|
We studied the 1990 calendar year because it was the
last year that the Quebec civil code required a psychiatrist, where
possible, to perform competency assessments. Subsequently, the civil
code was amended to allow other physicians to perform competency
assessments. In years up to and including 1990, our services therefore
experienced a high rate of referral for competency assessment, often
initiated by the suspicion of abuse or neglect. In addition, referral
sources such as the GARU, the CCT, and the CLSC were all relatively
familiar with the elder-abuse problem. These 2 factors likely increased
the sample prevalence of abuse and neglect.
The study included a total of 126 new assessments that were completed
during the calendar year. Table
1 presents the summary statistics that describe the sample.
The geriatric psychiatrist who attended each subject reviewed the
hospital record and completed a data form capturing variables selected
on the basis of clinical experience, review of the literature (2),
and systematic data availability.
Patient variables included demographics such as age, sex, marital
status, and living situation (that is, alone, spouse, nonspouse
family, friend(s), other caregiver, or supervised setting). Further,
we obtained diagnostic and psychometric data such as the DSM-III-R
psychiatric diagnosis and the DSM-III-R Global Assessment of Functioning
(GAF) scale score (34) and the Mini-Mental State Examination (MMSE)
score (35). The MMSE was not used in the presence of marked aphasia.
Due to the diversity and multiplicity of medical diagnoses seen,
cell sizes for medical disorders, groups of disorders, and their
severity were too small to interpret. Other patient variables included
the presence of a caregiver, the presence of significant social
isolation, the presence and type of chronic behaviour problems,
the presence of a history of abuse or neglect toward the patient,
the presence of a history of abuse or neglect by the patient toward
others, and the presence of a history of family discord or hostility.
Available data on caregivers and abusers were not systematic and
were not analyzed. Abuse and neglect variables included the presence,
type, and estimated severity of abuse or neglect.
We calculated the point prevalence of 4 types of abuse and neglect.
With respect to nominal independent variables, we compared abused
and nonabused groups using either odds ratio (OR), Yates continuity-corrected
chi square, or Fisher’s exact P test. With respect to ratio
and ordinal variables such as age, GAF and MMSE, we compared abused
and nonabused groups using parametric and nonparametric tests such
as the t-test and the Mann-Whitney U test. Finally, we performed
a multivariate analysis using logistic regression with backward
elimination. Variables were retained on the basis of standard modelling
strategy: variable specification based on plausibility, prior knowledge,
and bivariate OR, followed by an assessment of interaction, confounding,
and precision (36).
The mean age of the sample was 77.9 years (range 61 to 94 years,
SD 6.7) and the mean GAF score was 43.0 (range 5 to 95, SD 14.7).
The mean MMSE score for the 106 subjects to whom it was administered
was 22.3 (range 2 to 30, SD 6.4). Table
1 presents the other statistics that describe the study sample.
Diagnoses totalled more than 100% because of the frequent coexistence
of multiple diagnoses. Although most terms follow DSM-III-R categories
of disorder, we use the term “chronic cognitive disorder”
to include dementia of any cause and amnestic syndrome, primarily
because of the small cell sizes of the latter disorder.
Table 2 presents the
prevalence of 4 types of abuse and neglect, multiple abuse, and
clinical estimates of the severity of abuse.
Comparison of Abused and Nonabused Subjects With Respect to
Numerical Variables. In the abused and nonabused groups, respectively,
mean age was 76.4 years (SD 7.2 years) and 78.2 years (SD 6.6 years);
mean GAF score was 42.0 (SD 12.2) and 43.2 (SD 15.2); and mean MMSE
was 19.9 (SD 6.6) and 22.8 (SD 6.2).
These differences were minor and not statistically significant
on separate variances t-test and Mann-Whitney U tests, although
there was a trend toward slightly lower MMSE scores in the abused
group (P = 0.066).
Abuse Prevalence in Demographic and Diagnostic Subgroups.
Table 3 presents the
bivariate analysis of abuse and neglect by nominal independent variables.
The data on living situation showed a gradient of prevalence: lowest
in subjects living with spouses (2/32 or 6.3%) or in supervised
settings (1/16 or 6.3%); and progressively higher in subjects living
alone (10/61 or 16.4%), with nonspouse family (3/12 or 25%), friends
(1/2 or 50%), and others (3/3 or 100%). For reasons of similarity
and small cell sizes, some living situation groups were combined.
Because abuse was lowest and equivalent in the supervised-setting
and living-with-spouse groups, these were combined and designated
the reference group. The living-alone group was retained because
it included a mixture of vulnerable and autonomous individuals.
The nonspouse family, friends, or other groups were combined.
The data on marital status also showed a gradient of prevalence,
lowest in married subjects (2/34 or 5.9%) and progressively higher
in never-married (2/22 or 9.1%), widowed (14/62 or 22.6%), and divorced
and separated subjects (2/8 or 25%). Again, for reasons of similarity
and small cell sizes, some marital status groups were combined.
Because abuse was lowest in the married group, it was retained as
the reference group. We retained the never-married group because
of its distinct character. The divorced or separated and widowed
groups were combined owing to small cell sizes, the shared quality
of disruption of a previously stable family situation, and similar
prevalence of abuse. Although social isolation appeared strongly
associated with abuse, this was retrospectively judged in a nonblind
manner and may be biased because an abused individual without social
support to stop the abuse is likely to be considered socially isolated.
We excluded it from further analysis. The data on sex showed a nonsignificant
trend toward more abuse of women than of men.
