every (CAPE-)prognostic index, survival chances can be read directly from Fig. 2. 4. Discussion Gender and age, and patient behavior as measured on a behavioral rating scale (the BOP) were found to be helpful in estimating survival chances in institutionalized patients with dementia. Subscales measuring physical disability, apathy and depressive behavior carried independent prognostic value. The first two scales were mentioned before as being of prognostic importance (Diesfeldt, 1975; Jacobs et al., 1978). The association between the orientation and communication subscale' and survival was only small (Diesfeldt, 1975) or even absent (Jacobs et al., 1978). Observations in non-institutionalized patients also emphasized the prognostic value of scales measuring physical impairment (Martin et al., 1987) and impairment of ADL activities (Knopman et al., 1988). On the item-level our results are consistent with Diesfeldt (1980), who found that all items of the Physical Disability scale and seven items of the dependency scale were associated with l-year survival. The five selected items in Model 1 are all easily interpretable, and most of them are also incorporated in other rating scales. Table 4 and Fig. 1 showed that the fit of Model 1 was satisfactory. We would be very interested to see how adequately this 'internally validated'
111
P.T.M. van Dijk et al./Arch. Gerontol. Geriatr. 18 (1994) 101-113
survival probability 100%
80%
Calculation of prognostic index "-,~
_
gender
if male
age
61 - 70 71 - 80 81
"..~ %. 60%
"& x\
"...
5 year"....2 "Q..
~
X year',\ ~ year 4\ ~
"-", '... "~ ~..
40%
\'\
X
\\ & "\
'~ ~
needs help when walking occupied in useful activity restless at night utters physical complaints socializes with ~ otherpatients
""-. \\ X Add relevant X...........Q.. X& ~scores
20%
0%
~
,
I
10
,
i
20
....fh.........© 30 prognostic index
-~ 40
7 3 7
- 90
10
some 3 much 6 sometimes 2 never 4 sometimes 2 often 5 sometimes 1 often 3 sometimes 1 never 3 pro.gnostic [ ~ Inoex
1
50
B
60
Fig. 2. Prognostic index according to Model 2 and the predicted 1-, 2- and 5-year survival chances. Examples: a woman of 65 years needs much assistance when walking and is sometimes 'restless at night'. She has favorable scores on the other three items. Her prognostic index is 3 + 6 + 2 = 11. In Fig. 2 it can be seen that her predicted 1-, 2-, and 5-year survival chances are 83%, 78'Y,,and 48%, respectively. A man of 85 years who sometimes "utters physical complaints" and is only sometimes occupied in useful activity. and who has a score of two on the other three items has a prognostic index of 7 + 10 + 6 + 2 + 5 + 1 + 3 = 34, and his survival chances are only 23%, 7'7,,and 0%.
m o d e l c a n predict survival in o t h e r n u r s i n g homes. W e t h i n k o u r clientele is rather representative for p s y c h o g e r i a t r i c n u r s i n g h o m e s in the N e t h e r l a n d s . T h e m e a n age o f d e m e n t e d p a t i e n t s a d m i t t e d in the N e t h e r l a n d s in 1986 (78 years for m e n a n d 80 years for w o m e n ) was slightly lower t h a n o u r clientele, a n d relatively m o r e m e n (33%) were a d m i t t e d (SIVIS, 1986). Scores o n the B O P - - r a t i n g scales on d e m e n t i a patients were n o t essentially different f r o m figures f r o m o t h e r n u r s i n g h o m e s in the N e t h e r l a n d s , n e i t h e r did the 2-year survival rates (Diesfeldt, 1975; J a c o b s et al., 1978; v a n Dijk et al., 1991). D u t c h n u r s i n g h o m e s are r a t h e r c o m p a r a b l e to A m e r i c a n s k i l l e d - n u r s i n g facilities ( C o o l s a n d v a n der Meer, 1988), a l t h o u g h the availability a n d k i n d o f m e d i c a l care in D u t c h n u r s i n g h o m e s is different from that a b r o a d (Cools, 1993). Differences in case-mix m a y decrease the g e n e r a l i z a b i l i t y o f the m o d e l to other settings. W e therefore investigated the i m p a c t o f p r e v i o u s residence o n the results. W h e n a d d e d to M o d e l 1, n e i t h e r the v a r i a b l e ' c o m i n g f r o m o w n house" ( i n d i c a t i n g p a t i e n t s with less c o - m o r b i d i t y ) , n o r the v a r i a b l e ' c o m i n g from a h o s p i t a l ' (in-
112
P . T . M . van Di/k et al. / Arch. Ger¢mtol. Geriutr. 18 (1994) 101-113
dicating patients with more co-morbidity) had rate ratios significantly different from 1: their rate ratios were 1.1 (95% confidence interval 0.9-1.4) and 0.9 (0.7-1.1), respectively. This suggests that for an estimation of survival chances the severity of impairment (in combination with gender and age) is much more important than previous residence; this finding corroborates the possibility of using our results in other nursing homes. Variables such as reason for admission and problems in adaptation to the sudden change in living situation are not considered in the BOP. Furthermore medical symptoms and diagnoses undoubtedly also contain prognostic information. The fact that the item 'utters physical complaints' (because of physical origin or not) was selected is already an indication for this. Although co-morbidity and behavioral items are mutually related - - many diseases result in physical disability and dependency, and an increased severity of dementia may cause a higher vulnerability for acquiring all sorts of diseases-, co-morbidity might also give independent prognostic information. For instance, the 2-year survival rate for patients with a very poor prognosis according to Model 2 was 30%. Patients in this group with parkinsonism, a heart failure, or a respiratory tract infection had 2-year survival rates of about 15%. This rates was 32% in the absence of these diseases. Thus, clinicians must be aware that the prognosis of the patient may be considerably worse when severe co-morbidity is present. Our results can also be of help for health planners. Patients in our nursing home had a 2-year survival rate of 56%. This gives a direct indication of turnover in nursing homes. Together with demographic and epidemiologic data, future demand and capacity for the current admission strategy can be assessed. It is also possible to estimate 2-year survival rates, when the admission policy is changed, e.g. in such a way that only more severely demented patients are admitted. If, for instance, the mean scores on all BOP-items at admission increased with 0.5 points, the mean prognostic index according to Model 1 would have such an increase, that the expected 2-year survival rate would decrease from 56 to 41%. In conclusion, behavioral items (which in many nursing homes are assessed for other purposes also) contain interesting prognostic information and can be combined in multivariate prediction models. Using these models, health care planners might be able to estimate more adequately the need of beds in the future for patients with dementia. Also, these models can support the nursing home physician in getting the best possible insight into the prognosis of a patient. The more accurate this prognosis, the more well-founded the nursing home physician can decide to perform a diagnostic evaluation or a surgical procedure, which may be of great risk and inconvenience to the patient initially, but improve the quality of life afterwards. A better life-expectancy increases the chance that the expected long-term benefit for the patient exceeds the instantaneous risk and inconvenience of the diagnostic or surgical procedure, and thus weights in favour of an active approach (Rango, 1985).
5. Acknowledgements We thank the Psychogeriatric Center Stadzicht and its medical and psychological staff for their generous permission to use the medical records of their patients. This
P.T.M. van Dijk et al./Arch. Gerontol. Geriatr. 18 (1994) 101-113
113
study was supported by a grant from the Netherlands' Health research promotion programme (SGO). 6. References American Psychiatric Association (1980): Diagnostic and Statistic Manual of Mental Disorders (DSMIll). Washington: American Psychiatric Association. Revised version: 1987. Barclay, L.L., Zemcov, A., Blass, J.P. and McDowell F.H. (1985): Factors associated with duration of survival in Alzheimer's disease. Biol. Psychiatry, 20, 86-93, Cools, H.J.M. and Van der Meer, J.W.M. (1988): Infection control in a skilled nursing facility: a 6-year survey. J. Hosp. Inf., 12, 117-124. Cools, H.J.M. (1993): Adding life to years (in Dutch). AVO, The Hague. Diesfeldt H.F.A. (1975): Predicting survival and longevity of mentally impaired elderly patients (in Dutch). T. Soc. Geneesk,, 57, 343-350. Diesfeldt H.F,A. (1980): Activities of daily living, cognitive disturbances and survival in psychogeriatric patients (in Dutch). Gerontologie, 11, 205-212. Diesfeldt H.F.A. (1981): The BOP, a report on ten years' experience with a Dutch geriatric rating scale (in Dutch). Gerontologie, 12, 139-147. Jacobs, M., Trommel, J. and Gips, C.H. (1978): A rating scale for geriatric patients: need of care, age groups and one-year survival of psychogeriatric patients (in Dutch). Ned. Tildschr. Gerontol., 9, 29-34. Kalbfleisch, J.D. and Prentice, R.L. (1980): The Statistical Analysis of Failure Time Data. John Wiley and Sons. Knopman, D,S., Kitto, J., Deinard, S. and Heiring, J. (1988): Longitudinal study of death and institutionalization in patients with primary degenerative dementia. J. Am. Geriatr. Soc., 36, 108-112. Lee, E.T. (1980): Statistical Methods for Survival Analysis. Lifetime Learning Publications. Martin, D.C., Miller, J.K., Kapoor, W., Arena, V.C. and Boiler, F. (1987): A controlled study of survival with dementia. Arch. Neurol., 44, 1122-1126. McLaren, S.M., Barry, F., Gamsu, C.V. and McPherson, F.M. (1986): Prediction of survival by three psychological measures. Br. J. Clin. Psychol., 25, 223-224. Meer, B. and Baker, F. (1966): The Stockton Geriatric Rating Scale. J. Gerontol., 21, 392-403. Moran, S.M., Cockram, L.U, Walker, B. and McPherson, F.M. (1990): Prediction of survival by the Clifton Assessment Procedures for the Elderly (CAPE), Br. J. Clin. Psychol., 29, 225-226. Pattie, A. (1988): Measuring levels of disability - - the Clifton Assessment Procedures for the Elderly. In: Psychological Assessment of the Elderly, pp. 61-81. Editors: J.P. Wattis and 1. Hindmarch. Churchill Livingstone, London. Rango, N. (1985): The nursing home resident with dementia. Clinical care, ethics and policy implications. Ann. Intern. Meal., 102, 835-841. SIVIS Jaarboek 1986. SIVIS Utrecht. SIVIS Jaarboek 1990. SIVIS Utrecht. Van Dijk, P.T.M., Dippel, D.W.J. and Habbema, J.D.F. (1991): Survival in patients with dementia. J. Am. Geriatr. Soe., 39, 603-610. Van Dijk, P.T.M., Van de Sande, H.J., Dippel, D.W.J. and Habbema, J.D.F. ( 1992): The nature of excess mortality in nursing home patients with dementia. J. Gerontol.: Med. Sci., 47, M28-34.