With respect to diagnosis, there were nonsignificant trends toward
increased abuse in subjects with chronic cognitive disorders, alcohol
abuse, and behaviour problems. The trend toward increased abuse
with chronic cognitive disorders is consistent with the trend toward
lower MMSE scores in the abused group.
|
|
An unexpected finding was the decreased frequency of abuse in
the presence of major affective disorder. As ours was a clinically
derived sample, this may be due to referral bias. We could not examine
several diagnostic categories, such as affective disorder with psychosis,
adjustment disorder, anxiety disorder, and benzodiazepine abuse
due to small numbers.
We studied variables with bivariate associations with a significance
level of <= 0.25 (21) and variables of potential importance,
such as chronic cognitive disorder and GAF and age quartiles. We
used backward elimination.
Several variables were excluded from the logistic regression because
of irrelevance, bias, questionable validity, or missing values.
We excluded diagnostic groups with cell sizes under 10 subjects.
Similarly, MMSE was excluded owing to 20 missing values. Probable
bias excluded social isolation. Living situation and marital status
were regrouped as described above.
We examined the GAF score for linearity by dividing into quartiles
and examining quartile-specific ORs with the highest quartile as
the reference group. The ORs for the quartiles were nonlinear. The
highest 3 quartiles were approximately equivalent (ORs from 0.8
to 1.0) and, consequently, were combined and designated the reference
group for the multivariate analysis. The lowest quartile, representing
GAF < 35, showed an OR of 1.7 when compared with the highest
quartile, suggesting a threshold of function below which there is
increased abuse.
Similarly, age was examined for linearity by examining quartile-specific
ORs with the youngest quartile as the reference group. The first
3 quartiles were almost equivalent (ORs from 1.0 to 1.35) and therefore
were combined. The fourth quartile, representing age > 83 years,
showed an OR of 0.47.
Diagnostic groups whose multivariate ORs for abuse were < 1
were removed from the model because they more likely reflect referral
bias than protective factors. These included depression and delirium.
On multivariate analysis, because of high collinearity (that is,
intercorrelation), living situation and marital status were significantly
associated with abuse only when the other was not entered into the
model. We retained both variables because they describe 2 related
but different processes. Whereas living situation describes a current
state, marital status captures a history of family disruption (divorced,
separated, or widowed) vs relative stability (married or never married).
Further, removal of marital status from the model resulted in a
large change in the OR for living with nonspouse, friends, or others
(6.1 to 10.5) and the OR for female sex (3.3 to 2.6), suggesting
meaningful confounding. Removing alcohol abuse from the model resulted
in a change in the OR for female sex, from 2.6 to 3.3, suggesting
meaningful confounding. For this reason, and because alcohol abuse
is a plausible psychosocial risk factor, it was retained in the
model. No statistically significant interactions among any variables
were found. Table 4
describes the final model.
Our study has several limitations. The nonblinded retrospective
design and lack of structured instrument or screening policy for
abuse may have biased our results, and certain cases may have been
missed. Our clinically derived psychiatric service sample and the
selection bias of the Quebec civil code would result in higher prevalence
figures than those reported for other populations. The small sample
size limits the power and possibly the generalizability of our results.
We found suspected or confirmed elder abuse in 16% of our sample.
The most common form was financial (13%), followed by neglect (6%),
emotional abuse (4%), and physical abuse (2%).
Taken separately on bivariate analysis, 2 factors had meaningful
and significant associations with abuse. These were living situation
with nonspouse family, friends, and others (OR 10.5) and marital
status of divorced, separated, or widowed (OR 4.7). Factors that
showed suggestive trends on bivariate analysis included female sex
(OR 4.1), alcohol abuse (OR 2.1), behaviour problems (OR 1.9), and
chronic cognitive disorder (OR 1.4).
The complex interrelations of correlates, such as living situation,
marital status, sex, and others made multivariate logistic regression
important, despite our limited sample size. Although several variables
were not statistically significant, they were retained in the model
because they reduced the confounding or improved the precision of
other variables (21). Living situation with nonspouse family, friends,
or others was associated with the highest OR for abuse (OR 6.1)
on multivariate analysis. Marital status of divorced, separated,
or widowed, however, showed a trend toward additional association
(OR 2.4), even after living situation was controlled. This suggests
that both living situation and disruption of a previously stable
family unit may be associated with abuse. This is consistent with
studies that have reported an association of abuse with poor social
network (25) and caregiver social isolation (10). In particular,
Paveza and others (30) found that violence toward individuals with
Alzheimer’s disease was associated with the individual living
with immediate family but without a spouse and postulated that caregiving
may be particularly burdensome when the patient is residing within
the family with the long-term bond of marriage missing. In our study,
however, 2 other possibilities must be considered. First, in our
clinical sample it is possible that individuals suffering from social
isolation or family disruption were more likely to be known to and
referred by community services. Second, our most prevalent form
of abuse was financial, which has been associated with distant relatives
or nonrelatives rather than spouses (33). The situation of living
alone appeared to have a weak and nonsignificant association with
abuse (OR 1.6), most likely because this group represents a heterogeneous
mixture of vulnerable and autonomous individuals.
The bivariate trend of association of abuse with female sex (OR
4.1) was intriguing but did not exclude the possibility that women
are at higher risk because they live longer and marry earlier than
do men and are therefore more likely to be older, widowed, and isolated.
The decrease in OR for female sex from bivariate to multivariate
analysis (OR 2.6) suggested that such factors may account for some
of the variance, but that even after they are controlled, female
sex may be associated with abuse.
Alcohol abuse on the part of the patient—a plausible risk
factor on both biological and psychosocial grounds—showed a
trend toward association, even after other factors were controlled
(OR 2.2). This may be due to chance, but it is consistent with studies
that have shown an association between elder abuse and caregiver
alcohol abuse and between caregiver alcohol abuse and care receiver
alcohol abuse (17,18).
|
